You are on page 1of 14

34

The Psychology of Survey


Response
Norbert Schwarz

Since the beginning of public opinion surveys, 1977; Tourangeau, 1984; Strack & Martin,
researchers have been aware that minor 1987). Respondents’ first need is to under-
changes in question wording, format, or order stand the question posed to determine what
can profoundly affect respondents’ answers information they are asked to provide. If the
(Cantril, 1944; Payne, 1951). Nevertheless, question is an attitude question, they may
the field has long been characterized by two either retrieve a previously formed judgment
largely separate streams: rigorous theories of from memory or form a judgment on the spot.
sampling on the one hand, and an experience Because a previously formed judgment that
based ‘art of asking questions’ on the other fits the specifics of the question asked is rarely
hand. This changed since the early 1980s, accessible, judgment formation during the
thanks to a collaboration of survey method- interview setting is the most common case. To
ologists and cognitive psychologists, who form a judgment, respondents need to retrieve
brought theories of language comprehension, relevant information from memory. Usually,
memory, and judgment to bear on the response they will also need to retrieve or construct
process (for reviews see Sudman, Bradburn, & some standard against which the attitude
Schwarz, 1996; Schwarz, 1999; Tourangeau, object is evaluated. Once a ‘private’ judgment
Rips, & Rasinski, 2000; and the contributions is formed in respondents’ minds, they have to
in Sirken et al., 1999). This chapter highlights communicate it to the researcher. Unless the
key lessons learned, with particular attention question is asked in an open response format,
to the cognitive and communicative processes they need to format their judgment to fit
underlying answers to attitude questions. For the response alternatives. Finally, respondents
a complementary review of the processes may wish to edit their response before they
underlying behavioral reports, see Schwarz communicate it, due to self-presentation and
and Oyserman (2001). social desirability concerns.
Answering a survey question entails several The following sections review these tasks
distinct tasks (Cannell, Marquis, & Laurent, in more detail. Throughout, the emphasis is
THE PSYCHOLOGY OF SURVEY RESPONSE 375

on the theoretical conceptualization of the without saying’ (such as, ‘taking a shower’
underlying processes, with illustrations from in the above example). Third, a maxim of
selected experimental results. manner holds that the contribution should
be clear rather than obscure, ambiguous or
wordy. This maxim entails an interpretabil-
QUESTION COMPREHENSION ity presumption: research participants can
assume that the researcher ‘chose his wording
Survey textbooks rightly emphasize that so they can understand what he meant—and
researchers should avoid unfamiliar terms and can do so quickly’ (Clark & Schober, 1992,
complex syntax in writing survey questions p. 27). Hence, the most obvious meaning
(see Bradburn, Sudman, & Wansink, 2004, seems likely to be the correct one; if an
for advice; and Belson, 1981, for common obvious meaning is not apparent, respon-
problems). This focus on the literal or dents may consult the immediate context to
semantic meaning of the question, however, determine one. The researcher’s contributions
misses an important point: Language compre- to the conversation include formal aspects
hension is not about words per se, but about of questionnaire design, like the response
speaker meaning (Clark & Schober, 1992). alternatives, and respondents draw on these
When asked, ‘What have you done today?’ features in interpreting the question. Finally,
respondents surely understand the words. But a maxim of quality enjoins speakers not
to provide an answer, they need to determine to say anything they believe to be false
what kind of activities the researcher is or lack adequate evidence for. These rules
interested in—should they report that they of cooperative conversational conduct are
had a cup of coffee or took a shower, or is essential for understanding how respondents
that not what the researcher had in mind? make sense of the questions asked of them (for
Providing an informative answer requires reviews see Clark & Schober, 1992; Schwarz,
extensive inferences about the questioner’s 1996).
likely intention to determine the pragmatic
meaning of the question.
Question context
To draw these inferences, respondents rely
on the tacit assumptions of conversational In natural conversations, we draw on the con-
conduct described by Paul Grice (1975), a tent of the ongoing conversation to interpret
philosopher of language. According to Grice’s the next utterance, as licensed by the maxim
analysis, conversations proceed according to of relation. The same holds for survey inter-
a co-operativeness principle, which can be views, where respondents draw on the content
expressed in the form of four maxims. First, of preceding questions. Hence, a question
a maxim of relation enjoins speakers to about ‘drugs’ acquires a different meaning in
make their contribution relevant to the aims the context of health questions rather than
of the ongoing conversation. This maxim crime questions. As a particularly informa-
licenses the use of contextual information tive example, consider research in which
in question interpretation and invites respon- respondents are asked to report their opinion
dents to relate the question to the context about a highly obscure, or even completely
of the ongoing exchange. Second, a maxim fictitious, issue, such as the ‘Agricultural
of quantity requests speakers to make their Trade Act of 1978’ (e.g., Schuman & Presser,
contribution as informative as is required, 1981; Bishop, Oldendick, & Tuchfarber,
but not more informative than is required. 1986). Apparently confirming public opin-
It invites respondents to provide information ion researchers’ nightmares, about 30% of
the questioner seems interested in, rather than respondents report an opinion on such topics,
other information that may come to mind. presumably in the absence of any knowledge.
Moreover, it discourages the reiteration of Yet, their answers may be more meaningful
information provided earlier, or that ‘goes than assumed.
376 THE SAGE HANDBOOK OF PUBLIC OPINION RESEARCH

From a conversational point of view, the different concepts in mind, requiring different
sheer fact that a question about some issue and non-redundant answers. Again, the same
is asked presupposes that it exists—or else holds for survey interviews. Strack et al.
asking the question would violate every norm (1991) asked respondents to report on their
of conversational conduct. Respondents have general happiness and satisfaction with life.
no reason to believe that the researcher When these questions were presented as the
would ask a meaningless question and hence last and first questions of two separate ques-
draw on contextual information to make tionnaires, presented by different researchers,
sense of it. Once they have assigned a the two questions correlated, with r = 0.96.
particular meaning to the issue, they have no But when both questions were presented as the
difficulty reporting a subjectively meaningful last two questions of the first questionnaire,
opinion. Supporting this assumption, Strack, respondents differentiated between happiness
Schwarz, and Wänke (1991) observed that and satisfaction, resulting in different mean
German university students reported different reports and a significantly lower correlation
attitudes toward the introduction of a fictitious of r = 0.75. Hence, closely related questions
‘educational contribution,’ depending on the can elicit a differential interpretation when
content of a preceding question. Some presented in close proximity, reflecting that
students were asked to estimate the average we don’t assume that people ask the same
tuition fees that students have to pay at thing twice. This can attenuate the internal
US universities (in contrast to Germany, consistency of multi-item scales.
where university education is free), whereas In natural conversations, we further draw
others had to estimate the amount of money on our knowledge about the speaker to
that the Swedish government pays every infer the intended meaning of a question.
student as financial support. As expected, Again, the same applies to research situations,
respondents inferred that the fictitious ‘edu- where the researcher’s affiliation may provide
cational contribution’ pertained to students important clues. Norenzayan and Schwarz
having to pay money when it followed the (1999) presented respondents with newspaper
tuition question, but to students receiving accounts of mass murders and asked them
money when it followed the financial support to explain why the mass murder occurred.
question. Accordingly, they reported a more Depending on conditions, the questionnaire
favorable attitude toward the introduction of was printed on the letterhead of an alleged
an ‘educational contribution’ in the former ‘Institute for Personality Research’ or ‘Insti-
than in the latter case—hardly a meaningless tute for Social Research.’As expected, respon-
response. dents’ explanations showed more attention to
While the maxim of relation licenses the personality variables or to social-contextual
use of preceding questions in interpreting variables, depending on whether they thought
subsequent ones, the maxim of quantity the researcher was a personality psychologist
encourages respondents not to reiterate infor- or a social scientist. As requested by norms
mation they have already provided earlier. of conversational conduct, respondents took
This results in unique meaning shifts when the researcher’s affiliation into account to
closely related questions with overlapping determine which information would be most
meaning are presented. In daily life, one relevant to the questioner’s likely epistemic
may give the exact same answer when a interest.
friend asks how happy one is with life In sum, respondents make systematic use
and another asks how satisfied one is— of contextual information. In face-to-face
after all, these are closely related concepts and telephone interviews, this information is
and one’s friends may just use different limited to what has been provided earlier in
words. But if the same friend asks first the interview. In self-administered surveys,
how happy one is and then follows up however, respondents can read ahead and can
with how satisfied one is, he/she may have peruse subsequent questions in an effort to
THE PSYCHOLOGY OF SURVEY RESPONSE 377

make sense of preceding ones, giving rise successful would you say you have been
to systematic influences of later questions in life?’ accompanied by a rating scale
on earlier ones (e.g., Schwarz & Hippler, that ranges from ‘not at all successful’
1995). to ‘extremely successful.’ To answer this
question, they have to determine what the
researcher means by ‘not at all successful’—
Response alternatives
does this term refer to the absence of
Particularly relevant sources of contextual outstanding achievements or to the presence
information are the response alternatives of explicit failures? To do so, they may draw
that accompany a question. This has been on a feature that the researcher is unlikely
most extensively addressed in comparisons of to consider informative, namely the numeric
open and closed response formats. Suppose values of the rating scale. Schwarz, Knäuper,
respondents are asked in an open format, Hippler, Noelle-Neumann, and Clark (1991)
‘What have you done today?’ To give a presented the above success-in-life question
meaningful answer, they have to determine with an 11-point rating scale that ranged
which activities may be of interest to the either from 0 (‘not at all successful’) to
researcher. Observing the maxim of quan- 10 (‘extremely successful’), or from −5
tity, they are likely to omit activities that (‘not at all successful’) to +5 (‘extremely
the researcher is obviously aware of (e.g., successful’). Whereas 34% of the respondents
‘I gave a survey interview’) or may take for endorsed a value between 0 and 5 on the
granted anyway (e.g., ‘I took a shower’). If 0 to 10 scale, only 13% endorsed one of
respondents were given a list of activities the formally equivalent values between −5
that included giving an interview and taking and 0 on the −5 to +5 scale. Figure 34.1
a shower, most respondents would endorse shows the response shift underlying these
them. At the same time, such a list would differences.
reduce the likelihood that respondents report Subsequent experiments indicated that this
activities that are not represented on the list shift is due to differential interpretations
(see Schuman & Presser, 1981; Schwarz & of ‘not at all successful.’ When combined
Hippler, 1991, for reviews). Both of these with the numeric value ‘0,’ respondents
questions form effects reflect that response inferred that the term refers to the absence
alternatives can clarify the intended meaning of outstanding achievements. However, when
of a question and may remind respondents the same term was combined with the numeric
of material that they may otherwise not value ‘−5,’ and the scale offered ‘0’ as the
consider. mid-point, they inferred that it refers to the
These processes can result in pronounced presence of explicit failures. In general, a
and systematic differences between open format that ranges from negative to positive
and closed question formats. For example, numbers conveys that the researcher has
Schuman and Presser (1981, pp. 105ff.) asked a bipolar dimension in mind, where the
respondents what they consider ‘the most two poles refer to the presence of opposite
important thing for children to prepare them attributes. In contrast, a format that uses
for life.’ Whereas 62% picked ‘To think for only positive numbers conveys that the
themselves’ when this alternative was offered researcher has a unipolar dimension in mind,
as part of a list, only 5% provided an answer referring to different degrees of the same
that could be assigned to this category in an attribute.
open response format. Other ‘formal’ characteristics of response
Response alternatives can even affect scales, like the graphical layout of rat-
question interpretation when they are purely ing scales or the numeric values of frequency
‘formal’ in nature and apparently devoid of scales, can have a similarly profound impact
any substantive information. As an exam- on respondents’ question interpretation (for a
ple, suppose respondents are asked, ‘How review see Schwarz, 1996, chap. 5). While the
378 THE SAGE HANDBOOK OF PUBLIC OPINION RESEARCH

40
35
30
25
20
15
10
5
0
0; −5 1; −4 2; −3 3; −2 4; −1 5; 0 6; +1 7; +2 8; +3 9; +4 10; +5

0 to 10 Scale −5 to +5 Scale

Figure 34.1 Response distribution as a function of the numerical values of the


rating scale
Note: Shown is the percentage of respondents who chose the respective scale value when asked,
‘How successful would you say you have been in life?’ 0 or −5 = ‘not at all successful’; 10 or
+5 = ‘extremely successful.’ Adapted from Schwarz, Knäuper, Hippler, Noelle-Neumann, &
Clark (1991)

researcher’s selection of these formats is often otherwise is evident when more choice is
based on convenience or in-house traditions, provided.
respondents assume that every contribution
of the researcher is relevant to their task,
Cognitive pretests
consistent with the maxim of relation. They
therefore draw on formal characteristics of the As these findings indicate, question com-
research instrument in making sense of the prehension is not about words—it is about
question asked. speaker meaning. To determine the speaker’s
Moreover, respondents work within the intended meaning, respondents pay close
constraints imposed by the question format attention to contextual information, bringing
(Schuman & Presser, 1981). This is partic- the tacit assumptions that govern conver-
ularly apparent with regard to ‘don’t know’ sations in daily life (Grice, 1975) to the
(DK) or ‘no opinion’ responses. Standard research situation (Schwarz, 1996). That
survey questions usually omit ‘no opinion’ their responses are systematically affected
as an explicit option, but interviewers accept by minor features of the research instru-
this response when volunteered. Experimental ment highlights how closely they attend
studies (e.g., Schuman & Presser, 1981) to the specifics at hand in their quest to
have consistently found that respondents provide informative answers. Unfortunately,
are more likely to report not having an their efforts are rarely appreciated by the
opinion when a DK option is explicitly researcher, who considers these features
offered (see Schwarz & Hippler, 1991, for a substantively irrelevant and treats their influ-
review). Similarly, many respondents prefer ence as an undesirable artifact. Nor are
a middle alternative between two extreme researchers likely to note these influences
positions when offered, but endorse another in regular surveys, where control conditions
option when the middle alternative is omitted with different question formats are missing,
(e.g., Schuman & Presser, 1981). Thus, most or in regular field pretests, where problems
respondents assume that the rules of the can only be identified when respondents give
game call for working within the categories obviously meaningless answers or complain
offered, even though a desire to answer about the questions asked. What is needed
THE PSYCHOLOGY OF SURVEY RESPONSE 379

are pretest procedures that deliberately probe a judgment. Instead, they truncate the search
respondents’ understanding of a question in process as soon as enough information has
the context and format in which it will be come to mind to form the judgment with
presented in the actual interview. sufficient subjective certainty. Accordingly,
Several such cognitive interviewing pro- the judgment is disproportionately influenced
cedures have been developed (see Sudman by the information that is most accessible at
et al., 1996, chap. 2, and the contributions that point in time, e.g., because it has just
in Schwarz & Sudman, 1996). Most widely been used in answering a previous question
used are verbal protocols, in the form of (Bodenhausen & Wyer, 1987).
concurrent or retrospective think-aloud proce- Contextual influences on information
dures. In addition, respondents are often asked accessibility are the major contributor to the
to paraphrase the question, thus providing emergence of question order effects in survey
insight into their interpretation of question research, along with contextual influences
meaning (see DeMaio & Rothgeb, 1996, for on question interpretation. The underlying
common methods). An alternative, but less processes have been conceptualized in
sensitive, approach involves detailed analyses several related models that are consistent
of field pretests, known as behavior coding with current theorizing in social cognition
(see Fowler & Cannell, 1996). Based on (e.g., Feldman & Lynch, 1988; Tourangeau,
insights from cognitive interviews, Lessler 1992). The next section summarizes the
and Forsyth (1996) developed a detailed most comprehensive model (Schwarz &
coding scheme that allows researchers to Bless, 1992a, 2007), which identifies the
identify likely question problems in advance. conditions under which question order effects
Cognitive pretests that illuminate respon- emerge and predicts their direction, size, and
dents’ understanding of a question within its generalization across related items.
intended context can be conducted with a
relatively small number of respondents, and
provide the best available safeguard against Information accessibility and use:
later surprises.
The emergence, direction, and size
of question order effects
RECALLING INFORMATION AND Attitude questions ask respondents to provide
FORMING A JUDGMENT an evaluative judgment. To do so, respondents
need to form a mental representation of
Once respondents have determined what the the target (i.e., the object of judgment), as
researcher is interested in, they need to recall well as a mental representation of some
relevant information from memory. In some standard against which the target is evaluated.
cases, respondents may have direct access Both representations are context dependent
to a recently formed relevant judgment that and include information that is chronically
they can offer as an answer. In most cases, accessible as well as information that is
however, they will not find an appropriate only temporarily accessible—for example,
answer readily stored in memory, and will because it was brought to mind by preced-
need to form a judgment on the spot, drawing ing questions. How accessible information
on whatever information comes to mind at that influences the judgment depends on how it
time. What renders this process problematic is used. Information that is included in the
for public opinion research is its high context temporary representation formed of the target
dependency: what information comes to mind, results in assimilation effects; that is, the
and how it is used, is strongly influenced by inclusion of positive (negative) information
the specifics of the research instrument. results in a more positive (negative) judgment.
As a general rule, people never retrieve all The size of assimilation effects increases
information that may potentially be relevant to with the amount and extremity of temporarily
380 THE SAGE HANDBOOK OF PUBLIC OPINION RESEARCH

accessible information and decreases with the married respondents now reported lower
amount and extremity of chronically accessi- general life-satisfaction, whereas unhappily
ble information included in the representation married respondents reported higher life-
of the target (Bless, Schwarz, & Wänke, satisfaction, indicating that they excluded
2003). the positive (negative) marital information
For example, Schwarz, Strack, and Mai from the representation formed of their lives
(1991) asked respondents to report their in general. These diverging effects reduced
marital satisfaction and their general life- the correlation to r = 0.18, from r = 0.67
satisfaction in different question orders. when the same questions were asked in
When the general life-satisfaction question the same order without a joint lead-in.
was asked first, it correlated with marital A control condition in which the general life-
satisfaction r = 0.32. Reversing the question satisfaction question was reworded to, ‘Aside
order, however, increased this correlation from your marriage, which you already told
to r = 0.67. This reflects that the marital us about, how satisfied are you with your
satisfaction question brought marriage related life in general?’ resulted in a highly similar
information to mind that respondents included correlation of r = 0.20 (Schwarz et al.,
in the representation formed of their lives 1991).
in general. This increase in correlation was In addition, respondents may use excluded
attenuated, r = 0.43, when questions about information in constructing a standard of com-
three different life-domains (job, leisure time, parison or scale anchor. If the implications
and marriage) preceded the general question, of the temporarily accessible information are
thus bringing more diverse material to mind. more extreme than the implications of the
Parallel influences were observed in the mean chronically accessible information used in
reports. Happily married respondents reported constructing a standard, they result in a more
higher, and unhappily married respondents extreme standard, eliciting contrast effects for
reported lower, general life-satisfaction when that reason. The size of comparison based
their attention was drawn to their marriage by contrast effects increases with the extremity
the preceding question. and amount of temporarily accessible infor-
However, the same piece of accessible mation used in constructing the standard,
information may also elicit a contrast effect; and decreases with the amount and extremity
that is, a more negative (positive) judgment, of chronically accessible information used
the more positive (negative) information is in making this construction (Bless et al.,
brought to mind. This is the case when 2003). Comparison based contrast effects
the information is excluded from, rather generalize to all targets to which the standard
than included in, the cognitive representation is applicable.
formed of the target. For example, the above As an example, consider the impact of
study included a condition in which the mari- political scandals on assessments of the
tal satisfaction and life-satisfaction questions trustworthiness of politicians. Not surpris-
were introduced with a joint lead-in that read, ingly, thinking about a politician who was
‘We now have two questions about your life. involved in a scandal, say Richard Nixon,
The first pertains to your marriage and the decreases trust in politicians in general. This
second to your life in general.’ This lead- reflects that the exemplar is included in the
in was designed to evoke the conversational representation formed of the target ‘politicians
maxim of quantity, which enjoins speakers in general.’ If the trustworthiness question
to avoid redundancy when answering related pertains to a specific politician, however—say
questions. Accordingly, respondents who Bill Clinton—the primed exemplar cannot be
had just reported on their marriage should included in the representation formed of the
now disregard this aspect of their lives target—after all, Bill Clinton is not Richard
when answering the general life-satisfaction Nixon. In this case, Richard Nixon may
question. Confirming this prediction, happily serve as a standard of comparison, relative
THE PSYCHOLOGY OF SURVEY RESPONSE 381

to which Bill Clinton seems very trustworthy. these patterns are to be expected when we
An experiment with German exemplars con- take into account that recalling extreme and
firmed these predictions (Schwarz & Bless, vivid media examples drives the general and
1992b): Thinking about a politician who the specific judgments in opposite directions,
was involved in a scandal decreased the as predicted on theoretical grounds.
trustworthiness of politicians in general, but Given the crucial role of inclusion/
increased the trustworthiness of all specific exclusion operations in the construction of
exemplars assessed, as shown in Figure 34.2. mental representations, it is important to
In general, the same information is likely to understand their determinants. When thinking
result in assimilation effects in the evaluation about a topic, people generally assume that
of superordinate target categories (which whatever comes to mind bears on what they
allow for the inclusion of all information are thinking about—or why else would it
pertaining to subordinate categories), but come to mind now? Hence, the default is
in contrast effects in the evaluation of to include information that comes to mind
lateral target categories (which are mutually in the representation of the target. This
exclusive). These judgmental processes are renders assimilation effects more likely than
reflected in a wide range of discrepancies contrast effects. In fact, assimilation effects
between general and specific judgments (sometimes referred to as carry-over effects)
in public opinion research. For example, dominate the survey literature, and many
Americans distrust Congress in general, but models intended to account for question order
trust their own representatives (e.g., Erikson, effects don’t even offer a mechanism for
Luttbeg, & Tedin, 1988). Similarly, members the conceptualization of contrast effects (e.g.,
of minority groups consistently report high Zaller, 1992), which severely limits their
levels of discrimination against their group; usefulness as general theoretical frameworks.
yet they also report that their own personal Whereas inclusion is the more common
experience was more benign. In each case, default, the exclusion of information needs

6
5.6
5 4.9
5

4
3.4

0
Political Class Individual Politicians
Control Scandal question asked

Figure 34.2 Trust in politicians in general and in individual politicians as a function of a


preceding scandal question
Note: Respondents rated the trustworthiness of German politicians in general (‘political class’) or
of three individual politicians (averaged); 9 = ‘very trustworthy.’ Adapted from Schwarz &
Bless (1992b)
382 THE SAGE HANDBOOK OF PUBLIC OPINION RESEARCH

to be triggered by salient features of the always the case and some questions are too
question answering process. Any variable that mushy with regard to the conceptual variables
influences the categorization of information to allow strong predictions. Sudman et al.
can also determine the emergence of assimi- (1996, chap. 5) provide a more extensive
lation and contrast effects. These variables can discussion of these issues, with detailed
be conceptualized as bearing on three implicit examples.
decisions that respondents have to make with Question order effects are attenuated when
regard to topic related information. related questions are separated by a sufficient
The first decision bears on why this number of buffer items, which renders the
information comes to mind. Information that information brought to mind by the earlier
seems to come to mind for the ‘wrong question less accessible (for a review, see
reason,’ e.g., because respondents are aware Wänke & Schwarz, 1997). Similarly, pre-
of the potential influence of a preceding ceding questions are less likely to influence
question, is likely to be excluded. This is the judgments of older respondents, due to
more common in experiments that present age-related decline in memory. This age-
allegedly separate and independent tasks than sensitivity of question order effects can
in survey interviews, where the questions are undermine cohort comparisons (for examples
seen as part of the same ongoing conversation. see Schwarz & Knäuper, 2000).
The second decision pertains to whether the
information that comes to mind bears on
Response order effects
the target of judgment or not. The content
of the context question, the superordinate Another major source of context effects in
or lateral nature of the target category, the attitude measurement is the order in which
extremity of the information or its representa- response alternatives are presented. Response
tiveness for the target category are relevant at order effects are most reliably obtained when
this stage (see Schwarz & Bless, 1992a, 2007, a question presents several plausible response
for examples). Finally, conversational norms options (see Sudman et al., 1996, chap. 6,
of nonredundancy may elicit the exclusion for detailed discussions). To understand the
of previously provided information, as seen underlying processes, suppose a person is
above (Schwarz et al., 1991). asked to provide a few good reasons why
Whenever any of these decisions results ‘divorce should be easier to obtain.’ The
in the exclusion of information from the person can easily do so, but he or she could just
representation formed of the target, it will as easily provide some reasons why ‘divorce
elicit a contrast effect. Whether this contrast should be more difficult to obtain.’When such
effect is limited to the target, or generalizes alternatives are juxtaposed within a question
across related targets, depends on whether (as in ‘Should divorce be easier to obtain
the excluded information is merely subtracted or more difficult to obtain?’), the outcome
from the representation of the target or used depends on which alternative is considered
in constructing a standard against which the first. When respondents consider ‘easier’ and
target is evaluated. Conversely, whenever the generate some supportive thoughts, they are
information that comes to mind is included likely to truncate the search process and
in the representation formed of the target, endorse this response option; but had they
it results in an assimilation effect. Hence, considered ‘more difficult’ the same process
the inclusion/exclusion model provides a would have resulted in an endorsement of that
coherent conceptualization of the emergence, option. Again, respondents’judgment is based
direction, size, and generalization of context on the temporary representation formed of the
effects in attitude measurement. These effects attitude object, which is, in part, a function of
can be reliably produced when the questions the response option they consider first.
are written to clearly operationalize the Which option respondents consider first
relevant variables. Of course, this is not depends on the order and mode in which
THE PSYCHOLOGY OF SURVEY RESPONSE 383

the response alternatives are presented more than a dozen experiments and different
(Krosnick & Alwin, 1987). When presented self-report measures of attitude strength,
in writing, respondents elaborate on the Krosnick and Schuman (1988) found no
implications of the response options in the evidence that context effects are stronger for
order presented. In this mode, an alternative weak attitudes, except for the not surpris-
that elicits supporting thoughts is more likely ing finding that respondents with a weak
to be endorsed when presented early rather attitude are more likely to choose a middle
than late on the list, giving rise to primacy alternative. This observation contrasts with
effects. In contrast, when the alternatives are findings in other domains, where strongly
read to respondents, their opportunity to think held attitudes have been found to be more
about the early ones is limited by the need stable over time and less likely to change
to listen to later ones. In this case, they are in response to persuasive messages (see
more likely to work backwards, thinking first Krosnick & Abelson, 1992, for a review).
about the last alternative read to them. When From a social cognition perspective, this
this alternative elicits supporting thoughts, it apparent discrepancy is not surprising. First,
is likely to be endorsed, giving rise to recency many context effects reflect differences in
effects. Hence, a given alternative is more respondents’ inferences about the pragmatic
likely to be endorsed when presented early meaning of the question, resulting in answers
rather than late in a visual format (primacy to somewhat different questions. These effects
effect), but when presented late rather than are unlikely to be influenced by attitude
early in an auditory format (recency effect). strength. Second, people process persuasive
Sudman et al. (1996, chap. 6) review these messages and survey interviews with different
processes in more depth. background assumptions. They understand
Response order effects are more that persuaders want to influence them and
pronounced for older and less educated are prepared to scrutinize their arguments
respondents (see Knäuper, 1999, for (Friestad & Wright, 1994). In contrast, they
a meta-analysis), whose limited cognitive assume that survey interviews are intended
resources further enhance the focus on a single to assess rather than change their opinions.
response alternative. This age-sensitivity of Moreover, the thoughts brought to mind by
response order effects can invite misleading preceding questions are their own thoughts,
conclusions about cohort differences in the and drawing on these thoughts seems less
reported attitude, suggesting, for example, problematic than accepting the arguments
that older respondents are more liberal presented by others, who may be following
than younger respondents under one order their own agenda. As a result, the question–
condition, but more conservative under the answer sequence of survey interviews can
other (Schwarz & Knäuper, 2000). elicit attitude and behavior change that goes
far beyond what skilled persuaders can hope
to accomplish (e.g., Morwitz, Johnson, &
Attitude strength
Schmittlein, 1993).
Survey researchers have long assumed that Finally, it is worth noting that mental
attitudes vary in their degree of ‘strength,’ construal models of attitude judgment, like
‘centrality,’ or ‘crystallization,’ and that con- the inclusion/exclusion model (Schwarz &
text effects are limited to attitudes that are Bless, 1992a) presented above, can account
weak and not (yet) crystallized (e.g., Cantril, for stability as well as change in attitude
1944; Converse, 1964; see Krosnick & judgments. Empirically, an attitude is con-
Abelson, 1992, for a review). Despite its sidered ‘stable’ when respondents provide
popularity, this hypothesis ‘has clearly been highly similar judgments at two points in
disconfirmed’as Krosnick and Abelson (1992, time. From a construal perspective, this
p. 193) concluded after an extensive review. is to be expected when the question is
In the most comprehensive test, based on presented in the same or a highly similar
384 THE SAGE HANDBOOK OF PUBLIC OPINION RESEARCH

context or when the amount and extremity may reflect the impact of several distinct
of chronically accessible information exceeds processes. For example, white respondents
the amount and extremity of contextual have frequently been found to mute neg-
information. Construal models can therefore ative sentiments about African-Americans
conceptualize the context dependency as well when the interviewer is black rather than
as independency of attitude judgments and white (e.g., Williams, 1964; Hatchett &
specify the conditions under which each may Schuman, 1976). From a social desirability
be observed (see Schwarz & Bohner, 2001). perspective, the answers they provide to
the interviewer are assumed not to reflect
their ‘true’ attitudes. However, the friendly
FORMATTING THE RESPONSE conversation with a middle class African-
American interviewer may itself serve as
Having formed a judgment, respondents input into the attitude judgment, resulting
usually need to format it to fit the response in (temporary) ‘real’ attitude change, much
alternatives provided by the researcher. An as incidental exposure to the names or
issue that needs particular attention at this pictures of liked African-Americans has been
stage emerges when respondents are asked to found to affect attitudes toward the group in
rate several items along the same rating scale. laboratory experiments (e.g., Bodenhausen,
As numerous experiments have demonstrated, Schwarz, Bless, & Wänke, 1995). Hence, the
respondents use the most extreme items of a impact of social desirability per se is often
set to anchor the endpoints of the scale (see difficult to isolate in survey data. Moreover,
Parducci, 1965; Ostrom & Upshaw, 1968). social desirability certainly affects everyday
As a result, a given item will be rated as less behavior, including interracial interactions. Its
extreme if presented in the context of a more influence should therefore not be regarded as
extreme one than if presented in the context a simple artifact of survey interviewing; nor
of a less extreme one. This undermines the may we want to eliminate it when our goal is
comparability of ratings across studies when to predict everyday behavior.
the same item is presented in the context of
other, differentially extreme items.
CONCLUDING REMARKS

EDITING THE RESPONSE As this selective review illustrates, research


into the cognitive and communicative pro-
Finally, respondents may want to edit cesses underlying survey responses has
their response before they communicate it, moved far beyond the ‘art of asking questions’
reflecting considerations of social desirability (Payne, 1951) and increasingly provides a
and self-presentation. This is particularly scientific grounding for questionnaire design.
likely when the question is threatening (see The accumulating insights provide a useful
DeMaio, 1984, for a review). Editing is more framework for the wording and ordering
pronounced in face-to-face interviews than of questions and the choice of response
in self-administered questionnaires, which alternatives (Sudman et al., 1996; Tourangeau
provide a higher degree of confidentiality et al., 2000). Moreover, the new techniques
(e.g., Smith, 1979; Krysan, Schuman, Scott, & of cognitive interviewing allow us to identify
Beatty, 1994). All methods designed to reduce problems before a survey goes into the field
socially desirable responding address one of (Schwarz & Sudman, 1996).
these two factors (see Bradburn et al., 2004, At the same time, there is no hope for
chap. 3, for a review and advice). a silver bullet that eliminates the context
Although socially desirable responding is dependency of respondents’ answers. Instead,
undoubtedly a threat to the validity of survey the emerging picture is consistent with current
results, many of the more robust findings research in social and cognitive psychology
THE PSYCHOLOGY OF SURVEY RESPONSE 385

that emphasizes the situated and constructive Cannell, C. F., Marquis, K. H., & Laurent, A. (1977).
nature of human judgment (for reviews see A summary of studies of interviewing methodology.
Schwarz, 2000; Smith & Semin, 2004). From Vital and Health Statistics, Series 2, No. 69 (DHEW
this perspective, judgment stands in the ser- Publication No. HRA 77-1343). Washington, DC:
vice of action, and merely retrieving opinions Government Printing Office.
Cantril, H. (1944) Gauging public opinion. Princeton, NJ:
formed in the past may not serve us well.
Princeton University Press.
To guide action, a useful system of judgment
Clark, H. H., & Schober, M. F. (1992). Asking questions
should be informed by past experience, but be and influencing answers. In J. M. Tanur (Ed.),
highly sensitive to the specifics of the present; Questions about questions (pp. 15–48). New York:
it should overweight recent experience and Russell-Sage.
experience from similar situations, and take Converse, P. E. (1964). The nature of belief systems in
current goals and concerns into account—in the mass public. In D. E. Apter (Ed.), Ideology and
short, it should be highly context sensitive. discontent (pp. 206–261). New York: Free Press.
Unfortunately, respondents bring this context DeMaio, T. J. (1984). Social desirability and survey
sensitivity to the interview, where it creates a measurement: A review. In C. F. Turner & E. Martin
serious challenge for public opinion research; (Eds.), Surveying subjective phenomena (Vol. 2,
to the extent that respondents’ judgments pp. 257–281). New York: Russell Sage.
reflect the immediate context created in DeMaio, T. J., & Rothgeb, J. M. (1996). Cognitive
interviewing techniques: In the lab and in the field.
the survey interview, generalizations to the
In N. Schwarz & S. Sudman (Eds.), Answering
population are fraught with uncertainty. There questions: Methodology for determining cognitive
is no reliable solution to this challenge. The and communicative processes in survey research
best we can do is to increase the odds that (pp. 177–196). San Francisco: Jossey Bass.
we become aware of contextual influences by Erikson, R. S., Luttbeg, N. R., & Tedin, K.T. (1988).
embedding experiments in regular surveys. American public opinion (3rd ed.). New York:
Macmillan.
Feldman, J. M., & Lynch, J. G. (1988). Self-generated
REFERENCES validity and other effects of measurement on belief,
attitude, intention, and behavior. Journal of Applied
Belson, W. A. (1981). The design and understanding of Psychology, 73, 421–435.
survey questions. Aldershot: Gower. Fowler, F. J., & Cannell, C. F. (1996). Using behavioral
Bishop, G. F., Oldendick, R. W., & Tuchfarber, A. J. coding to identify cognitive problems with survey
(1986). Opinions on fictitious issues: the pressure to questions. In N. Schwarz & S. Sudman (Eds.),
answer survey questions. Public Opinion Quarterly, Answering questions: Methodology for determining
50, 240–250. cognitive and communicative processes in survey
Bless, H., Schwarz, N., & Wänke, M. (2003). The research (pp. 15–36). San Francisco: Jossey Bass.
size of context effects in social judgment. In Friestad, M., & Wright, P. (1994). The persuasion
J. P. Forgas, K. D. Williams & W. von Hippel (Eds.), knowledge model: How people cope with persuasion
Social judgments: Implicit and explicit processes attempts. Journal of Consumer Research, 21,
(pp. 180–197). Cambridge, UK: Cambridge University 1–31.
Press. Grice, H. P. (1975). Logic and conversation. In P. Cole, &
Bodenhausen, G. V., & Wyer, R. S. (1987). Social J. L. Morgan (Eds.), Syntax and semantics, Vol. 3:
cognition and social reality: Information acquisition Speech acts (pp. 41–58). New York: Academic Press.
and use in the laboratory and the real world. In Hatchett, S., & Schuman, H. (1976). White respondents
H. J. Hippler, N. Schwarz & S. Sudman (Eds.), Social and race of interviewer effects. Public Opinion
information processing and survey methodology Quarterly, 39, 523–528.
(pp. 6–41). New York: Springer Verlag. Knäuper, B. (1999). The impact of age and education
Bodenhausen, G. V., Schwarz, N., Bless, H., & Wänke, M. on response order effects in attitude measurement.
(1995). Effects of atypical exemplars on racial beliefs: Public Opinion Quarterly, 63, 347–370.
Enlightened racism or generalized appraisals? Journal Krosnick, J. A., & Abelson, R. P. (1992). The case for
of Experimental Social Psychology, 31, 48–63. measuring attitude strength. In J. M. Tanur (Ed.),
Bradburn, N., Sudman, S., & Wansink, B. (2004). Asking Questions about questions (pp. 177–203). New York:
questions (2nd ed.). San Francisco: Jossey Bass. Russell-Sage.
386 THE SAGE HANDBOOK OF PUBLIC OPINION RESEARCH

Krosnick, J. A., & Alwin, D. F. (1987). An evaluation effects. Personality and Social Psychology Bulletin, 18,
of a cognitive theory of response order effects in 574–579.
survey measurement. Public Opinion Quarterly, 51, Schwarz, N., & Bless, H. (2007). Mental construal pro-
201–219. cesses: The inclusion/exclusion model. In D. A. Stapel &
Krosnick, J. A., & Schuman, H. (1988). Attitude J. Suls (Eds.), Assimilation and contrast in social psy-
intensity, importance, and certainty and susceptibility chology (pp. 119–141) Philadelphia, PA: Psychology
to response effects. Journal of Personality and Social Press.
Psychology, 54, 940–952. Schwarz, N., & Bohner, G. (2001). The construction
Krysan, M., Schuman, H., Scott, L. J., & Beatty, P. (1994). of attitudes. In A. Tesser & N. Schwarz (Eds.),
Response rates and response content in mail versus Blackwell handbook of social psychology: Intraindi-
face-to-face surveys. Public Opinion Quarterly, 58, vidual processes (Vol.1, pp. 436–457). Oxford, UK:
381–399. Blackwell.
Lessler, J. T., & Forsyth, B. H. (1996). A coding Schwarz, N., & Hippler, H. J. (1991). Response
system for appraising questionnaires. In N. Schwarz & alternatives: The impact of their choice and ordering.
S. Sudman (Eds.), Answering questions: Method- In P. Biemer, R. Groves, N. Mathiowetz, & S. Sudman
ology for determining cognitive and communica- (Eds.), Measurement error in surveys (pp. 41–56).
tive processes in survey research (pp. 259–292). Chichester: Wiley.
San Francisco: Jossey Bass. Schwarz, N., & Hippler, H. J. (1995). Subsequent ques-
Morwitz, V., Johnson, E., & Schmittlein, D. (1993). tions may influence answers to preceding questions
Does measuring intent change behavior? Journal of in mail surveys. Public Opinion Quarterly, 59, 93–97.
Consumer Research, 20, 46–61. Schwarz, N., & Knäuper, B. (2000). Cognition,
Norenzayan, A., & Schwarz, N. (1999). Telling what they aging, and self-reports. In D. Park & N. Schwarz
want to know: Participants tailor causal attributions (Eds.), Cognitive aging. A primer (pp. 233–252).
to researchers’ interests. European Journal of Social Philadelphia, PA: Psychology Press.
Psychology, 29, 1011–1020. Schwarz, N., Knäuper, B., Hippler, H. J., Noelle-
Ostrom, T. M., & Upshaw, H. S. (1968). Psychological Neumann, E., & Clark, F. (1991). Rating scales:
perspective and attitude change. In A. C. Greenwald, Numeric values may change the meaning of scale
T. C. Brock, & T. M. Ostrom (Eds.), Psychological labels. Public Opinion Quarterly, 55, 570–582.
foundations of attitudes (pp. 217–242). New York: Schwarz, N., & Oyserman, D. (2001). Asking questions
Academic Press. about behavior: Cognition, communication and
Parducci, A. (1965). Category judgments: A range- questionnaire construction. American Journal of
frequency model. Psychological Review, 72, Evaluation, 22, 127–160.
407–418. Schwarz, N., Strack, F., & Mai, H.P. (1991). Assimi-
Payne, S. L. (1951). The art of asking questions. lation and contrast effects in part-whole question
Princeton: Princeton University Press. sequences: A conversational logic analysis. Public
Schuman, H., & Presser, S. (1981). Questions and Opinion Quarterly, 55, 3–23.
answers in attitude surveys. New York: Academic Schwarz, N., & Sudman, S. (1996). Answering
Press. questions: Methodology for determining cognitive
Schwarz, N. (1996). Cognition and communication: and communicative processes in survey research.
Judgmental biases, research methods, and the logic San Francisco: Jossey-Bass.
of conversation. Hillsdale, NJ: Erlbaum. Sirken, M., Hermann, D., Schechter, S., Schwarz, N.,
Schwarz, N. (1999). Self-reports: How the questions Tanur, J., & Tourangeau, R. (Eds.) (1999). Cognition
shape the answers. American Psychologist, 54, and survey research. New York: Wiley.
93–105. Smith, E. R., & Semin, G. R. (2004). Socially situated
Schwarz, N. (2000). Agenda 2000: Attitudes and social cognition: Cognition in its social context. Advances in
judgment—Warmer, more social, and less conscious. Experimental Social Psychology, 36, 53–117.
European Journal of Social Psychology, 30, 149–176. Smith, T. W. (1979). Happiness. Social Psychology
Schwarz, N., & Bless, H. (1992a). Constructing reality Quarterly, 42, 18–30.
and its alternatives: Assimilation and contrast effects Strack, F., & Martin, L. (1987). Thinking, judging, and
in social judgment. In L. L. Martin & A. Tesser (Eds.), communicating: A process account of context effects
The construction of social judgments (pp. 217–245). in attitude surveys. In H. J. Hippler, N. Schwarz, &
Hillsdale, NJ: Erlbaum. S. Sudman (Eds.), Social information processing and
Schwarz, N., & Bless, H. (1992b). Scandals and the survey methodology (pp. 123–148). New York:
public’s trust in politicians: Assimilation and contrast Springer Verlag.
THE PSYCHOLOGY OF SURVEY RESPONSE 387

Strack, F., Schwarz, N., & Wänke, M. (1991). Semantic psychological research (pp. 35–47). New York:
and pragmatic aspects of context effects in social and Springer Verlag.
psychological research. Social Cognition, 9, 111–125. Tourangeau, R., Rips, L. J., & Rasinski, K. (2000).
Sudman, S., Bradburn, N. M., & Schwarz, N. (1996). The psychology of survey response. Cambridge:
Thinking about answers: The application of cognitive Cambridge University Press.
processes to survey methodology. San Francisco, CA: Wänke, M., & Schwarz, N. (1997). Reducing question
Jossey-Bass. order effects: The operation of buffer items. In
Tourangeau, R. (1984). Cognitive science and survey L. Lyberg, P. Biemer, M. Collins, E. DeLeeuw,
methods: A cognitive perspective. In T. Jabine, C. Dippo, & N. Schwarz (Eds.), Survey measurement
M. Straf, J. Tanur, & R. Tourangeau (Eds.), Cognitive and process quality (pp. 115–140). Chichester, UK:
aspects of survey methodology: Building a bridge Wiley.
between disciplines (pp. 73–100). Washington, DC: Williams, J. A. (1964). Interviewer-respondent interac-
National Academy Press. tion: A study of bias in the information interview.
Tourangeau, R. (1992). Attitudes as memory structures: Sociometry, 27, 338–352.
belief sampling and context effects. In N. Schwarz & Zaller, J. R. (1992). The nature and origins of mass
S. Sudman (Eds.), Context effects in social and opinion. Cambridge: Cambridge University Press.

You might also like