Solutions to problems of nonexistence of parameter estimates and sparse data bias in Poisson regression

Poisson regression can be challenging with sparse data, in particular with certain data constellations where maximum likelihood estimates of regression coefficients do not exist. This paper provides a comprehensive evaluation of methods that give finite regression coefficients when maximum likelihood estimates do not exist, including Firth’s general approach to bias reduction, exact conditional Poisson regression, and a Bayesian estimator using weakly informative priors that can be obtained via data augmentation. Furthermore, we include in our evaluation a new proposal for a modification of Firth’s approach, improving its performance for predictions without compromising its attractive bias-correcting properties for regression coefficients. We illustrate the issue of the nonexistence of maximum likelihood estimates with a dataset arising from the recent outbreak of COVID-19 and an example from implant dentistry. All methods are evaluated in a comprehensive simulation study under a variety of realistic scenarios, evaluating their performance for prediction and estimation. To conclude, while exact conditional Poisson regression may be confined to small data sets only, both the modification of Firth’s approach and the Bayesian estimator are universally applicable solutions with attractive properties for prediction and estimation. While the Bayesian method needs specification of prior variances for the regression coefficients, the modified Firth approach does not require any user input.

Use of two-point models in “Model choice in time-series studies of air pollution and mortality”

In this work, a new technique is proposed to study short-term exposure and adverse health effects. The presented approach uses hierarchical clusters with the following structure: each pair of two sequential days in 1 year is embedded in the year. We have 183 clusters per year with the embedded structure <year:2 days>. Time-series analysis is conducted using a conditional Poisson regression with the constructed clusters as a stratum. Unmeasured confounders such as seasonal and long-term trends are not modelled but are controlled by the structure of the clusters. The proposed technique is illustrated using four freely accessible databases, which contain complex simulated data. These data are available as the compressed R workspace files. Results based on the simulated data were very close to the truth based on the presented methodology. In addition, the case-crossover method with 1-month and 2-week window, and a conditional Poisson regression on 3-day clusters as a stratum, was also applied to the simulated data. Difficulties (high type I error rate) were observed for the case-crossover method in the presence of high concurvity in the simulated data. The proposed methods using various forms of a stratum were further applied to the Chicago mortality data. The considered methods have often different qualitative and quantitative estimations.