Analytical strategies for the marble burying test: avoiding impossible predictions and invalid p-values

Area of Science:

• Behavioral Neuroscience
• Animal Behavior Analysis
• Statistical Modeling in Biology

Background:

• The marble burying test is a common method to assess anxiety-related and repetitive behaviors in rodents.
• Data from this test are count data, characterized by non-negative integers with upper and lower bounds.
• Traditional normal linear models (e.g., t-test, ANOVA) are frequently misapplied to this data, violating assumptions of unboundedness and constant variance.

Purpose of the Study:

• To highlight the limitations of using normal linear models for analyzing marble burying test data.
• To introduce and advocate for the use of generalised linear models (GLMs) as a more appropriate statistical framework.

Main Methods:

• Demonstration of the statistical problems arising from applying normal linear models to bounded count data.
• Introduction and explanation of generalised linear models (GLMs) as a suitable alternative.

Main Results:

• Normal linear models yield problematic confidence intervals (including impossible values) and misleading p-values when applied to marble burying data.
• GLMs are shown to be a more appropriate statistical tool for analyzing count data from behavioral studies.

Conclusions:

• Generalised linear models are specifically designed for non-Gaussian data, including count data.
• GLMs offer straightforward application and interpretation, leading to more valid and reliable inferences in behavioral research.
• The adoption of GLMs will improve the statistical rigor of studies utilizing the marble burying test.