Reflection on modern methods: visualizing the effects of collinearity in distributed lag models.

Area of Science:

• Epidemiology
• Biostatistics
• Environmental Health

Background:

• Collinearity poses challenges in regression analysis, particularly when assessing exposure effects over time.
• Constrained distributed lag models help mitigate but do not fully resolve collinearity issues.
• The impact of remaining collinearity in distributed lag models is often overlooked.

Purpose of the Study:

• To illustrate the effects of collinearity within distributed lag models.
• To develop a tool for assessing the influence of collinearity on study results.

Main Methods:

• Simulations were conducted using various exposure effect scenarios to visualize lagged effect curves.
• Analysis of three real-world datasets: air pollution cohort study, air pollution time series, and temperature-mortality time series.
• Development and implementation of the 'collin' R package for collinearity assessment.

Main Results:

• Collinearity can lead to unexpected findings, such as statistically significant associations in the opposite direction of expectation.
• It can incorrectly indicate specific time periods as more influential than others.
• The 'collin' R package provides a visual method to explore these collinearity consequences.

Conclusions:

• Collinearity remains a potential confounder in distributed lag models, even after applying mitigation techniques.
• The 'collin' R package offers a valuable visual tool for researchers to evaluate the impact of collinearity on their analyses.
• Researchers should consider using this tool to ensure the robustness of findings derived from distributed lag models.