A New Covariate Ratio Procedure for Estimating Treatment Differences with Applications to Climax I and II Experiments

View More View Less
  • 1 Colorado State University, Fort Collins, CO 80523
© Get Permissions Rent on DeepDyve
Restricted access

Abstract

A new covariate ratio procedure is presented for estimating a treatment-induced effect. The procedure 1) allows for uncontrolled natural variability, 2) adjusts for disproportionate allocation of non-treated and treated experimental units, 3) diminishes the influence in an objective manner of an individual value corresponding to any experimental unit, and 4) accounts for differential treatment effects, i.e., a simple location or scale change is not assumed. This procedure is applied to specific meteorologically defined partitions involving data of the Climax I and II experiments. Results based on the pooled data indicate a 32% precipitation increase for the −20 to − 11°C 500 mb temperature partition, a 49% increase for the 190 to 250° 700 mb wind direction partition, and a 13% increase for the total sample. Comparisons based on Monte Carlo simulations (re-randomization) indicate that this procedure yields estimates which are more stable (precise) than corresponding estimates based on the double ratio.

Abstract

A new covariate ratio procedure is presented for estimating a treatment-induced effect. The procedure 1) allows for uncontrolled natural variability, 2) adjusts for disproportionate allocation of non-treated and treated experimental units, 3) diminishes the influence in an objective manner of an individual value corresponding to any experimental unit, and 4) accounts for differential treatment effects, i.e., a simple location or scale change is not assumed. This procedure is applied to specific meteorologically defined partitions involving data of the Climax I and II experiments. Results based on the pooled data indicate a 32% precipitation increase for the −20 to − 11°C 500 mb temperature partition, a 49% increase for the 190 to 250° 700 mb wind direction partition, and a 13% increase for the total sample. Comparisons based on Monte Carlo simulations (re-randomization) indicate that this procedure yields estimates which are more stable (precise) than corresponding estimates based on the double ratio.

Save