Background. Higher education has been posited to protect against cognitive decline, either because the rate of decline is slower in the more highly educated or the start of decline is delayed. Latent growth models provide improved methodology to examine this issue.

Methods. The sample consisted of 887 participants aged 70–93 years in 1991 and followed up in 1994 and 1998. Latent growth models and standard regression techniques were used to examine the rate of cognitive decline in four cognitive measures while controlling for health status and sex. A delayed start model was examined by incorporating interaction effects in a regression model.

Results. Neither the latent growth models nor the regression techniques revealed a slower rate of decline for the more highly educated. The proportion of the highly educated showing no change was no larger than the proportion of the less well educated. There were no significant age by education interaction effects, no chronologically later accelerations in the rate of change as a function of education, and no differences in rate of decline between the first measurement interval and the second.

Conclusions. Education may not protect against cognitive decline although it is associated with long-term individual differences in level of functioning. The discrepancy between our study and others may be attributable to attrition effects, follow-up length, sample age, scaling artefacts and negative publication bias. Most importantly, practice effects may favour the better educated and hence account for the supposed protective effect in many longitudinal studies of cognitive change.