Statistical analysis of modified complete randomized designs: applications to chemo-orientation studies

Area of Science:

• Experimental design
• Statistical analysis
• Animal behavior

Background:

• Randomized Complete Block Design (RCBD) is common but complete randomization is often impractical.
• Restricted randomization is used when resetting experimental apparatus is costly.
• Incorrect analysis of restricted randomization data can inflate Type I error rates.

Purpose of the Study:

• Demonstrate the consequences of using standard F-statistic methods with restricted randomization.
• Introduce split-plot analysis of variance as a more powerful alternative for restricted randomization.
• Validate the alternative method using simulated and real experimental data.

Main Methods:

• Simulated experiments to assess Type I error rates under different randomization and analysis methods.
• Statistical analysis comparing F-statistic approach with split-plot analysis of variance.
• Application of the alternative method to data from virgin male Periplaneta americana orientation studies.

Main Results:

• Standard F-statistic analysis under restricted randomization increases the risk of false positives.
• Split-plot analysis of variance provides a more powerful and accurate method for restricted randomization.
• Simulated and real-world data confirm the benefits of the alternative approach.

Conclusions:

• Treating restricted randomization data as complete randomization inflates Type I errors.
• Split-plot analysis of variance is a recommended statistical approach for experiments using restricted randomization.
• The findings are applicable to various fields, including animal orientation and chemical ecology.