Gumbel based p-value approximations for spatial scan statistics
2010

Improving P-Value Precision in Spatial Scan Statistics

Sample size: 245 publication Evidence: high

Author Information

Author(s): Allyson M Abrams, Ken Kleinman, Martin Kulldorff

Primary Institution: Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Health Care Institute

Hypothesis

Can a Gumbel distribution provide more precise p-values for spatial scan statistics compared to traditional Monte Carlo methods?

Conclusion

The Gumbel distribution can yield more precise p-values with less computational effort than traditional Monte Carlo hypothesis testing.

Supporting Evidence

  • The Gumbel distribution provides a better fit for the tail of the distribution of test statistics.
  • Using fewer Monte Carlo replicates with the Gumbel method maintains statistical power.
  • Traditional Monte Carlo methods require significantly more computational resources for small p-values.

Takeaway

This study shows a new way to calculate p-values that is faster and more accurate, helping researchers find disease clusters more efficiently.

Methodology

The study used computer simulations to fit a Gumbel distribution to test statistics from Monte Carlo replicates under the null hypothesis.

Potential Biases

Using distributions other than Gumbel can lead to anti-conservative bias in p-values.

Limitations

The method may not perform well in scenarios with very small sample sizes or when the maximum scanning window size is limited.

Participant Demographics

The study focused on 245 counties in the Northeast United States.

Statistical Information

Statistical Significance

p<0.05

Digital Object Identifier (DOI)

10.1186/1476-072X-9-61

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication