Mixed models for assessing correlation in the presence of replication

Area of Science:

• Environmental Science
• Statistics
• Biostatistics

Background:

• Assessing correlation with multiple measurements per variable is common in environmental science.
• Introductory statistics texts often lack coverage of this topic.
• Existing ad hoc methods can yield invalid conclusions and complicate correlation measure selection.

Purpose of the Study:

• To provide a practical statistical method for correlation assessment with multiple measurements.
• To reanalyze existing data using PROC MIXED in SAS for linked replicate measurements.
• To extend methods for unlinked replicate measurements, illustrated with benzene concentration data.

Main Methods:

• Reanalysis of Lam et al.'s data using PROC MIXED in SAS for linked measurements.
• Extension of Lam et al.'s maximum likelihood estimation method to unlinked measurements.
• Application to a study correlating indoor and outdoor benzene concentrations.

Main Results:

• Demonstrated obtaining parameter estimates with minimal SAS code for linked data.
• Successfully extended the methodology to handle unlinked replicate measurements.
• Provided a practical framework for analyzing environmental correlation data.

Conclusions:

• The PROC MIXED approach in SAS offers a accessible solution for correlation analysis with linked measurements.
• The extended method effectively addresses scenarios with unlinked replicate measurements.
• This work provides valuable statistical tools for environmental science research involving complex measurement designs.