It is often thought that standard regression models, like multiple linear regression and logistic regression, cannot be used for the analysis of longitudinal data. The reason is that the observations are not independent of each other. Without missing data, however, the story is a bit more intricate. Standard regression models, in that case, do provide consistent parameter estimates. However, asymptotic standard errors obtained from such standard models are wrong, invalidating test statistics, p-values, and conclusions. In this research project an alternative to the asymptotic theoretical standard errors is investigated: the cluster bootstrap. This methodology is investigated for continuous and binary response variables, under various forms of missing data, in combination with multiple imputation of missing values, and as a model selection tool.

Share on Facebook Share on X Share on LinkedIn Share by WhatsApp Share by Mastodon