Cohort-Sequential Design

A standard single-cohort longitudinal study follows one group of people continuously across the whole age range, which takes as long as the range itself. A cohort-sequential design instead recruits several overlapping age cohorts and follows each for only a short span, then statistically links their pieces into one long trajectory. This compresses calendar time dramatically — a ten-year curve might be assembled in three years — at the cost of relying on the assumption that the cohorts share a common developmental path, an assumption a single-cohort study does not need.

The key assumption is convergence, or linkage: that the different birth cohorts lie on the same age trajectory, so that age differences across cohorts validly stand in for within-person change. The threat to it is an age-by-cohort interaction, where cohorts grew up in different conditions and follow different curves. Miyazaki and Raudenbush provide formal multilevel tests that examine the overlapping age region for such interaction. If no age-by-cohort interaction is detected, the segments can be joined; if it is detected, the cohorts cannot be naively linked and the assembled curve would be misleading.

Only partially. Like all age-period-cohort settings, the design faces the fundamental confounding of age, period, and cohort, and it cannot fully disentangle developmental (age) change from historical (period) and generational (cohort) change without additional assumptions. What it does is use overlap to test whether cohort effects are large enough to invalidate linkage; when they are negligible, the assembled curve can be interpreted as an age trajectory. When cohort effects are substantial, the design signals the problem rather than solving it, and a formal age-period-cohort analysis may be required instead.