A note on the variance of paired comparisons estimates

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• The paired comparisons method is used to estimate treatment effectiveness.
• It is similar to odds ratios derived from independent groups.
• Variance computation for paired comparisons has been assumed to be the same as for independent groups.

Purpose of the Study:

• To demonstrate that the variance for paired comparisons estimates is lower than for odds ratios based on independent groups.
• To distinguish between 'real' paired comparisons and odds ratios from independent data.
• To analyze the impact of data dependencies on variance in treatment effectiveness estimation.

Main Methods:

• Utilizing a simple binomial model and linear approximation.
• Comparing variance calculations for paired comparisons versus independent group odds ratios.
• Illustrating with examples of both independent and paired data in traffic crash fatality analysis.

Main Results:

• Paired comparisons estimates exhibit lower variance compared to odds ratios from independent groups.
• Dependencies between paired data points reduce the overall variance.
• The study provides a formula for variance in independent cases and explains variance reduction in paired cases.

Conclusions:

• The variance of paired comparisons estimates is demonstrably lower than that of odds ratios from independent groups.
• Accounting for data dependencies in paired comparisons leads to more precise effectiveness estimates.
• This statistical insight is crucial for accurate analysis in various fields, including traffic safety research.