|
Abstract:
|
Regression analysis of empirical data often suffers the problem of bad fit, e.g., low R2, and the interpretation of the parameter estimates is inconsistent from sample to sample leading to conflicting empirical evidence. A modeling strategy to provide a better fit locally is to use the forward search algorithm to separate the sample into several subgroups such that a well-fit regression line is obtained within each group. The main objectives of this study are to investigate the conditions that the strategy works well, and to examine a number of stopping rules that are used to determine the subgroups. With reliable estimates, the significance of parameter becomes meaningful. The strategy is then applied to study the effect of distributor dependence on the supplier-distributor commitment in the manufacturing industry where various empirical studies reported conflicting conclusions.
|
