Some statistical issues related to multiple linear regression modeling of beach bacteria concentrations

Area of Science:

• Environmental science
• Statistics
• Ecology

Background:

• Multiple linear regression (MLR) is widely used for predicting beach bacteria concentrations.
• Previous studies often inadequately addressed key issues like serial correlation and model selection criteria.

Purpose of the Study:

• To highlight the importance of addressing serial correlations in time-series data for MLR models.
• To recommend appropriate criteria for model selection and assessment in predicting beach bacteria concentrations.

Main Methods:

• Investigated the impact of serial correlations and time-series effects within a statistically rigorous framework.
• Evaluated the utility of interaction terms in reducing bias in reduced models.
• Proposed comprehensive statistics for assessing model predictive capacity, distinct from goodness-of-fit.

Main Results:

• Serial correlations require careful attention and statistically rigorous testing and adjustment.
• Joint criteria of R(2) and Cp-statistic are recommended for model selection, not t-statistics of the full model.
• Interaction terms can reduce bias but may offer limited numerical improvement.
• R(2) and error counts alone are insufficient for objective model assessment.

Conclusions:

• Inadequate handling of serial correlations and inappropriate model selection can lead to erroneous conclusions.
• A statistically rigorous approach to time-series effects and comprehensive model assessment are essential for accurate beach bacteria prediction.
• Previous studies on beach bacteria modeling using MLR may require re-evaluation.