Exact logistic regression

Question

Stats Experts.

What constitutes a small sample size for using exact logistic regression?

How would I confirm that I need to use logistic regression versus exact logistic regression? I believe my work fits this category but I am soliciting for a knowledgeable person's input to weigh in and offer your insight.

Is SAS the only statistical system that offers a true exact logistic regression option for analysis?

Thank you in advance for your insight.

Answer 1

Determining what constitutes a "small" sample size for using exact logistic regression can vary depending on the specific context and guidelines set by the statistical software you are using. However, generally speaking, exact logistic regression is typically considered when dealing with small sample sizes, especially when:

1. You have a small number of events (positive outcomes) relative to the number of predictors or independent variables. In such cases, traditional logistic regression might not yield reliable estimates.
2. You have sparse data, which means some combinations of predictor variables have very few or zero events or observations.
3. The assumptions of asymptotic methods (like ordinary logistic regression) may not be met due to small sample sizes.

To confirm whether you should use logistic regression versus exact logistic regression, you should consider the following:

1. Sample Size: Assess whether you have a small sample size, especially if you have a limited number of events compared to the number of predictors.
2. Data Sparsity: Check if your data exhibits sparsity, meaning some predictor combinations have few or no occurrences of the outcome.
3. Software Capabilities: Verify whether the statistical software you are using supports exact logistic regression. Not all statistical packages offer this option.

Regarding your question about SAS, as of my last knowledge update in September 2021, SAS does provide options for exact logistic regression using procedures like PROC FREQ and PROC LOGISTIC. However, it's always a good practice to consult the latest SAS documentation or contact SAS support to ensure you have the most up-to-date information on the availability and usage of exact logistic regression in SAS.

Keep in mind that exact logistic regression can be computationally intensive, especially for larger datasets, so it's important to weigh the benefits of exact inference against the computational cost when deciding to use it. Additionally, consulting with a statistician or data analyst experienced in your specific research area can provide valuable insights into the choice of regression method for your particular study.