| Major objectives of this dissertation research were: (1) to determine if regression models of Iowa corn yields derived from data stratified by area (east and west) and time periods (1957-1963 and 1964-1970) and from statewide data were different from each other; (2) to compare their yield predictability in different data sets; and (3) to test the model selection process by using a statewide model with 90% of the data to predict yields of the remaining 10% data. Corn yield and related data from 2657 observations at 678 sites in 15 Iowa counties represented most soil association areas.;The final models were used to estimate yields ((')Y) of all observed values (Y(,i)) in each data set. The prototype model, Y(,i) = (')Y(,i), gave the statistics for these tests: F-test, transfer hypothesis, R('2), bias, relative bias, and relative standard deviation. Other comparisons were based on standard partial regression coefficients, latent roots and corresponding vectors, and relative gains in precision.;The comparisons proved that: (1) area and time trends were different as shown by significant differences among the 1/4-data set models; each was best for its own data set but poor for others; (2) interaction models for Time2 and statewise data were not significantly different, but the Time2 and West-Time2 data; and (3) since no difference occurred between observed and predicted yields of the independent 10% data set (predicted by the 90%-data model), the model selection process was valid.;Models for predicting Iowa corn yields have the following relative precision: East-Time2 and West-Time2 in its own area > Time2 > statewide (Iowa).;Stratification generated 8 data sets: East-Time1, East-Time2, West-Time1, and West-Time2, each with about 1/4 of the observations, and East, West, Time1, and Time2, each with about 1/2 of the observations. Quadratic models (only linear or quadratic functions) were derived for each of the 8 data sets; interaction models (interactions added) were derived for the 1/4-data sets, Time2 ( 1/2-data set), statewide data set, and 90% of the statewide data. |
