Wellbeing Measurement
Interview Method

May 30, 2008

Abstract

The influence of interview method on life-satisfaction judgments was examined in the German Socio-Economic Panel data (N = 44,386 respondents, 335,984 responses). The main finding was that reported life-satisfaction was higher in oral interviews than in mailed questionnaires. This finding indicates that wellbeing levels based on oral interviews are inflated by a tendency to respond socially desirable. A number of limitations are discussed.

Introduction

A common concern in survey research is that possibility of social desirable responding. Studies of responses on anonymous paper-pencil questionnaires show little evidence of social desirable responding (WellbeingMeasurementSocialDesirability). However, it is possible that social desirability has a pronounced effect on reports of wellbeing when respondents have to report their wellbeing to an interviewer or when the report is made in the presence of a third person. Few studies have systematically examined this question. The data from the German Socio-Economic Panel (SOEP) provide a unique opportunity to examine this question in a large, representative sample. Moreover, the SOEP data have been used in numerous research reports. Thus, even if the results cannot be generalized to other surveys, the results are relevant. [Although I am reporting my own analyses, the data are freely available to researchers who sign a contract with the DIW. Feel free to run your own analyses and contact me if your results do not replicate the findings reported on this website]

Method and Results

Respondents were asked to report their life-satisfaction on an 11-point scale ranging from 0 = totally dissatisfied to 10 = totally satisfied (SOEP Info Documentation). A total of 335,984 responses from 44,386 respondents (average 7.60 responses per respondent). For most responses, the interview method was also recorded. Table 1 presents the main interview methods, number of respondents, and number of responses for the most common interview methods. These methods cover 86% of all responses.

Table 1

Method

N Respondents

N Responses 

Mean

SD

Oral Interview

  25,348         

  117,405 

  7.12

1.63

Written Q. with Interviewer

  17,253

    77,506

  6.89

1.65

Written Q without Interviewer

  10,207

     18686

  7.10

1.88

CAPI (since 1998)

  13,906

    42,972

  7.25

1.52

Mailed-questionnaire

   6,974

    33,396

  6.78

1.71

The results in Table 1 simply average across all responses. To control for potential selection effects, it is possible to compare responses by the same individuals who used different methods. These results are reported in Table 2.

Table 2

Method 1

Method 2

N Respondents 

Mean 1

Mean 2

Correlation        .

Oral Interview

Written Q with Interviewer

     8,448         

  7.19a

  6.78b

    .50

Oral Interview

Written Q. without Interviewer

     5,828

  7.08a

  7.02c

    .46

Oral Interview

CAPI

     6,296

  7.17a

  7.07d

    .59

Oral Interview

Mailed questionnaire

    3,408

  7.22a

  6.64e

    .42

Written Q. with Interviewer

Written Q. without Interviewer

     5,869

  6.83b

  7.03c

    .57

Written Q. with Interviewer

CAPI

     3,444

  6.88b

  7.21d

    .50

Written Q. with Interviewer

Mailed questionnaire

     3,031

  6.89b

  6.72e

    .50

Written Q. without Interviewer

CAPI

     1,878

  7.14c

  7.09d

    .39

Written Q. without Interviewer

Mailed questionnaire

     1,803

  7.12c

  6.68e

    .41

CAPI

Mailed-questionnaire

        772

  7.46d 

  6.66e 

    .48              .

Table 2 replicates most of the mean differences reported in Table 1. This finding suggests that the results are not due to selection effects, but rather effects of the assessment method. In addition, Table 2 shows the correlations between the different methods. Differences in correlations may reveal method effects on the rank order of respondents, which would occur if a subgroup of respondents changes their responses in response to a method. For example, only some individuals respond socially desirable in an oral interview. However, many other confounding factors can influence these correlations. These preliminary analyses show correlations in a fairly narrow range from .39 to .57. The correlation between oral interview and mailed questionnaire is at the low end of this range, suggesting that method effects attenuate estimates of stability.

The main pattern of results is that oral interview, CAPI, and written Q without interview produce similar means. In comparison, mailed questionnaires and written Q without interviewer produce lower means.

Conclusion

Consistent with a social desirability effect, mailed questionnaires produced lower means than most interview methods. However, the comparison of the means for respondents who completed written questionnaires with or without the interviewer does not show a social desirability effect. The largest difference exists between CAPI and mailed back questionnaires (mean difference = .47). Although this effect is small by standard criteria, a half-point difference in means can produce misleading conclusions about wellbeing. As a result, caution is warranted when results from surveys with different assessment methods are compared.

Limitations

1. The results are based on a single data set. It is unclear whether these results generalize to wellbeing reports in general.
2. The analyses are limited to a question about life-satisfaction. Other questions could produce different results.
3. The results are based on repeated responses in a panel study and may differ from results when respondents answer these questions only once.
4. The results are correlational. It is possible that assessment methods were influenced by wellbeing levels.