Bootstrap-based methods for testing factor-by-curve interactions in generalized additive models: assessing prefrontal...

Area of Science:

• Statistics
• Computational Neuroscience
• Behavioral Science

Background:

• The effect of continuous variables on outcomes often differs across distinct groups.
• Generalized additive models (GAMs) can model complex relationships but require methods to test interactions.
• Understanding factor-by-curve interactions is crucial in various scientific fields.

Purpose of the Study:

• To develop and validate statistical methods for estimating and testing factor-by-curve interactions in generalized additive models.
• To assess the performance of bootstrap-based tests for interaction terms.
• To apply the methodology to analyze neural activity in monkeys during decision-making.

Main Methods:

• Utilized a local scoring algorithm with local linear kernel smoothers for model estimation.
• Proposed two bootstrap-based procedures (likelihood ratio test and direct estimation) for testing interaction terms.
• Employed binning techniques to enhance computational efficiency for estimation and testing.
• Conducted a simulation study to validate the bootstrap tests.

Main Results:

• The simulation study demonstrated the validity of the proposed bootstrap-based tests.
• The methodology was successfully applied to analyze prefrontal cortex neural activity in monkeys.
• The statistical procedure effectively identified neural correlates of decision-making strategies.

Conclusions:

• The developed statistical framework accurately estimates and tests factor-by-curve interactions in GAMs.
• Bootstrap-based tests provide a reliable method for assessing interaction significance.
• This approach offers valuable insights into the neural basis of complex behaviors like decision-making.