Estimates of birth intervals in Pakistan, with and without the WFS restrictions.

This exercise in examining the proximate determinants of birth intervals in Pakistan showed that significant bias in estimates was introduced when data of births were restricted to the last closed and open birth intervals. The reason for the bias was a sample restricted to a small number of births in the past 40 months. Analysis of socioeconomic determinants would be most greatly affected by the selection bias. Intermediate variables may be less biased due to a closer association with fertility. Data were obtained from the fertility portion of the 1979-80 Population, Labor Force, and Migration Survey in Pakistan and the Pakistan World Fertility Survey (WFS), which had half the number of reproductive histories. The data reflected a pattern whereby the proportions of births were smaller the farther away from the survey date. Logit regression methodology was adopted as modeled by Rindfuss, Bumpass, and Palmore and pertained to three durations of birth intervals spanning 2-12 years before the survey: 2-6 years, 2-5 years, and 2-4 years preceding the survey and 2, 3, and 4-8 birth orders. Models estimates were obtained for 96 logistic regressions for 3 sets of birth order intervals, 4 sets of birth segments (17-22, 23-28, 29-34, and 35-40 months), 4 sets of time periods, and 2 sets of restrictions (imposing WFS restrictions or not). The 96 estimates were compared to estimates based on two birth intervals 2-12 years prior to the survey without WFS restrictions. A higher proportion of unbiased results was obtained with the longer period preceding the interview and an unrestricted sample. For example, 95% of the betas are within the confidence interval for the period 2-6 years preceding the survey compared to 87% for 2-5 years and 80% for 2-4 years. Restrictions to the last closed and open birth intervals yielded 26-52% of betas falling in the confidence intervals. 57% of the betas for proximate intervals fell within confidence intervals, compared to only 48% for socioeconomic factors.