I have paired data (GWAS case/control study) and I have heard using conditional logistic regression or generalized linear mixed models (GLMM) is appropriate. Which should I use in this case? Why would you use one over the other. More importantly can you guys point me towards resources for doing these methods in R? I'm finding a lot of material for SAS, which I do not prefer. I can provide more details if necessary.

  • The conditional logistic regression applies fixed effects (in the context of econometrics),

    $$ logit(p_{ij})=\boldsymbol x_{ij}^{'}\boldsymbol\beta+u_i.$$ where each pair of subjects has an individual intercept ($u_i$). It can be implemented with clogit() of package survival or clogistic() of package Epi.

  • Generalized linear mixed models (GLMM) for binary data can adopt link functions like logit, probit and cloglog. The mixed logistic regression is as,

    $$ logit(p_{ij})=\boldsymbol x_{ij}^{'}\boldsymbol\beta+\boldsymbol z_{ij}^{'}\boldsymbol u_i$$ where $\boldsymbol u_i$ are random variables and can have the distribution assumption (e.g. normal distribution). Of course you can use a random intercept model, i.e. $\boldsymbol z_{ij}^{'}=1$ and $\boldsymbol u_i$ is a scalar. You can estimate GLMM using glmer() of package lme4.

  • As to the choice between conditional logistic regression and GLMM for binary data, some people are in favor of conditonal (fixed-effects) logistic regression and GLMM with probit link, but against fixed-effects probit or GLMM with logit link. The reason may be that some of the consistency properties break down, especially with small within-cluster sample size ($n_i=2$ for your case).

    You can find the clarification of fixed effects and random effects (and marginal models) in different contexts here.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.