On the Small Sample Properties of Norm-restricted Maximum Likelihood Estimators for Logistic Regression Models