Abstract
In this study, the effects of multicollinearity on the maximum likelihood estimator are analyzed in the probit regression model. It is known that the near-linear dependencies in the design matrix affect the maximum likelihood estimation negatively, namely, the standard errors become so large so that the estimations are said to be inconsistent. Therefore, a new jackknifed ridge estimator is introduced as an alternative to the maximum likelihood technique and the well-known ridge estimator. The mean squared error properties of the listed estimators are investigated theoretically. In order to evaluate the performance of the estimators, a Monte Carlo simulation study is designed, and simulated mean squared error and squared bias are used as performance criteria. Finally, the benefits of the new estimator are illustrated via a real data application.
Disclosure statement
No potential conflict of interest was reported by the author(s).