The relationship of select task variables to response patterns and response quality in a survey of consumer purchasing
One of the common tasks in survey research is the measurement of behavioral frequency. This is particularly true of marketing surveys, where researchers often wish to know the frequency with which respondents have engaged in some shopping, purchase or consumption behavior. In addition, market researchers are often interested in the intensity of that behavior; for example, the amount purchased or consumed. Because these behavioral data are so frequently desired and have important uses, researchers want these measurements to yield data that are as close as possible to true values. Hence, there is a concern with developing survey procedures that are capable of obtaining the best possible data. Also, there is substantial interest in understanding how design of the survey task can affect the nature and quality of such data. Historically, survey researchers have not investigated the possibility that respondents might use various cognitive procedures to generate responses to survey questions about behaviors. Studies involving socially threatening questions about behavior have ignored cognition and have relied on a motivational model as the means of obtaining more complete and better responses from respondents. On the other hand studies involving non-threatening topics have placed emphasis on the performance of a cognitive task, recall, as the basis for response quality. However, these studies have relied upon a memory model that presumes a single cognitive procedure, that is, episodic processing whereby respondents recall individual events and the characteristics of those events to generate answers to survey questions. The focus of this research is non-threatening behaviors. The research questions the conception that reporting of non-threatening behaviors is based only on recalling and enumerating individual events which may be forgotten or misplaced in time. It is proposed here that respondents do not always enumerate events to obtain frequency estimates; that they sometimes begin by estimating a rate of frequency. At least four different factors seem plausible as determinants of which method is used- the number of events to be reported (which is a function of their rate of occurrence and time period), their vividness, the regularity of the events, and the nature of the question itself. To the extent that question form or time frame selection can influence the recall task, these factors might also influence response effects. Subsequent hypotheses about behavioral frequency and intensity data are studied in the context of department store purchases as measured by different survey question forms. The research draws on an assumption that different question forms (manipulation of the survey task) lead respondents to perform different cognitive procedures in preparing their answers, and that these differences impact the nature and quality of reporting. Task variables manipulated include the following: (1) the time frame over which behavior is measured (one, three, six, or twelve months); (2) the form of the behavioral frequency measure (how many times, how often or the time interval between purchases); (3) the form of the expenditure measure (direct measurement of total expenditure or inference of total expenditure from purchase frequency and expenditure per purchase); and, (4) the form of the expenditure per purchase measure (average expenditure or most recent expenditure). These variables are studied in a consumer survey of 1112 respondents. Question forms are manipulated both between and within respondents for measurement of purchases at two major department stores. All respondents are credit card holders at one of the stores, and store record data are available for comparison with reported credit card expenditures. Three major groups of hypotheses are investigated.These concern (1) means and variances of reported purchase frequency under different question forms, (2) means and variances of reported expenditure under different measurement procedures, and (3) response accuracy in expenditure data under different measurement procedures.Results indicate that all hypotheses relating to frequency estimates based on the time interval between purchases are supported. That is, interval based estimates produce higher means and variances of purchase frequency and expenditure than non-interval based estimates, and this may be associated with positive response bias. However, other hypotheses generally are not supported. It is concluded that question form can influence the cognitive process used in responding to survey questions and the nature and quality of response. Regarding failed hypotheses, though, a call is made for research that more carefully studies response processes.