Sample Size for Multilevel Logistic Regression Models
Author Information
Author(s): Moineddin Rahim, Matheson Flora I, Glazier Richard H
Primary Institution: Centre for Research on Inner City Health, St. Michael's Hospital, Toronto, Canada
Hypothesis
What is the effect of varying sample size on the accuracy of estimates in multilevel logistic regression models?
Conclusion
Using a minimum group size of 50 with at least 50 groups is recommended to produce valid estimates for multilevel logistic regression models.
Supporting Evidence
- Estimates of fixed effect parameters are unbiased for 100 groups with group size of 50 or higher.
- Low prevalent events require larger sample sizes with at least a minimum of 100 groups and 50 individuals per group.
- The biases for both fixed and random effects are severe for group size of 5.
- Standard errors for fixed effect parameters are unbiased while for variance covariance components are underestimated.
Takeaway
To get good results when studying groups of people, you need to have enough groups and enough people in each group. If there aren't enough, the results might not be right.
Methodology
Simulation studies were used to assess the effect of varying sample sizes at both individual and group levels on the accuracy of estimates in multilevel logistic regression models.
Potential Biases
Non-convergence can occur when estimating too many random components that are close to zero.
Limitations
The study did not explore the effect of different estimation procedures on the accuracy of parameter estimates.
Statistical Information
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website