Objective To investigate how lifetime socioeconomic position (SEP) is associated with later-life crystallised cognitive function, accounting for different missing data mechanisms.

Participants A nationally representative population sample born in 1946 (MRC National Survey of Health and Development; NSHD, N=5362), and a sample of British civil servants (Whitehall II; WHII, N=10 308).

Methods Novel structured statistical approach to distinguish between accumulation and sensitive period life course models using SEP measures from childhood, early-adulthood and midlife. Results of complete case (CC) (assuming missing completely at random), multiple imputation (MI) (missing at random) and a Heckman selection model (missing not at random) were compared.

Outcomes National Adult Reading Test, age 53 (NSHD); Mill Hill Test, age 55–79 (WHII).

Results NSHD: After adjusting for childhood cognitive function, the best fitting model was an accumulation model allowing SEP at each time point to have its own estimate. However estimates varied by missing data method (women: childhood SEP: CC: coefficient=1.11 (95% CI 0.15 to 2.06), MI: coefficient=1.82 (95% CI 0.87 to 2.76), Heckman: coefficient=0.70 (95% CI −0.38 to 1.78)). WHII (not adjusted for childhood cognition): the best fitting model represented accumulation in adulthood only, with childhood SEP not significant.

Conclusion Despite adjustment for childhood cognitive score, childhood SEP remains important in NSHD, whereas in Whitehall II childhood SEP was not associated with cognitive function. These differences may be due to recall bias of early SEP in WHII. Our findings demonstrate the utility of the method for distinguishing models of how SEP across the life course influences cognition and the importance of dealing with missing data.