Abstract
In the linear regression model for regressing the mean yield of a crop (wheat) on predictor variables which are based on meteorological records from official weather stations and a technology time trend, the classical approach for estimating variances of the parameters and variances of prediction are not strictly applicable for two reasons. 1) There is a sampling error induced by using a sample of weather stations in the state to predict the mean yield for all the fields in the state for some future year. 2) There is a correlation between the residuals in the model and the predictor variables that arises from the fact that the weather is measured from locations in the state (weather stations) that are geographically different from the locations where the yield measurements are taken (the farms). To account for these two difficulties, the least squares estimation of parameters and variances will be modified to allow for the sampling error and used to develop an asymptotic theory which allows for the correlation by viewing the problem as a linear model with errors of measurement in the predictor variables. In addition, an analysis of variance technique is investigated which takes account of both the above problems and provides an alternative to the least squares approach. A Monte Carlo study is included to compare the different approaches.
Booker, Jane Broadt (1978). Mean yield prediction when predictor variables are subject to error. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -235430.