Trend detection in the presence of covariates: Stagewise versus multiple regression

The assessment of trend in environmental quality is an important task for environmental managers and scientists. Much recent research has focused on the proper methodology for the assessment of trend, for example, when the data are correlated or censored. An equally important problem is discussed in this paper, that of the specifying underlying models that relate the environmental quality time series to other time dependent covariates and how to adjust for these covariates. Two basic approaches for the adjustment for covariates are stagewise regression and multiple regression. We use simulation, analysis and a data set from the Chesapeake Bay, U.S.A., to show that the approach based on stagewise regression can yield misleading results. We show that the multiple regression approach is generally more powerful than the stagewise method.