To the political scientist, the study of values represents an important challenge. On the one hand, the concept of citizen values has been used as a key component in theories of political behavior. On the other hand, it is not always clear exactly what is meant by values, how they differ from other kinds of political judgments, and exactly how to measure them. I will attempt to address these problems by presenting a working model of the study of values and by arguing that consideration of the components of this model can enhance the way values are used to explain political behavior. While I will draw from a number of sources, including my own research, a major source for many of the ideas comes from the chapters in this volume.

A value is a complex construct. From one perspective, values are considered to be attributes of a collective, that is, representing some shared norm, understanding, or view of the world (Hofstede, 1983; Triandis, Bontempo, Betancourt, and Bond, 1986), perhaps based on common experience (Inglehart, 1977). From another perspective, values have been located within individuals, and are associated with individual experiences and perceptions (Rasinski, 1987; Rokeach, 1973). I argue that both of these views must be considered in order to fully understand the role of values in political behavior. Even so, much research focuses on one or the other view.

National surveys generally steer clear of questions that might offend respondents, keeping to relatively impersonal factual questions or items about abstract attitude issues. But almost every survey includes some items that are likely to seem intrusive or too personal to some respondents. For instance, it is quite common for surveys to ask the respondents about their income. Here's a representative income question, taken from the CPS:

1. (Including overtime pay, tips and commissions,) what are (name's/your) usual monthly earnings on (this job/(your/his/her) MAIN job), before taxes or other deductions? [CPS]

Many surveys try to soften the blow a bit by letting respondents choose from broad income groupings, but the CPS item asks for an exact dollar amount. The census Long Form – sent to one-sixth of the U.S. population – also includes detailed income questions about multiple sources of income (salaries, royalties, Social Security payments, and so on) that seek exact dollar amounts for every adult member of the household.

Even when a survey doesn't ask about the respondent's income, it almost certainly includes other standard demographic questions, and it isn't hard to imagine circumstances in which questions about age, say, or marital status might provoke embarrassment. In addition, surveys often begin by asking respondents for a roster of household members.

A glance at several national surveys reveals a large number of questions that ask respondents about the time of an event. For example, the National Crime Survey asks Question (1) for each crime victimization that the respondent reports:

1. (1) In what month did (this/the first) incident happen? (Show calendar if necessary. Encourage respondent to give exact month.) [NCS]

We presented similar questions in Chapter 3. Equally common are questions, such as (2), concerning the duration of a state or event:

1. (2) How much time did you lose because of injuries? [NCS]

These probes for a time point or duration, common as they are, aren't the only questions that require respondents to consider when or how long ago events occurred. Many survey items specify a reference period and ask whether or not an event of a given type occurred within it, or they specify a duration and ask whether an event or state lasted for a longer or shorter time. In answering Questions (3) and (4), for example, a respondent doesn't have to supply a date or duration but must still determine the time of the occurrence or the length of time:

1. (3) Did the accident happen during the past 2 years or before that time? [HIS]

2. (4) IF RETIRED, IN SCHOOL, KEEPING HOUSE, OR OTHER: Did you ever work for as long as one year? [GSS]

Surveys aim to discover facts about the respondents – often biographical facts about them. Respondents can sometimes answer factual questions on the basis of external records or documents, such as sales receipts, deposit slips, birth certificates, and the like, but more often they have to rely on memory for the information. Even easy questions like “What is your name?” or “Where do you live?” require respondents to consult their memory. Some survey questions, such as (1), from the 1988 National Health Interview Survey, may require memory feats of considerable difficulty:

1. (1) In what year did you first have an episode of back pain lasting a week or more? [HIS]

The greater the demands a question places on memory, the less accurate the respondents' answers and, all else being equal, the less accurate the survey estimates derived from them.

Attitude questions, like factual questions, require the respondents to search their memory. When the General Social Survey asks whether “you would approve of an adult male punching a stranger if the stranger was beating up a woman and the man saw it,” the respondent must search his or her memory for relevant information before making a response. But attitude questions pose special problems because the respondent may never have considered the issue prior to the survey. For this reason, we postpone considering attitude questions until Chapter 6, and we concentrate here on memory for biographical facts about the respondents.

During the past decade, cognitive theories have swept through survey methodology as they had earlier swept through psychology in 1970s and 1980s. Although the adoption of cognitive concepts and methods may seem quite rapid, it's still uncertain how much impact these ideas will ultimately have.

The first effort to foster collaboration between cognitive scientists and survey researchers took place in the late 1970s, when the British Social Science Research Council sponsored a conference on retrospective data in surveys. The conference brought psychologists and survey researchers together, apparently for the first time, to explore measurement problems in surveys; the immediate product of the conference was a volume edited by Moss and Goldstein (1979) focusing on problems of recall. Similar efforts to spark collaborations between cognitive psychologists and survey researchers began in the United States in 1980 and reached an early watershed in 1983, when the Committee on National Statistics, with support from the National Science Foundation (NSF), sponsored a seminar on cognitive aspects of survey methodology (see Jabine, Straf, Tanur, & Tourangeau, 1984). This seminar marked the beginning of long-term institutional support for the interdisciplinary movement, in the form of NSF grants to researchers outside the government and interagency transfers from NSF to the National Center for Health Statistics (NCHS). Much of the research reviewed in this book was carried out under grants from NSF, contracts from NCHS, or both.

Survey designers don't need to be reminded that the wording of the questions has an important impact on the results. Respondents can misinterpret even well-formulated questions, and when that happens, the question the respondent answers may not be the one the researcher intended to ask. Because of this obvious danger, the questions on national surveys are often subjected to empirical pretests. For example, the questionnaire designers may conduct cognitive interviews or focus groups in which they probe respondents' understanding of the questions and invite them to describe how they go about answering them (see Willis, DeMaio, & Harris-Kojetin, 1999, for a survey of these methods; we present a briefer discussion of them in Chapter 11). This practice is useful in bringing to light problems the designers may have overlooked.

This chapter looks at those aspects of survey questions that make them difficult for respondents to understand. These aspects are of many different sorts, ranging from features of grammar and word meaning to the broader situation in which the respondent and interviewer find themselves. Grammar can come into play either because the sentence is structurally ambiguous or because it includes complex clauses that respondents cannot parse. As an example of structural ambiguity, Item (1) asks respondents whether they agree or disagree with this statement:

1. (1) Given the world situation, the government protects too many documents by classifying them as SECRET and TOP SECRET. [GSS]

Over the last 25 years, the application of computer technology to the collection of survey data has revolutionized the survey industry. From its modest beginnings in 1971, when Chilton Research Services conducted the first computer-assisted telephone interviewing (CATI) survey for AT&T, CATI has become the most commonly used method for collecting data by telephone in the United States and Europe. In a CATI survey, the questionnaire takes the form of a computer program that displays the items to the interviewer on the computer screen; the interviewer then reads the questions to the respondent and enters the answers by pressing the appropriate keys. And, just as CATI has become the standard method for collecting data in telephone surveys, computer-assisted personal interviewing (CAPI) has become the standard method for face-to-face interviews. In the United States, one major federal survey after another – among them the National Medical Expenditure Survey, the National Longitudinal Survey of Youth, the Current Population Survey, the Medicare Current Beneficiary Survey, the National Survey of Family Growth, and the Survey of Consumer Finances – has converted from paper-and-pencil interviews to CAPI as its primary mode of data collection. CAPI uses essentially the same technology as CATI, except that the programs are loaded onto portable laptop computers that can be used in face-to-face interviews. Couper and Nicholls (1998) provide an excellent summary of the history of these developments.

This book examines surveys from a psychological perspective. It proposes a theory about how respondents answer questions in surveys, reviews the relevant psychological and survey literatures, and traces out the implications of the theories and findings for survey practice. We hope the book appeals to a variety of audiences, including survey researchers, methodologists, statisticians, and others who are involved in designing and carrying out surveys; political scientists, social psychologists, and others who study public opinion or who use data from public opinion surveys; cognitive psychologists and other researchers who are interested in everyday memory and judgment processes; and demographers, market researchers, sociologists, and anyone else who uses survey data and is curious about how such data come into being.

Although we have written the book to be read from cover to cover, we recognize that not every reader will share our enthusiasm for all the topics the book includes. Readers who are most interested in public opinion may want to focus on Chapters 1, 2, 6, 7, and 8, skipping or skimming the other chapters. Those who care mostly about survey data on factual matters may want to focus instead on Chapters 1–5, 9, and 10. Those who are most interested in traditional issues in survey methodology, such as question order effects and differences across methods of data collection, may want to concentrate on Chapters 4, 7, 9, 10, and 11.

The view of attitudes presented in the preceding chapter implies that attitude judgments are temporary constructions (Wilson & Hodges, 1992). They may rest on long-standing evaluations and beliefs about a topic, but the judgments themselves must often be created in response to a question, sometimes quite quickly. Moreover, the judgments called for by attitude questions are rarely absolute but are typically made in relation to some standard, generally an implicit one. It is hardly surprising, then, that attitude judgments turn out to be quite context-dependent. As survey researchers have demonstrated repeatedly, the same question often produces quite different answers, depending on the context (see Bradburn, 1982, and Schuman & Presser, 1981, for numerous survey examples). To cite just one particularly well-documented example, the level of support expressed for legal abortions when the “woman is married and does not want any more children” depends on whether that question comes before or after a similar question on abortion when “there is a strong chance of a serious defect in the baby” (Schuman, 1992; Schuman & Presser, 1981). Asking the birth defect question first can greatly reduce support for abortion for married women who simply don't want any more children.

According to the belief-sampling model, the stability of attitude responses across different occasions or contexts largely reflects three factors – the homogeneity of the underlying pool of considerations related to the topic, the overlap in the samples of considerations that are the basis for responses to the different questions (or the same question in different contexts), and the reliability of the process of evaluating these considerations.