Reconstructing women’s reproductive histories: are survey data reliable?

Francesca Rinesi, Istituto Nazionale di Statistica (ISTAT)
Marina Attili, Istituto Nazionale di Statistica (ISTAT)
Claudia Iaccarino, Istituto Nazionale di Statistica (ISTAT)

During the last decades increasing importance has been given to the study of reproductive histories based on event history approaches. Given the increasing supply of survey data collected retrospectively and their wide use in scientific research, it is becoming crucial to assess their reliability. By comparing sample surveys with exhaustive sources, or different sample surveys with each other, several scholars have noticed that data collected from sample surveys with retrospective questions on reproductive histories tend to severely overestimate/underestimate fertility levels of selected subgroups of women. This paper precisely aims to evaluate the reliability and accuracy of women reproductive histories collected with the Follow-up Survey on Births, by comparing self-reported information on childbearing histories with evidence that comes both from the Sample Survey on Births (“main survey”) and from population registers (list of all live-births from women resident in Italy for selected years). The present study does not focus on the representativeness of the sample1, but on problems that relate to the “incorrect” registration of childbearing histories. The approach used in this paper has its main strength in that the comparison of fertility histories from three independent data sources (but referring to the same women) takes place at individual level. This comparison makes it possible not only to evaluate the Follow-up Survey’s data quality, but also to correct its records in case of clear inconsistency. Lastly, we took into account if interviewer effects had a substantial impact in the follow-up data quality. It must be noticed that interviewers were paid per completed interview, regardless of its length. This may push some of the interviewers to deliberately misreport the number of children living in the household in order shorten the length of the interview. However, we found no significant interviewer effects.