Abstract

Analysis of water quality data from a paired watershed design is needed to determine if a best fertilizer management practice reduces a specific water quality variable compared to a conventional fertilizer management practice. This study examines an existing recommended method of analysis for paired watershed designs, simple analysis of covariance (ANCOVA) on time aggregated data, then offers two autoregression analyses (AR) as alternatives. The first approach models the sequence of paired differences and estimates its 95% confidence band. The second approach develops individual watershed AR models then examines the joint 95% confidence interval about the predicted difference. A reliability analysis on the water quality data reveals that the data for the controlled watershed, i.e., the covariate, has a sizable measurement error, a factor that is not considered in the usual ANCOVA model. The AR methods avoid the measurement error and other inherent problems with the published recommended method. Graphically both AR analyses are similar and reveal three distinct trend phases: a period of continued similarity; a period of transition; and a period of sustained change. The model for the sequence of paired differences is the easier one of the two AR methods to use and interpret because its trend model of splined linear segments readily defines each response phase. Hence, we recommend it over the given alternatives. It offers water resources researchers an effective and readily adoptable analysis option.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Share

COinS
 
Apr 30th, 2:00 PM

AN AUTOREGRESSION MODEL FOR A PAIRED WATERSHED COMPARISON

Analysis of water quality data from a paired watershed design is needed to determine if a best fertilizer management practice reduces a specific water quality variable compared to a conventional fertilizer management practice. This study examines an existing recommended method of analysis for paired watershed designs, simple analysis of covariance (ANCOVA) on time aggregated data, then offers two autoregression analyses (AR) as alternatives. The first approach models the sequence of paired differences and estimates its 95% confidence band. The second approach develops individual watershed AR models then examines the joint 95% confidence interval about the predicted difference. A reliability analysis on the water quality data reveals that the data for the controlled watershed, i.e., the covariate, has a sizable measurement error, a factor that is not considered in the usual ANCOVA model. The AR methods avoid the measurement error and other inherent problems with the published recommended method. Graphically both AR analyses are similar and reveal three distinct trend phases: a period of continued similarity; a period of transition; and a period of sustained change. The model for the sequence of paired differences is the easier one of the two AR methods to use and interpret because its trend model of splined linear segments readily defines each response phase. Hence, we recommend it over the given alternatives. It offers water resources researchers an effective and readily adoptable analysis option.