This paper deals with the problem of nonstationarity of regressors in binary choice model. The limit distribution of the ML-estimator is mixed normal, but restriction testing shall not be based on standard t-statistic. The results of the conducted Monte Carlo experiment demonstrate that the true size of the restriction test is far from the significance level. Therefore, the t-Student statistic should be modified and this paper proposes its modification. The results of the Monte Carlo investigation point to the superiority of the new statistic.