Longitudinal studies examining changes in physical functioning with advancing age among very old people are plagued by high death rates, which can lead to biased estimates. This study was conducted to analyse changes in physical functioning among the oldest old with three distinct methods which differ in how they handle dropout due to death. The sample consisted of 3992 persons aged 90 or over in the Vitality 90+ Study who were followed up on average for 2.5 years (range 0–13 years). A generalized estimating equation (GEE) with independent ‘working’ correlation, a linear mixed-effects (LME) model and a joint model consisting of longitudinal and survival submodels were used to estimate the effect of age on physical functioning over 13 years of follow-up. We observed significant age-related decline in physical functioning, which furthermore accelerated significantly with age. The average rate of decline differed markedly between the models: the GEE-based estimate for linear decline among survivors was about one-third of the average individual decline in the joint model and half the decline indicated by the LME model. In conclusion, the three methods yield substantially different views on decline in physical functioning: the GEE model may be useful for considering the effect of intervention measures on the outcome among living people, whereas the LME model is biased regarding studying outcomes associated with death. The joint model may be valuable for predicting the future characteristics of the oldest old and planning elderly care as life expectancy continues gradually to rise.