With pseudo-panel analysis biases may remain if cell sample sizes are too small (Deaton,
1985). Verbeek and Nijman (1992, 1993) find that cell sample sizes greater than 100
observations are generally sufficient to produce estimates with small biases from pseudo-panels.
In addition, heterogeneity across cohorts can produce biased pseudo-panel estimators in the
absence of cohort-specific fixed effects. As in Pencavel (1998), specifications with and without
various cohort fixed effects are presented here to show how alternative sources of variation affect
elasticity estimates.