The Legacy of Charles Cannell
Summary of Cannell’s Key Contributions
Between 1960 and 1990, Charles Cannell and his colleagues documented significant survey response errors, created a model of the survey response process, developed a theory of the interview and a set of interviewing techniques to encourage accurate reporting, and designed methods for observing interviewer and respondent behavior. These integrated accomplishments have had a noteworthy impact on thinking about interviewer-respondent interaction, but they deserve more attention from students of the survey interview now. In this chapter, we summarize Cannell's work, examine its evolution, and suggest ways in which it can inform current research and practice.
Documenting Errors in Self-Reports
In order to judge if interviewing procedures lead to better self-reports of behavior, we need to understand when self-reports are more trustworthy. Cannell's productive collaboration on "validity studies" with the U.S. National Center for Health Statistics (NCHS) early in the 1960s built the foundation for subsequent research on interviewing. The studies were record check efforts: interviews with people who were known - by virtue of information in medical records - to have been hospitalized or to have visited a physician. Respondents in these investigations failed to report these documented experiences in predictable ways. For example, as the amount of time between the date of the interview and the medical encounters increased, reporting of hospitalizations and doctor visits declined (cf. the much earlier work on memory by Ebbinghaus 1913). Shorter hospital stays were less likely to be reported than longer ones. Hospitalizations associated with embarrassing health conditions were less likely to be reported (Cannell, Fisher, and Bakker 1965; Cannell and Fowler 1963). Other investigations funded by NCHS around this time focused on reporting of chronic health conditions, comparing survey reports to medical records (Balamuth and Shapiro 1965; Madow 1967). These studies also found that health experiences were under-reported.
Such findings led to the following general inferences: (1) self-reports of experiences will decline in accuracy as the experiences themselves recede in time; (2) self-reports of experiences will decline in accuracy as their subjective salience to the respondent declines; (3) self-reports of experiences will be less accurate if they are embarrassing ("socially undesirable"). Interviewing procedures that led to more reporting of experiences that are difficult to remember or uncomfortable to report can be said to have produced more accurate data. These generalizations were in line with contemporaneous work by other investigators outside the health research sphere (Ferber 1966; Neter and Waksberg 1964).
The under-reporting concept was further extended to phenomena lacking external validation information. As Marquis, Cannell, and Laurent (1972:1) wrote, "Since higher reporting does seem to suggest more accurate reporting, the number of reported events has been used as a dependent variable in later studies where better validity data have not been available." This benchmark formed the basis for judging the quality of data produced by different interviewing approaches.
Modeling the Survey Response Process
Cannell developed a theory of the response process. Survey methodologists will recognize components of the cognitive approach to survey response (Tourangeau, Rips, and Rasinski 2000) in a model originally sketched by Cannell, Marquis, and Laurent (1977:53) and revised in Cannell, Miller, and Oksenberg (1981). As can be seen in Figure 2.1, the response process begins with the respondent's comprehension of the question. Beyond the consideration of question wording and sentence structure complexity, Cannell and colleagues' concept of comprehension also involves broader issues of interpretation, factoring in consideration of the intended meaning of questions.
Step 2 of the model, labeled here as cognitive processing, involves retrieval, beginning with the identification of the relevant information needed for searching one's memory. Step 3 involves the process of evaluation of the retrieved information with respect to its accuracy. Step 4 involves the assessment of that same information with respect to other goals (which may lead to inaccurate reporting), and Step 5 identifies the actual response.
An interesting feature of the model is that it suggests where the response process can go astray, resulting in less than complete or accurate information. Cannell and his colleagues believed that deviation from the linear phases 1 through 5 could be due to insufficient retrieval effort or motivation (possibly embarrassment) on the part of the respondent, or due to extraneous situational cues (e.g., unhelpful interviewer behavior). Cannell came to rely on the idea of "interference" in memory (McGeoch 1932; Tulving 1972), which suggests, in the survey context, that events are not simply forgotten, but are hard to retrieve because succeeding events have "buried" them. Encouraging more retrieval effort in the
Diagram of respondent's (R) question-answering process. Source: Cannell, Miller, and Oksenberg (1981).
interview, therefore, is a worthwhile approach. Both cognitive and motivational difficulties can hamper respondents from progressing linearly through the steps to accurate response. The model's depiction of a "detour" to inaccurate answering is the intellectual forerunner of the concept of "satisficing" in survey response (Krosnick 1991). How questions are designed and how interviewers administer them are keys to respondents' staying on the "straight and narrow" path.