Pair-Feeding Study Designs Can Create Biases and Inflate Type I Error Rates: A Simulation Study

Area of Science:

• Biostatistics
• Experimental Design
• Animal Research

Background:

• Pair-feeding is a common experimental design to isolate treatment effects from food intake variations.
• Investigators often overlook the statistical implications of pair-feeding, assuming equivalent food intake.
• The impact of pair-feeding on type I error rates (T1Er) has not been previously quantified.

Purpose of the Study:

• To quantify the impact of pair-feeding on type I error rates in experimental studies.
• To evaluate whether pair-feeding designs inflate statistical significance when food intake is not accounted for.
• To determine methods for mitigating inflated type I error rates associated with pair-feeding.

Main Methods:

• Monte Carlo simulations were employed to model animal weight and food intake.
• Animals were randomized into pair-fed and non-pair-fed groups.
• Pair-feeding involved truncating food intake to match non-pair-fed controls (individually or by group average); analyses were conducted with and without adjustment for food intake.

Main Results:

• Both individual and group pair-feeding significantly inflated type I error rates in unadjusted models, with rates ranging from 0.12 to 0.71.
• Statistical adjustment for food intake effectively reduced type I error rates to approximately 0.05.
• Unadjusted analyses in pair-feeding studies are prone to false positive findings.

Conclusions:

• Pair-feeding study designs, when unadjusted for food intake, can lead to inflated type I error rates.
• Adjusting statistical analyses for actual food intake is essential to correct for inflation and maintain accurate error rates.
• This study highlights the importance of careful statistical analysis in pair-feeding experiments to ensure the validity of treatment effect conclusions.