-68- 
VI. CONCLUSION 
The results presented above are, in many respects, not generalizable to survey research at 
large. A comparison of data quality among longitudinal surveys with different designs is needed to 
evaluate the data in perspective. The extent to which they provide a contribution to the field of survey 
methodology depends in part on where the trend in survey research will go. Virtually all surveys work 
in part with retrospective data collection but rarely as extensive as in the Life History Study 
considered here. Hardly any survey has similarly high interviewer workloads yet many institutions 
work with more or less stable interviewer staffs, and a professionalization of interviewing has been 
discussed in the social sciences for a long time. 
The first question I will address is whether the data analyzed here are affected by an 
interviewer bias. My results speak to differences in data quality and therefore to the comparability 
of cases across interviewer samples. I have shown that interviewers may influence the number of job 
spells coded in an interview (see Appendix F). I find 5 significant coefficients for individual 
interviewers, with effect sizes between 17 and 30%. These interviewers recorded less job spells than 
others, controlling for other variables. In addition, the more interviewers work, the more job spells 
we find in the unedited interviews. These effects were eliminated during data edition. The same 
analysis performed after data editing shows no effects of interviewing or individual interviewers on 
the number of job spells." 
Individual interviewers certainly do influence the number of errors counted by my measures. 
Partly, data editing can remove these influences, but may introduce editing effects such as the fact that 
more job spells were added for interviews with tape recording. Also, data editing is costly and error 
prone. Where call-backs were impossible, where no tape recording could provide additional 
information, missing data on the event sequences had to be imputed. 
Interviewer effects on interview time can be both beneficial and harmful with respect to data 
quality. First, interviewers may become more efficient in leading respondents through the ques- 
tionnaire, avoiding confusion and supporting the retrieval process. Also, where life courses are 
complex with respect to the instrument, and depending on the volume of the information transmitted, 
additional time enhances data quality. Second, intervie wers may push respondents through the event 
As explained above, a comparison of visible errors before and after data editing is not 
usesul since all visible errors were eliminated. One could compute the duration of events before 
and after editing in order to make that comparison. Still, for events with missing data on the event 
sequence (see table 1) no duration can be calculated.