Use the Pearson residuals versus event probabilities to assess the appropriateness of the fitted probit model. Any unusual pattern or trend in the Pearson residual plot indicates that the fitted probit model may be inappropriate.
Note that the Pearson residuals account for the binomial response variable. Because the response variable follows a binomial distribution, the variability is usually highest at stress levels where approximately 50% of the units fail. Ordinary residuals, like those used in regression and ANOVA, do not account for this non-constant variance due to the binomial response variable.
For the windshield data, the Pearson residuals are randomly scattered about zero. Therefore, the plot does not indicate any problems with the selected probit model.