Designing Surveys: A Guide to

Decisions and Procedures 3rd Edition

(eBook PDF)
Visit to download the full and correct content document:
https://ebooksecure.com/download/designing-surveys-a-guide-to-decisions-and-proce
dures-3rd-edition-ebook-pdf/
 Types of Nonprobability Samples
Comparing Probability and Nonprobability Samples
Guidelines for Good Sampling
How Good Must the Sample Be?
General Advice
Supplemental Reading
6. Sampling II: Population Definition and Sampling Frames
Defining the Survey Population
Framing the Population
For More In-Depth Reading
7. Sampling III: Sample Size and Sample Design
Sampling Error Illustrated
Confidence Interval Approach to Sample Size
Power Analysis Approach to Sample Size
Nonstatistical Approaches to Sample Size
Stratified Sample Designs
Cluster Sampling to Improve Cost-Effectiveness
Computing Sampling Errors
Additional Resources
Methodology Appendix 1: Using Models in Sampling
8. Questionnaire Development I: Measurement Error and Question Writing
Measurement Error and Response Error
A Model of the Response Process
Response Task Problems
Questionnaire Design as Process
Factors in Questionnaire Development
Writing Questions
The Structure of Survey Questions
The Use of Qualifiers in Survey Questions
Response Categories
Rating Scales
Avoiding or Identifying Common Weaknesses in Survey Questions
Supplemental Readings
9. Questionnaire Development II: Questionnaire Structure
Introducing the Survey
What Questions Should the Questionnaire Begin With?
Grouping Questions Into Sections
Questionnaire Length and Respondent Burden
Formatting Instruments for Multimode Data Collection
Supplemental Reading
Methodology Appendix 2: Questionnaire Evaluation Workshop
10. Questionnaire Development III: Pretesting
Objectives of Pretesting
Types of Response Problems

8
 Samples for Pretests
Pretesting Procedures
Conventional Pretests and Interviewer Debriefings
Postinterview Interviews
Behavior Coding
Cognitive Interviews
Respondent Debriefing
Expert Panel
Assessing Interviewer Tasks
Experimental Design for Question Evaluation
Revising and Retesting
Pilot Tests
Some Last Advice
For More In-Depth Reading
Methodology Appendix 3: Cognitive Interviewing Workshop
11. Data Collection II: Controlling Error in Data Collection
Measures of Survey Quality
Unit Nonresponse
Recent Increases in Nonresponse
Item Nonresponse
Balancing Survey Goals and Ethical Obligations to Participants
Controlling Error in Data Collection
Interviewer-Administered Surveys
Computer-Assisted Data Collection
Minimizing Item Nonresponse
Interviewer Effects
Self-Administered Surveys: Mail and Internet
Data Collection Costs and Contingencies: Planning for the Unexpected
For More In-Depth Reading
Methodology Appendix 4: An Overview of Organization Surveys
12. Postsurvey Statistical Adjustments and the Methodology Report
Nonresponse Bias
Data Adjustments: Weighting and Imputation
The Methodology Report
Surveys in Other National and Cultural Contexts
Supplemental Reading
Appendix A: University of Maryland Undergraduate Student Survey
Appendix B: Maryland Crime Survey
Appendix C: AAPOR Code of Professional Ethics
Appendix D: Internet Resources
References
Index

9
10
 PREFACE

T
here were 9 years between the first (1996) and second (2005) editions of this book. Although fewer years have
elapsed since that last edition than between the first two, the revisions are substantially larger, reflecting the
evolution of a field driven by rapid technological and societal change, but also by the effect of advances in
survey methodology. In addition to the necessary updates, we have expanded the technical range of the book.
First, the chapters on sampling (Chapters 5 7) have been completely restructured and expanded. This revision
is consistent with the increased focus on survey error (Chapter 2) as a framework for thinking about each stage
of survey design and implementation. Second, we have added four Methodology Appendixes to address some
topics of special interest (Using Models in Sampling and An Overview of Organization Surveys) and to
provide in-depth workshop treatments on topics that are crucial to containing measurement error, often the
major threat to survey quality (Questionnaire Evaluation Workshop and Cognitive Interviewing Workshop).
In this edition, we have also taken a broader view of the needs of diverse researchers who may employ the
survey method. We use examples drawn from a wide range of social research—including sociological, political,
educational, public health, marketing, and business applications—and we consider issues faced by researchers
in varying contexts, including academic versus nonacademic research, surveys of organizations versus
households or individuals, and cross-cultural surveys.
The treatments of data collection (Chapters 4 and 11) and questionnaire development (Chapters 8–10)
reflect a wider range of methodologies (including online, cell phone, intercept, and multimode surveys), as
well as contributions to survey practice from recent survey methods research (e.g., the use of cognitive models
for understanding the response process and effective design of response categories and rating scales). Finally,
from Chapter 1 (Survey Practice) through Chapter 12 (Postsurvey Statistical Adjustments and the
Methodology Report), we have integrated the ethical treatment of research subjects into survey design,
questionnaire construction, and data collection.
The guidance of our editor, Vicki Knight, has been essential in this effort. The technical assistance
provided by Nadra Garas was critical, and the efficient support of Kalie Koscielak and Lyndsi Stephens of
SAGE Publications contributed greatly to the timely editing and processing of the manuscript. We appreciate
the extensive, helpful comments on a draft of the book provided by reviewers engaged by SAGE Publications:
Jennifer Reid Keene, University of Nevada, Las Vegas; Lisa House, University of Florida; Julio Borquez,
University of Michigan, Dearborn; Yasar Yesilcay, University of Florida; Chadwick L. Menning, Ball State
University; Brenda D. Phillips, Oklahoma State University; Bob Jeffery, Sheffield Hallam University; Michael
F. Cassidy, Marymount University; Young Ik Cho, University of Wisconsin, Milwaukee; Richard J. Harris,
University of Texas at San Antonio; Valentin Ekiaka Nzai, Texas A&M University, Kingsville; and Mary-
Kate Lizotte, Birmingham-Southern College.
In particular, we want to recognize our intellectual debt to the late Seymour Sudman, a seminal figure in
modern survey research, who was our teacher and mentor for many years, beginning when we were all, in
various capacities, at the University of Illinois Survey Research Laboratory. This book would not have been
possible without the guidance of his scholarship and his example as a researcher.

11
 ABOUT THE AUTHORS

Johnny Blair is an independent consultant in survey methodology. Previously, he was a Principal Scientist at
Abt Associates and a manager of survey operations at the University of Maryland Survey Research Center and
the University of Illinois (Urbana-Champaign) Survey Research Laboratory. Over a 40-year career in survey
research, he has designed and/or implemented surveys for health (including HIV high-risk populations),
education (including large-scale student assessments), environment (including contingent valuation), and
criminal victimization (including proxy reporting) surveys, among other areas. He has conducted
methodological research on sampling rare populations, measurement error in proxy reporting, cognitive and
usability testing of computer-based student writing assessments, and data quality in converted refusal
interviews. He has been involved in a decade-long program of research on cognitive interview pretesting, most
recently on the theory of pretest sample size and the validation of pretest problem identification. He has been
a member of the editorial board of Public Opinion Quarterly, has served on several National Research Council
Panels, and has been a consultant to many federal agencies, academic organizations, law firms, and other
companies. Since 1996, he has served on the Design and Analysis Committee (DAC) for the National
Assessment of Educational Progress, NAEP, the Nation’s Report Card.

Ronald F. Czaja is Associate Professor Emeritus of Sociology and Anthropology at North Carolina State
University. He taught courses in both undergraduate and graduate research methodology and medical
sociology. His methodological research focused on sampling rare populations, response effects in surveys, and
the cognitive aspects of questionnaire design. From 1969 to 1990, he worked at the Survey Research
Laboratory, University of Illinois at Chicago, as project coordinator, co-head of sampling, assistant director,
and principal investigator.

Edward A. Blair is the Michael J. Cemo Professor of Marketing & Entrepreneurship and Chair of the
Department of Marketing & Entrepreneurship in the C. T. Bauer College of Business, University of
Houston. He has served as chair of the American Statistical Association Committee on Energy Statistics,
which advises the U.S. Energy Information Administration on the gathering, analysis, and dissemination of
information. He also has served on the U.S. Census Bureau Census Advisory Committee of Professional
Associations and as a National Science Foundation panelist in Innovation and Organizational Change. His
research has been published in journals such as the Journal of Marketing, the Journal of Marketing Research, the
Journal of Consumer Research, Public Opinion Quarterly, Sociological Methods and Research, the Journal of
Advertising Research, the Journal of the Academy of Marketing Science, the Journal of Retailing, and elsewhere. He
has served on the editorial boards of the Journal of Marketing Research, the Journal of the Academy of Marketing
Science, and the Journal of Business Research and as national conference chair for the American Marketing
Association (AMA). In addition to teaching university classes, Dr. Blair has taught in a variety of professional
programs, including the AMA’s School of Marketing Research (for research professionals) and Advanced
Research Techniques Forum.

12
 ONE
SURVEY PRACTICE

I
t would be difficult to name another social science method that has so quickly and pervasively penetrated our
society as the sample survey. In fewer than two generations, the notion of relying on relatively small samples
to measure attitudes and behaviors has grown from a little-noted curiosity to the dominant data collection
practice. Surveys are used by academic researchers, governments, businesses, political parties, media, and
anyone who wants insights into what people are thinking and doing. Survey data underlie our knowledge
regarding

• demographic characteristics of the population;

• economic variables such as household incomes, unemployment rates, business hiring intentions, and
consumer confidence;
• education variables such as levels of educational attainment, dropout rates, and educational practices in
schools;
• health variables such as access to health care, immunization rates, nutritional practices, obesity rates, and
engagement in health risk behaviors;
• political variables such as party identification, candidate preferences, voting intentions, and public opinions
on policy issues;
• marketing variables such as product preferences and purchase intentions;
• and many more.

The requirements of these diverse survey applications have naturally spawned a wide range of performance
practices. How a survey is designed and implemented for a federal agency is vastly different than one for a
newspaper poll.
A large and rapidly expanding survey literature reflects both the numerous applications of survey data
collection and the inherently multidisciplinary nature of survey research. To successfully design and
implement a survey, we need to understand the basics of a few disciplines and techniques. At some points, we
can rely on scientific understanding and training; at others, we need a knowledge of accepted practices and,
throughout the process, a healthy dose of common sense.

13
WHAT IS A SURVEY?

Surveys collect information by interviewing a sample of respondents from a well-defined population. The
survey population may comprise individuals, households, organizations, or any element of interest. The
boundaries of the population may be defined by demographic characteristics (e.g., persons 18 years of age or
older), geographic boundaries (residing in Maryland), behaviors (who voted in the last election), intentions
(and intend to vote in the next election), or other characteristics. The population should be defined so that its
members can be unequivocally identified. In addition, we must be convinced that the majority of respondents
will know the information we ask them to provide. It makes little sense to ask people questions, such as the
net worth of their family that many in the targeted population, maybe most, will not be able to answer.
Surveys can be conducted in person, by phone, by mail, or over the Internet, among other methods. The
identifying characteristic of a survey interview is the use of a fixed questionnaire with prespecified questions.
The questions are most often, but not always, in a closed format in which a set of response alternatives is
specified. Using a fixed questionnaire allows a researcher to control the interview without being present,
which allows the interview to be conducted at relatively low cost, either through self-administration by
respondents (as in Internet or mail surveys) or through administration by interviewers who, although trained,
are typically paid at a modest rate. The resulting data are then entered into a data file for statistical analysis.
Surveys are, of course, not the only method used by social researchers to gather data. Alternatives include
observation, depth interviews, focus groups, panels, and experiments. Key points of comparison between
surveys and other methods, as well as examples of how other methods are sometimes used to support survey
design, are as follows:

• As the term suggests, we gather observational data by observing events rather than by asking questions.
Observation can capture information about inanimate phenomena that can’t be questioned directly, and
observation doesn’t suffer from respondents misunderstanding the question, forgetting what happened, or
distorting their answers to make a good impression. However, for observational data to be feasible, the
phenomenon of interest must be observable; mental states such as attitudes or intentions are out. After an
election, we can observe how precincts voted, but before the election, we cannot observe how they intend to
vote. We can observe how precincts voted, but not why they voted that way. It also may not be cost-effective
to gather observational data. We can observe how someone uses public transportation by following that person
for a month, but it is much less costly to ask him or her about last month’s behavior.
In some situations, observation may be used to learn more about a particular population while developing
plans for a survey. For example, in preparing for a survey of parents about their children’s dietary habits, we
observed kids at a few school lunches. Behaviors such as discarding, sharing, or exchanging foods between
children led to some improvements in the parent survey. More important, it showed that some of the
children’s eating could not be fully reported by the parents.

• Depth interviews, like surveys, gather data through questioning. However, depth interviews do not use a
fixed questionnaire. The interviewer usually has a list of topics to be covered and may use fixed questions to
get respondents started on these topics, but the overall goal is to let respondents express their thoughts freely
and to probe as needed. This approach is good for getting deep, detailed, complex information that doesn’t
work well in a survey. However, these interviews usually must be administered in person, and a highly skilled
interviewer is needed to manage the unstructured interaction, resulting in much higher cost than a survey
interview and consequently less ability to interview a broad sample. So, for example, if you want to know how
top officials in city government interact and make decisions, then you might do depth interviews with a small
number of city leaders, but if you want to know how the broad electorate rates the performance of city
government using standard questions, a survey will be more useful.
In planning a survey, depth interviews can be used to uncover issues that are important to include in the
questionnaire, learn how potential respondents think about the topic, develop response alternatives, and learn
how the population might react to survey procedures (such as those for gaining cooperation). This can be
especially useful when the population and/or topics are not familiar to the researcher.

• Focus groups, like depth interviews, do not use a fixed questionnaire. Unlike depth interviews, which are
conducted one-on-one, focus groups facilitate interaction among group participants. This interaction can
inform the researcher about reasons for differences in a population on an issue, or other social dynamics. As

14
with depth interviews, focus groups are sometimes used in the survey planning process to uncover issues that
are important to include in the questionnaire, learn how potential respondents think about the topic, develop
response alternatives, and learn how the population might react to survey procedures.

• Panels are groups of research participants that provide information over time. This information can be
collected through observation or self-reports, including diaries where people record their behavior over time or
repeated surveys of the same respondents. The greatest strength of panels is that measures of change over time
are precise and not subject to variation due to shifting samples of respondents. The greatest weakness of panel
research is the fact that many people are not willing to accept the commitment required by a panel, while
many others drop out after a short stay, so it is difficult to keep panels representative of the population. Also,
panels tend to be more expensive than most onetime surveys because panel members must be given some
reward for their ongoing effort, while most surveys do not reward respondents. Therefore, if a researcher
wants to study how political preferences evolve through a campaign, she might recruit a panel of registered
voters and track them, but if her focus is on accurate estimates of candidate preference at any given point in
time, she is more likely to use a series of independent surveys.

• An experiment is a study in which the researcher actively manipulates one or more experimental variables,
then measures the effects of these manipulations on a dependent variable of interest, which can be measured
by either observation or self-report. For example, a researcher interested in knowing which of two
advertisements has a stronger effect on willingness to buy a product could conduct a survey that measures
respondents’ awareness of each ad and willingness to buy, and correlate willingness to buy with awareness of
each ad. Alternately, the researcher could show participants either one ad or the other, with random
assignment, and measure subsequent willingness to buy. The goal of experimentation is to verify that the
observed relationships are causal, not just correlational.
In survey research, experiments have proved to be powerful tools for studying effects of survey question
wording on response. Much of what we know about how to write good survey questions is a product of
experimental research, in which alternative versions of questions intended to measure the same construct are
compared. For example, through experimentation, researchers learned that asking whether respondents
thought a certain behavior should be “allowed” or “forbidden” can produce different response distributions,
even when the alternative wordings logically mean the same thing. There is a very large literature on this
response effects research. Classic works include Sudman and Bradburn (1974), Schuman and Presser (1981),
and Tourangeau, Rips, and Rasinski (2000).
Exhibit 1.1 summarizes how surveys relate to other data collection methods.

Exhibit 1.1 Other Data Collection Methods

Method Strength vs. Surveys Weakness vs. Surveys

Observation Not subject to reporting bias Can’t measure mental states; not efficient for
measuring infrequent behaviors

Depth Can probe freely and go into depth Expensive, poor population coverage
interviews

Focus Can probe freely and go into depth; can Expensive, poor population coverage
groups see social dynamics

Panels Shows changes over time Expensive; a limited number of people will
participate

Experiments Strong test of causation Difficult to do outside lab

15
16
THE COMBINATION OF DISCIPLINES

Survey research is inherently interdisciplinary. Sampling and estimation have a theoretical basis in probability
theory and statistics; to select an efficient sample requires some knowledge of those areas. Data collection
involves persuasion of respondents and then, on some level, social interaction between them and interviewers.
Developing questionnaires and conducting interviews require writing skills to construct questions that elicit
desired information using language that respondents can easily understand and do not find too difficult to
answer. Interviews or questionnaires that use computers or the Internet require programming or other
specialized skills. Very few survey professionals have hands-on expertise in all of these areas, but they do have
a basic understanding of what needs to be done to successfully implement each part of a survey.
Unlike some scientific or scholarly enterprises, surveys are usually a team effort of many people with diverse
skills. One can find examples of surveys designed and implemented by the lone researcher, but they are the
exception. Even if the researcher who formulates the research questions also designs the questionnaire and
analyzes the data, that person will almost always use help in sample design, data collection, and database
construction. Whether a survey is done by a research organization or as a class project, there is division of
labor, coordination of tasks, and management of the costs.
To design and implement a quality survey within available resources, the practitioner relies on a relatively
small number of statistical principles and practical guidelines. The goal of this book is to explain those
fundamentals and illustrate how they are applied to effectively conduct small- to moderate-scale surveys.

17
THE SOCIAL AND SOCIETAL CONTEXT

Survey practices and methodology change as our knowledge grows and our experience increases. And, of
course, just as in every other field, changes in technology affect survey design and implementation. It is also
important to be sensitive to the impact of societal, demographic, and cultural changes on survey practice.
For example, 40 years ago, most national surveys in the United States were conducted only in English. The
proportion of the population that was fluent only in some other language was considered small enough that
the cost of interviewing in those languages could not be justified. The decrease in coverage was too small to be
of concern. Today, any important national survey allows for, at a minimum, Spanish-language interviews and
often interviews in additional languages. This allows inclusion of both people who do not speak English at all
and those who, although they can communicate in English, are much more at ease in their first language.
Likewise, many states, or smaller areas, have large enclaves of non-English-language groups. The California
Health Interview Survey is conducted in several Asian languages, partly for reasons of coverage of the entire
state population, but also because of the need to sample enough people in some of these groups for their
separate analysis.
The need to accommodate different languages is just part of a larger imperative to be aware of different
cultural norms. A more detailed discussion is beyond the scope of this book, but this is just one example of
how changes within society affect how we go about conducting surveys.
General social norms can also have important effects. People’s willingness to allow a stranger into their
homes has greatly changed since the middle of the last century when in-person household interviews were the
norm. Such surveys are still conducted, but the costs and procedures necessary to make them successful have
limited the practice to only the most well-funded programs. Similarly, in recent decades, the rise of
telemarketing and telephone screening devices has affected researchers’ success in contacting members of the
general population by telephone and, once contacted, securing their participation in telephone surveys. The
rise of cell phones and the Internet continues the accelerating technological change that can be a benefit or an
obstacle to conducting surveys. One indicator of the impact of these factors on survey practice can be seen in
the shifting proportions of surveys administered by mail, phone, web, or in person, as shown in Figure 1.1.
Societal changes can occur at different rates in different parts of the population. Some new technologies, for
example, may be initially adopted more heavily by younger than older people or by the more affluent than less
affluent. Such patterns can become fixed or be only a step along the way to wider diffusion through the
population. How important it is for a survey designer to take account of some technical development or
changing social norm may depend on what population the survey targets.

Figure 1.1 Survey Mode by Year

18
Source: Adapted from information in the newsletter Survey Research, a publication of the Survey Research Laboratory, University of Illinois,
Chicago and Urbana-Champaign.

19
ETHICAL TREATMENT OF SAMPLE MEMBERS

A subtext in the discussion of the societal context of a survey is that the survey designer must accommodate
the potential survey respondents. Respondents are volunteers we depend on. They are not obligated to
participate, but the extent to which they agree to participate affects the survey’s success, its cost, and, in some
ways, its quality. Later, we will consider in detail methods to elicit the cooperation of sample members.
However, the methods we can use are not just determined by their effectiveness. There are ethical boundaries
we cannot cross. There are two concepts central to our treatment of respondents: informed consent and
protection of confidentiality.
While we apply extensive efforts to obtain respondents’ cooperation in the survey, the respondents’
agreement must be reasonably informed. This means that we must not mislead respondents as to the nature
and purpose of the research. We must honestly answer their questions about the project, including who is
sponsoring it, its major purposes, the amount of time and effort that will be required of respondents, the
general nature of the subject matter, and the use that will be made of the data. We must not badger or try to
intimidate respondents either into participating or into answering particular questions after they agree to be
interviewed.
Once respondents have agreed to be interviewed, we then assume an obligation to protect the
confidentiality of their answers. This is true whether or not we have explicitly told respondents we will do so.
Results or data sets that permit the identification of individual respondents should never be made available to
others.
These ethical guidelines are recognized by the major professional organizations of survey researchers and
are typically overseen by human subjects review committees at universities and other organizations that engage
in population research.
These obligations are no less applicable when a project is conducted by a class or a large team of researchers
than when a single researcher is involved. In fact, additional cautions may need to be observed in the former
situation because there are additional opportunities for inadvertent breaches of these ethical guidelines when
many people are privy to the sample and the data.
Revealing or discussing an individual respondent’s answers outside of the research group is inappropriate.
Also, it is not proper to recontact survey respondents for purposes not related to the research for which they
originally agreed to participate. The sample list used for a survey should not be made available to others (even
other legitimate researchers) without the additional consent of the respondents. If the data are made available
to another party, all identifiers that would permit linking answers to individuals should be removed.

20
APPROACH AND OVERVIEW

The main concern in designing and conducting a survey is to achieve the research or other data collection
objectives within available resources. Sometimes the initial objectives may be adjusted along the way to
accommodate resource constraints or practical obstacles, but we must not lose sight of them. For example,
assume that during planning, we determine that the budget is insufficient for both the preferred sample size
and number of survey questions. In such a circumstance, it may be necessary either to reduce the sample size
and forego separate analyses of some subgroups that will not be possible with fewer respondents or to reduce
the length of the interview and eliminate topics of secondary interest. But we still proceed toward a clear set of
objectives. Along with those objectives, we need a sense of how good the survey needs to be for our purposes.
“Good” can be specified in a number of ways, which we will return to. Many conceptions of survey quality
involve the idea of accuracy or, conversely, of error. In the next chapter, we consider the types of error surveys
are subject to. Following this, we provide an overview of survey planning and implementation, from
determining the main survey methods to the decisions and tasks at each stage of carrying out the survey. The
means of collecting the data is, along with sample size, the main factor in survey cost. Factors in choosing a
method of data collection are covered in the next chapter. Three chapters each are devoted to sampling and
then to questionnaire development. Following these topics is a chapter on conducting data collection while
controlling the main sources of error in that stage of the survey. The last chapter describes the key post–data
collection activities, including preparing a report of the survey methodology.
Two Methodology Appendixes deal with topics of particular importance, at a greater level of detail:
Questionnaire Evaluation Workshop and Cognitive Interviewing Workshop. Two other Methodology
Appendixes focus on Using Models in Sampling and An Overview of Organization Surveys. Throughout the
book, we emphasize the value of resources available on the Internet. Appendix D lists several sources that are
of particular importance themselves or that provide links to additional information.

21
 TWO
SURVEY ERROR

S
urvey researchers use the term error to refer to “deviations of obtained survey results from those that are true
reflections of the population” (Groves, 1989 p. 6). In this chapter, we provide a framework for understanding:

• The types of error that occur in surveys

• The sources of error in each stage of the survey
• The idea of total survey error

Survey findings are essentially probabilistic generalizations. The sampling process provides a statistical basis
for making inferences from the survey sample to the population. The implementation of the sampling and
data collection are subject to error that can erode the strength of those inferences.
We can visualize how this erosion happens if we imagine the perfect sample survey. The survey design and
questionnaire satisfy all the research goals. A list is available that includes accurate information about every
population member for sampling purposes. The selected sample precisely mirrors all facets of the population
and its myriad subgroups. Each question in the instrument is absolutely clear and captures the dimension of
interest exactly. Every person selected for the sample is contacted and immediately agrees to participate in the
study. The interviewers conduct the interview flawlessly and never—by their behavior or even their mere
presence—affect respondents’ answers. The respondents understand every question exactly as the researcher
intended, know all the requested information, and always answer truthfully and completely. Their responses
are faithfully recorded and entered, without error, into a computer file. The resulting data set is a model of
validity and reliability.
Except for trivial examples, we cannot find such a paragon. Each step in conducting a survey has the
potential to move us away from this ideal, sometimes a little, sometimes a great deal. Just as all the processes
and players in our survey can contribute to obtaining accurate information about the target population, so can
each reduce that accuracy. We speak of these potential reductions in accuracy as sources of survey error.1
Every survey contains survey errors, most of which cannot be totally eliminated within the limits of our
resources.
To design and implement an effective survey, we must recognize which sources of error pose the greatest
threat to survey quality and how the design and implementation of the survey will affect its exposure to various
sources of error. The critical sources of error will differ depending on the survey objectives, topic, population,
and its methods of sampling and data collection.

22
TYPES AND SOURCES OF ERROR

When we work with survey data, we typically want our data to provide an accurate representation of the
broader population. For example, assume that your friend is running for election to the local school district
board, and you agree to do a survey to help her learn what people think about the issues. If your data show
that 64% of the respondents want the school district to place more emphasis on reading skills, you want this
64% figure to be an accurate reflection of feelings among the population at large. There are three reasons why
it might not be:

• Sampling error, also known as sample variance

• Sample bias
• Nonsampling error

Sampling Error

Sampling error refers to the fact that samples don’t always reflect a population’s true characteristics, even if
the samples are drawn using fair procedures. For example, if you flip a coin 10 times, then 10 more times,
then 10 more times, and so on, you won’t always get five “heads” out of 10 flips. The percentage of “heads” in
any given sample will be affected by chance variation. In the same way, if you ask 100 people whether the
school district should cut other programs to place more emphasis on reading, the percentage that answer
affirmatively will be affected by chance variation in the composition of the sample.
The level of sampling error is controlled by sample size. Ultimately, if the sample encompasses the entire
population (i.e., the sample is a complete census of the population), then sampling error would be zero,
because the sample equals the population. More generally, as samples get larger and larger, the distribution of
possible sample outcomes gets tighter and tighter around the true population figure, as long as the samples are
not biased in some way. To put it another way, larger samples have less chance of producing results that are
uncharacteristic of the population as a whole. We expand upon these points in Chapters 5, 6, and 7 and show
how to set sample size based on acceptable levels of sampling error.

Sample Bias

Sample bias refers to the possibility that members of a sample differ from the larger population in some
systematic fashion. For example, if you are conducting a survey related to a school bond election, and you limit
interviews to people with school children, then your results may not reflect opinions among voters at large; if
nothing else, people with school children may be more likely to support bonds to pay for school
improvements. Similarly, if your sample contains a disproportionately large percentage of older people because
you didn’t make the callbacks needed to find younger people, or if your sampling procedure underrepresents
married versus single people, then you are likely to have sample bias.
Sample bias can arise in three general ways. Coverage bias will occur if some segment of the population is
improperly excluded from consideration in the sample or is not available through the method employed in the
research. Conducting interviews near school grounds with parents as they arrive to pick up their children and
implicitly limiting the sample to people with school children is an example of potential coverage bias if the
population of interest is the broader electorate. Likewise, conducting a telephone survey implicitly limits the
sample to people with telephones, and conducting an online survey implicitly limits the sample to people with
online access, either of which will exclude some low-income households, resulting in potential coverage bias if
these households differ from the broader population on the items being measured.
Selection bias will occur if some population groups are given disproportionately high or low chances of
selection. For example, if you conduct telephone interviews using conventional household landlines, and you
take the first adult to answer the phone, then married people will only have one chance in two of being
selected within their household, while single people have one chance in one, and married people will thus tend
to be underrepresented in a sample of possible voters.
Finally, even if the sample is fairly drawn, nonresponse bias will occur if failure to respond is
disproportionate across groups. For example, if you send e-mails inviting people to click through to a survey
on issues related to the local schools, people who are less interested in those issues are probably less likely to

23
respond. If one of the topics on the survey concerns willingness to pay higher taxes to support the schools,
there may be substantial differences between respondents and nonrespondents on this topic, resulting in
nonresponse bias.
Unlike sampling error, sample bias is not controlled by sample size. Increasing the sample size does nothing
to remove systematic biases. Rather, sample bias is controlled by defining the population of interest prior to
drawing the sample, attempting to maximize population coverage, selecting a sample that fairly represents the
entire population, and obtaining data from as much of the selected sample as possible.

Nonsampling Error

Nonsampling error consists of all error sources unrelated to the sampling of respondents. Sources of
nonsampling error include interviewer error related to the administration of the survey, response error related
to the accuracy of response as given, and coding error related to the accuracy of response as recorded. All of
these may result in either random or systematic errors in the data. If random, they will add to random
sampling error to increase the level of random variation in the data. If systematic, they will be a source of bias
in the data, much like sample bias.
One source of interviewer error is cheating; the interviewer fails to administer the questionnaire, or portions
of the questionnaire, and simply fabricates the data. Cheating is controlled through some form of validation,
in which a supervisor or a third party verifies that the interview was conducted and key questions were asked.
This is typically done by telephone; when the interview is conducted, whether face-to-face or by phone, the
respondent is asked for a telephone number and told that a supervisor may call to verify the interview for
quality control purposes. Typically, 10% to 20% of interviews are validated. For surveys that are done on
location, validation may be done on site by a supervisor. For telephone surveys that are conducted in
centralized telephone interviewing facilities, validation may be replaced by monitoring interviews as they are
conducted, so the interviewer never knows when the supervisor will be listening in.
Interviewer error also may stem from question administration error, in which the interviewer does
something other than read the question as intended, or probing error. Regarding errors in question
administration, the interviewer may skip a question that should not be skipped, ask a question that should be
skipped, omit part of the question, misread the question, add to the question, or add some vocal inflection or
body language that conveys a point of view. Any of these actions, of course, may influence the response. With
respect to probing, if the respondent expresses confusion about the question or gives an inadequate answer,
interviewers are usually instructed to give neutral probes that do not presume an answer or convey a point of
view; for example:

• If the respondent says he or she did not follow the question, simply reread it.
• If the respondent asks about the meaning of some term, read a preset definition or say “whatever it means
to you.”
• If the respondent gives an answer to an open-ended question that is insufficiently specific, ask, “Can you
be more specific?” or “When you say X, what do you mean?”
• If the respondent gives an answer to a closed-ended question that does not match the categories, ask,
“Would you say [reread categories]?”

If the interviewer fails to probe an answer that should be probed, probes in a manner that changes the
question, presumes a certain answer, or leads the respondent toward a certain answer, then any of these
actions may influence the quality of response.
Question administration error and probing error are controlled through interviewer training. When first
hired, interviewers receive training that covers general principles such as “ask the question as written” and
when and how to probe. Subsequently, interviewers are given specific instructions and training for any survey
that they work on. This training normally includes at least one practice interview conducted under a
supervisor’s direction. The better the training, the lower the exposure to errors of this type. Also, data
collection may be spread across multiple interviewers, so any idiosyncrasies among interviewers manifest
themselves as variance across interviewers rather than bias from a single data source.
The next general source of nonsampling error is response error. Response error may occur for at least three
different reasons. Comprehension error will occur if the respondent does not understand a question correctly
or if different respondents understand it in different ways. For example, in a survey of school district issues, if

24
you ask respondents what priority they would give to programs for the development of reading skills, will they
know what you mean by programs and skills? Will some of them think about instruction for preschoolers while
others think about older students with reading deficiencies? Will this affect their answers? Comprehension
errors often stem from questions that are poorly thought out or poorly written, and they are controlled
through careful question design as well as pretesting the questionnaire to ensure that respondents understand
the questions as intended.
Knowledge error may occur if respondents do not know the answer to the question, or cannot recall it
accurately, but still provide an answer to the best of their ability. For example, if you ask parents how often
their children fail to eat all of their lunches at school, will parents know the answer? If you ask parents how
many hours per evening their children have spent on homework over the past month, will they be able to
recall accurately over this time frame? As with other forms of nonsampling error, knowledge errors may be
either random or systematic and hence a source of either increased variance or bias in the data. These errors
are generally controlled by getting the right respondent, screening for knowledge if appropriate, writing
questions that are realistic about what people might know, and matching the time frame of the question to a
plausible recall period for the phenomenon in question.
Reporting error may occur if respondents hesitate to provide accurate answers, perhaps because certain
responses are socially desirable or undesirable, or perhaps because respondents wish to maintain privacy. For
example, if you ask parents to report their children’s grades, they may be tempted to give answers that are
higher than the truth. If you ask parents to report how often their children have been given disciplinary
actions at school, they may underreport. Reporting errors may be random but often are a source of systematic
bias for sensitive questions These errors are controlled in a variety of ways, including professionalism in the
survey to convince respondents that you care whether they answer but not what they answer, or using self-
report methods so the respondent does not have to present himself or herself to an interviewer.
The final source of nonsampling error is coding error. For open-ended questions, the coder may
misinterpret the answer and assign the wrong numerical code to be entered in the data. For categorical
questions, the interviewer (or respondent) may mark the wrong category. Also, for paper questionnaires,
interviewers or coders may select the correct code or category, but at the data entry stage, someone may enter
data incorrectly. To control such errors, open-ended questions may be assigned to at least two coders to
identify and resolve inconsistencies. Likewise, data may be double-entered to identify and resolve
inconsistencies. Also, questionnaires are edited to verify that the answers appear to be correctly recorded and
to check for interitem consistency where applicable (this process is automated for computer-based
questionnaires). In some cases, respondents may be recontacted to resolve apparent errors or inconsistencies.
Overall, nonsampling error is controlled, to the extent possible, by doing a good job of training and
supervising interviewers, using good questionnaire design, and exercising good control over coding and data
entry.

25
Another random document with
no related content on Scribd:
Text Page of Didot’s Racine, Paris, 1801 256
Firmin Didot. From Engraving by Pierre Gustave Eugène Staal
(1817–1882) 256
William Morris. From Portrait by G. F. Watts, R. A., in the National
Portrait Gallery, London. Painted in 1880 258
Sir Edward Burne-Jones, Bart. From a Photograph at the British
Museum 260
Text Page of Kelmscott Chaucer, 1896 262
Title Page of Doves Bible, London, 1905 265
Text Page of Doves Bible, London, 1905 267
The Sala Michelangiolo, in the Laurenziana Library, Florence 276
Dott. Comm. Guido Biagi, in 1924 278
Vestibule of the Laurenziana Library, Florence 280
Miniature Page from the Biblia Amiatina, R. Lau. Bibl. Cod. Amiatinus
I 288
Antonio Magliabecchi 293
Library Slips used by George Eliot while working on Romola in
Magliabecchian Library, Florence 296
 CHAPTER I

In Quest of the Perfect Book

 I
IN QUEST OF THE PERFECT BOOK

“Here is a fine volume,” a friend remarked, handing me a copy of The Ideal

Book, written and printed by Cobden-Sanderson at the Doves Press.
“It is,” I assented readily, turning the leaves, and enjoying the composite
beauty of the careful typography, and the perfect impression upon the soft,
handmade paper with the satisfaction one always feels when face to face with a
work of art. “Have you read it?”
“Why—no,” he answered. “I picked it up in London, and they told me it
was a rare volume. You don’t necessarily read rare books, do you?”
My friend is a cultivated man, and his attitude toward his latest acquisition
irritated me; yet after thirty years of similar disappointments I should not have
been surprised. How few, even among those interested in books, recognize the
fine, artistic touches that constitute the difference between the commonplace
and the distinguished! The volume under discussion was written by an
authority foremost in the art of bookmaking; its producer was one of the few
great master-printers and binders in the history of the world; yet the only
significance it possessed to its owner was the fact that some one in whom he
had confidence had told him it was rare! Being rare, he coveted the treasure,
and acquired it with no greater understanding than if it had been a piece of
Chinese jade.
“What makes you think this is a fine book?” I inquired, deliberately
changing the approach.
He laughed consciously. “It cost me nine guineas—and I like the looks of
it.”
Restraint was required not to say something that might have affected our
friendship unpleasantly, and friendship is a precious thing.
“Do something for me,” I asked quietly. “That is a short book. Read it
through, even though it is rare, and then let us continue this conversation we
have just begun.”
 A few days later he invited me to dine with him at his club. “I asked you
here,” he said, “because I don’t want any one, even my family, to hear what I
am going to admit to you. I have read that book, and I’d rather not know what
you thought of my consummate ignorance of what really enters into the
building of a well-made volume—the choice of type, the use of decoration, the
arrangement of margins. Why, bookmaking is an art! Perhaps I should have
known that, but I never stopped to think about it.”
One does have to stop and think about a well-made book in order to
comprehend the difference between printing that is merely printing and that
which is based upon art in its broadest sense and upon centuries of precedent.
It does require more than a gleam of intelligence to grasp the idea that the
basis of every volume ought to be the thought expressed by the writer; that the
type, the illustrations, the decorations, the paper, the binding, simply combine
to form the vehicle to convey that expression to the reader. When, however,
this fact is once absorbed, one cannot fail to understand that if these various
parts, which compositely comprise the whole, fail to harmonize with the
subject and with each other, then the vehicle does not perform its full and
proper function.
I wondered afterward if I had not been a bit too superior in my attitude
toward my friend. As a matter of fact, printing as an art has returned to its
own only within the last quarter-century. Looking back to 1891, when I began
to serve my apprenticeship under John Wilson at the old University Press in
Cambridge, Massachusetts, the broadness of the profession that I was
adopting as my life’s work had not as yet unfolded its unlimited possibilities.
At that time the three great American printers were John Wilson, Theodore L.
De Vinne, and Henry O. Houghton. The volumes produced under their
supervision were perfect examples of the best bookmaking of the period, yet
no one of these three men looked upon printing as an art. It was William
Morris who in modern times first joined these two words together by the
publication of his magnificent Kelmscott volumes. Such type, such
decorations, such presswork, such sheer, composite beauty!
This was in 1895. Morris, in one leap, became the most famous printer in
the world. Every one tried to produce similar volumes, and the resulting
productions, made without appreciating the significance of decoration
combined with type, were about as bad as they could be. I doubt if, at the
present moment, there exists a single one of these sham Kelmscotts made in
America that the printer or the publisher cares to have recalled to him.
 When the first flair of Morris’ popularity passed away, and his volumes
were judged on the basis of real bookmaking, they were classified as
marvelously beautiful objets d’art rather than books—composites of Burne-
Jones, the designer, and William Morris, the decorator-printer, co-workers in
sister arts; but from the very beginning Morris’ innovations showed the world
that printing still belonged among the fine arts. The Kelmscott books awoke in
me an overwhelming desire to put myself into the volumes I produced. I
realized that no man can give of himself beyond what he possesses, and that to
make my ambition worth accomplishing I must absorb and make a part of
myself the beauty of the ancient manuscripts and the early printed books. This
led me to take up an exhaustive study of the history of printing.
 JOHN GUTENBERG, c. 1400–1468
From Engraving by Alphonse Descaves
Bibliothèque Nationale, Paris

Until then Gutenberg’s name, in my mind, had been preëminent. As I

proceeded, however, I came to know that he was not really the “inventor” of
printing, as I had always thought him to be; that he was the one who first
foresaw the wonderful power of movable types as a material expression of the
thought of man, rather than the creator of anything previously unknown. I
discovered that the Greeks and the Romans had printed from stamps centuries
earlier, and that the Chinese and the Koreans had cut individual characters in
metal.
 I well remember the thrill I experienced when I first realized—and at the
time thought my discovery was original!—that, had the Chinese or the
Saracens possessed Gutenberg’s wit to join these letters together into words,
the art of printing must have found its way to Constantinople, which would
have thus become the center of culture and learning in the fifteenth century.

From this point on, my quest seemed a part of an Arabian Nights’ tale.
Cautiously opening a door, I would find myself in a room containing treasures
of absorbing interest. From this room there were doors leading in different
directions into other rooms even more richly filled; and thus onward, with
seemingly no end, to the fascinating rewards that came through effort and
perseverance.
Germany, although it had produced Gutenberg, was not sufficiently
developed as a nation to make his work complete. The open door led me away
from Germany into Italy, where literary zeal was at its height. The life and
customs of the Italian people of the fifteenth century were spread out before
me. In my imagination I could see the velvet-gowned agents of the wealthy
patrons of the arts searching out old manuscripts and giving commissions to
the scribes to prepare hand-lettered copies for their masters’ libraries. I could
mingle with the masses and discover how eager they were to learn the truth in
the matter of religion, and the cause and the remedies of moral and material
evils by which they felt themselves oppressed. I could share with them their
expectant enthusiasm and confidence that the advent of the printing press
would afford opportunity to study description and argument where previously
they had merely gazed at pictorial design. I could sense the desire of the
people for books, not to place in cabinets, but to read in order to know; and I
could understand why workmen who had served apprenticeships in Germany
so quickly sought out Italy, the country where princes would naturally become
patrons of the new art, where manuscripts were ready for copy, and where a
public existed eager to purchase their products.
While striving to sense the significance of the conflicting elements I felt
around me, I found much of interest in watching the scribes fulfilling their
commissions to prepare copies of original manuscripts, becoming familiar for
the first time with the primitive methods of book manufacture and
distribution. A monastery possessed an original manuscript of value. In its
scriptorium (the writing office) one might find perhaps twenty or thirty monks
seated at desks, each with a sheet of parchment spread out before him, upon
which he inscribed the words that came to him in the droning, singsong voice
of the reader selected for the duty because of his familiarity with the subject
matter of the volume. The number of desks the scriptorium could accommodate
determined the size of this early “edition.”
When these copies were completed, exchanges were made with other
monasteries that possessed other original manuscripts, of which copies had
been made in a similar manner. I was even more interested in the work of the
secular scribes, usually executed at their homes, for it was to these men that
the commissions were given for the beautiful humanistic volumes. As they had
taken up the art of hand lettering from choice or natural aptitude instead of as
a part of monastic routine, they were greater artists and produced volumes of
surpassing beauty. A still greater interest in studying this art of hand lettering
lay in the knowledge that it soon must become a lost art, for no one could
doubt that the printing press had come to stay.
Then, turning to the office of Aldus, I pause for a moment to read the
legend placed conspicuously over the door:
Whoever thou art, thou art earnestly requested by Aldus to state thy business briefly
and to take thy departure promptly. In this way thou mayest be of service even as was
Hercules to the weary Atlas, for this is a place of work for all who may enter
 ALDUS MANUTIUS, 1450–1515
From Engraving at the British Museum

But inside the printing office I find Aldus and his associates talking of
other things than the books in process of manufacture. They are discussing the
sudden change of attitude on the part of the wealthy patrons of the arts who,
after welcoming the invention of printing, soon became alarmed by the
enthusiasm of the people, and promptly reversed their position. No wonder
that Aldus should be concerned as to the outcome! The patrons of the arts
represented the culture and wealth and political power of Italy, and they now
discovered in the new invention an actual menace. To them the magnificent
illuminated volumes of the fifteenth century were not merely examples of
decoration, but they represented the tribute that this cultured class paid to the
thought conveyed, through the medium of the written page, from the author
to the world. This jewel of thought they considered more valuable than any
costly gem. They perpetuated it by having it written out on parchment by the
most accomplished scribes; they enriched it by illuminated embellishments
executed by the most famous artists; they protected it with bindings in which
they actually inlaid gold and silver and jewels. To have this thought cheapened
by reproduction through the commonplace medium of mechanical printing
wounded their æsthetic sense. It was an expression of real love of the book
that prompted Bisticci, the agent of so powerful a patron as the Duke of
Urbino, to write of the Duke’s splendid collection in the latter part of the
fifteenth century:
In that library the books are all beautiful in a superlative degree, and all written
by the pen. There is not a single one of them printed, for it would have been a shame to
have one of that sort.
Aldus is not alarmed by the solicitude of the patrons for the beauty of the
book. He has always known that in order to exist at all the printed book must
compete with the written volume; and he has demonstrated that, by supplying
to the accomplished illuminators sheets carefully printed on parchment, he can
produce volumes of exquisite beauty, of which no collector need be ashamed.
Aldus knows that there are other reasons behind the change of front on the
part of the patrons. Libraries made up of priceless manuscript volumes are
symbols of wealth, and through wealth comes power. With the multiplication
of printed books this prestige will be lessened, as the masses will be enabled to
possess the same gems of thought in less extravagant and expensive form. If,
moreover, the people are enabled to read, criticism, the sole property of the
scholars, will come into their hands, and when they once learn self-reliance
from their new intellectual development they are certain to attack dogma and
political oppression, even at the risk of martyrdom. The princes and patrons
of Italy are intelligent enough to know that their self-centered political power
is doomed if the new art of printing secures a firm foothold.
What a relief to such a man as Aldus when it became fully demonstrated
that the desire on the part of the people to secure books in order to learn was
too great to be overcome by official mandate or insidious propaganda! With
what silent satisfaction did he settle back to continue his splendid work! The
patrons, in order to show what a poor thing the printed book really was, gave
orders to the scribes and the illuminators to prepare volumes for them in such
quantities that the art of hand lettering received a powerful impetus, as a result
of which the hand letters themselves attained their highest point of perfection.
This final struggle on the part of the wealthy overlords resulted only in
redoubling the efforts of the artist master-printers to match the beauty of the
written volumes with the products from their presses.
These Arabian Nights’ experiences occupied me from 1895, when Morris
demonstrated the unlimited possibilities of printing as an art, until 1901, when
I first visited Italy and gave myself an opportunity to become personally
acquainted with the historical landmarks of printing, which previously I had
known only from study. In Florence it was my great good fortune to become
intimately acquainted with the late Doctor Guido Biagi, at that time librarian
of the Laurenziana and the Riccardi libraries, and the custodian of the Medici,
the Michelangelo, and the da Vinci archives. I like to think of him as I first saw
him then, sitting on a bench in front of one of the carved plutei designed by
Michelangelo, in the wonderful Sala di Michelangiolo in the Laurenziana Library,
studying a beautifully illuminated volume resting before him, which was
fastened to the desk by one of the famous old chains. He greeted me with an
old-school courtesy. When he discovered my genuine interest in the books he
loved, and realized that I came as a student eager to listen to the master’s word,
his face lighted up and we were at once friends.
 Dott. Comm. GUIDO BIAGI
Seated at one of the plutei in the
Laurenziana Library, Florence (1906)

In the quarter of a century which passed from this meeting until his death
we were fellow-students, and during that period I never succeeded in
exhausting the vast store of knowledge he possessed, even though he gave of
it with the freest generosity. From him I learned for the first time of the far-
reaching influence of the humanistic movement upon everything that had to
do with the litteræ humaniores, and this new knowledge enabled me to crystallize
much that previously had been fugitive. “The humanist,” Doctor Biagi
explained to me, “whether ancient or modern, is one who holds himself open
to receive Truth, unprejudiced as to its source, and—what is more important
—after having received Truth realizes his obligation to the world to give it out
again, made richer by his personal interpretation.”
This humanistic movement was the forerunner and the essence of the
Renaissance, being in reality a revolt against the barrenness of mediævalism.
Until then ignorance, superstition, and tradition had confined intellectual life
on all sides, but the little band of humanists, headed by Petrarch, put forth a
claim for the mental freedom of man and for the full development of his
being. As a part of this claim they demanded the recognition of the rich
humanities of Greece and Rome, which were proscribed by the Church. If this
claim had been postponed another fifty years, the actual manuscripts of many
of the present standard classics would have been lost to the world.
The significance of the humanistic movement in its bearing upon the
Quest of the Perfect Book is that the invention of printing fitted exactly into
the Petrarchian scheme by making it possible for the people to secure volumes
that previously, in their manuscript form, could be owned only by the wealthy
patrons. This was the point at which Doctor Biagi’s revelation and my previous
study met. The Laurenziana Library contains more copies of the so-called
humanistic manuscripts, produced in response to the final efforts on the part
of patrons to thwart the increasing popularity of the new art of printing, than
any other single library. Doctor Biagi proudly showed me some of these
treasures, notably Antonio Sinibaldi’s Virgil. The contrast between the hand
lettering in these volumes and the best I had ever seen before was startling.
Here was a hand letter, developed under the most romantic and dramatic
conditions, which represented the apotheosis of the art. The thought flashed
through my mind that all the types in existence up to this point had been
based upon previous hand lettering less beautiful and not so perfect in
execution.
“Why is it,” I demanded excitedly, “that no type has ever been designed
based upon this hand lettering at its highest point of perfection?”
 HAND-WRITTEN HUMANISTIC CHARACTERS
From Sinibaldi’s Virgil, 1485
Laurenziana Library, Florence (12 × 8 inches)

Doctor Biagi looked at me and shrugged his shoulders. “This, my friend,”

he answered, smiling, “is your opportunity.”
At this point began one of the most fascinating and absorbing adventures
in which any one interested in books could possibly engage. At some time, I
suppose, in the life of every typographer comes the ambition to design a
special type, so it was natural that the idea contained in Doctor Biagi’s remark
should suggest possibilities which filled me with enthusiasm. I was familiar
with the history of the best special faces, and had learned how difficult each
ambitious designer had found the task of translating drawings into so rigid a
medium as metal; so I reverted soberly and with deep respect to the subject of
type design from the beginning.
In studying the early fonts of type, I found them exact counterfeits of the
best existing forms of hand lettering at that time employed by the scribes. The
first Italic font cut by Aldus, for instance, is said to be based upon the thin,
inclined handwriting of Petrarch. The contrast between these slavish copies of
hand-lettered models and the mechanical precision of characters turned out by
modern type founders made a deep impression. Of the two I preferred the
freedom of the earliest types, but appreciated how ill adapted these models
were to the requirements of typography. A hand-lettered page, even with the
inevitable irregularities, is pleasing because the scribe makes a slight variation
in forming the various characters. When, however, an imperfect letter is cut in
metal, and repeated many times upon the same page, the irregularity forces
itself unpleasantly upon the eye. Nicolas Jenson was the first to realize this,
and in his famous Roman type he made an exact interpretation of what the
scribe intended to accomplish in each of the letters, instead of copying any
single hand letter, or making a composite of many hand designs of the same
character. For this reason the Jenson type has not only served as the basis of
the best standard Roman fonts down to the present time, but has also proved
the inspiration for later designs of distinctive type faces, such as William
Morris’ Golden type, and Emery Walker’s Doves type.
 Specimen Page of proposed Edition of Dante. To be
printed by Bertieri, of Milan, in Humanistic Type (8¼ × 6)

William Morris’ experience is an excellent illustration of the difficulties a

designer experiences. He has left a record of how he studied the Jenson type
with great care, enlarging it by photography, and redrawing it over and over
again before he began designing his own letter. When he actually produced his
Golden type the design was far too much inclined to the Gothic to resemble
the model he selected. His Troy and Chaucer types that followed showed the
strong effect of the German influence that the types of Schoeffer, Mentelin,
and Gunther Zainer made upon him. The Doves type is based flatly upon the
Jenson model; yet it is an absolutely original face, retaining all the charm of the
model, to which is added the artistic genius of the designer. Each receives its
personality from the understanding and interpretation of the creator (pages 22,
23).

Jenson’s Roman Type.

From Cicero: Rhetorica, Venice, 1470 (Exact size)
 Emery Walker’s Doves Type.
From Paradise Regained, London, 1905 (Exact size)

From this I came to realize that it is no more necessary for a type designer
to express his individuality by adding or subtracting from his model than for a
portrait painter to change the features of his subject because some other artist
has previously painted it. Wordsworth once said that the true portrait of a man
shows him, not as he looks at any one moment of his life, but as he really
looks all the time. This is equally true of a hand letter, and explains the vast
differences in the cut of the same type face by various foundries and for the
typesetting machines. All this convinced me that, if I were to make the
humanistic letters the model for my new type, I must follow the example of
Emery Walker rather than that of William Morris.
 During the days spent in the small, cell-like alcove which had been turned
over for my use in the Laurenziana Library, I came so wholly under the
influence of the peculiar atmosphere of antiquity that I felt myself under an
obsession of which I have not been conscious before or since. My enthusiasm
was abnormal, my efforts tireless. The world outside seemed very far away, the
past seemed very near, and I was indifferent to everything except the task
before me. This curious experience was perhaps an explanation of how the
monks had been able to apply themselves so unceasingly to their prodigious
labors, which seem beyond the bounds of human endurance.
My work at first was confined to a study of the humanistic volumes in the
Laurenziana Library, and the selection of the best examples to be taken as final
models for the various letters. From photographed reproductions of selected
manuscript pages, I took out fifty examples of each letter. Of these fifty,
perhaps a half-dozen would be almost identical, and from these I learned the
exact design the scribe endeavored to repeat. I also decided to introduce the
innovation of having several characters for certain letters that repeated most
frequently, in order to preserve the individuality of the hand lettering, and still
keep my design within the rigid limitations of type. Of the letter e, for instance,
eight different designs were finally selected; there were five a’s, two m’s, and so
on (see illustration at page 32).
After becoming familiar with the individual letters as shown in the
Laurenziana humanistic volumes, I went on to Milan and the Ambrosiana
Library, with a letter from Doctor Biagi addressed to the librarian, Monsignor
Ceriani, explaining the work upon which I was engaged, and seeking his co-
operation. It would be impossible to estimate Ceriani’s age at that time, but he
was very old. He was above middle height, his frame was slight, his eyes
penetrating and burning with a fire that showed at a glance how affected he
was by the influence to which I have already referred. His skin resembled in
color and texture the very parchment of the volumes he handled with such
affection, and in his religious habit he seemed the embodiment of ancient
learning.
After expressing his deep interest in my undertaking, he turned to a
publication upon which he himself was engaged, the reproduction in facsimile
of the earliest known manuscript of Homer’s Iliad. The actual work on this, he
explained, was being carried on by his assistant, a younger priest whom he