Professional Documents
Culture Documents
has little to do with a lack of interest by researchers; rather, poli- • Outsiders might not care about ending unethical
ticians responding to advocacy groups have blocked funding or behavior, but instead use the exposure of unethical
permission to do research.3 In 1970, the United States Con- behavior to advance their own goals.
gress “failed to follow its usual review process dictated by the
• Managers will try to protect the organization and dis-
Controlled Substances Act that requires scientific evaluation
credit the whistle-blower.
and testimony before legislative action. It declared cannabis ille-
gal in the absence of such evidence” (Bostwick, 2012:181). • The whistle-blower often experiences emotional dis-
Since then, researchers have been prohibited from studying it. tress and strained relations, and even lawsuits.
As Bostwick (2012: 182) noted, “the political climate at the fed- • Future employers may not trust the whistle-blower
eral level has essentially quashed the type of research that is and may avoid hiring him or her.
routine before commercial introduction of new drugs.” As a
result, the gradual introduction of “medical marijuana” and can- A whistle-blower needs to be prepared to make sacri-
nabis legalization by several states has proceeded based on fices—loss of a job or no promotions, lowered pay, an
opposing political pressure, not based on neutral evidence from undesirable transfer, abandonment by friends at work, or
scientific study. Likewise, in 1996 the U.S. Congress enacted legal costs. There is no guarantee that doing the ethical-
strict restrictions and cut Center for Disease Control research on moral thing will end the unethical behavior or protect an
the issue of firearms and violence, and funding dropped by 96 honest researcher from retaliation.4
percent. This effectively stopped government-funded research
on the issue until a Presidential order in 2014 in the wake of the
“Sandy Hook” mass shooting of elementary school children. As
a result, the amount of research-based knowledge has been 3.4: Politics and Social
miniscule. In both situations, politicians and special interest
groups feared that objective, neutral research might uncover Research
findings that could contradict their ideological position. They
preferred a situation of ignorance, where their ideological posi- 3.4 Recognize the impact that politics and large
tion would prevail in the absence of evidence, to a climate of corporations have on research
open inquiry and research-based truth seeking. The ideals of a free, open, and democratic society include
Pay attention to who paid for research and when doing advancing and sharing knowledge. People have a right to
sponsored research, negotiate conditions for releasing find-
study and inquire into any question and to share their find-
ings prior to beginning the study or signing a contract. It is best
ings publicly. Ethical issues largely address moral concerns
to begin with an explicit guarantee that you will only conduct
and standards of professional conduct that are usually under
ethical research. It is legitimate to delay the release of findings
to protect the identity of informants, to maintain access to a
the researcher’s control. Political concerns can also influence
research site, or to protect your personal safety. It is not legiti- and interfere with the research process. Organized advocacy
mate to censor findings because a sponsor does not want to groups, powerful interests, government officials, or politi-
look bad or wants to protect its reputation. The researcher cians may try to restrict or control the direction of research.
directly involved and knowledgeable about a study shoulders a In the past, powerful political interests and groups
responsibility for both conducting the research and making its have tried to stop research or the spread of legitimate
findings public. research findings to advance their own political goals.
They have used their political power to threaten research-
ers or their employers, to cut off research funds, to harass
Blowing the Whistle Whistle-blowing occurs individual researchers and ruin their careers, and to censor
when a researcher informs an external audience of a seri- publication of findings that they disliked.
ous ethical problem that is being ignored. It is never a first Politically powerful groups have directed research
step; rather it occurs after the researcher has repeatedly funds away from studying questions that researchers saw
attempted to inform superiors and fix the problem inter- as important and toward studies that supported the policy
nally. The whistle-blowing researcher must believe that the positions of their political views. Members of the U.S. Con-
situation is a serious breach of ethics and the organization gress have targeted and removed funds for social research
will not end it without public pressure. Whistle-blowing is projects that panels of independent scientists evaluated as
risky in several ways: being well designed and critical to advanced knowledge.
Often the reason has been that the politicians personally
• Outsiders may not be interested in the ethical abuse
disliked the study topics (e.g., sexual behavior of teens, ille-
and simply ignore it.
gal drug users, voting behavior). Politicians are not the only
3
See American Psychological Association (Crawford [2014], Gutting [2012],
4
Jamieson [2013], and Underwood [2013] on gun violence). See Bostwick (2012), See Yong, Ledford, and Van Noorden (2013) on whistle-blowing and scientific
Crawford (2014), and Harris (2010) on medical marijuana. fraud.
Becoming an Ethical Researcher 67
ones who block the free flow of knowledge. Large corpora- WRITING PROMPT
tions have threatened individual researchers with lawsuits
Was Crime Research Censored?
for delivering expert testimony in public about research
findings that revealed to the public the corporation’s bad Studies showed how politicians have redirected crime research
toward certain issues and away from others. Do you consider this
conduct, or have hired researchers to discredit studies with a negative form of government censorship, or is this an example of
findings unfavorable for their financial interests.5 good representative democracy in action? Why?
Example Study
performance dashboard and can be viewed by
your instructor.
Learning Objectives
4.1 Identify reasons why we use samples in 4.5 Identify the three main types and uses of
research random sampling
4.2 Describe four nonprobability sampling 4.6 Identify potential challenges presented in
techniques in research sampling for research and how to effectively
manage those difficulties
4.3 List the specialized vocabulary used in the
process of random sampling 4.7 Examine how sampling errors lead to
incorrect inferences
4.4 Evaluate the three main steps that are
followed in simple random sampling
Many teens in large urban areas of the United States, espe- they carry weapons, the most common response is self-
cially in areas with high concentrations of crime and pov- defense. Yet, prior studies indicated that many teens hold
erty, carry illegal guns or other weapons. They are no more misperceptions and often overestimate the behavior of
likely than teens in other countries to get into fights, but peers regarding sexual activity, drug use, and weapons.
they are far more likely to have deadly weapons, so con- Hemenway and colleagues (2011) conducted a survey
frontations are more likely to be deadly. When asked why of teens in the Boston area in 2008 to learn whether
70
Sampling 71
4.1: Why Do We Use clarify and deepen our understanding of specified areas of
social life.
Samples?
4.1 Identify reasons why we use samples in research
4.2: Types and Applications
When we draw a sample, we examine the small subset, or
the sample. We do this because we lack the time and of Nonrandom Samples
resources or we do not need to look at every single unit or
4.2 Describe four nonprobability sampling techniques
case in the large collection, or population. If we sample
in research
carefully, we can reach highly accurate conclusions about
the whole population from the sample alone. We have sev- Random samples are best to use when we want to create an
eral ways to sample. Two factors influence the type of sam- accurate representation of a population, but they can be
pling design to use, whether the data are quantitative or difficult to conduct. Researchers unable to draw random
qualitative, and the purpose of the study. sample or who have goals that differ from creating a repre-
In quantitative research, we put a lot of effort into sentative sample often use nonprobability sampling tech-
sampling design because the goal is to produce genu- niques. Four such techniques are: convenience, quota,
inely representative sample, which has all the features of purposive, and snowball sampling.
the population from which it came. A representative
sample enables us to make highly accurate generaliza-
tions about the entire population from using the sample 4.2.1: Convenience Sampling
data alone. From probability theory in applied mathe- Convenience sampling (also called accidental or haphaz-
matics we know that using a random selection process ard sampling) is appealing because it is easy, cheap, and
72 Chapter 4
fast. Its biggest drawback is that it frequently produces tend to pick p eople who look “normal” to them and avoid
highly unrepresentative samples; plus, it lacks the depth unattractive, very busy, or inarticulate people. Their sample
and context sensitivity desired for qualitative research. If is for entertainment purposes and not for serious research.
you haphazardly select convenient cases, your sample may
Have you watched a television show that asks you to call
distort what is in the population (see Figure 4.1). in your opinion or visited a web page that asked you a
few survey questions?
Figure 4.1 Representative and Nonrepresentative These too are convenience samples. Only some people who
Samples of 6 out of 18
are watching television or visiting the web page respond.
Representative Samples Allow Accurate Generalization to
Even if the number who do so is large (e.g., 500,000), we
the Population
cannot generalize accurately from sample to the popula-
Representative tion. Like a person-on-the-street interview, such samples
can seriously distort what is in the population. It is impor-
tant to remember that the method of sample selection (not
the number responding) is the most important factor. A
large convenience sample should not be confused with a
Nonrepresentative true representative sample.
Of 32 adults and children in the street scene, select 10 for the sample:
• Next, decide how many units to get for each category. An interesting historical case illustrates the limitations of
• After you fix the categories and number of units in quota sampling. George Gallup’s American Institute of Public
each category, select units by any method. Opinion used quota sampling. It successfully predicted the
outcomes of the 1936, 1940, and 1944 U.S. presidential elec-
For example, you are interested in a sample of 80 shoppers
tions. However, in 1948, Gallup predicted wrong and said
at a grocery store. You think gender and age are important
Thomas Dewey would win over Harry Truman. The incorrect
aspects of diversity. You select 10 males and 10 females
prediction had several causes (e.g., many voters were unde-
under age 30, 10 of each gender aged 30–39, 10 of each gen-
cided, interviewing stopped early), but the main reason was
der aged 40–50, and 10 of each gender over age 60. You
that quota sampling categories did not represent all geo-
might interview the first 12 males who walk into the store,
graphical areas and all types of people who actually voted.
asking each his age. Once you have 10 who are under 30
years of age, you have to skip all other males in that age
group because you have filled your quota.
Quota sampling is an improvement over convenience
4.2.3: Purposive Sampling
sampling because it ensures that some major differences Purposive sampling is a widely accepted special sampling
within the population will also appear in a sample. In con- technique. It is appropriate when the goal is other than get-
venience sampling, everyone selected might be of the same ting a representative sample of the whole population. Pur-
age, gender, or racial category. However, quota sampling posive sampling requires having a very specific purpose in
has limitations and can yield a nonrepresentative sample. mind and using judgment to select cases, and is sometimes
A first limitation is due to the selection process for called judgmental sampling. In a way, it is convenience
placing cases into quota categories. Quota sampling relies sampling for a highly targeted, clearly defined population.
on a convenience selection process. This means, we might We use it in two situations:
select only people who “act friendly,” are easy to reach, or • To select especially informative cases.
who want to be part of the study into the sample. A second
• To select cases from a specific, difficult-to-reach
limitation is little diversity in a set of quota categories.
population.
Quota sampling categories only capture the diversity of a
few predetermined population characteristics. A popula- We may want to pick cases that have richer information. For
tion might differ in 20 ways, but quota samples rarely example, you want to examine magazine content for cultural
include more than three characteristics. Let us say a quota themes. You select two specific popular women’s magazines
sample of grocery shoppers includes combinations of gen- to study because they are most trend setting rather than
der (male/female), age (over/under 50), race (white/non- select a representative sample of all women’s magazines.
white), and shopping companion (none/with others). For To study a targeted group, we may use many diverse
this quota sample, you must find enough people in each methods to identify as many cases as possible. For exam-
combination. A third limitation is the sample size for each ple, you want to study people under 30 years old who use
quota category. Often, you set the number of cases to select wheelchairs in the Seattle metropolitan area. Without a list
for each quota category without having knowledge of the of wheelchair users, you cannot use a random sampling
true population. For example, you set a quota of 10 percent method. To use purposive sampling, you use many diverse
of the grocery shopper sample to be males under age 30. forms of information to get a sizable collection (maybe of
Yet, perhaps they actually make up 18 percent of the gro- 60 names). To get the names, you may go to locations
cery shopping population. where wheelchairs are sold and repaired or ask knowl-
edgeable local experts (e.g., health workers, other wheel-
chair users, or disability advocate groups).
Example Study
Purposive Sample
In Italy and Japan, unmarried adults in their 30s often live at
home with their parents. In fact, more than half of Italian men,
aged 25–35 still live at home with their mothers. In Japan, 60
percent of unmarried men and 70 percent of unmarried
women aged 30–34 remain at home with their parents. There
is even a term, “parasite single,” to describe the situation. Yet
Bad Samples Often Yield Inaccurate Results in Scandinavia, teens as young as 16 live independently of
74 Chapter 4
their parents. Katherine Newman (2008) wanted to find out • Drug dealers and suppliers who work together to form
what such relationships mean to those involved, and adopted a distribution network
a comparative, qualitative approach. She wanted to uncover
• People on a college campus who have had sexual rela-
the subjective, cultural understandings of autonomy and inde-
tions with one another
pendence of “delayed departure” and “accelerated indepen-
dence.” She assembled a research team from three “delayed The crucial feature is that each person or case has a connec-
departure” countries (Spain, Italy, and Japan) and two “accel- tion with the others. The linkage can be either direct or
erated independence” countries (Denmark and Sweden). indirect. Members of the network may not directly know
She used judgment sampling to obtain 49–52 interviews or interact with all others in the network. Rather, taken as a
per country, a total of 250 interviews. Qualitative interviews were
whole, each is part of a larger linked web.
conducted with unmarried adults over 22, some co-residing
with parents, others living independently, and parents of an
unmarried adult. The interviews included people in both urban
and rural areas and from different regions of each country. The
sample included parents and adult children in the same families
when possible, and in different families when a parent and child Example Study
in the same family could not be interviewed. She says (p. 651)
“we cannot claim that these samples are representative in any
Snowball Sample
definitive way.” The interviews by native speakers took place in
homes and cafes. She found that in the “delayed departure”
countries the parents had almost uniformly experienced “a
before (childhood) and an after (adulthood) that was marked by
clear behavioral changes in their lives” (p. 652). The clear divid-
ing line included marriage, full-time employment, and childbear-
ing. For the young adults today, adult maturity was marked less
by outward behavioral changes than a slowly evolving inner feel-
ing of more responsibility and making independent decisions.
This occurs while still living in the natal home. The unmarried
adults and parents both recognized that staying at home was
often an economic necessity and required creating new parent–
Anju Mary Paul (2011) studied “stepwise international migra-
child relationships. In “accelerated independence” countries,
tion.” It is when people with limited income and skills from
attending high school or university away from home where gov-
low-income countries engage in a multistage process of inter-
ernment aid covers all expenses is common. Remove the eco-
national labor migration. They work in intermediate countries
nomic necessity and physical separation from the natal family
as a conscious strategy to reach a desired final destination
occurs early and brings slightly more internal, subjective auton-
country. As they work outside their home country, the
omy and emotional separation for the unmarried adults. Some
migrants increase savings and gain work experience and edu-
youth said that they felt somewhat less close to their family, and
cational certifications. They also build a network of overseas
believed family relations were stronger in other countries.
contacts with the goal of accumulating resources to gain
entry into a desirable destination country. Dr. Paul conducted
in-depth interviews with 95 prospective, current, and former
Filipino domestic workers in the Philippines, Hong Kong, and
4.2.4: Snowball Sampling Singapore about their migration and destination decisions.
Snowball sampling is a special technique used when we She approached local nongovernmental organizations that
sought to improve the welfare of migrant workers and asked
want to capture an already-existing network. Its name
for help in finding potential interviewees. She also approached
comes from an analogy to the way a snowball increases in
Filipino domestic workers in Singapore and Hong Kong during
size: It begins small but gets larger as you roll it, and it
their off days when they gathered in downtown shopping cen-
picks up additional snow. In this multistage technique, you ters. In the Philippines and Singapore, she included clients of
start with one or a few cases, then spread out based on recruitment agencies. She used snowball sampling by ask-
direct or indirect linkages to the initial case. We often use ing the first contact interviewees to refer her to other Filipinos
snowball sampling if we want to sample a social network they knew. Her final sample had 27 women in the Philippines,
of people or linked organizations. Networks for which 26 women and 2 men in Hong Kong, and 40 women in Singa-
researchers used snowball sampling include the following: pore. She conducted semistructured interviews about deci-
sions to leave the Philippines and work overseas as domestic
• Scientists around the world who are investigating the workers. Dr. Paul learned that many migrants (40 percent) had
same issue worked two or more years in several other low or middle
• The elites of a medium-sized city who consult with countries in Asia or the Middle East as a conscious strategy to
one another reach their goal of entering a high-income Western country.
Sampling 75
Each change in country was a movement upward, with better the students in a classroom, all employees currently work-
wages and working conditions. Most of the intermediate ing the second shift at factory number three of Tom’s shoe
countries in Asia and the Middle East had policies to facilitate company on March 30 of this year), you have to refine the
the entry of foreign domestic workers on renewable short- population to be very specific (i.e., the target population)
term contracts. However, none offered the option of perma-
before you can draw a sample.
nent residence.
Once you designate a target population, you must
create a list of all its sampling elements, or the sampling
frame. There are many types of sampling frames: tele-
4.3: The Terminology Used phone directories, tax records, driver’s license records,
and so on. In the opening study about teens carrying
to Discuss Random guns, the researchers had a complete list of schools and
English classes for the sampling frames. The researchers
Sampling did not create a list of all individual students because
all students in the same English class were in the sam-
4.3 List the specialized vocabulary used in the process ple. Listing the elements is often difficult because no
of random sampling good list of the elements in a population exists. A good
A random sampling method will produce a sample (i.e., sampling frame is crucial to accurate sampling. If there
small collection of cases) that most accurately represents a is a mismatch between the sampling frame and the pop-
far larger population. The process of random sampling has ulation, it can create major errors and cause invalid
a specialized vocabulary (see Figure 4.3). sampling.
You recall the three terms: population, sample, and Any statistical characteristic of an entire population
universe. The case or unit of analysis in the population is a (e.g., the percentage of city residents who smoke cigarettes,
sampling element. It can be a person, a group, an organi- the average height of all women over the age of 21, the per-
zation, a written document or symbolic message, or even a cent of people who believe in UFOs) is a population
social action (e.g., an arrest, a divorce, or a kiss). Three parameter. If you have all the elements in a population,
terms with similar meanings can be a source of confusion. you accurately compute a parameter with absolute accu-
We select the sample from a population, but usually we tar- racy. For very large populations (e.g., an entire nation), you
get a more specific and concrete collection of sampling ele- never have all elements, so you use information in a sam-
ments within the overall population. ple to estimate the population parameter. If you end up
For example: taking a statistics class and hear about “parameter estima-
tion,” this is where it comes from.
• Universe: all people in Florida The sample will be smaller than a target population,
• Population: all adults in the Miami metro area and a sampling ratio indicates the percentage of the target
• Target population: people aged 18–88 who had a population that is in a sample. To compute this, we simply
permanent address in Dade County, Florida, in divide sample size by target population. For example, a
September 2016, and who spoke English, Spanish, target population has 50,000 people and you draw a sam-
or Haitian Creole ple of 150 from it. Your sampling ratio is 150/50,000 =
0.003, or 0.3 percent. If the target population is 500 hospi-
The population is more an idea than it is something con- tals and you sample 100 of them, your sampling ratio is
crete. Except for small or specialized populations (e.g., all 100/500 = 0.20, or 20 percent.
Sample
Sampling Process
Sampling
Frame
76 Chapter 4
Sampling with a Random Selection Process As Landon over Franklin D. Roosevelt. In this election, the Liter-
you learned, random samples are most likely to represent ary Digest was very wrong; Franklin D. Roosevelt won by a
the population. However, sampling with a random selec- landslide. The prediction was wrong because the sampling
tion processes requires a lot more work than nonrandom frame did not accurately represent all voters. It excluded
people without telephones or automobiles, as much as 65
ones. In statistics, the word random refers to a random
percent of the population in 1936, during the worst part of
selection process, one that gives each element in a popula-
the Great Depression. More importantly, this excluded seg-
tion an equal (or known) probability of being selected. Two
ment of the voting population (lower income) tended to favor
critical features of true random processes are: Roosevelt. The magazine had been accurate in earlier elec-
1. They are purely mechanical or mathematical without tions because people with higher and lower incomes did not
human involvement. differ in how they voted. In addition, before the Depression,
more lower-income people could afford telephones and
2. They allow us to calculate the probability of outcomes
automobiles. We can learn two lessons from the Literary
with great precision.
Digest mistake. First, the sampling frame is crucial. Second,
A random process enables us to estimate mathematically the the size of a sample is less important than whether or not it
degree of match between the sample and the population, or accurately represents the population. An excellent sample of
estimate the sampling error. Whenever we sample and do 3,000 can produce more accurate predictions about the 300
not have the entire population, the sample might deviate million in the U.S. population than a nonrepresentative sam-
ple of 30 million people.
from the entire population. Sampling error indicates the size
of this deviation or mismatch. Later in this chapter, we will
consider ways to minimize the sampling error.
WRITING PROMPT
The several kinds of random samples have three key
Sample Size ZE
features:
Many people focus on how many units are in a sample but ignore
• You must begin with an accurate sampling frame or the sample selection process. Explain why the method of selecting
list of elements in the target population. units to be included in a sample is equally or more important than
sample size.
• You must use a random selection process without
subjective human decisions (e.g., a computer program, The response entered here will appear in the
random number table). performance dashboard and can be viewed by
your instructor.
• You must identify and pick a particular sampling ele-
ment, rarely using substitutions.
Submit
For example, if you use a telephone directory as your sam-
pling frame (actually is it inaccurate, as you will see later)
and sample names for a telephone survey, you must reach 4.4: Producing a Simple
the specific sampled household or person. This means call-
ing back many times before giving up and going to a sub- Random Sample
stitute randomly selected alternative.
4.4 Evaluate the three main steps that are followed in
simple random sampling
The simple random sample is the basis from which other
Learning from History types of random samples are modeled. In simple random
sampling, there are three main steps to follow:
The Famous Literary Digest 1. Develop an accurate sampling frame.
Mistake 2. Select elements from the frame based on a mathemati-
The Literary Digest was a major U.S. magazine that pre-
cally random selection procedure.
dicted presidential elections in the 1920s and 1930s. The 3. Locate the exact selected elements to be in the sample.
magazine sent postcards to people before the U.S. presi-
In practice, after we develop a sampling frame we begin
dential elections. The magazine staff created a sampling
by numbering each element in the sampling frame, from 1
frame by taking names from automobile registration and
telephone directories. People returned the postcards indi-
to the last element. Next, we obtain a set of randomly
cating the candidate they supported. The magazine cor- generated numbers, from 1 to the largest number in the
rectly predicted election outcomes in 1920, 1924, 1928, and sampling frame. Most people do this with a computer
1932. The magazine’s success with predictions was well program that asks for the size of sampling frame and the
known. In 1936, the magazine increased its sample size to size of the sample; the program then produces a list of
10 million. The magazine predicted a huge victory for Alf random numbers. There are many such inexpensive or
Sampling 77
42 58 1
43 57 1
45 55 2
46 54 4
47 53 8
48 52 12 Number of blue and white marbles that were
49 51 21 randomly drawn from a jar of 5,000 marbles
50 50 31 with 100 drawn each time, repeated 130
51 49 20 times for 130 independent random samples.
52 48 13
53 47 9
54 46 5
55 45 2
57 43 1
Total 130
Number of Samples
31 *
30 *
29 *
28 *
27 *
26 *
25 *
24 *
23 *
22 *
21 * *
20 * * *
19 * * *
18 * * *
17 * * *
16 * * *
15 * * *
14 * * *
13 * * * *
12 * * * * *
11 * * * * *
10 * * * * *
9 * * * * * *
8 * * * * * * *
7 * * * * * * *
6 * * * * * * *
5 * * * * * * * *
4 * * * * * * * * *
3 * * * * * * * * *
2 * * * * * * * * * * *
1 * * * * * * * * * * * * * *
42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57
Number of Blue Marbles in a Sample
that are close to the population parameter. The mathemat- A central idea in random sampling is: We do not have
ics supporting random sampling enables us to estimate 100 percent accuracy 100 percent of the time. In reality, such
how close results are to the population parameter. The accuracy is virtually impossible in most endeavors. Life is
same logic lets us calculate sampling errors and estimate full of risk and chance. Some activities have a high risk of
the probability that a particular sample is unrepresenta- serious mishap or death, such as skydiving, speeding
tive. In short, information from one sample allows us to while driving drunk, or defusing a live bomb. Other activi-
estimate a sampling distribution of many samples. The ties have a very low risk such as walking to the local store,
mathematics of sampling distributions can tell us the sitting in a classroom, or eating your lunch. We cannot pre-
chances that sample results deviate from the true popula- dict an outcome with 100 percent accuracy, but we know
tion parameter. the probability of an outcome varies by type of activity.
Sampling 79
Number of Samples
31 *
30 *
29 *
28 *
27 *
26 *
25 *
24 *
23 *
22 *
21 * *
20 * * *
19 * * *
18 * * *
17 * * *
16 * * *
15 * * *
14 * * *
13 * * * *
12 * * * * *
11 * * * * *
10 * * * * *
9 * * * * * *
8 * * * * * * *
7 * * * * * * *
6 * * * * * * *
5 * * * * * * * *
4 * * * * * * * * *
3 * * * * * * * * *
2 * * * * * * * * * * *
1 * * * * * * * * * * * * * *
42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57
Number of Blue Marbles in a Sample
6. Sampling distribution get the sampling interval. It tells you to skip over five
7. Sampling element names then pick the sixth one for the sample. Repeat the
8. Sampling error process until 300 names are in the sample. In many
9. Sampling ratio instances, the sample produced with systematic sampling
10. Sampling frame will differ little from random selection.
11. Target population Avoid using systematic sampling if elements in the
sampling frame are grouped or organized in a pattern,
12. Universe
because the resulting sample can deviate significantly from
that produced by random selection. Grouped elements
WRITING PROMPT make it possible to skip over elements with a sampling
interval. For example, the elements are individual names
Sampling Distributions
grouped into four-person family units. With a sampling
A sampling distribution is a collection, or distribution, of many sepa-
interval of 16, the sample would contain one family mem-
rate samples drawn from the same population. Explain how a collec-
tion of many samples can help us learn more about a population ber from every fourth family unit and regularly skip over
than just taking one sample. other people from the same family unit or people in adja-
cent family units (see Table 4.1).
The response entered here will appear in the
performance dashboard and can be viewed by
your instructor.
Table 4.1 Problems with Systematic Sampling
of Cyclical Data
Submit
Case
1 Husband 7 Husband
of Random Samples 4
5
Wife
Husband
10a
11
Wife
Husband
4.5 Identify the three main types and uses of random 6a Wife 12 Wife
sampling Random start = 2; Sampling interval = 4.
a
Selected into sample.
Although we have talked about the random sample gener-
ally, there are actually several types:
Making It Practical: How to Draw Simple Ran-
• systematic
dom and Systematic Samples The processes of
• stratified
drawing a simple random and a systematic sample are
• cluster sampling slightly different, but they usually yield very similar results
All are slight variations on the true, simple random sample. (see Table 4.2).
WRITING PROMPT schools from each stratum to get eight schools. Next, she ran-
domly selected a single class from each grade level at each
Systematic Sampling
school. It was the largest required course (usually E
nglish). As a
Since systematic sampling is a type of “short-cut” method to get a result, she administered the survey in 32 classrooms, where 72.8
representative sample, in what ways is it faster or easier than draw- percent of the students completed the survey. This yielded 982
ing a true random sample? What are some of its limitations?
completed surveys. She then examined rates of hazardous
The response entered here will appear in the drinkers and other factors. She found that the strength of individ-
performance dashboard and can be viewed by ual-level risk and protective factors varied by type of neighbor-
your instructor. hood that fed into a school.
Submit
16 184
WRITING PROMPT
Sample of 200
employees
Stratified Sampling
4.5.3: A Cluster Sample clusters. In the study that opened this chapter, the authors
began with the population of all Boston area high schools.
A cluster is a grouping of the elements in the population. In
They had a sampling frame of all English classes by grade
cluster sampling, we treat each cluster as temporary sam-
for the schools. They sampled one English class from each
pling element. Frequently, there is no good sampling frame
grade in each school. They did not need to have a sampling
for a smaller unit, but there is a good sampling frame for
(a) (b)
Step 2: Randomly select a city block or section of the county.
also explain violence levels. If people trust and believe in the fair- a good random sample, we need a complete sampling
ness of the law, the courts, police, and law enforcement, they frame but published telephone directories are not accept-
will call the police and rely on the legal system to resolve con- able. We miss four kinds of people using a telephone direc-
flicts. However, if they have high levels of legal cynicism, i.e., tory as the sampling frame:
they distrust and fear the police and believe the legal system is
unfair or ineffective, they will instead rely on nonlegal methods to • People without telephones
resolve local conflicts (i.e., illegal gangs or individual acts of vio- • People who have recently moved
lence). To examine the legal cynicism, the authors looked at sur-
• People with unlisted numbers
vey data from a large-scale study, the Project on Human
Development in Chicago Neighborhoods (PHDCN) that used • People who only use a cell phone
cluster sampling. The PHDCN grouped the 865 census tracts Any telephone interview study will miss people without
in the City of Chicago into 343 homogenous “neighborhood phones (e.g., poor, uneducated, and transient people), but
clusters.” Each cluster had similar social-economic features,
this is usually not a big problem since nearly 97 percent of
natural boundaries (e.g., railway tracks, major highways), and
people have phones. As more people got phones, the per-
was of equal size, about 8,000 people. The researchers drew a
centage of them with unlisted numbers also grew. In some
random sample of 80 from the 384 clusters. In each of these 80
sampled neighborhoods, they randomly selected city blocks. urban areas, over 50 percent of phone numbers are unlisted.
Next, they went down the randomly selected blocks and com- In addition, people change their residences, so directories
piled a list of all dwellings on the blocks. This created a list of often have numbers for people who have left and do not
roughly 40,000 dwellings. They next went to a sampled dweling include those who have recently moved into an area. One
and screened people to randomly select 8,782 residents for report suggests that as of 2014 about 40 percent of Ameri-
interviews. The multi-stage cluster sampling had four levels: cans only have a cell phone, no landline.*
1. neighborhood clusters,
2. city blocks, 4.6.1: Random-Digit Dialing
3. dwellings on city blocks, and Random-digit dialing (RDD) avoids the problems of tele-
4. residents in the dwellings. phone directories by randomly sampling possible tele-
The researchers found that even in neighborhoods that phone numbers, not a list of people with telephones. This
had improved with regard to structural factors, the level of vio- avoids the bias of using listed numbers. RDD is not diffi-
lence remained high if the residents expressed high levels of cult, and there are several specialized computer programs
legal cynicism. designed to make the calls. However, it takes time and can
frustrate the person doing the calling. Many of the num-
bers may not be operating. As with any sampling that uses
WRITING PROMPT the telephone, you must retry reaching a selected, working
Neighborhoods as Clusters number many times (5–6 attempts at different times) before
giving up on it.
Imagine that you want to sample individuals in a medium-sized city
and your first cluster was the neighborhood, how might you continue Making It Practical: How Random-Digit Dial-
after you drew a random sample of 15 neighborhoods (from among
55) of the city to get to the individual level? Would you want the 15 ing Works In the United States, a telephone number
neighborhoods that are similar to each other, or very different types has three parts: a three-digit area code, a three-digit ex-
of neighborhoods? Why? change number or central office code, and a four-digit
The response entered here will appear in the number. It is easy to create a list of all possible phone num-
performance dashboard and can be viewed by bers by getting a list of active area codes and three-digit
your instructor. phone exchanges in the area code. Possible phone numbers
in an exchange go from 0000 to 9999. In RDD, a computer
Submit randomly selects a number (0000 to 9999) in an exchange
and makes the call. Some selected numbers are out of ser-
vice, disconnected, pay phones, or numbers for businesses.
4.6: Sampling in Difficult Only some numbers are what you want—a working resi-
or Specialized Situations dential phone number. Until you call, it is not possible to
know whether the number is a working residential num-
4.6 Identify potential challenges presented in ber. This means spending a lot of time getting disconnected
sampling for research and how to effectively numbers, numbers for businesses, and so forth. The sam-
manage those difficulties pling element in RDD is the phone number, not the person
or the household. Several families or individuals can share The purpose of randomly selecting one person from the
the same phone number, or each person may have a sepa- household is to avoid the bias of including only certain
rate phone number or more than one number. This means types of people, whomever always answers the phone or
that after contacting a person at a residential phone, a sec- door first. However, within household sampling often
ond stage is necessary, within-household sampling, to adds time, because the individual of the household ran-
select the person who you will interview. domly selected may not be immediately available.
WRITING PROMPT
Random Digit Dialing
4.6.3: Sampling Hidden
Why is Random Digit Dialing (RDD) widely used in telephone survey
Populations
research? What are some of the limitations or drawbacks of RDD? Sometimes a particular type of person is very difficult to
locate, making sampling more complicated. A hidden
The response entered here will appear in the
performance dashboard and can be viewed by population includes people who engage in concealed
your instructor. activities. They are often central in the studies of deviant or
stigmatized behavior. Examples of hidden populations
Submit include users of illegal drugs, prostitutes, people with
HIV/AIDS, people on parole, or homeless people.
We need to make adjustments to sample people in a
4.6.2: Within-Household Sampling hidden population because they are more difficult to
locate than the general population of visible and accessi-
We use within-household sampling to address situations
ble people. To sample hidden populations, we need to be
in which we sample households. Building on the idea of
creative and use nonprobability sample techniques, such
cluster sampling, the household is a kind of cluster with
as purposive or snowball sampling. As you have seen,
multiple sampling elements, or individuals. We want to
there are many different types of samples and the type of
avoid a possible bias of having one member of the house-
sample is based on the study purpose and type of data
hold always being in a sample because he/she is the one
being gathered.
who regularly answers the phone, or mail, or door. One sit-
uation occurs with telephone interviewing when we have
one phone number for several people in a household. This
is a common situation when we use RDD and have a sam-
pling frame that is a list of phone numbers. Another situa-
Example Study
tion occurs with face-to-face or mail surveys when we want Hidden Populations
to sample the individuals in a town, city, or state but do not
have a current, complete list of each person in the area. It Draus and associates (2005) sampled a hidden population in
a field research study of illicit drug users in four rural Ohio
might be possible to create a list of all addresses (i.e., all
counties. The researchers used a version of snowball
houses, apartments, and similar residential locations).
sampling, respondent-driven sampling (RDS), which is used
In both situations, instead of picking the first person
when members of a hidden population maintain contact with
who answers a phone, or who comes to the door or who one another.
happens to read the mail first, we sample individuals within RDS begins by identifying an eligible case or partici-
a household. For within-household sampling, we first deter- pant. The researchers give this person, called a “seed,” refer-
mine the number of eligible members in the household, such ral coupons to distribute among other eligible people who
as adults over a certain age. If there is only one person, we engage in the same activity. For each successful referral, the
choose that person. If there are two or more people, we ran- “seed” receives some money. This process is repeated with
domly pick one. Let us say you consider all adults over 18 several waves of new recruits until a point of saturation (no
years old who reside in a household. There are three eligible new people). Draus and associates interviewed a drug-using
people at one house: a 50-year-old woman, a 53-year-old participant who was paid $50 for an initial two-hour interview
and $35 for an hour-long follow-up interview. The partici-
man, and a 20-year-old woman. To select a person randomly,
pants received three referral coupons at the end of the initial
we use selection rules, such as the following:
interview. They got $10 for each eligible participant they
• If the last sampled person interviewed was male, select referred who completed an initial interview. No participant
received more than three referral coupons. Sometimes this
a female (or vice versa).
yielded no new participants, but at other times more than
• If there is one male or one female, interview that person. three people were recruited. In one case, a young man heard
• If there are two females/males, select the oldest one about the study at a local tattoo parlor. He called the study
first, next time select the youngest. office in July 2003. He (participant 157) had been a powder
Sampling 87
sampling error will be tiny. If you have a small sample (80 size, shrinks. Without getting into the complexity of this,
randomly selected from 80,000) and there is great diversity three simple principles can help you make sense of what is
among cases, the sampling error will be very large and important:
your sample may not accurately represent the population.
1. For small populations (less than 500), you need one
half or more in your sample, and the required sample
4.7.2: Sample Size size grows very fast as the population size gets smaller.
A large sample size alone does not guarantee a representa- 2. For target populations over 5,000, you need 17.5 per-
tive sample. A large sample without random sampling or cent or 27 percent of the population, depending on
with an inaccurate sampling frame will be less representa- degree of confidence. Sample size changes very little as
tive than a smaller sample that uses random sampling and the population size grows larger.
has an excellent sampling frame. 3. Once your target population is over 250,000, the sam-
Calculating sample size mathematically requires ple size hardly changes at all.
making assumptions, estimating population size and
Practically, this means that if you want to sample a very
diversity, knowing how many variables you plan to
small target population, such as the 50 employees of the
examine, determining how confident you want to be, and
local fast food outlet, just include everyone, or a sample
the degree of accuracy you require. The mathematics and
ratio of close to 100 percent. However, to sample a city of
calculations are beyond the level of this text. Most people
250,000 people, you can be equally accurate with a sam-
just use “rules of thumb.” These are rough approxima-
ple of 1,063 in your sample, or a sampling ratio of 0.4 per-
tions based on target population size, assuming you
cent. The main idea about sample sizes is that the smaller
examine one or a few variables, there is moderate popula-
the population, the bigger the sampling ratio must be for an
tion diversity, and you need medium accuracy (such as
accurate sample. Larger populations permit smaller sam-
getting within 3 percent of the population parameter).
pling ratios for equally good samples because as the pop-
Table 4.2 provides sample sizes for a range of target pop-
ulation gets bigger, returns in accuracy for sample size
ulation sizes (50 to 100 million).
quickly shrink. This is why random sampling is so pow-
erful and efficient when you want estimates about large
populations.
Table 4.6 Sample Sizes for Two Levels of Confidence and
Various Population Sizes
4.7.3: Confidence Intervals
Target Population 95 Percent 99 Percent
Size Confident Confident The confidence interval expresses a familiar idea. When
50 48 49 reporters discuss polling results, they say “the margin of
200 168 180 error being plus or minus 2 percentage points.” They are
500 340 393 using a simple version of confidence intervals. After you
1,000 516 648 draw a random sample, you may look at a characteristic or
5,000 879 1347 measure you get from the sample, such as the average
25,000 1023 1717
income or the percentage saying “agree” to a question. From
100,000 1056 1810
the above discussion of sampling error, you know this does
250,000 1063 1830
not indicate that the sample measure (income or percent
agree) is identical to what is in the population, or the popula-
1,000,000 1066 1840
tion parameter. You use the sample results to estimate the
100,000,000 1067 1843
population parameter.
The mathematics around a sampling distribution
Notice from the table that the sampling ratio is very helps us estimate a zone or range around what we find in a
large when the target population is small. When your sample within which the population parameter will be.
population is under 500, you will need about one-half the Interval in confidence interval is the zone or range around
population in your sample to be highly confident and what you found in sample. Confidence in confidence interval
accurate. If your population is under 100, you might as refers to the probability that the population parameter falls
well take everyone in the sample. However, if your popu- within the interval. A typical level of confidence is
lation is 25,000, you need less than 10 percent. This per- 95 percent. It means you can be 95 confident that the true
centage gets smaller as the population grows in size. As population parameter falls within the interval. A higher
the population size increases, the number in the sample level of confidence (99 percent) requires you to have a
size might grow too, but proportion from the population slightly larger sample size, everything else being identical.
that is in the sample, or the ratio of sample to population Sample size affects the confidence interval. Everything else
Sampling 89
being the same, as your sample size gets bigger, the interval you draw a second sample. This time you increase the sam-
gets narrower. ple size to 500 eligible voters. You find that 51 percent sup-
The mathematical calculations for sampling errors or port the referendum, but your confidence interval with this
confidence intervals build on the ideas of the sampling dis- larger sample size is narrower. It is 1.5 percent above or
tribution. For example, you cannot say, “There are pre- below, or from 50.5 to 53.5 percent. It now looks as if the
cisely 2,500 red marbles in the jar based on a random vote will be close, but the referendum is likely to pass (see
sample.” However, you can say, “I am 95 percent certain Figure 4.7).
that the population parameter of red marbles is between
2,450 and 2,550.” You can combine characteristics of the
sample (e.g., its size and variation) to set up an interval Figure 4.7 Confidence Interval with Sample of 100. 99%
Confident
and level of confidence. Let us say you have two identical 99% confidence interval with a sample size of 100
samples except that one is larger. A larger sample will have 55.6
48.4
a smaller sampling error and a narrower confidence inter-
val. A narrow confidence interval lets you be more precise
52% estimate
when estimating the population parameter.
An example illustrates the basic ideas. You will see 99% confidence interval with a sample size of 500
how sample size reduces the sample error, which in turn 53.5
50.5
affects the confidence interval.
Example
Let us say you want to know how many people are likely to
support a referendum to add a tax for a new school in a city WRITING PROMPT
that has 5,000 eligible voters (i.e., a target population). You Confidence Intervals
want to be 99 percent confident. You get a good sampling Give an example from an area of daily life (e.g., sports, cooking,
frame and use simple random sampling. At first, you draw a politics, weather, travel, appointments) where the core idea of
random sample of 100 people and find that 52 percent sup- confidence intervals can be found, and explain how it illustrates
the concept.
port the referendum. Your confidence interval is 3.6 percent
above or below that number, or 48.4 to 55.6 percent sup- The response entered here will appear in the
port. In other words, you can be 99 percent certain that the performance dashboard and can be viewed by
your instructor.
population parameter (true vote intention in the population)
lies between 48.4 and 55.6. The vote is too close to call,
and it could go either way. You want to be more certain, so Submit
s ampling, and specific research techniques are interde- 4. Purposive sampling is a valuable type of nonprobability
pendent. Unfortunately, the constraints of presenting infor- sample that is appropriate to use, especially in qualitative
mation in a textbook necessitate presenting each part studies, when the goal is to focus on a specifically tar-
separately, in sequence. In practice, researchers think about geted set of cases or units.
data collection when they design research and develop 5. Snowball sampling is a valuable type of nonprobability
measures for variables. Likewise, sampling issues influ- sample. Its special design enables it to capture cases or
units that are within of an interlinked network.
ence research design, measurement of variables, and data
collection strategies. As you will see in future chapters,
good social research depends on simultaneously control-
ling quality at several steps—research design, conceptual- The Terminology Used to Discuss
ization, measurement, sampling, and data collection and Random Sampling
handling. Making major errors at any one stage could 1. Random sampling has its own terminology. In addition to
make an entire research project worthless. the previously introduced terms of universe, population,
and sample, some of key terms include target population,
sampling element, and sampling frame. The terms refer
respectfully to the specific, concrete population of a study,
Quick Review the cases or units of that population, and a specific list of
every element.
Why Do We Use Samples? 2. Three other important terms are sampling ratio,
1. We often invest at lot of effort into sampling design for population parameter, and sampling error. The sampling
quantitative studies because our goal is to create a genu- ratio is the size of the sample relative to the size of the
inely representative sample (i.e., a sample that has all the population. The population parameter is some character-
features of the population) that will enable us to make istic of interest in the entire population; we often estimate it
accurate generalizations about the entire population using using the sample. The sampling error is the degree of mis-
sample data. The best representative samples use a ran- match or deviation of a sample from the population. In
dom selection process. theory, a perfectly representative sample has a sampling
2. The data from a well-designed random sample are equally error of zero. While this is very rare, we can use statistics
or even more accurate than if we tried to reach everyone in with a true random sample to estimate the size of the sam-
the population. In addition, the samples are highly efficient pling error.
in terms of time and cost.
3. A census is an official government count carried out at
periodic intervals. Usually, it attempts to count everyone, Producing a Simple Random
but this is less accurate than a well-designed random Sample
sample.
1. The process of producing a simple random sample
4. In qualitative research, our goal differs from getting a rep- requires that we first list all elements of the target popula-
resentative sample of a large population, so instead of tion in a sampling frame and then use a true mathemati-
using random sampling we use nonrandom types of sam- cally random process to select some elements from the
pling. Our goal is usually to learn how a small collection of frame.
cases, units, or activities can illuminate key features of
2. Some people have the misconception that a bigger sam-
social life. Some types of nonrandom sampling are better
ple is always best. If the sampling frame is inaccurate, or if
designed for this purpose.
a nonrandom selection process is used, the sample can
be significantly “off” or unrepresentative of the population,
Types and Applications of even if the sample size is very large. However, for two
samples of the same target population that use equally
Nonrandom Samples accurate sampling frames and equally random selection
1. Random samples are best to use when the goal is to cre- processes, the sample that is larger is likely to be more
ate an accurate representation of a population, but they representative of the population.
can be difficult to conduct and are not appropriate for all 3. The sampling distribution is a collection of different ran-
purposes. Four nonprobability techniques are also used. dom samples taken from the same population, and then
2. Convenience or haphazard samples are very quick and plotted on a graph. Used with the mathematics of proba-
easy, but can be very inaccurate and nonrepresentative. bilities, it enables us to estimate the size of sampling error
This makes them inappropriate for most purposes. and the location of the population parameter. Although
3. Quota sampling yields a somewhat representative sample, random sampling is not 100 percent accurate each time,
and it much easier to produce than random sampling, but using it enables us to calculate the probability that a spe-
it has important limitations. cific sample deviates from the population.
Sampling 91
Types and Uses of Random 2. When sampling households, either as a type of cluster or
by phone where one phone is shared by a household, we
Samples need to sample randomly within the household. This helps
1. Systematic sampling is a less complicated version of the ensure that one household member who may regularly
simple random sample. You still need to have a good sam- answer is not oversampled.
pling frame, but instead of a random selection process you 3. Some populations are “hidden” because they engage in
create a sampling interval. After picking a random starting socially disapproved or illegal behavior, are highly transient,
place, you select sampling elements using the sampling or lack a stable residence (e.g., homeless). Sampling such
interval. So long as there is no cycle, the results are only populations requires additional efforts that vary with the cir-
slightly less accurate than the simple random sample. cumstances of the particular hidden population.
2. In stratified sampling, you first subdivide the sampling
frame into categories, or strata, then you draw a random
sample from each (or can take the entire population of a
strata if it is small). This requires you to have knowledge of
strata in the population and be able to subdivide it. This
Shared Writing: Samples You Can Believe In
can be more accurate than a simple random sample.
Sampling principles are the same whether you are sampling
3. Cluster sampling is a type of multi-stage sampling process.
leaves from a forest floor, fish from the ocean, products on an
It is often used when there is no good sampling frame for the
assembly line, or the residents of a town. In your opinion, what
final sample elements, but there are good sample frames for
makes drawing a true random sample of individual people in a
clusters of the elements. In addition, clusters can be nested town more difficult or complicated than sampling leaves, fish, or
within one another. You sample clusters then once you have products? When reading about a sample of individual people in a
clusters, you can create a sampling frame for elements study, what features of the sample would you be looking for to
within that cluster for another stage of sampling. give you great confidence that the sample truly represents the
entire population?
Learning Objectives
5.1 Analyze how measures in social life shape 5.5 Describe how principles of reliability and
research outcomes and larger social issues validity operate for qualitative data
5.2 Describe how measurement in the social 5.6 List the characteristics of the four levels of
world extends the range of our natural senses measurement as they relate to quantitative
measurement
5.3 List four ways in which systematic data
may be collected 5.7 Apply the processes involved in
the construction and use of indexes
5.4 Apply reliability and measurement validity
and scales
to create good measures
Are you a “redneck”? Do you know “rednecks?” What is a white S outherners, she used purposive and snowball
“redneck anyway?” Carla Shirley (2010) conducted a sampling to contact and secure cooperation from 42
study to address this question. She chose four rural com- respondents. She conducted 1–3 hour in-depth interviews
munities in two eastern Mississippi counties. After con- with an equal number of white men and white women
tacting community development clubs and churches aged 24 to 84 in private settings. Half of the respondents
explaining that she was doing research on the lives rural had lived in their community for their entire lives. The
92
Measuring Social Life 93
remainder came from other rural communities or metro- measure types, amounts, frequency, intensity, duration, loca-
politan areas of the American South. She asked a range of tion, and so forth about the concepts you wish to study.
questions about the meaning of being a white Southerner Measures can profoundly shape both research out-
and the types of people called “rednecks.” She also asked comes and larger social issues. Consider intelligence. Psy-
for a definition, whether the term’s meaning had changed, chologists debate what intelligence means and how to
and whether media portrayals of rednecks were accurate. measure it. Most intelligence tests used in schools, on job
Two-thirds of respondents said that besides being a white applications, or in statements about racial or other inherited
Southerner, a redneck had a set of values or lifestyle. Nearly superiority measure only one type of intelligence, analytic
all (84 percent) said the media and non-Southerners associate reasoning (i.e., a capacity to think abstractly and to infer
the term with a Southern white person who is backwards, logically). Most experts agree that humans possess multiple
uneducated, and racially prejudiced. Other “redneck” charac- types of intelligence besides the analytic type, such as prac-
teristics include being of lower social status and someone tical, creative, social-interpersonal, emotional, body-kines-
who has lived in the same rural community for his or her thetic, musical, or spatial intelligences. If there are many
entire life. Interestingly, 95 percent of respondents said the types of intelligence but schools and businesses are only
term refers more to males than females, especially a male who looking at one type, then schools and businesses are limited
chews tobacco, drives a pick-up truck, and spends a lot of lei- in how they are evaluating, promoting, and recognizing
sure time fishing and hunting. Negative characteristics people’s contributions. The way we measure intelligence
include being self-centered and rude, combative, and racist. influences our ability to value diverse human abilities.
The more negative term “white trash” overlapped somewhat Here is another example. Human service agencies allo-
with that of redneck. Over one-third of the respondents iden- cate assistance from social programs (e.g., subsidized hous-
tified male relatives or themselves as “rednecks.” Many ing, food aid, health care, child care, etc.) to people identified
acknowledged that outsiders used the term more broadly as being poor. Government agencies shift funding to an area
than its use among white Southerners. The term “redneck” based on the number of poor people living there. Politicians
has multiple meanings, but its core refers to a white, rural and economists argue over rising and falling poverty rates.
male Southerner of lower social status. To study any topic,
including rednecks, you first must clarify the concept’s mean- 5.1.1: Who Is Poor?
ing and think seriously about how you might measure it.
Some of what we want to measure (e.g., age, height, techniques (such as a questionnaire) to capture the
skin tone, eye shape, etc.) is visible, but we cannot directly ideas. Logic links ideas to the measures used to gather
observe many other things of interest (e.g., employee sat- empirical evidence. You may begin with a very abstract
isfaction, poverty, a child’s self-esteem, religious commit- idea (“quality of life”), link it to a less abstract idea
ment, desire to purchase, or quality of life). Just as natural (“being socially engaged”), then link it to a specific mea-
scientists invent indirect ways to observe the “invisible” surement act (answers to a survey question, “How often
forces of the physical world, social researchers create do you see friends socially?”). In a qualitative data
measures to reveal difficult-to-see aspects of the social study, you try to make sense of data by actively creating
world. new, or adapting existing concepts. You simultaneously
gather data and link data to ideas that clarify the data’s
meaning. You may first develop an idea (such as “din-
5.3: Do We Measure with ing in isolation”) based on gathering the data (observe
many people eating silently alone at a nursing home).
Numbers or Words? The idea, once developed, may influence your subse-
quent observations.
5.3 List four ways in which systematic data may
be collected
All researchers use careful, systematic methods to collect
5.3.1: Two Parts of the
data but the process varies in four ways depending on Measurement Process
whether the data are quantitative or qualitative: timing, Measurement builds on two processes: conceptualization
direction, form, and linkages. and operationalization. Often our ideas are fuzzy around
1. Timing. For quantitative data, first convert concepts the edges and not fully developed within; we conceptual-
into variables, next convert variables into specific mea- ize to sharpen and clarify ideas so that other people can
surement actions at a planning stage before from gath- understand them better. During the process of converting
ering or analyzing data. For qualitative data, create mental images or ideas into words, we develop a concep-
measures of concepts while collecting data. The pro- tual definition. This expresses what we mean by the idea.
cesses of thinking about concepts, collecting data, and To create a definition, we think carefully, are observant,
starting to analyze qualitative data all blur together. consult with others, read what others have written about
Measuring qualitative data is integrated with other the idea or related ones, and try out several possible defini-
research activities and not a separate stage. tions, making many readjustments to improve clarity. A
definition should identify the boundaries of the idea and
2. Direction. Recall the inductive versus deductive direc-
its core meaning.
tions of research. Most quantitative data research fol-
Let us say you want to develop a conceptual definition
lows a deductive route: Start with the abstract idea and
of discrimination. Perhaps you think it means a “negative
end with visible empirical data. Most qualitative data
action.” To begin the conceptualization process, consider
research follows an inductive route: Start with empiri-
your personal experiences, your inner thoughts about it,
cal data and end with a mix of ideas and data. In both,
discussions you have had with other people, and what you
the process is interactive, with measuring processes
read about it in the literature. Reflect on what you know,
influencing ideas and vice versa.
ask others what they think, and look up multiple defini-
3. Data form. In a quantitative data study, measures pro-
tions. As you ponder over the meaning of discrimination,
duce data in the form of numbers. You go from an
the core of the idea should become clearer. Eventually you
abstract idea to a data collection technique that will
may collect several alternative potential definitions and
yield precise numerical information. In a qualitative
will need to sort through them. You settle on:
data study, data might be in the form of numbers, but
more often, they are written or spoken words, actions, “Discrimination is the act of treating people unequally
sounds, symbols, physical objects, or visual images simply because they belong to a social category or are the
(e.g., maps, photographs, videos, etc.). Instead of con- member of a group.”
verting all observations into a single medium, num- You conclude that discrimination involves ways of
bers, you leave the data in many diverse shapes, sizes, behaving toward a type of people, and it refers to “others”
and forms. They stay as words, images, quotes, and or to an outgroup (a group to which a person does not
descriptions instead of all becoming numbers. belong). As your thinking expands, you may consider vari-
4. Linkages. In all research, you link ideas to observable ous types of discrimination based on various social
empirical data. In a quantitative data study, you reflect categories or groups—racial, religious, age, gender, lin-
on and refine ideas and then create specific measurement guistic, national origin, sexual orientation, and so forth.
96 Chapter 5
As you conceptualize, consider the unit of analysis 1. You are treated with less courtesy than other people.
and develop a measure for a unit of analysis that best fits 2. You are treated with less respect than other people.
the conceptual definition. For example, you think discrimi- 3. You receive poorer service than other people at restau-
nation is an action of individuals, and this makes the indi- rants or stores.
vidual the unit of analysis. However, organizations or
4. People act as if they think you are not smart.
institutions (e.g., families, clubs, churches, companies,
5. People act as if they are afraid of you.
schools, or media outlets) may also treat people in out-
groups unequally (e.g., not hiring an outgroup member). If 6. People act as if they think you are dishonest.
this is your focus, the group or organization becomes the 7. People act as if you are not as good as they are.
unit of analysis. To conceptualize and develop a measure, 8. You are called names or insulted.
you decide on a unit of analysis. Do you want to look at 9. You are threatened or harassed.
discrimination only as individual actions or as actions by
groups, organizations, and institutions? If a person’s answers indicated that many of these
During the conceptualization process, it is important actions occurred to him or her on a regular basis, the
to distinguish the idea, or concept, from closely related researchers considered it as empirical evidence of routine
ones. Ideas overlap with others and blur into one another. discrimination. Social scientists have created measures of
Good measurement requires separating the concept you many widely used ideas, such as social distance.
want to study from others that are closely related. For
example:
How are prejudice, racism, and stereotypes similar to or differ-
ent from discrimination? Learning from History
5.3.2: Conceptualization and Measuring Social Distance
Operationalization The famous sociologist Emory Borgadus (1882–1973) wrote
275 books and articles. Twenty-seven dealt with the idea of
Conceptualization requires thinking carefully about the social distance. In a 1925 article (“Social Distance and Its
concept. You might define discrimination as “a negative Origins,” Journal of Applied Sociology 9:216–226), he outlined
action of unequal treatment by a person directed toward the concept of social distance. He saw social distance as a
members of an outgroup that relies on stereotypes.” This is force that influenced most social relationships and indicated
more precise than your initial idea, “negative action.” It the degree of social-emotional closeness and trust in others.
now has links to other ideas, such as outgroup and stereo- He tried to capture how close or distant people felt toward
type. Repeatedly reevaluate each part of the definition. members of different racial and ethnic categories. Researchers
Consider the term “negative action.” Can a positive action still use variations of it today. To measure social distance,
be a kind of discrimination? Can there be positive discrimi- Borgadus asked people whether they were willing to interact
and establish specific social relationships with members of
nation, or unequal treatment in favor of a group based on
racial and ethnic groups other than their own. He asked peo-
stereotypes? Conceptualization is a process that requires
ple how they would feel to have a member of group X
systematical thinking, stating ideas clearly, and using pre-
cise terms. a. in close kinship marriage
Operationalization links a conceptual definition to a b. in my club as a personal chum
specific set of measurement procedures, or its operational c. in my street as a neighbor
definition. An operational definition restates the concep- d. as a fellow employee in my occupation
tual definition as measures. It might be one or more survey e. as a fellow citizen in my country
questions, a way to observe events in a field setting, or f. as a visitor only in my country
counting the appearance of symbols in a video. It is a spe- g. I would exclude all members of X from my country.
cific activity in which we observe, document, or represent a
conceptual definition. In the original list, X stood for 30 racial-nationality groups.
After people rated how they felt toward each group, Borgadus
It is not always necessary to invent a new measure of a
developed a picture of social distance among groups. The
concept if existing ones capture the idea adequately. Many
results showed that the dominant white majority felt very socially
measures of racial discrimination are already in use. For
distant from some groups and socially close to others. Borgadus
example, Gee et al. (2007) wanted to see whether Asian created an average distance score for each of the 30 groups. In
Americans who experienced more everyday discrimina- subsequent years, researchers replicated the study and found
tion suffered worse health. Their operational definition of slight changes in how distant social groups feel from one another
“everyday discrimination” was how people answered nine (see Kleg and Yamamoto, 1998; Parrillo & Donoghue, 2013).
survey questions about routine unfair treatment:
Measuring Social Life 97
Operational
Operationalization Operationalization Level
Besides the two main variables, the researchers con- s elf-aware and reflect continuously while you are in the
sidered other factors that often shape views on same-sex middle of doing research. You must simultaneously docu-
issues (e.g., liberal vs. conservative political views, educa- ment both the data and the process of how you gathered
tion level, and age) and had survey questions related to the data, including emerging ideas about the data.
each. They wanted to learn whether the religious factor During conceptualization, you “interrogate” or ask
was a more powerful, separate force from the nonreligious theoretically related questions about the data:
factors (e.g., political views, age, income) in predicting
Is this a case of idea Z? What is the sequence of events,
African American views regarding same-sex marriage. and could it be different? Why did this happen here and
In a study with quantitative data, you move from the not somewhere else?
abstract concept toward a concrete measure. Begin by con-
ceptualizing, i.e., create a clear conceptual definition for As in quantitative research, you conceptualize by
each variable. Next, operationalize by giving each an opera- developing clear, explicit definitions. The definitions are
tional definition. Next, gather empirical data following the more abstract than your direct observations but are tied to
measurement operations described in the operational defi- specific data. In effect, you anchor the conceptual defini-
nition. Lastly, test the hypothesis empirically with the data. tions in the specific words, events, or actions that are the
The empirical tests are logically connected back to the con- data. Because qualitative measurement is integrated with
ceptual level. In this way, empirical tests provide evidence to other parts of a study and not a separate step, it may be
support or refute a conceptual hypothesis (see Figure 5.1). more difficult to carry out.
Although the above discussion emphasized quantita-
5.3.4: Qualitative Conceptualization tive–qualitative differences, in practice, neither type of
research adheres to a rigid process. For both, ideas and
and Operationalization data mutually influence one another. For both, we draw on
In research using quantitative data, you give abstract ideas ideas from beyond a specific research setting and blend
a conceptual definition early in the research process. By past techniques and concepts with new ones that emerge
contrast, in qualitative data research you may begin by during data collection. In this way, ideas and evidence
using partly developed, rudimentary working ideas dur- become mutually interdependent. Remember, all opera-
ing the data collection process. As you gather and analyze tionalization is a process of connecting ideas with data.
qualitative data (i.e., field notes, photos and maps, histori-
cal documents, etc.), you rethink and refine your initial
ideas and develop new ones based on your data. As you
try to “make sense” of the data, you adjust or create clearer Example Study
definitions of the ideas. It is an iterative, back-and-forth,
process of gathering data, refining ideas, gathering more Operationalizing Social Ties
data, then again refining ideas. Eventually, you connect Matthew Desmond (2012a, 2012b) studied evictions in
the ideas together into theoretical relationships. Often this urban neighborhoods. Desmond noted that past studies
is grounded theory. The process requires you to be showed how the poor often rely on kinship networks to
Measuring Social Life 99
Read More
Summary Review
In the Green Street Mobile Home Park and in a north side
impoverished neighborhood in Milwaukee, Desmond estab-
lished relationships with several families—11 of these families Table 5.1 Steps in Quantitative and Qualitative
went through evictions. Past ethnographic studies of low- Conceptualization and Operationalization
income neighborhoods showed that poor people relied on kin
Quantitative Qualitative
and nonkin social networks. Because many poor people are
1. Conceptualize variables by 1. G
ather empirical data and
loners with few kin or have kin without resources, Desmond
developing a clear, complete simultaneously think about
wondered how low-income people used such networks in a written conceptual definition concepts to organize and
time of crisis, such as eviction. for the core idea of each. make sense of the data.
Build on past theories, Develop clear definitions for
Desmond operationalized major concepts of his study by
consider the definitions others each concept by drawing on
describing specific residents and their experiences. He have used, and be clear past readings, newly created
described what he saw and offered quotes of what he heard, and logical. ideas, or from the ideas
used by the people you are
grounding his concepts in specific events and individuals. In
studying.
the trailer park, he introduced us to several people.
2. Operationalize variables by 2. As you gather data, be very
They include Teddy, a 52-year-old half-paralyzed man creating specific activities to aware of processes you use
who received disability payments; Pam and Ned, 32 and 42, measure each. This opera- to make sense of the data
respectfully, a white couple who raised five children working tional definition closely and interpret details of the
matches how you have data. Reflect on and describe
odd jobs for cash; and Tina, a 40-year-old white single mother defined the variable in its this process of linking ideas
of three with a seasonal phone answering job and who relies conceptual definition. to specific observations in
on unemployment for half the year. In the inner city black the data.
neighborhood, we learned of Arleen, a 38-year-old black 3. Gather empirical data using 3. Review and refine your defini-
the specific measurement tions and the descriptions of
women and mother of six who receives welfare. We also activities of your operational how you gathered data and
learned about Lamar, a 48-year-old wheelchair-bound black definition; this links data to made sense of it.
man and single father of two who receives welfare, and Ches- the conceptual definition.
ter and Myesha, a black couple, both 33, who support two
100 Chapter 5
measures pick up less detailed information. If you do not mea- a ccurate estimates of the population parameter than a
sure specific information, things other than the concept of inter- sample with a large sampling error. In a parallel manner,
est might slip in. The general principle is to try to measure at the for the measurement process you want specific empirical
most precise level possible. measures that represent abstract concepts that you cannot
3. Use multiple indicators. You can improve reliability by using directly observe. If you have a valid measure, it deviates
multiple indicators because two (or more) indicators of the same very little from the concept it represents.
concept are better than one. Using multiple measures is a widely
accepted principle of good measurement. For example, I create
three indicators of employee satisfaction. My first indicator is an
5.4.1: Relationship of Reliability
attitude question on a survey. I ask research participants their and Validity
beliefs and feelings about different job areas. For a second indi-
We want both reliability and validity in measures. Reliabil-
cator, I observe research participants at work. I note whether
ity is necessary for validity and is easier to achieve than
they smile, appear to be happy, maintain good relationships with
coworkers and customers, or appear stressed, complain fre-
validity. Although reliability is necessary for a valid mea-
quently, and act gruff. Lastly, I examine employment records for sure, it does not guarantee that a measure will be valid. It is
absenteeism, disciplinary actions, length of service, and turn- not a sufficient condition for validity. A measure can pro-
over rates. By using the three separate indicators—survey, duce the same result over and over, but what it measures
observation, and written records—I learn about satisfaction. If all may not match the definition of the construct (i.e., validity).
three show consistent satisfaction (high or low), I gain confi- A reliable measure can be invalid.
dence that I have a dependable measure. Multiple indicators let Here is a simple example: You step onto a scale to get
you measure different aspects of the concept (such as employee weighed. The weight registered by the scale is the same each
satisfaction with pay, with workplace conditions, with supervi- time you get on and off. The scale is a reliable measure. You
sion) each with its own indicator. In addition, one indicator may
go to another scale—an “official” one that measures true
be imperfect or unstable, but several indicators are unlikely to
weight—and it says that your weight is much heavier. The
have the same (systematic) error.
first scale was reliable (i.e., dependable and consistent), but it
4. Use pilot studies and replication. Trying pilot versions of a
did not give a valid measure of weight. A diagram shows the
measure can improve its reliability. Develop one or more draft or
relationship between reliability and validity using the anal-
preliminary versions of a measure and try them out before using
the final version. Of course, this takes more time and effort. You
ogy of a target. The bull’s-eye represents a good fit between a
can also replicate the measures other researchers have used. If measure and the definition of the construct (see Figure 5.2).
you find measures from past research, you can build on and use
them, citing the source, of course.
WRITING PROMPT
Validity is more difficult to achieve than reliability. We Reliable Measures
never achieve absolute validity because it links invisible,
Select one aspect from your daily life that is measured and you
abstract ideas to specific empirical observations. A gap check occasionally. It can be part of physical or social world. How
always exists between our mental images about the world reliable is this measure? How would you know if it stopped being a
and the concrete reality we experience. Nonetheless, some reliable measure? How would using multiple indicators help you gain
confidence in the reliability of your measure?
measures are more valid than others are.
Measurement validity and sampling error are similar. The response entered here will appear in the
In sampling, you want minimal sampling error—i.e., you performance dashboard and can be viewed by
your instructor.
want a specific sample, for which you have data, to repre-
sent the population that you cannot directly observe. A
sample with a small sampling error provides more Submit
Content validity Content validity is a special contact, with members of other cultural groups provides
type of face validity. The conceptual definition of a direct information about the values, lifestyles, and experi-
sophisticated idea is a type of mental “space” that con- ences of members of those groups.” Information obtained in
tains areas with secondary or reinforcing ideas. Mea- this way is likely to be more favorable and accurate than infor-
sures that capture all the areas of the conceptual space mation gained through other, less direct sources. 2 Without
have high content validity. For example, your concept of direct contact, people often rely on socialization, peers, and
the media—all that tend to be sources of stereotypes. Such
discrimination has three aspects (e.g., interpersonal,
indirect information and images tend to be less accurate and
housing, employment). A measure that captures all three
less favorable toward outgroups that are minorities. Ellison,
aspects has content validity.
Shin, and Leal (2011) explored the contact hypothesis for
Criterion validity Criterion validity uses a stan- relations between non-Latinos and the Latino population of
dard or criterion to indicate a concept. An indicator ’s the United States. They conceptualized the independent vari-
validity depends on comparing it with another mea- able, contact, to include non-Latinos having personal contact
with a Latino in four areas—in high school, as co-workers, as
sure of the same construct in which you have strong
friends or acquaintances, or as relatives. Their dependent
confidence.
variable on views about Latinos had four dimensions: accep-
tance of negative stereotypes (e.g., lazy, unintelligent, lack a
strong family), degree of awareness and respect for Latino
Summary Review achievements, high social distance measured by the Borga-
dus Social Distance Scale, and views on immigration. Thus,
they conceptualized both the independent and dependent
variables as having content in multiple areas. They used one
Table 5.2 Measurement Validity Types
or more survey questions to ask about each content area of
Validity Type Features both variables. This enabled them to measure both the
Face There is widespread agreement among degree of contact and favorable–unfavorable views. They
informed people that the measure is valid also measured whether the general relationship held for all
Content Measure captures the entire meaning of all content areas in the independent and dependent variables.
parts of a concept’s definition
The researchers also controlled for several alternative expla-
Criterion Measure agrees with highly trusted other nations, such as living in an area with many Latinos. They
measures of the same concept
found having personal contact with Latinos, especially as a
friend or relative, had a positive effect on all areas of the
dependent variable, including immigration views. Living in an
area with many Latinos by itself did not increase favorable
views; having personal contact was essential. Another finding
Example Study that was not part of the main hypothesis was that among
non-Latinos, African Americans had more positive views
Content Validity and the toward Latinos than the white Anglos. Although the contact
hypothesis was developed in the United States to examine
Contact Hypothesis racial intergroup interactions, it has been tested across a
Scholars have examined “contact hypothesis” for 50 years,
and it is central to our understanding intergroup relations. The
hypothesis says, “contact, particularly close and sustained 2
Ellison, Shin, and Leal (2011 ) pp. 938–39
Measuring Social Life 103
wide range of settings and for many issues, from defusing the Validity links a concept to empirical measures. To be
Israeli–Arab conflict, to “straight” Dutch teens attitudes about valid, qualitative data measures should have authenticity.
gays and lesbians, to South Korean college students’ views The goal is less to match an abstract concept with a single,
regarding and willingness to interact with foreign students in fixed, standard version of reality, than to match concepts
their country.3
with the understandings of the people being studied in a
manner that “rings true” to their life experiences.
old a person is in years, months, days, hours, and minutes. arbitrary zeros, not true zeros. This can create confusion
Age can also be a set of discrete categories, such as infancy, and lead to mistakes. For example, a rise in temperature
childhood, adolescence, young adulthood, middle age, and from 30° to 60° is not a doubling of the temperature. The
old age. Education can be continuous, indicating years of zero in measuring temperature is arbitrary because 0° is
schooling, or a set of discrete categories by degree/diploma, not the absence of all heat. With an arbitrary zero, the
such as less than high school, high school diploma, more numbers on a thermometer might double, but the actual
than high school but less than four-year college degree, four- temperature does not double.
year college degree, and graduate degree.
Practical Reasons to Conceptualize and
While it is possible to collapse most continuous vari-
Measure Variables at Higher Levels of Mea-
ables into a few discrete categories, you cannot do the
surement You can always collapse a high level of
opposite—turn a discrete variable into a continuous mea-
measurement to a low level, but the reverse is not true. If
sure. For example, sex, religion, and marital status cannot
you measure a concept very precisely, you can decide
be conceptualized as continuous. Nevertheless, you can
later to “throw away” or ignore some of the precision. But
shift to related concepts that you conceptualize as continu-
you cannot measure a concept with little precision and
ous. Sex is discrete, but the “degree of femininity” and
then make it more precise later. You can turn a ratio-level
“degree of masculinity” are continuous variables. A spe-
measure (annual family income) into ordinal level (High,
cific religion (e.g., Catholic, Lutheran, or Baptist) is dis-
Medium, or Low income) or nominal level (same as or
crete, but the degree of commitment to religion is
different from others). However, the process does not
continuous. Marital status is discrete, but the number of
work in the opposite way. You cannot measure something
years a person has been married is continuous.
at the nominal level and then reorganize it to be ordinal,
The four levels of measurement, from lowest (discrete
interval, or ratio.
and least precise) to highest precision, are nominal, ordi-
nal, interval, and ratio. Each level provides different types
of information. Discrete variables are at the nominal and
ordinal levels, whereas you can measure continuous vari-
Summary Review
ables at the interval or ratio levels.
What are the levels of measurement? Table 5.3 Characteristics of the Four Levels
of Measurement
Compare Your Thoughts
Distance between
Nominal measures only indicate a difference among cate- Different Categories True
gories. Examples include gender: male or female; religion: Level Categories Ranked Measured Zero
Protestant, Catholic, Jew, Muslim, or other; racial heritage: Nominal Yes No No No
African, Asian, Caucasian, Latino, other; state/province or Ordinal Yes Yes No No
region of residence: Illinois, Ontario, New York, Texas, or Interval Yes Yes Yes No
Midwest, Northeast, and so on. Ordinal measures indicate Ratio Yes Yes Yes Yes
a difference among categories, and the categories can be
ordered or ranked. Examples include letter grades: A, B, C,
D, E; opinion measures: Strongly Agree, Agree, Disagree,
Strongly Disagree; quality ratings: Excellent, Very Good, When you use an ordinal measure, try to have at least
Good, Fair, Poor. Interval measures do everything the first five ordinal categories and obtain multiple observations
two do, plus they specify the distance between categories. for each. Interval measures can be confusing because
Examples include Fahrenheit or Celsius temperature: 5°, some measures, like temperature, use arbitrary zeros.
45°, 90°; and IQ scores: 95, 110, 125. Ratio measures do Arbitrary zeros in interval measures confuse many peo-
everything all the other levels do, plus they have a “true ple. The zeros are “arbitrary” or just there to keep score.
zero.” A true zero means a score of zero indicates a value They do not indicate a real zero. The temperature can be
of zero or nothing. Having a true zero makes it possible to zero, or below zero, but zero is an arbitrary number. Com-
talk about relationships in terms of proportion or ratios, pare 0°C with 0°F—they are different temperatures.
such as twice as much. Examples include money income: Because the zeros are arbitrary, doubling the degrees in
$10, $100, $500; years of formal schooling: 1 year, 10 one system does not double the degrees in the other. It
years, 13 years; age: 18 years, 32 years, 64 years. In most makes no sense to say that it is “twice as warm” if the
situations, the distinction between interval and ratio levels temperature rises from 2° to 4°, from 15° to 30°, or from
makes little difference. Some interval-level variables have 40° to 80° (see Figure 5.3).
Measuring Social Life 105
Protestant Catholic Muslim Jewish
Religious Faith
The categories differ, but we do not assume that one is “better/high” or “worse/lower” than others.
Ordinal Level
Very Happy Happy Neither Sad Very Sad
Degree of Happiness
Interval Level
60 70 80 90 100 110 120 130 140
IQ Score
We can precisely determine the size of differences among score, but there is no absolute zero.
Ratio Level
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76 78
Age
We can precisely determine the size of differences among scores, plus a true or absolute zero exists.
List four examples of nominal-level social variables not mentioned in and Use Indexes
and Scales?
the discussion. Explain why each of your nominal-level example vari-
ables cannot be ranked as better/worse higher/lower.
Now list four examples of ordinal-level social variables. Identify
five categories from high/best to low/worse for each of them, where 5.7 Apply the processes involved in the construction
the categories are about equal distance from one another. and use of indexes and scales
The response entered here will appear in the Scales and indexes are information-filled quantitative data
performance dashboard and can be viewed by
your instructor.
measures. Researchers have created hundreds of indexes
and scales to measure the prestige of occupations, the
adjustment of people to a marriage, the intensity of group
Submit
interaction, the level of social activity in a community, the
106 Chapter 5
degree to which a state’s sexual assault laws reflect feminist How much schooling did you complete?
values, the level of socioeconomic development of a nation, _____ 0–6 years _____ 6–10 years _____ 11–12 years
and much more. You can borrow already created scales/ _____ 12–15 years _____ 16–18 years _____ 18 or more years
indexes or create your own. In this section, we look at the
Exhaustive means all possibilities are included in the
principles of scale and index construction and explore a few
measure of a variable. If you measure religion and ask
major types. Before we begin, keep two things in mind:
whether a person is Catholic, Protestant, or Jewish, it is not
1. We can develop a measure of every social phenome- exclusive. A Buddhist, a Muslim, or an agnostic does not fit
non. We can measure some concepts directly and pro- anywhere. You must have variable categories to include
duce precise numerical values (e.g., family income), every possible situation. For example, Catholic, Protestant,
whereas for other concepts we use proxies to measure Jewish, or other is both exclusive and mutually exclusive.
a variable indirectly and it may not be as precise (e.g., The following question about years of schooling is mutu-
predisposition to commit a crime). ally exclusive but it is not exhaustive:
2. You can learn a great deal by looking at measures cre-
ated by other researchers. You do not have to start How much schooling did you complete?
from scratch but can use a previously used scale or _____ 6–10 years _____ 11–12 years _____ 13–15 years
index, or you can modify it. _____ 16–18 years _____ 19 years
Social researchers do not use a consistent nomencla- People with less than 6 years or more than 19 years of
ture to distinguish between index and scale. One person’s schooling are not included.
scale can be another’s index. Both produce ordinal- or In addition to being mutually exclusive and exhaus-
interval-level measures. Also, you can combine scale and tive, scales and indexes should be unidimensional or have
index techniques to create a single measure. Compared to one dimension. They all measure a single concept. Unidi-
basic single indicator measures, scales and indexes provide mensionality says that if you combine several specific
more information, increase measurement reliability and pieces of information into a single score or measure, all the
validity, and organize and condense complex information. parts should act alike and measure the same core concept.
For most purposes, you can treat scales and indexes as Many indexes combine subparts of a concept into a sin-
interchangeable, although they have distinct features. gle measure. This appears to contradict the principle of uni-
A scale creates an ordinal measure of intensity, direc- dimensionality. It does not contradict the principle because
tion, level, or potency by arranging responses or observa- you can define concepts at different levels of abstraction.
tions along a continuum. It can be a single indicator or You can define a general, abstract concept (e.g., happiness)
multiple indicators. An index combines information from as having subparts (happiness with health, with job, with
multiple separate indicators into a one score, often a sum marriage). Each subpart is an aspect of the concept’s con-
of their values. Most indexes add the numerical values of tent. One subpart, marital happiness, is at a lower level of
several items to yield a score at the interval level of mea- abstraction; yet, it too may have subparts (e.g., happy with
surement. You can combine several indicators measured communication in a marital relationship, happy with sexual
with scales into a single composite index measure. intimacy, happy with sharing household tasks, and so forth).
A measure can indicate a unidimensional construct in
one situation but measure one part of a different, more
5.7.1: Mutually Exclusive, abstract concept in another situation. General happiness is
Exhaustive, and Unidimensional more abstract than marital happiness, which is more
Two features of all good measurement are that variable cate- abstract than happiness with communication in a m arriage.
gories are mutually exclusive and exhaustive. For example,
a variable measuring type of religion—with the attributes
Christian, non-Christian, and Jewish—is not mutually exclu- 5.7.2: Index Construction
sive. Judaism is both a non-Christian religion and a Jewish News reports in the United States regularly mention the
religion. A Jewish person fits into both the non-Christian and Federal Bureau of Investigation (FBI) crime index, the con-
the Jewish category. Likewise, type of city as river port city, sumer price index (CPI), the index of leading economic
state capital, and interstate exit lacks mutually exclusive indicators, and the consumer confidence index. The FBI
attributes. One city could be all three (a river port state capi- index is the sum of police reports on seven so-called index
tal with an interstate exit), any one of the three, or none of the crimes (criminal homicide, aggravated assault, forcible
three. For numerical data, you do not want any overlap. The rape, robbery, burglary, larceny of $50 or more, and auto
following question about years of school is not mutually theft). It began with the Uniform Crime Report in 1930. The
exclusive (it is exhaustive) Persons with exactly 6, 12 or 18 CPI, which is a measure of inflation, is created by totaling
years of schooling can fit into two categories.: the cost of buying a list of goods and services (e.g., food,
Measuring Social Life 107
rent, and utilities) and comparing the total to the cost of B = the total black population of the geographic entity
buying the same list in the previous year. The U.S. Bureau (e.g., the entire city)
of Labor Statistics has used a CPI since 1919; wage increases, wi = the white population of the ith area
union contracts, and social security payments are based on W = the total white population of the large geographic
it. The Conference Board, a non-governmental organiza- entity
tion, produces the index of leading economic indicators The index of dissimilarity also shows us historical patterns of
(LEI) and the Consumer Confidence Index (CCI). The LEI segregation. For example, the black–white D index for Saint
tries to predict economic conditions in the near future by Louis was 39.1 in 1900, it rose to 54.3 by 1910, and to 92.6 in
adding scores on 11 items. The 11 items include average 1940, dropped to 89.3 in 1970 and to 74.3 by 2000 but rose
hours in employee work weeks, new unemployment to 78.0 in 2010. Researchers modified and extended the index
claims, new orders for consumer goods, new equipment to other topics (e.g., gender segregation in occupations).
orders, building permits, stock prices (S&P 500), and so Table 5.4 illustrates the D indexes for racial relations in 10 U.S.
forth. The consumer confidence index is based on a monthly cities based on the 2010 census.
survey of a random sample of 5,000 households that has
been conducted since 1985. It contains questions about the Table 5.4 Index of Dissimilarity for 14 U.S. Cities, 2010
current situation and about future expectations. It is the
The table shows the index of dissimilarity scores. Black–white
sum of respondents’ appraisal of current business condi- segregation was highest in Milwaukee and lowest in Seattle.
tions, current employment conditions, and expectations six Asian–white segregation was lower than black–white
months hence regarding respondents’ employment condi- segregation in all cities, and white–Hispanic segregation was
highest in New York City.
tions, total family income, and business conditions. Respon-
dent answers are scored positive, negative, or neutral. City/Metro White vs. White vs. White vs.
Area Black Asian Hispanic
To create an index, you combine two or more items into
Atlanta 59.0 48.5 49.5
a single numerical score. Indexes measure the most desir-
Boston 64.0 45.4 59.6
able place to live (based on unemployment, commuting
time, crime rate, recreation opportunities, weather, and so Chicago 76.4 44.9 56.3
on), the degree of crime (based on combining the occurrence Dallas 56.6 46.6 50.3
of different specific crimes), and a person’s mental health Denver 62.6 33.4 48.8
(based on the person’s adjustment in various areas of life). Detroit 75.3 50.6 52.5
Also called a summative or composite index, you add sev- Houston 61.4 50.4 52.5
eral specific numerical measures that represent parts of one Milwaukee 81.5 40.7 57.0
concept. Some indexes involve more than simply adding Minneapolis 52.9 42.8 42.5
items and are adjusted to give measure with a 0 to 1 score. New York City 78.0 42.3 62.0
Philadelphia 68.4 42.3 55.1
San Francisco 62.0 46.6 49.6
5.7.3: Two Index Construction Issues their feelings or rating by checking a point on a line that runs
from one extreme to another. This conveys the idea of a con-
1. Count Items Equally or Weigh Them? If you find an index,
tinuum. Numbers on a line can help people think about quan-
unless it is otherwise stated, assume an unweighted
tities. When using a scale, you assume that people with the
index, i.e., each item has equal weight. Unless you have a
same subjective feeling mark the visual line at the same place.
very good reason to do otherwise, add up the items with-
Figure 5.4 is an example of a “feeling thermometer” scale.
out modification, as if each were multiplied by 1 (or –1 for
items that are negative). In a weighted index, you value
or weight the items differently. The size of the weights Figure 5.4 Feeling Thermometer Graphic Rating Scale
depends upon your assumptions, conceptual definition, A graphic that is used to measure feelings
or specialized statistical techniques. Weighting can pro- 100 Very Warm
duce different scores than an unweighted index. For
90
example, in a weighted index of a desirable place to live,
the percentage of days with sunshine might be weighted 80
one-half the importance of a low crime rate or quality of 70
public schools but the same as amount of park land or
60
number of museums.
2. Missing Data. Missing data can become an issue when 50 Neither Warm nor Cold
constructing an index, threatening validity, and reli- 40
ability. For example, you construct an index of the
degree of societal development for 50 nations using 30
choices creates a crude measure and does not capture com- The index offers a score for each person, ranging from 0
plex distinctions. Usually you want to use four to six (strongly disagree to all) to 36 (strongly agree to all nine).
answer categories. If you use six categories, you can reduce
Making It Practical: Likert Scale and Mea-
to four or two categories after you collect the data. But, if
suring Self-Esteem You have heard about self-
you collect data with just two categories, you cannot make
esteem and may have also heard that certain social
your data more precise later. Keep the number of answer
problems or personal issues stem from a lack of self-esteem.
choices to under nine. More distinctions than that are
Morris Rosenberg operationalized the idea into a widely
rarely meaningful, and people may become confused.
used measure in 1965 (Rosenberg, 1965). In the measure,
Always balance the answer choices (e.g., “strongly agree,”
people read a list of 10 statements dealing with general
“agree” with “strongly disagree,” “disagree”).
feelings and then answer using a Likert Scale from strongly
Should you use a “don’t know,” “undecided,”
agree to strongly disagree (see Table 5.5). Take this oppor-
“unsure,” “no opinion” category in addition to the direc-
tunity to measure your own self-esteem with it.
tional categories (e.g., “disagree,” “agree”) in a Likert
scale? Researchers are divided on this issue, but in most
situations include the “don’t know” category. It is better to Table 5.5 Rosenberg Self-Esteem Scale
have people who are uncertain or without an opinion indi- Strongly Strongly
cate that than to force them to guess. Agree Agree Disagree Disagree
Statements (SA) (A) (D) (SD)
Another issue arises when you have a long set of Lik-
1. On the whole, I am ______ _____ ______ ______
ert scale questions on the same issue. If answer choices are
satisfied with myself.
set up such that answering one way (such as “strongly
2. *At times, I think I am ______ _____ ______ ______
agree”) always indicates the same position on the issue, it no good at all.
can create problems. For example, you ask nine questions 3. I feel that I have a ______ _____ ______ ______
about the issue of a woman’s right to a legal abortion. If number of good
qualities.
you word all nine such that the response “strongly oppose”
4. I am able to do things ______ _____ ______ ______
means strong opposition to abortion, you might create a as well as most other
“response set” problem. Some people stop paying close people.
attention after several similar questions or have a tendency 5. *I feel I do not have ______ _____ ______ ______
much to be proud of.
to agree or disagree to similar Likert scale answer choices.
6. *I certainly feel use- ______ _____ ______ ______
The solution is to reverse the wording of some questions. less at times.
For example, you could phrase three questions so that
7. I feel that I’m a per- ______ _____ ______ ______
“strongly oppose” indicates opposition to legal abortion, son of worth, at least
and six so that “strongly oppose” indicates support for on an equal plane
with others.
legal abortion, then mix the questions. For example, after
8. *I wish I could have ______ _____ ______ ______
the question “Do you support a woman’s choice for legal more respect for
abortion if she is pregnant due to rape?” you ask, “Do you myself.
oppose laws that restrict a woman’s access to legal abor- 9. *All in all, I am inclined ______ _____ ______ ______
to feel that I am a
tion clinics?” A person with a strong antiabortion view failure.
must switch from repeating “strongly disagree” to say 10. I take a positive atti- ______ _____ ______ ______
“strongly agree” to be consistent. tude toward myself.
You can combine several Likert-scaled items into an Scoring: SA = 3, A = 2, D = 1, SD = 0. Items marked with an asterisk (*) are reverse
scored: SA = 0, A = 1, D = 2, SD = 3. Sum the scores for the 10 items. The higher your
index if all measure a single concept. Consider the Self- score, the higher your self-esteem.
WRITING PROMPT scale, strongly predicted whether the person favored putting
limits on all legal immigration and specifically Mexican immigra-
Likert Scale tion. Their study confirmed findings from other studies that
The Likert Scale is one of the most widely used formats for measur- white Anglo respondents applied their feelings regarding race-
ing attitudes. Create a Likert scale for measuring an attitude of inter- ethnicity when they took positions on immigration policies.
est to you. Give it five answer categories and number scoring.
1. The first cluster was older and evenly divided between 5 YES NO YES YES
males and females. They drank different beverages but 6 YES NO YES NO
tended toward wine, and they drank less than the other 7 YES NO NO YES
two clusters. Getting drunk was not a motivation to drink 8 YES NO NO NO
alcohol in this cluster. 9 NO YES YES YES
2. The second cluster was much younger and predominantly 10 NO YES YES NO
male. They mostly drank beer or spirits, and a majority said 11 NO YES NO YES
that getting drunk was their prime motivation for drinking.
12 NO YES NO NO
3. The third cluster was mixed male and female. They drank 13 NO NO YES YES
more than cluster 1 but less than cluster 2. When indoors
14 NO NO YES NO
they often drank wine, and when outdoors they drank
15 NO NO NO YES
either beer or wine. Getting drunk was a motivation for
16 NO NO NO NO
some in this group but not for the majority.
112 Chapter 5
At one extreme (sequence 1) is the child who knows all the uestions. A Guttman scale allows you to measure how well
q
four items. At the opposite extreme (sequence 16) is the child the data fit a hierarchical pattern by seeing how many people
who knows none of the four. With a Guttman scale, you answer in the predicted pattern and how many answer in
hypothesize combinations that will be frequent and fit a other ways. Various statistical measures can tell you how
structured pattern for many participants. Let us say you think “scalable” the data are ranging from 0 to 100 percent (i.e.,
the pattern is the following order: age, phone number, how well they fit the hierarchical pattern among items that
teacher marital status, and mayor’s name. Most children you hypothesized). A score of 100 percent indicates that
know their age even if nothing else, but few who know the everyone’s answer fits the hierarchical or scaled pattern and
mayor’s name do not also know the lower or “easier” 0 occurs with a random pattern, or an absence of hierarchy.
Example Study
Guttman Scaling and Neighborhood Preference
Xie and Zhou (2012) hypothesized that whites’ preference for hypothetical neighborhoods. Each of the neighborhoods had
housing in a racially mixed neighborhood with African Americans, varying numbers of blacks: 0, 1, 3, 5, and 8 of 14 immediate
based on the mix of percentage of African Americans versus neighbors, representing a mix of 0, 7, 21, 36, or 57 percent
whites, might fit the Guttman scaling pattern. They examined blacks. The percentage of white respondents willing to move
survey data on approximately 8,900 adults from the 1990s taken into each type of neighborhood indicated the whites’ neighbor-
from respondents in Detroit, Atlanta, Los Angeles, and Boston hood preferences.
metropolitan areas. The survey showed five images of 14 The authors hypothesized that if responses fit a Guttman
houses, i.e., five neighborhoods, with the houses filled-in white scale, any respondent willing to move to a neighborhood with a
or black to represent the race of the family in the house. It asked higher level of blacks’ presence would also be willing to move to
white respondents to express their willingness to live in the five a neighborhood with a lower level. This rank-order divides the
respondents into six categories. Five conform to a Guttman hier-
archical scale of varying tolerance of black neighbors. The last
category (#6) is for those who did not fit the Guttman scale
Table 5.7 Whites’ Preference for Different Racial requirement. We can see this in Table 5.8. For example, category
Mixes in a Neighborhood 1 are whites who cannot tolerate a single black out of 14 neigh-
Percentage of White Respondents bors. Category 2 consists of whites who can tolerate only one
Willing to Move into the Neighborhood black neighbor but not two or more black neighbors; and so
Percentage
of Blacks in a by City* forth. The next set of four columns to the right show percentages
Neighborhood Detroit Atlanta Los Angeles Boston of whites in the categories for the four metropolitan areas. The
0% 96% 96% 95% 94% last row (category 6) shows that the data fit the Guttman scale
7% 87% 88% 94% 92% very well. There are fewer whites in category 6, about 5 percent
than in any other. It appears that whites in Detroit and Atlanta are
21% 70% 74% 89% 85%
less tolerant of black neighbors than those in Los Angles and
36% 43% 50% 73% 62%
Boston. We see this clearly from the percentage of category 1. It
57% 29% 32% 59% 46%
is much higher for Detroit and Atlanta than for Los Angeles or
* rounded to nearest whole percent
Boston.
Table 5.8 Guttman Scale Pattern and Data for Whites’ Residential Preferences
Percentage of Whites with Category
GUTTMAN PATTERN Percentage of Blacks in Neighborhood Response by City
Category 0% 7% 21% 36% 57% Detroit Atlanta Los Angeles Boston
1 Y N N N N 10.5% 9.3% 3.1% 4.9%
2 Y Y N N N 18.1% 15.0% 6.8% 8.0%
3 Y Y Y N N 26.7% 22.9% 15.2% 22.2%
4 Y Y Y Y N 13.9% 18.2% 14.2% 14.4%
5 Y Y Y Y Y 26.6% 30.0% 55.9% 44.3%
6 Not Guttman Scalable 4.3% 4.4% 4.8% 6.2%
Measuring Social Life 113
5. The Guttman Scale is a hierarchically structured set of Describe a news item in which a measure of an aspect of the
responses. If research participants respond to a set of social world was contested. What was the measure, and why was
questions or relations in a manner that indicates a it criticized?
sequence of levels, one within the other, it fits the response Examine two of the new items presented by fellow class-
pattern of the Guttman Scale. mates. Do you agree with the criticism of the measure, and if so
can you suggest a better measure?
Learning Objectives
6.1 Identify the ubiquitousness of social surveys 6.5 List the advantages and disadvantages
of various survey formats
6.2 List the three stages of conducting a
research survey 6.6 Apply best practices for preparing
the interviewer to administer social
6.3 Evaluate strategies for developing effective
surveys
survey questions
6.7 Analyze some of the ethical issues in
6.4 Analyze some of the features of an effective
conducting survey research
questionnaire
Seventeen nations (including Argentina, Belgium, Brazil, then, many states passed laws banning same-sex mar-
Canada, Denmark, France, the Netherlands, New Zealand, riage. In November 2003, the issue exploded after the
Norway, South Africa, Spain, Sweden, and the United Massachusetts Supreme Court ruled in Goodridge v.
Kingdom) recognize marriages between people of the Department of Public Health that it was unconstitutional to
same sex. In the United States, this has been a heated, ban same-sex marriages in that state. In the 2004
divisive political issue, which surfaced in 1996 when the P residential election, social-religious conservatives
U.S. Congress passed the Defense of Marriage Act. Since pushed same-sex marriage to the forefront, displacing the
116
The Survey 117
a bortion issue. Since the first legalization in Massachu- that people are willing and able to answer without great
setts in 2003, the courts in 35 of the 50 states have difficulty. Examples of these types of questions include:
approved or overturned bans on same-sex marriage. By how much schooling a person completed or whether a
2015 only 13 states either still had bans in place or had person favors or opposes same-sex marriage. To learn
mixed court rulings. Finally, in July 2015, in a split deci- about things people are unaware of or are unwilling to
sion, the U.S. Supreme Court ruled that states cannot ban self-report on, such as illegal behaviors, require special
same sex marriage. adjustments to surveys or may not be possible using the
Beyond the court battles, advocacy campaigns, and survey technique.
political rhetoric, you may ask: What do ordinary Ameri- In a social survey, respondents hear and give an
cans think about the issue? Dozens of polls and surveys tell answer to the exact same questions. The questions may be
us that 40 to 60 percent of the public oppose legal same- about past behaviors, experiences, opinions, and character-
sex marriage, 30 to 53 percent favor it, with 1 to 15 percent istics. You can use one survey to measure many variables
uncertain. The most ardent opponents tend to be men, and test multiple hypotheses at the same time. You can test
older people, those who say religion is central in their hypotheses by looking for patterns in the data (answers
lives, people with less schooling, those with strong anti- given). For example, you might hypothesize that a per-
abortion views, and people who want to deny atheists son’s views on other issues (e.g., gun control, immigration,
legal rights. Results vary somewhat by when and how a spanking a child) are associated with how a person feels
survey asked the question. Did it ask about same-sex about same-sex marriage. In short, one attitude predicts
marriage, gay marriage, homosexual marriage, or the the other. Statistical techniques allow you to see whether
marriage of gay men and lesbians? Did it ask about per- an association exists between other views and feelings
mitting or banning such marriages? Did it place the issue about same-sex marriage.
at the federal level as a constitutional amendment or as a
state law, or did it ask about a personal moral position?
Did it ask about civil unions, specific legal rights for gay
people, or only about marriage? With this issue, as with
6.1.1: How Does an Opinion Poll
others, if you want to know what people think and why, Differ from a Social Survey?
you must first understand how the survey research The difference is minor. An opinion poll is a type of
method operates. survey; it is a short survey about opinions on current
issues. Most polls look at a sample over a short time
period, such as one week. Many polling organizations
(Roper, Gallup), media organizations (CNN, ABC
6.1: What Is a Social News/Washington Post, New York Times/CBS, Fox News),
political organizations, and research centers (Pew
Survey? Research Center, National Opinion Research Center) con-
duct polls on current issues, such as same-sex marriage.
6.1 Identify the ubiquitousness of social surveys
Most reputable polling organizations use techniques simi-
Nearly everyone has completed a survey or read about lar to those used by professional academic survey
survey results. The social survey may be too familiar researchers.
and popular. Many people say “do a survey” to get In addition to polls, there are other types of surveys.
information when they should ask, “What is the most Some use samples, others do not; some survey the public,
appropriate research technique for learning about this and others focus on a specific group. Beyond asking about
issue?” A survey may be the method appropriate opinions on current issues, a survey may ask about
depending on what you want to find out. Also, because knowledge, social background characteristics, general
surveys are ubiquitous and asking questions is easy, beliefs, or behaviors. For example, a business may con-
people find it easy to create a survey. Unfortunately, it is duct a survey to learn about employee job satisfaction or
even easier to create a survey that produces misleading customer product preferences, and a medical clinic may
or worthless results. A good survey that yields accurate conduct a survey to document patient health behaviors.
data requires serious thought and effort. In this chapter, Polls rarely have more than a dozen questions, but sur-
you will learn what makes a quality social survey, limi- veys can have over a hundred questions. Survey research-
tations of the survey method, and how to conduct your ers measure many variables simultaneously and analyze
own survey. survey data to test hypotheses, explore relationships
Survey data come from self-reports. It requires you to among variables, and document people’s thoughts and
phrase the research question and variables as questions actions.
118 Chapter 6
Example Study
Views on Same-Sex Marriage
Between August 11 and 17, 2009, Princeton Survey Research same-sex marriage. Over time, the public moved to be more
Associates International conducted 2,010 t elephone supportive of same-sex marriages. In 2001, only about one-third
interviews for the Pew Forum on Religion & Public Life (see of American adults supported same-sex marriage (35 percent),
Figure 6.1). while 57 percent opposed it. By 2014, a majority (54 percent)
A total of 1,510 respondents over the age of 18 were inter- supported it (see Figure 6.2).
viewed on a landline telephone, and 500 were interviewed on a
cell phone. The sampling used a random digit dialing (RDD)
sample of landline and cell phone numbers in all 50 U.S. states Figure 6.2 U.S. Opinion Trends Regarding Same-Sex
and the District of Columbia. Telephone interviewers asked the Marriage, 1996–2014
following two questions (in random order) with Likert scale
Source: PEW Research Center
answer choices:
80
Favor Same-Sex Marriage
• Do you strongly favor, favor, oppose, or strongly oppose Oppose Same-Sex Marriage
70
allowing gay and lesbian couples to marry legally?
60
• Do you strongly favor, favor, oppose, or strongly oppose
allowing gay and lesbian couples to enter into legal agree- 50
Percent
ments with each other that would give them many of the 40
same rights as married couples? 30
Among respondents, a majority (57 percent) favored allow- 20
ing gay and lesbian couples to enter into legal agreements (civil
10
unions) with each other that give them many of the same rights as
0
married couples. In the survey, opponents of same-sex marriage
96
01
03
04
06
08
09
10
11
12
13
14
15
16
05
07
outnumbered supporters, 53 percent opposed allowing gays and
19
20
20
20
20
20
20
20
20
20
20
20
20
20
20
20
lesbians to marry legally, compared with 39 percent who support Year
(SOURCE: “Political Survey” Pew Research Center For The People & The Press Final Topline, © May 2013 Pew Research Center.
The survey question asked by the National Opinion Research Center in the General Social Survey.
Do you agree or disagree? Homosexual couples should have the right to marry one another.
Values Categories
1 STRONGLY AGREE
2 AGREE
3 NEITHER AGREE NOR DISAGREE
4 DISAGREE
5 STRONGLY DISAGREE
0 NAP
8 CANT CHOOSE
9 NA
6.1.2: Survey Data and Cause-Effect must rule out the alternative of age as a cause of health
problems before you can say that widowhood causes
Explanations more health problems. You will want to ask about age in
A cause-effect explanation based on survey data differs addition to marital status and health. As you plan a sur-
somewhat from other research techniques, such as the vey, you need to measure variables from the main hypoth-
experiment. Recall that to say one variable causes another, esis (dependent and independent variables) and variables
three conditions must be met: that represent potential alternative explanations (control
variables).
1. The independent variable must come before the depen-
dent in time
Making It Practical: Survey Research and
2. The two variables must be associated, or correlated
Control Variables Control variables measure
with one another
variables from alternative explanations that compete
3. There is no alternative cause for the relationship, or
with the primary hypothesis you wish to test. Let us say
there is no spuriousness
you think that gender influences differences in opinion
Survey data are sometimes called correlational, about same-sex marriage. An alternative explanation
because they best satisfy the second condition of causal- might be that race, or how deeply religious a person
ity. Meeting the first condition, time order, can be compli- feels, causes the opinion. Recall the study by Sherkat, de
cated with survey data because most survey data are Vries, and Creek (2010) on African Americans, religion,
collected at a single time point. Without data at multiple and views on same-sex marriage. You learned that reli-
time points, we rely on logic to show that information gious views and attendance explained greater African
from one survey question (e.g., father’s occupation while American opposition to same-sex marriage. In that study,
growing up) occurred earlier than that from another (the the authors looked at many independent variables
person’s current income). Meeting the third condition, no besides religious attendance and view, such as educa-
alternative cause, requires thinking of variables that could tion, age, gender, political belief, that influence views on
be possible alternative causes and measuring them in the same-sex marriage. The authors found that religion had
survey. These are control variables, because we can sta- the largest impact on African American views. They
tistically control for, or take into account, their effects looked at all variables and used statistical techniques to
using statistics. For example, you find that widowed peo- see which one had the greatest predictive power. Reli-
ple have more health problems than married people do. gious attendance and views were the most powerful fac-
Before you say that being widowed itself is the cause of tors for African Americans, but the variables were not as
poor health, you must consider alternative causes that important for whites for whom political beliefs were the
might make the relationship spurious. If most widowed dominant factor predicting views on same-sex marriage
people are older than most married people are, then you (see Table 6.1).
Table 6.1 View on Homosexual Marriage Based on Gender, Age, and Religiosity
Gender of Respondent
View on Gay Marriage Male Female Total
Agree 24.0% 34.1% 29.6%
Neither 13.3% 15.6% 14.6%
Disagree 62.7% 50.2% 55.7% Gender gap 5 12.5%
Total (N) 525 659 1184
100.0% 100.0% 100.0%
Age of Respondent
Submit
Step 2:
• Plan how to record data
• Pilot test survey instrument
6.2: How Do We Conduct
a Survey? Step 3:
6.2 List the three stages of conducting a research survey • Decide on target population
• Get sampling frame
Once you decide that the survey is an appropriate method • Decide on sample size
for gathering data to test a hypothesis, you proceed • Select sample
through three stages start-up, implementation and data
analysis. We can subdivide the stages into six steps (see
Figure 6.3). Step 4:
• Locate respondents
6.2.1: Start-Up Stage • Conduct interviews
• Carefully record data
In this stage, you address the following three questions:
Who will be the respondents of your survey?
What information do you want to learn from them?
Step 5:
How can you effectively get that information? • Enter data into computers
The type of respondent influences the topics you ask about • Recheck all data
• Perform statistical analysis on data
and question wording. At the beginning, you need to think
about the respondents. Topics relevant in a survey of nurs-
ing home residents may not be relevant to a survey of college
students. Survey questions about the working conditions of Step 6:
part-time workers at a fast food outlet may differ from ques- • Describe methods and findings
tions on the same topic in a survey of medical doctors. in research report
• Present findings to others for
Second, you must be very clear about exactly what you
critique and evaluation
want to learn from each question. Survey questions are the
The Survey 121
operationalization of variables and depend on how well and clarity. After all respondents have completed all ques-
you conceptualized them. Before gathering data, you tionnaires, you then organize the recorded data and pre-
should consider what the data might look like and how you pare them for statistical analysis.
intend to use the results. Newcomers to survey research can Large-scale survey research, such as with 2,000 respon-
be disappointed because they discover that the data do not dents located across a wide geographic area and asking 100
allow them to answer their research question when they questions, can be very complex and expensive. It requires
failed to think clearly about what the results of the survey coordinating many people and has dozens of steps. Such a
questionnaire. Too often they ask questions unrelated to large survey research project requires excellent organiza-
their true concern or fail to ask specific enough questions. tion and accurate recordkeeping to keep track of each
To prepare a survey, follow these steps: respondent, questionnaire, and interviewer. Similar proce-
First, create an instrument—a survey questionnaire or dures apply for a small-scale survey, such as distributing
interview schedule—to measure variables. Respondents questionnaires to 80 people in one location with 20 ques-
may read the questions themselves and mark answers on a tions, but they are more manageable.
questionnaire. Alternatively, you may prepare an inter- Whether you conduct a large- or small-scale survey,
view schedule—a set of survey questions designed so that assign an identification number to each respondent. Place
an interviewer can read them to a respondent. The inter- the number on each questionnaire and then check com-
view may be by phone or face-to-face. To simplify the dis- pleted questionnaires against a list of sampled respon-
cussion, I will use only the term questionnaire. dents. You should review the responses on each
Once you decide on who the respondents will be, questionnaire and transfer data from questionnaires to a
exactly what you want to measure, it is time to start writ- format for statistical analysis. Also, be sure to store the
ing questions. Expect to write and rewrite questions sev- original questionnaires (physical copies or electronic ones)
eral times for clarity and completeness. It is important to in a secure place. Meticulous bookkeeping and labeling are
organize carefully the flow of questions on a questionnaire. essential. Otherwise, you may find that valuable data and
Base the flow of survey questions on the research question, your efforts are lost through sloppiness.
respondents, and the survey format (discussed later in this
chapter), and think ahead to how you will record and orga-
nize data for statistical analysis.
6.2.3: Data Analysis Stage
After you have written the questions and before collect- In this stage, you have all the data so you are ready for the
ing the data, it is always best first to conduct a short pilot test analysis, interpretation, and reporting of survey data. This
or “dry run” of the survey questionnaire. Pilot tests can stage in a survey differs little from the types of analysis
increase question clarity. Use a small set of respondents who and reporting you would use for other sources of quantita-
are similar to those in the final survey. After they answer, tive data (e.g., an experiment, content analysis).
briefly interview them and ask the pilot respondents whether
the questions were clear, whether they interpreted the ques-
tions with the intended meaning, and whether the answer
choices offered were sufficient. Based on pilot test feedback,
6.3: Writing Good Survey
you may want to reword the questions or answer choices or Questions
decide to reorganize items in the questionnaire for clarity. In
6.3 Evaluate strategies for developing effective survey
this start-up stage, you also draw the sample of respondents.
questions
Excellent communication is essential to writing quality
6.2.2: Implementation Stage survey questions. Two core principles guide writing sur-
After selecting respondents, writing questions, and revising vey questions: avoid confusion, and keep the respondent’s per-
the questionnaire, you are ready to collect data. Many new- spective in mind. Good survey questions provide a valid
comers are surprised that planning and preparation require and reliable measure of variables. They also help respon-
much time. After you have located the sampled respondents dents feel that they understand exactly what you are ask-
in person, by telephone, over the Internet, or by mail, you ing in a question and that their answers are meaningful.
must provide each with information about the survey and When questions fail to mesh well with a respondent’s
instructions on how to complete it. Survey questions follow viewpoint or respondents find them confusing, the survey
a simple stimulus/response or question/answer pattern. questions will not produce high-quality data.
You will need to create a system to record all responses You face a dilemma in survey research. You want each
clearly and accurately immediately after respondents give respondent to hear the exact same question, because the
them. After a respondent finishes and you thank him or her, standard procedure is to measure each variable the same
you should also quickly review responses for completeness way across many people. On the other hand, if respondents
122 Chapter 6
have diverse backgrounds and use different frames of use neutral words. If you use words with a lot of emotional
r eference, the exact same wording may not carry the exact “
baggage,” respondents may react to the emotionally laden
same meaning to all respondents. However, if you tailor words rather than directly answer your question. Asking “What
question wording to each respondent, it will make com- do you think about a policy to pay murderous terrorists who
parisons very difficult, because you will not know whether threaten to steal the freedoms of peace-loving people?” is full of
the question wording or differences among the respon- emotional words—such as murderous, freedoms, steal, and
dents account for variation in answers. peace. It is not always easy to know what words have emotional
Writing good survey questions takes practice, patience, baggage or to avoid them. Survey questions refer to same-sex
and creativity, even for experienced and skilled profession- marriage as “gender-neutral marriage,” “equal marriage,” “gay
als. You can get a sense of the principles of survey question marriage,” “lesbian marriage,” “homosexual marriage,” and
writing by looking at things to do or to avoid when you “same-gender marriage.” More people oppose “homosexual
write survey questions. marriage” than “gay marriage,” and more oppose “gay mar-
riage” than “same-gender marriage,” and more oppose it than
What are three things to do when writing questions? “equal marriage.” The phrase used can influence answers.
examine the hypothesis that educated people are more Now, imagine a very different type of respondent (e.g., much dif-
accepting of same-sex marriage, you need two questions, ferent age, racial-ethnic or cultural background, education level).
Rewrite the questions so that this very different type of respondent
one about each variable (education, view on same-sex mar- would also understand the questions.
riage). After you have the data, simple statistics allow you to
The response entered here will appear in the
see whether the two variables are associated with one
performance dashboard and can be viewed by
another. The wrong way to test the hypothesis between edu- your instructor.
cation and favoring same-sex marriage is to ask people, “Do
you think less educated people oppose same-sex marriage Submit
more than highly educated people?” This asks people their
opinion about the hypothesis and can tell you about people’s
beliefs about the variables. It does not reveal the actual rela- 6.3.1: What Are Leading Questions?
tionship among the two variables. The people’s beliefs might
You’ve probably heard about “leading questions,” but what
be right or wrong about the actual relationship.
are they? You should never intentionally use leading ques-
tions in an honest, ethical survey. A leading question prompts
WRITING PROMPT the respondent to pick one response over another. They fre-
Specific Survey Questions for Targeted Groups quently appear in dishonest surveys, in which someone tries
A dilemma when writing survey questions is between writing ques- to manipulate results or mislead people. They also can occur
tions that a highly specific and homogenous type of respondent can when an inexperienced survey question writer is unclear.
easily understand, and using the same survey questions for a large, There are many types of leading questions. In a good survey
diverse collection of respondents.
Pick an issue and write two narrow and specific survey ques- question, respondents do not know which answer you expect;
tions that would be well understood by group to which you belong. they feel free to state what they really think or feel.
Summary Review
Summary Review
How Do You Write Good Closed-Format 20–29). The ambiguous verbal choice is another type
Responses? Most surveys offer preset answers from of overlapping response category—e.g., “Are you
which a respondent chooses. Writing good answer choices satisfied with your job, or are there things you don’t
is just as important as writing a good question. The answer like about it?” It is not clear how a person who is
choices should have three features: generally satisfied but has a few minor complaints
• Mutually exclusive. Response categories do not over- would answer this.
lap. You can easily correct numerical ranges (e.g., • Exhaustive. This means that each respondent has a
5–10, 10–20, 20–30) that overlap (e.g., 5–9, 10–19, choice—a place to go. For example, asking respondents
126 Chapter 6
“Are you working or unemployed?” leaves out respondents a “don’t know” or “not certain” answer alter-
respondents who are not working but do not con- native. A full-filter question is a special type of contin-
sider themselves unemployed (e.g., full-time home- gency question (contingency questions are discussed later).
makers, people on vacation, students, people with It is a two-part question. It first asks whether respondents
disabilities, or retired people). When writing a ques- have an opinion, and then asks for the opinion among
tion, first think about what you want to know and those who say that they have an opinion.
then consider the circumstances of all possible Studies of survey formats vary, but several suggest
respondents. For example, if you ask about a respon- that without a “no opinion” choice, as in the standard
dent’s employment, do you want information on the question, some respondents will offer to answer a ques-
primary job or on all jobs? Do you want both full- tion, even if they are very uncertain or unaware of an issue.
and part-time work? Do you only want jobs for pay Many people find saying “I don’t know” or “I have no
or also unpaid and volunteer jobs as well? If some- opinion” difficult to assert or may feel embarrassed in
one is temporarily unemployed, do you want the last doing so. With a quasi-filter question, most such respon-
job he or she held? dents choose “don’t know” because it appears as a legiti-
• Balanced. This means you offer the favorable or unfa- mate response. A full-filtered question takes “no opinion”
vorable choices equally in a set of responses. A case of or “don’t know” options one step higher. You should use a
unbalanced choices is the question, “What kind of job full filter for issues about which many people may not be
is the mayor doing: outstanding, excellent, very good, informed or have a firm opinion. An option is to ask about
or satisfactory?” It offers three favorable and one an opinion using a quasi-filter question, then follow up all
neutral response choice. Another type of unbalanced those with an opinion with a second question about how
question omits information—e.g., “Which of the five strongly they feel.
candidates running for mayor do you favor: Eugene
Example
Oswego or one of the others?” You can balance
responses by offering bipolar opposites. Unless there What is your opinion about the issue of global warming? Do you feel
in the future it is going to be a major threat, a minor threat, or no real
is a specific purpose for doing otherwise, offer
threat to how our society will be, or do you have no opinion?
respondents equal polar opposites at each end of a
continuum. Asking, “How strongly you support a _____ Major Threat
ban on same-sex marriage? Do you strongly support _____ Minor Threat
it, somewhat support it, or just barely support it?” is _____ No Real Threat
a set of unbalanced answers. To make it balanced, _____ No opinion {Go directly to next question}
you could ask, “How do you feel about a ban on [ASK ONLY IF FIRST THREE ANSWERS ARE GIVEN]
same-sex marriage; do you support it, oppose it, or How strongly do you hold that opinion?
neither support or oppose it?” Do you hold the opinion _____ very strongly, _____ somewhat
strongly, or _____ not very strongly at all?
Should You Offer a “Don’t Know” or “No
Opinion” Response Choice? Professional survey
researchers debate whether to include choices for neutral,
middle, and nonattitudes (e.g., “not sure,” “don’t know,”
“undecided,” or “no opinion”) in closed-ended questions. Example Study
They want to avoid two errors:
Questionnaire Items from
• Getting a “no opinion” or “don’t know” response when the 2009 Pew Research
a respondent actually holds a nonneutral opinion
• Forcing a respondent to choose a position when he or she
Center Survey
has no opinion on an issue or knows nothing about it. This marital status question has mutually exclusive and
exhaustive answer choices.
We have three ways to address the “don’t know”
response in attitude questions: Are you currently married, living with a partner, divorced, sepa-
rated, widowed, or have you never been married? (IF R SAYS
1. standard-format “SINGLE,” PROBE TO DETERMINE WHICH CATEGORY IS
2. quasi-filter APPROPRIATE) {QID:MARITAL1}
3. full-filter questions. 1. Married
A standard-format question does not offer a “don’t 2. Living with a partner
know” choice, and respondents must volunteer their lack 3. Divorced
of knowledge or opinion. A quasi-filter question offers 4. Separated
The Survey 127
5. Widowed
The response entered here will appear in the
6. Never been married performance dashboard and can be viewed by
9. Don’t know/Refused (VOL.) your instructor.
Standard Form (%) Quasi-Filter (%) Full Filter (%) Instead, ask it this way:
Agree 48.2 27.7 22.9 “I want to know how many sporting events you
Disagree 38.2 29.5 20.9 attended last winter. Let’s go month by month. Think
No opinion 13.6* 42.8 56.3 back to December. Did you attend any sporting events
* Volunteered for which you paid an admission in December? Now,
Source: Adapted from Schuman and Presser (1981:116–125). Standard format is from think back to January. Did you attend any sporting
Fall 1978; quasi- and full-filter are from February 1977.
events in January?”
Try asking: did not. To reduce social desirability bias, you can phrase
“In a typical weekday during the past year, about how questions to make norm violation appear less objection-
long did you watch TV? What about in a typical weekend able, or present a wider range of behavior as acceptable or
day, how often?” give respondents “face-saving” alternatives. The National
Election Survey asked about voting in the following way to
You can then multiply the respondent’s answers by the
reduce the social desirability bias: “In talking to people
number of weekdays and weekend days to create an esti-
about elections, we often find that a lot of people were not
mate for annual TV watching.
able to vote because they weren’t registered, they were
How Can You Ask Respondents about Sensi- sick, or they just didn’t have time. Which of the following
tive Issues? Most respondents want to present a posi- best describes you?—One, I did not vote. Two, I thought
tive image of themselves. They may feel ashamed, about voting this time but didn’t. Three—I usually vote,
embarrassed, or afraid to give truthful answers regarding but didn’t this time. Four—I am sure I voted.”
unpleasant or unflattering behaviors or events. They may
What Are Contingency Questions? A contin-
find it emotionally painful to confront their own actions
gency question (also called screen or skip question) is a
honestly, let alone admit them to a stranger. People may
two-question sequence that increases relevance. A first
underreport or self-censor reports of behavior or attitudes
question selects respondents for whom the second ques-
they wish to hide or believe to violate social norms such as
tion is relevant. It screens in/out respondents who get the
having an illness or disability (e.g., cancer, mental illness,
second part. The following example is a contingency
venereal disease) or engaging in illegal or deviant behavior
question.
(e.g., evading taxes, taking illegal drugs, engaging in
uncommon sexual practices). They may be hesitant to 1. Did you vote in the mayoral election last April when Guo,
reveal their financial status (e.g., income, savings, or debts). Smith, and Lopez were candidates?
Alternatively, they may over report positive or generally [ ] Yes (GO TO QUESTION 2)
[ ] No (SKIP TO QUESTION 3)
accepted behaviors.
Researchers developed several techniques to increase 2. Which candidate did you vote for? _____ Guo _____ Smith
_____Lopez _____ Don’t remember
getting truthful answers to sensitive issues. One is to alter
the context and question wording to be less threatening. 3. What kind of overall job is the new mayor doing in your
opinion? _____ Excellent _____ Good _____ Fair _____
You should only ask about sensitive issues after a warm-
Poor
up, when respondents feel more trust in the survey or
interviewer. You can emphasize to respondents that you How Can You Avoid Specific Words That
want honest answers and reassure them of confidential- Affect Answers? Wording effects occur when a par-
ity. You can provide a context that makes it easier for ticular word evokes a response. Professional survey
respondents to answer and appear less unusual. For researchers recognize that particular words in a survey
example, rather than asking “Have you stolen from a questions may trigger strong feelings or have connotations
store?,” you can ask, “In past surveys, many people have that color answers. Because respondents react to one word
reported that at some point they took items from a store rather than thinking about the issue in a question, you
without paying. Have you ever taken something from a want to avoid such words in survey questions. It is easier
store without paying for it?” Another technique is first to to write survey questions if you have a large vocabulary,
ask about more serious activities, making the sensitive know the connotations and meanings of many words, and
question issue appear less unusual. A respondent may are sensitive to the vocabulary of respondents. In general,
hesitate to admit that he or she shoplifted. However, a you want to use simple vocabulary and grammar to mini-
question about shoplifting appears after several questions mize confusion. Unfortunately, it is not possible to know in
about armed robbery or burglary, respondents may admit advance whether a word or phrase will affect responses.
to shoplifting because it appears to be less serious than
the other crimes.
Social desirability bias occurs when respondents dis- Learning from History
tort answers to look good or to conform to social norms.
Many people over report being highly cultured (i.e., read- The Power of Words
ing books, attending high-culture events), giving money to Survey researchers have uncovered several powerful wording
charity, having a good marriage, loving their children, and effects in surveys. One well-documented effect is the differ-
so forth. For example, one study found that one-third of ence between forbid and not allow.1 Both terms mean the
people who said they gave money to a local charity in a
survey really did not. Because a norm says that one should
vote in elections, people often say they voted when they 1
See Foddy (1993) and Presser (1990).
The Survey 129
same thing, but many more people are willing to “not allow”
something than to “forbid” it. In general, less educated 6.4: How Can You Design
respondents are most influenced by minor wording differ-
ences. Certain words trigger an emotional reaction or have an Effective Questionnaire?
significant connotations that we are just beginning to learn
6.4 Analyze some of the features of an effective
about. Smith (1987) found large differences (e.g., twice as
questionnaire
much support) in U.S. survey responses depending on
whether a question asked about spending “to help the poor” Once you have created a collection of survey questions,
or “for welfare.” Heated political attacks on welfare in the you face two decisions in designing an effective question-
1970s and 1980s changed connotations of the word welfare, naire. First, how many questions can you include in the
and it took on negative connotations that it did not previously questionnaire, or total questionnaire length. Second, how
have. The once neutral word came to imply lazy and immoral should you organize or arrange the questions that are in
people as well as wasteful, ineffective, and expensive govern-
the questionnaire.
ment programs. Today, it is best to avoid using it. Likewise,
Hurwitz and Peffley (2005) discovered that in recent years
many Americans have come to associate the term inner city 6.4.1: Length of Survey
with negative racial stereotypes about African Americans.
or Questionnaire
Racially prejudiced whites gave negative responses when the
phrase “inner city” appeared in a survey question but neutral The length of a questionnaire depends on the format of
responses for the same issues when it did not appear. In a your survey and on respondent characteristics. A 3–5
2005 Pew Research survey, 51 percent of respondents said minute telephone interview is rarely a problem. You can
they favored “making it legal for doctors to give terminally ill often extend it to 10 minutes. Web surveys vary but few
patients the means to end their lives” but only 44 percent people spend more than 10 minutes taking them. Mail
favored “making it legal for doctors to assist terminally ill questionnaires are more variable. A short (three-page)
patients in committing suicide.” Both questions asked about questionnaire is appropriate for the general population.
the same thing, but the respondent reactions differed because
Some researchers used questionnaires as long as 10 pages
of the word “suicide.”
(about 100 items) with the general public, but responses
Respondents also can become confused about the
drop significantly for longer questionnaires. For highly
meaning or connotations of key words. One survey asked
respondents whether they thought television news was
educated respondents and a salient topic, using question-
“impartial.” Impartial is a ninth-grade vocabulary term, and naires of 15 pages may be possible. Many face-to-face
researchers assumed everyone knew its meaning. They later interviews last a half-hour. In special situations, research-
learned that fewer than half of the respondents had inter- ers have been able to conduct face-to-face interviews for
preted the word with its proper meaning. Over one-fourth as long as 3–5 hours.
had ignored it or had no idea of its meaning. Others gave it
unusual meanings, and one-tenth gave it a meaning directly
opposite to its true meaning. You need to be cautious,
6.4.2: Question Sequence
because although a few wording effects (e.g., the difference You face three survey question sequence issues when
between forbid and not allow) are known, we are still learning designing a questionnaire:
about the power of specific words to shape respondent
1. How to organize questions on a questionnaire
answers.
2. How to reduce question order effects
3. How to control context effects
crime, respondents may have considered drunk driving to question is, “How are you doing in your classes?” Most
be a less serious offense. By contrast, respondents asked respondents will assume that the second question only
about drunk driving first may have thought about it as means classes outside their major because they already
criminal behavior. When asked about crime in general, they answered about the major. If you wanted to ask about
still are thinking of drunk driving as a type of crime. classes overall, then you should place that question
before the question about the classes in the major.
How to Organize Questions on a Question-
naire Every questionnaire has opening, middle, and
ending questions. Sequence the questions in a way to mini- WRITING PROMPT
mize respondent discomfort and confusion. The early ques- Survey Question Order
tions should help a respondent feel positive and comfortable In many surveys, we ask about multiple topics. Can you think of two
with the survey process. After an introduction that explains topics for a survey where answering one topic first might shift how
the purpose of the survey, make opening questions pleasant, people think about a second topic? What might we do to find out
whether topic order makes a difference, and if so how to deal with
interesting, and easy to answer. Demographic questions the issue?
(e.g., age, education level, and so forth) are easy but not
interesting, and should go toward the end. In addition, place The response entered here will appear in the
performance dashboard and can be viewed by
questions on the same topic together and introduce the sec- your instructor.
tion with a short statement (e.g., “Now I would like to ask
you questions about housing”) to orient respondents. Ques-
Submit
tion topics should flow smoothly and logically. Organize
them to assist respondents’ memory and comfort levels.
How to Reduce Question Order Effects The How to Control Context Effects Respondents
order in which questions appear may influence a respon- tend to answer questions based on a context of preceding
dent’s answers. Order effects are strongest for people who questions and the interview setting. Context effects are
lack strong views, for less educated respondents, and for strongest for ambiguous or unclear questions. This is
older respondents or those with memory loss. For exam- because respondents draw on the context when interpret-
ple, support for a single woman’s having an abortion pre- ing a question. It is not always possible to control context
dictably rises if it comes after a question about abortion effects. A first step is to be aware of them. You want to ask
being acceptable when a fetus has serious defects, but not a more general question before a more specific one. It takes
when the question is by itself or it comes before a question a bit more work, but making two versions of the question-
about fetus defects. naire and altering topics to two random parts of a sample
allows you to check whether context effects are operating.
Making It Practical: The Effect of Previous
Questions Previous questions in a questionnaire influ-
ence later ones in two ways:
Example Study
1. Question content (i.e., the issue). This occurred in the
above example of my student’s study about drunk driv- Question Order Effects
ing and crime. In another case, researchers compared
In a study with 9,000 respondents, Wilson (2010) found that
three forms to ask how much a respondent followed
the ordering of survey questions influenced reports of interra-
politics: the question alone, after asking what the respon- cial prejudice expressed by American blacks and whites.
dent’s elected representative recently did, and after ask- When questions specifically deal with salient emotional cues,
ing about what the representative did and about the such as interracial hostility among blacks and whites, question
representative’s “public relations work” in the area. The context and racial group membership strengthen in-group and
percentage of respondents’ reporting that they followed out-group categorization. In-group members (e.g., blacks)
politics “now and then” or “hardly at all” was 21 per- tend to view members of an out-group (e.g., whites) as having
cent, 39 percent, and 29 percent, respectively, for the greater dislike toward them (e.g., whites dislike blacks) when
three forms. The second form apparently made respon- respondents were first asked about the in-group (e.g., blacks).
dent’s feel they knew little, but the last form gave respon- This produced a desire to favor one’s in-group over the out-
group—a contrast effect. When the opposite order occurred, a
dents an excuse for not knowing the first question—they
norm of “evenhandedness” suggests that evaluating an out-
could blame their representative for their ignorance.
group in a noncomparative context, the in-group member feels
2. Respondent’s response. A respondent having already a need to appear balanced or evenhanded, particularly in the
answered one part of an issue may assume no overlap. presence of an interviewer. When first asked if their own group
For example, a respondent is asked, “How are you dislikes the out-group, respondents felt compelled to say that
doing in the classes in your academic major?” The next more of the out-group members disliked them in a follow-up
The Survey 131
question. So, when black respondents were asked “Do whites Each survey format has advantages or disadvantages: self-
dislike blacks?” 41% said “yes.” When white respondents administered, mail, face-to-face-interview, phone inter-
were asked “Do blacks dislike whites?” 42.8% said “yes.” view, and web survey.
However, when asked about their own in-group first, percep-
tions shifted. When black respondents were asked the same
question after they first answered whether “blacks dislike
whites,” then 55.4% said “yes” instead of 41%. When white
6.5.1: Mail and Self-Administered
respondents were asked “Do blacks dislike whites?” after hav- Questionnaires
ing answered whether whites dislike blacks, 51% said “yes” Advantages Mail and self-administered question-
instead of 42.8%. In short, roughly 10 percent more people naires are popular because they are easy and inexpen-
said the out-group disliked their in-group in one situation over
sive. You distribute or mail questionnaires directly to
the other. It all depended upon the context, or which question
respondents, respondents read instructions and ques-
they answered first. (See Figure 6.4.)
tions, and then record their answers. If you use mail, you
can cover a wide geographical area. In addition, a mail
Figure 6.4 Race of Respondent by Question Order survey allows respondents to check personal records at
Interaction Pattern (Predicted Probability of Perceiving home if necessary, offers anonymity, and avoids inter-
More Dislike toward Out-Group Members)
viewer bias.
Answering Survey Questions Vary by Question Order and
Respondent’s Race Disadvantages Physical location is limited for dis-
60 tributed, self-administrated questionnaires. Distribution
by mail provides large geographic area, but people do not
55.4
55 always complete and return questionnaires and the biggest
51 problem with mail questionnaires is a low response rate.
Percentage
50 You might mail out 500 questionnaires but get back only
50. Increasing the number mailed out to 50,000 so that you
45
42.8 have 500 both becomes much more expensive and can cre-
41 ate a bias. The 10 percent of people who respond are
40
unlikely to be representative. Perhaps only people who are
highly interested in the survey topic or who have a lot of
35
Black White free time (e.g., unemployed, retired, traditional homemak-
Respondents Respondents ers) respond. The opinions, education, income, age, and
Whites dislike Blacks first asked other characteristics of those who respond may not ade-
Blacks dislike Whites first asked
quately reflect the entire sample. It might seriously distort
your data.
Another limitation is that you do not control the con-
ditions under which a person completes a mail question-
naire. A questionnaire completed during a drinking party
6.5: The Advantages and by a dozen laughing people may be returned along with
one filled out by an earnest respondent. With mail ques-
Disadvantages of Different tionnaires, no one is present to clarify questions or to
probe for more information when respondents give
Survey Formats incomplete answers. Someone other than the sampled
6.5 List the advantages and disadvantages of various respondent (e.g., spouse, new resident, etc.) may open the
survey formats mail and complete the questionnaire. Respondents can
complete the questionnaire weeks apart or answer ques-
tions in an order different from that intended by research-
ers. Incomplete questionnaires can also be a serious
problem.
A mail questionnaire format limits the kinds of ques-
tions that a researcher can use. Questions that require
visual aids (e.g., look at this picture and tell me what you
see), open-ended questions, many contingency questions,
and complex questions do poorly in mail questionnaires.
Likewise, mail questionnaires are ill suited for the illiterate
or semi-illiterate in a country’s main language.
132 Chapter 6
6.5.2: Telephone Interviews visual materials as well as the survey questions, and use
elaborate contingency questions.
Advantages The telephone interview is a popular
survey method because you can reach about 95 percent of Disadvantages A drawback with web surveys is that
the population by telephone. You call a respondent, ask not everyone has Internet access, especially low-income,
questions, and record answers. The sample of respondents elderly, and less educated people. Low response rates and
can come from lists, telephone directories, or random-digit incomplete surveys are also common problems with web
dialing (RDD) and can be from a wide geographical area. surveys. As incentives for web survey participation
A staff of interviewers can interview 1,500 respondents improve, these issues may become less important. Another
across a nation within a few days and, get response rates as disadvantage is a lack of control over the conditions under
high as 85 percent with 7 to 10 callbacks. The telephone which a person completes a web survey. Someone who is
interview is a flexible method and has most of the strengths not serious or someone other than the selected respondent
of face-to-face interviews. Interviewers pick a specific may complete the web survey. Also, as with mailed ques-
respondent, control the sequence of questions, and can use tionnaires, no one is present to clarify questions or to probe
some probes. A specific respondent answers the questions for more information when respondents give incomplete
alone. Interviewers can use contingency questions effec- answers.
tively, especially with computer-assisted telephone inter-
WRITING PROMPT
viewing (CATI) (discussed later in this chapter). Also,
supervisors can monitor interview quality by listening in. Different Survey Formats
If you use other people to conduct interview, you need or clarify misinterpretations. An interviewer helps to define
to train them with the questionnaire. Interviewers must the situation for respondents, offers guidance and deter-
become very familiar with the wording and purpose of mines whether respondents have the information sought,
each question and practice following the flow of question- understands what is expected, and is providing relevant and
naire items. They will also need to exhibit proper survey serious answers.
interview behavior.
Stage 3: The Exit In this last stage, the interviewer interviews in the office and in the field that are recorded
thanks the respondent and leaves. He or she then goes to a and critiqued, many practice interviews, and role playing.
quiet, private place to edit the questionnaire. During this Interviewers are taught about survey research and the role
time, other details like the date, time, and place of the inter- of the interviewer. They must become very familiar with
view; a thumbnail sketch of the respondent and interview the questionnaire and the purpose of questions.
situation; the respondent’s attitude (e.g., serious, angry, or
flippant); and any unusual circumstances (e.g., “Telephone 6.6.4: Using Probes in Interviews
rang at question 27 and respondent talked for four minutes
A good interviewer knows how and when to use probes.
before the interview started again”) should be recorded.
Probes clarify a respondent’s ambiguous or irrelevant
The interviewer will also note anything disruptive that
response and confirm that respondents understand the
happened during the interview (e.g., “Teenage son entered
questions as intended. This means interviewers need to
room, sat at opposite end, turned on television with the
understand the survey and recognize an irrelevant or inac-
volume loud, and watched a music video”). The inter-
curate answer. There are many types of probes. A three- to
viewer also records his or her personal feelings and any-
five-second pause is often effective, as is nonverbal com-
thing that was suspected (e.g., “Respondent became
munication (e.g., tilt of head, raised eyebrows, or eye con-
nervous and fidgeted, changed answers once each on ques-
tact). The interviewer can repeat the question or the reply
tions 14, 15, and 16 about his marriage”).
and then pause.
He or she can ask a neutral question, such as:
6.6.3: Training Interviewers
“Any other reasons?”
Perhaps someday you may get a job interviewing for a “Can you tell me more about that?”
professional survey organization. A large-scale survey “How do you mean?”
employs many interviewers. Few people other than pro- “Could you explain more for me?”
fessional survey researchers appreciate the difficulty of
Making It Practical: Using Probes
the interviewer’s job. A professional-quality interview
requires a carefully selected and trained interviewer. As Interviewer Question: What is your occupation?
with any employment situation, interviewers need ade- Respondent Answer: I work at General Motors.
quate pay and good supervision to perform consistently at
Probe: What is your job at General Motors? What type
their peak. Good interviewers are pleasant, honest, accu-
of work do you do there?
rate, mature, responsible, intelligent, stable, and moti-
Interviewer Question: How long have you been
vated. They are patient and calm. They have experience
unemployed?
with many types of people and possess poise and tact.
Face-to-face interviewers must have a nonthreatening Respondent Answer: A long time.
appearance. If the survey involves face-to-face interview- Probe: Could you tell me more specifically when your
ing in high-crime areas, the interviewers need to have current period of unemployment began?
proper “street smarts” and may require extra protection Interviewer Question: Considering the country as a
(e.g., a partner or assistant). whole, do you think we will have good times during
the next year, or bad times, or what?
Respondent Answer: Maybe good, maybe bad, it
depends, who knows?
Probe: What do you expect to happen?
Probes are not substitutes for writing clear ques- Gender also affects interviews in terms of both obvi-
tions or creating a framework of understanding for the ous issues, such as sexual behavior, and support for
respondent. Unless carefully stated, probes might shape gender-related collective action or gender equality. In gen-
the respondent’s answers. Yet flexible or conversational eral, interviewers of the same gender or ethnic-racial
interviewing, in which interviewers use many probes, group as the respondent tend to get the most accurate
can improve accuracy on questions about complex answers.
issues for which respondents do not clearly understand
basic terms or about which they have difficulty express-
ing their thoughts. For example, to the question, “Did Example Study
you do any work for money last week?,” a respondent
might hesitate and then reply, “Yes.” An interviewer can Interviewer Race Effects
probe, “Could you tell me exactly what work you did?” Are Subtle and Pervasive
The respondent may reply, “On Tuesday and Wednes-
Survey researchers have been long aware that an interviewer’s
day, I spent each afternoon helping my buddy John
race can influence how respondents answer racially sensitive
move into his new apartment. For that he gave me $50, questions. Other research documented that women or racial
but I didn’t have any other job or get paid for doing any- minorities tend do poorly on certain tests administered by
thing else.” If the question intended only to get reports members of outside groups (such as white males) because of
of regular paid employment, the probe revealed a a stereotype that they will do poorly. They feel great pressure
misunderstanding. to disconfirm the negative stereotype, but the pressure creates
test anxiety that lowers their test score. In contrast, when a
member of their same group administers the test, the stereo-
WRITING PROMPT
type threat is not activated, test anxiety is reduced, and they
Probes in Interviews score higher.
Does using many probes impede the goal of standardizations (i.e., To test whether the stereotype-triggered test anxiety also
each respondent hears the exact same survey question, so we operates in survey interviews, Davis and Silver (2003) conducted
know all answers from respondents are to the same question)? a telephone survey of whites and African Americans in the
Explain your answer. Detroit area. They wanted to see whether a survey interviewer’s
The response entered here will appear in the race affected how a respondent answered. The research ques-
performance dashboard and can be viewed by tion was, “Do African Americans score differently on survey
your instructor. knowledge questions depending on whether they think the tele-
phone interviewer is white or African American?” Results showed
Submit that white respondents answered the same irrespective of inter-
viewer’s race. However, African American respondents scored
higher on the knowledge questions when they believed their
interviewer was an African American. The authors concluded
6.6.5: Interviewer Bias that beyond widely known social desirability and racial confor-
mity issues in surveys, the race of an interviewer can create
Survey researchers prescribe interviewer behavior to
subtle anxiety based on negative stereotypes that influence how
reduce bias. Bias is when a particular interviewer ’s
a respondent answers many survey questions.
actions have influenced how a respondent answers, and
responses differ from what they would be if another
interviewer had asked the questions. Proper interviewer
behavior and exact question reading are critical, but there 6.6.6: Computer-Assisted Telephone
is a larger issue.
Interviewing
An interviewer’s uncontrolled visible characteris-
tics, including race, stature, age and gender, often affect
interviews and respondent answers. This means noting
the characteristics of both interviewers and respondents,
especially in questions about issues related to visible
characteristics. For example, African American and His-
panic American respondents tend to express different
policy positions on race- or ethnic-related issues depend-
ing on the apparent race or ethnicity of the interviewer. Most professional survey organizations that do phone
This occurs even with telephone interviews when a interviewing have installed computer-assisted telephone
respondent has clues about the interviewer’s race or interviewing (CATI) systems. With CATI, the interviewer
ethnicity. sits in front of a computer and makes calls. Wearing a
136 Chapter 6
headset and microphone, the interviewer reads the survey to gain entry into homes, persuade a person to vote
questions from a computer screen for the specific respon- in a certain way, or try to sell something in the guise of a
dent, then enters the answer via the keyboard. Once he or survey.
she enters an answer, the computer shows the next ques- The mass media report more on surveys than other
tion on the screen. CATI speeds interviewing and reduces types of social research; however, the way mass media
interviewer errors. It also eliminates the separate step of report on surveys permits abuse. Few people who read
entering information into a computer and speeds up data poll results in a newspaper or hear them on television real-
processing. The CATI system works well for contingency ize that ethical codes require certain details about the sur-
questions because the computer shows the questions vey method. The purpose of providing details on method
appropriate for a specific respondent; interviewers do not is to reduce misuse of survey research. Researchers urge
have to look for the next appropriate question. In addition, the media to include the information, but it is rarely
the computer can check an answer immediately after the included. In addition to omitting critical details on how a
interviewer enters it. For example, the interviewer enters survey was conducted, the media often report on weak,
an answer that is impossible or clearly an error (e.g., an H biased, and misleading surveys along with sound, rigor-
instead of an M for “Male” or F for “Female”). The com- ous, professional ones without any distinction. This only
puter catches the error immediately. increases public confusion and a distrust of all surveys.
Ethical in Survey Research? ommend to include when reporting any poll or survey:
137
Chapter 7
The Experiment
Learning Objectives
7.1 Identify how experiments are the strongest 7.4 Describe combining and managing as the
type of social research for testing cause– two main tasks in experimental designs
effect relationships
7.5 Report some of the issues that might
7.2 Report the need to make valid comparisons weaken internal and external validity
in experiments
7.6 Identify the importance of comparisons in
7.3 Analyze the function of independent and experimental research
dependent variables as used in
7.7 Analyze why ethical concerns often arise
experimental research
in experimental research
In terms of politics, many observers like journalists, policy party in the 2008 and 2012 elections were sharper than any
makers, informed-educated citizens, and researchers in the time in the past 40 years. Others claim that most ordinary
United States and in other countries see Americans as a people are not polarized and share many common beliefs.
deeply polarized society. Recent news stories report solid What influences perceptions of political polarization? Why
red states versus blue states, refusals to compromise, and do some perceive sharp partisan divides whereas others do
political gridlock. Poll data show that divisions by political not? Van Boven, Judd, and Sherman (2012) thought that
138
The Experiment 139
people who were intensely committed to a particular point The results revealed that the participants who had
of view and held extreme partisan attitudes might be pro- introspected about their attitudinal processes showed
jecting their own polarization onto others. In short, not more polarization projection than did participants in the
everyone is polarized; rather the people who feel most control condition. The study demonstrated the most
polarized see others in highly polarized terms. highly committed people overestimate the polarization
The researchers identified three social psychological among others. When polarized people are stimulated to
processes to account for this situation: think about why they hold their views (i.e., naïve realism),
they become even more polarized and see people who
1. a general tendency to categorize others
hold an opposing position as stereotypic caricatures. This
2. a shift from categorizing to a dividing of people be- especially happens when they believe they have exercised
tween in-groups (“the people like me”) and out-groups careful reasoning and believe those holding an opposing
(“others”) position engaged in biased, self-interested reasoning.
3. “naïve realism.” Naïve realism is a tendency people
have to see their own attitudes on partisan issues as un-
biased and rational, and the attitudes of people in out-
groups as influenced by bias, self-interest, and ideology. 7.1: When Are Experiments
Essentially, polarized people assume that they are not Most Useful?
biased but that others are. The researchers called these com-
bined processes polarization projection. They created a social 7.1 Identify how experiments are the strongest type of
experiment to test polarized projection using a fictional social research for testing cause-effect relationships
political issue that had the essential features of political par- Compared to the other social research techniques, experi-
tisan conflict: well-defined groups with polarized attitudes ments provide the strongest means for testing causal rela-
toward a policy concerning the allocation of scarce tionships. In many ways, a research experiment is an
resources. They told the experimental participants, under- extension of commonsense logic. The key difference is that
graduate students at a large public university, that the uni- everyday experiments are less careful and systematic than
versity administration was seriously considering adoption scientific ones. In commonsense language, an experiment
of a fictional Nonresident Attraction and Retention Pro- refers to two situations:
gram (NARP). The NARP would give nonresident, out-of-
Before-and-After Comparison You modify some-
state students (who pay higher tuition than in-state
thing and then compare an outcome to what existed prior
students) privileged benefits, such as priority access to
to the modification. For example, one morning, you try to
desirable dorms, priority class registration, and free print-
start your car. To your surprise, it does not start. You
ing. The researchers thought resident and nonresident stu-
“experiment” by cleaning off the battery connections and
dents would develop partisan attitudes toward NARP.
then try to start it again. You modified something (cleaned
In the experiment, 101 undergraduate students (41
the connections) and compared the outcome (whether the
men and 60 women) at the University of Colorado-Boulder
car started) to the prior situation (it did not start). Your
first completed a questionnaire on the NARP policy using
“hypothesis” was that the car did not start because of a
a 5-point Likert Scale. It asked questions such as what stu-
buildup of crud on the connections, and once you removed
dents liked or disliked about the policy and whether it vio-
the crud, the car would start.
lated their sense of fairness. The researchers next randomly
assigned participants into two conditions, either the intro- Side-by-Side Comparison You have two similar
spection group or the control group. They gave the intro- things, and then you modify one but not the other and
spection group time and cues to ponder why they held compare the two. For example, you watch a young boy
their views before they were asked to estimate the distribu- playing with a soft drink can. He vigorously shakes it. You
tion of all other students’ attitudes on campus. The control hold a can that is not shaken. You then open both cans for
group students just went to estimation. him. He laughs when the shaken can “explodes” with a
The researchers asked participants in the introspection fizzy mess, but the one you held does not explode. You
condition to describe their reasoning processes with ques- began with two similar things (soft drink cans) and modi-
tions like: What factors, different thought processes, and fied one (shook up a can) but not the other. Before you
experiences might have caused you to hold your stance? opened the cans, you hypothesized that the shaken can but
Did you consider if the proposal was in your self-interest? not the other one would make a fizzy mess.
Did you engage in careful and extensive thought? After- Compared to other social research methods, experi-
wards, the introspection participants estimated the distri- ments best satisfy the three conditions needed to demon-
bution of other students’ attitudes toward NARP. strate causality—temporal order, association, and no
140 Chapter 7
alternative explanations. In order to demonstrate causality are. You can study this issue using a survey or existing sta-
in an experiment, we follow three basic steps: tistics research, but not the experimental method. For a true
experiment, you would force some parents to have biracial
1. Start with a cause–effect hypothesis.
children and not others, force all the parents to raise their
2. Modify a situation or introduce a change.
children similarly, and then wait until the children grow up
3. Compare outcomes with and without the modification.
to examine their career choices. This is both unethical and
impractical. Despite its great strength at demonstrating
7.1.1: Questions You Can Answer causal relations, the experiment is limited in the questions
with the Experimental Method you can ask, the variables you can measure, and your ability
to generalize from one experiment to larger society.
Experimental research has a clear, simple logic that offers a
The ideal solution is to study an issue using both
very powerful way to focus narrowly and demonstrate
experimental and a nonexperimental methods and then
causal relations among a few variables in controlled situa-
combine what you learn from both. Maybe you are inter-
tions. Research questions most appropriate for an experi-
ested in attitudes toward people in wheelchairs. You could
ment fit its strengths and limitations that include the
conduct an experiment asking participants to respond to
following features:
photos of people in wheelchairs and not in wheelchairs,
• It can isolate and identify a causal mechanism. with the different research participants seeing the same
• It targets two or three specific variables, and it is nar- person in or not in a wheelchair. You could ask them sur-
row in scope. vey questions like the following: Would you hire this per-
• It is limited by the practical and ethical aspects of the son? How comfortable would you be if this person asked
situations you can impose on humans. you for a date? You could also conduct a social survey on
attitudes about disability issues. In addition, you could
In general, the social experiment is better for targeted conduct field research and observe people’s reactions to a
micro-level concerns (e.g., individual or small-group phe- person in a wheelchair in a natural setting, or you might be
nomena) than for complex macro-level issues containing in a wheelchair and carefully note the reactions of others to
many factors that operate together. In other words, they you. Our greatest confidence comes when many well-
are excellent when you can isolate few variables in a con- designed studies conducted by different researchers yield
trolled, small-scale setting but are not well suited for ques- similar results.
tions involving many diverse influences that operate across
an entire society or over decades. For example, to track WRITING PROMPT
changes in public attitudes about same-sex marriage across
True Experiments
the entire society over the past 20 years, a survey would be
much better than an experiment. We can, however, provide A true experiment often requires manipulating situations or aspects
of people lives but within practical and ethical limits. What practical
insights into larger issues when we integrate and synthe- or ethical limits can you see restricting your ability to carry out true
size the results from many narrowly focused experiments. experiment on social situations or relations?
participants who do not differ with regard to variables that in which each case has a known and equal chance of selec-
could be alternative explanations to your hypothesis. Let us tion. It obeys mathematical laws, which make precise cal-
say you want to learn whether completing a college course culations possible.
affects a person’s skill level. You have two groups of partici- Both random sampling and random assignment use a
pants: one group completed the course whereas the other mathematically random process. When you randomly
did not. To make a valid comparison, you want participants sample, you use a random process to select a smaller sub-
in the groups to be similar in every respect, except the one set of people (sample) from a larger pool (population).
you are examining in the study—taking the course. When you randomly assign, you use a random process to
sort a collection of participants into two or more groups.
Why is it important to have participants in the groups You can both randomly sample and randomly assign. You
to be similar in every respect? can first sample to obtain a set of participants (e.g., 150
students out of 15,000), then randomly assign to divide the
Compare Your Thoughts sampled participants into groups (e.g., divide the 150
Let’s imagine that participants who completed the course people into 3 groups of 50 each). Combining random sam-
are also two years older than those who did not. You could pling and random assignment is the ideal; however,
not know whether it was completing the course or being because of the extra effort required, it is rare in social
older that accounted for group differences in skill levels. experiments (see Figure 7.1).
If the purpose of random assignment is to create two
7.2.1: Random Assignment (or more) equivalent groups, you may ask whether it might
be simpler to match participants’ characteristics in each
and Random Sampling group. Professional researchers occasionally match partici-
Many experiments use random assignment, also called pants on certain characteristics, such as age and sex, but
randomization. To compare groups of participants, you do this is infrequent. True equal matching falls apart as the
not assign them based on your feelings, their personal number of relevant characteristics expand, and soon find-
preference, or their specific features. Random assignment ing exact matches is impossible. Individuals differ in thou-
is unbiased because your desire to confirm a hypothesis or sands of ways. You cannot know which traits are relevant
a participant’s personal interests is removed from the and influence your variables. Let us say you compare two
selection process. It is random in a statistical or mathemati- groups of 15 students. Group 1 has eight males. To match,
cal sense, not in the everyday sense of unplanned, haphaz- group 2 must also have eight males. Two males in group 1
ard, or accidental. In probability theory, random is a process are only children; the rest have one or more siblings. Three
Random Assignment
Step 1: Begin with a collection of subjects.
Step 2: Devise a method to randomize that is purely mechanical (e.g., flip a coin).
Step 3: Assign subjects with “Heads” to one group and “Tails” to the other group.
Example Study
Random assignment is simple in practice. Start with a collec- Was It a Gun or a Tool?
tion of people, such as 50 volunteers who show up to par-
ticipate in a study. Next, divide them into two or more
equal-sized groups using a true random process, such as
tossing a coin or throwing dice. If you toss a coin, you may
assign all for whom heads appear in one group and the rest
in another. You will probably have about 25 participants in
each group. What if you have 80 people and want to assign
them to four groups? You can use coin toss (every other toss
goes to a group), dice, or another pure random process, such
as a randomizing software application. The key feature of
the process is that all have an equal, one-in-four chance, of
ending up in a group. Nothing about a specific participant
or an experimenter’s preference affects who goes to which
In many situations, police officers must respond quickly and
group. It is entirely due to pure mathematical chance. accurately to a potentially dangerous situation, such as trying
to determine if an individual is holding a gun. In 2014, national
WRITING PROMPT outrage erupted when a white police officer shot an unarmed
Random Assignment or Matching black teen in Ferguson, Missouri. A few weeks later, a white
police officer shot and killed a 12-year-old black boy in a park
Imagine that you have 100 volunteers to be participants in an experi-
holding a toy gun just 2 seconds after pulling up in his squad
ment, and you want to randomly assign them into four groups of
equal size (i.e., 25 each). How would you randomly assign the 100 car. Earlier, in 1999, four white New York City police officers
into four groups? Be very specific in explaining what you would do. shot and killed an unarmed immigrant from West Africa who
was holding out his wallet which the police mistakenly thought
The response entered here will appear in the was a gun.
performance dashboard and can be viewed by
Payne (2001) created two experiments to learn whether
your instructor.
racial stereotypes about “dangerous people” interferes with the
person’s ability to make accurate split-second judgments.
Submit Payne built on past studies of priming that found people link
The Experiment 143
visual and other images to preexisting negative stereotypes. In many experimental designs, you measure the
The images prime or activate a negative response, often auto- dependent variable more than once, both before introduc-
matically and unconsciously, in people who hold stereotypes. ing the independent variable, in a pretest, and again after
If revealing a negative racial stereotype is socially inappropriate, introducing it, in a posttest.
people will try to control or hide its priming effects. However,
When introducing an independent variable, researchers
hiding the expression of a stereotype slows a person’s deci-
can use two or more groups that experience different situa-
sion making. For a short time period (perhaps several sec-
tions or a single participant group that experiences different
onds), people must reconsider their public responses in order
to make them more socially acceptable. situations one after the other. If there are two or more groups,
In the experiment, Payne had 31 white undergraduate it is an independent group design. In the study that opened
students complete an attitude survey. Next, they looked at this chapter, the independent variable was to have one
visual images on a computer. The experimenter told partici- group of participants “introspect” and seriously consider
pants he was measuring their speed and accuracy of identify- their reasoning for holding a partisan position, whereas the
ing visual images. Participants practiced using the equipment control group did not. The group that introspected was the
and classifying 48 photos. Payne then told them that they experimental group. When the independent variable has
would see a pair of photos. The first one was a warning that many different values, you may have more than one experi-
the second would soon appear. The first photo was always the mental group, called comparison groups, one for each level
face of either a white or a black man. The second photo was of
of the independent variable. At other times, experimenters
a handgun or a tool (hand drill or socket wrench). The first
use a repeated measures design. There is one group and the
photo appeared for 200 milliseconds; the second photo
same participants experience multiple situations over time
appeared for 200 milliseconds. After a participant responded,
the screen went blank. Participants did not have a fixed time by
as was the case in the study about viewing images of a black
which they had to respond. After the participants gave a or white face then a gun or tool.
response, the next pair of photos appeared. Participants iden-
tified 192 photos with the race of the first photo and tool or gun 7.3.1: Planning an Experiment
randomly mixed. Thus, Payne randomly mixed what each par-
ticipant saw instead of randomly assigning the participants. To plan an experiment, you must decide on a specific experi-
In a second experiment, everything was the same except mental design and plan what participants will experience
that Payne added pressure. If participants failed to respond from beginning to end. Planning includes decisions about
within 500 milliseconds after the second photo, they saw a dra- the number of groups, how and when to create independent
matic red warning and had 1 second to respond. In the first variable conditions, and how often to measure the depen-
experiment, the participants showed no differences in accuracy. dent variable. Overall, you need to plan and configure seven
In the second experiment, with time pressure, the error rate was parts of a design. Not all experiments have all seven parts,
much higher. Many of the participants who first saw the black but a few experimental designs have more than these seven.
man’s face mistook a hand tool for a gun. However, their errors
of seeing a tool as being a handgun did not increase after they 1. Independent variable
saw a white man’s face. The participants who made the most 2. Dependent variable
errors, i.e., mistaking a tool for a gun, were participants who most 3. Pretest
strongly accepted negative stereotypes about African Americans
4. Posttest
based on answers to the survey at the start of the study.
5. Experimental group
6. Control group
Analysis of Was It a Gun or a Tool? 7. Random assignment
In the experiment described in Example Study: Was It a
You also develop measures of the dependent variable
Gun or a Tool?, the independent variable is different prim-
and pilot test the experiment.
ing that the experimenter created by showing participants
either a white man or black man’s face.
Experimenters measure dependent variables in many Summary Review
ways, including response times, percent accurate scores,
social behaviors, attitudes, feelings, and beliefs. In the
experiment described in the Example Study, the dependent Steps in Conducting
variable was participant accuracy about whether a second an Experiment
photo showed a gun or a tool. You can measure dependent
variables by paper-and-pencil indicators, such as a 1. Begin with a straightforward hypothesis that is appropriate
questionnaire, by interviews, by observing behaviors for experimental research.
(making a choice or response time), or by physiological 2. Decide on an experimental design that will test the
responses (e.g., heartbeat or sweating palms). hypothesis within practical limitations.
144 Chapter 7
3. Decide how to introduce the independent variable or cre- misleading them with written or verbal instructions, by
ate situations to induce it. using the actions of helpers or confederates, or by arrang-
4. Develop a valid and reliable measure of the dependent ing a setting. Experimenters also invent false dependent
variable. variable measures to keep the true ones hidden. Obviously,
5. Set up an experimental setting and conduct a pilot test of using deception raises ethical concerns.
the variables.
6. Locate appropriate participants.
7. Randomly assign participants to groups and give them
7.4.1: Types of Experimental Design
instructions. To plan an experiment, you combine the parts of an experi-
8. Gather data for the pretest measure of the dependent ment (e.g., pretests, control groups, etc.) into an experi-
variable for all groups. mental design. Experimental designs vary in their
9. Introduce the independent variable to the experimental components: Some do not use random assignment, some
group only (or to relevant groups if there are multiple lack pretests, some do not have control groups, and others
experimental groups) and monitor all groups. have several experimental groups. In research reports,
10. Gather data for posttest measure of the dependent experimenters name widely used designs. To understand
variable. experimental design, it is helpful to learn the standard
11. Debrief the participants. This means you ask partici- designs that illustrate several ways to combine the parts of
pants what they thought was occurring and reveal the an experiment.
true purpose and situation you deceived them about in We can illustrate the standard experimental designs by
any aspect of the experiment. looking at variations on this simple example: You want to
12. Examine data collected and make comparisons between test whether servers (waiters and waitresses) receive big-
different groups using statistics and graphs to determine ger tips if they introduce themselves by first name before
whether the data support the hypothesis. taking an order and return 8 to 10 minutes after delivering
food to ask, “Is everything fine?” The independent variable
is server behavior. The dependent variable is the size of the
tip received.
7.4: How Do We Combine We can divide standard designs into three groups:
• True experimental
Parts into Experimental • Pre-experimental
Designs? • Quasi-experimental designs
Two-Group Posttest-Only Design This design ependent variable or improves their posttest score. Rich-
d
has all the parts of the classical design except a pretest. It is ard Solomon developed a design to address this issue by
similar to the static group comparison, a type of two-group combining the classical experimental design with the two-
pre-experimental design. The two-group pretest-only group posttest-only design and randomly assigning par-
design includes random assignment absent in the static ticipants to one of four groups.
group comparison. Random assignment improves the
Example. You randomly divide 80 newly hired serv-
chance that the groups are equivalent, but without a pre- ers into four groups and give them training. You instruct
test you cannot be as certain that the groups really began participants in groups 1, 2, and 4 not to introduce them-
equal on the dependent variable. selves or to return during the meal to check on the cus-
Example. You randomly divide 40 newly hired serv- tomers. You instruct participants in group 3 (experimental
ers into two groups and give all training. You instruct one group 2) to introduce themselves and to return during the
group not to introduce themselves by first name or to return meal to check on the customers. During the first month,
during the meal to check on the customers. You instruct you count average weekly tips for groups 1 and 2 only
participants in the other group to introduce themselves to (pretest). After the first month, you “retrain” group 1 par-
customers by first name and to check on the customers 10 ticipants (experimental group 1) henceforth to introduce
minutes after delivering food. You record average weekly themselves to customers by first name and to return dur-
tips for both groups (posttest score) (see Figure 7.3). ing the meal to check on the customers 10 minutes after
delivering food. All other groups continue as first in-
Solomon Four-Group Design It is possible that structed. During the second month, you record average
your pretest measure sensitizes participants to your weekly tips for all groups (posttest) (see Figure 7.4).
Month 1 Month 2
Randomly Group 1 Pretest Independent Variable Present Posttest
assign Serve food without (amount of tips) Self-introduction and return to (amount of tips)
participants to introduction or checking check on customer
training
Group 2 Pretest Independent Variable Absent Posttest
sessions
Serve food without (amount of tips) Continue to serve food without (amount of tips)
introduction or checking introduction or checking
Group 3 Independent Variable Present Posttest
Serve food with Continue self-introduction and (amount of tips)
introduction and checking return to check on customer
Group 4 Independent Variable Absent Posttest
Serve food without Continue to serve food without (amount of tips)
introduction and checking introduction or checking