Incorporating sources of correlation between outcomes: An introduction to mixed models

Area of Science:

• Veterinary statistics
• Biostatistics
• Animal research methodology

Background:

• Animal research frequently involves repeated measurements on the same subjects.
• These repeated measures yield correlated outcome data due to inherent animal characteristics.
• Standard statistical methods often assume independent observations, which are violated by this data structure.

Purpose of the Study:

• To highlight the challenges of analyzing correlated outcome data in animal research.
• To explain the consequences of using standard statistical methods inappropriately on such data.
• To introduce appropriate statistical approaches for handling correlated outcomes in animal studies.

Main Methods:

• Discussion of common study designs leading to correlated outcomes.
• Illustrative examples of statistical inference errors from incorrect analysis.
• Introduction to statistical methods designed for correlated data.

Main Results:

• Incorrect application of standard statistical methods leads to flawed statistical inference (p-values, confidence intervals).
• This can result in either an overstatement or understatement of statistical significance.
• Appropriate methods are available to correctly analyze correlated outcome data.

Conclusions:

• Proper statistical analysis is crucial for accurate interpretation of animal research findings with repeated measures.
• Researchers must utilize methods accounting for outcome correlation to ensure valid conclusions.
• Adopting specialized statistical techniques enhances the reliability of animal study results.