I would like to run a semi-symmetric bi-directional case-crossover method on some generated data using conditional logistic regression.

I generated data from Poisson distribution Poiss(lambda) with

$\textrm{log}(\lambda)=\lambda_0 \textrm{exp}(x\beta), $

where $\lambda_0$ and $\beta$ are known and $x$ is a known time series. Then for each timepoint I split generated value into pairs: case in this timepoint and control in referent period (that is if in generated data there are $5$ cases at day $11$, I get $10$ observations for this timepoint: $5$ $(1$, $x$ at day $11)$ and $5 (0$, $x$ at reference period for day $11)$). Now I run conditional logistic regression on this data, with separate stratum for each pair.

The question is, should the parameter estimates that I get from CLR be $\beta$ from parametrer of Poisson distribution? I found few articles with this procedure applied, but I can't figure out why estimetes from CLR and $\beta$ from Poisson distribution should be the same.

Reference articles:

Thanks for help


1 Answer 1


In a case-crossover study, the referent time periods are randomly sampled from the person-time at risk of each individual. Therefore, the odds ratio in a case-crossover estimates the incidence density ratio that would have been obtained in a cohort study - it is just an efficient sampling scheme, similar to an incidence density or risk set sampled case-control study, albeit one that doesn't require any reference people and automatically accounts for between-person confounding and within-person non-time-varying confounding. So, when you use conditional logistic regression your $\beta$ estimates are odds ratios which are in turn estimates of the incidence density ratios, which is what the $\beta$ parameters of your Poisson time-series are.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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