I read these from a classical paper:


With a discrete dependent variable, however, the R2 in the OLS regression is not a meaningful measure of the goodness-of-fit. The OLS estimate of the R2 is generally underestimated when the dependent variable is dichotomous.


For qualitative choice models, there is no way to quantify precisely the unexplained variance of the equation, as is done with the R2 in OLS regressions. The measures of goodness-of-fit that do exist provide only relative comparisons between models; they have no meaning in an absolute sense.


One such measure is the percentage of acceptances and denials correctly predicted. The evaluation of the percentage correctly predicted will depend upon the sample frequencies of the alternatives; thus, the most commonly assumed baseline for comparison is a logit with only a constant term, which in this model is equivalent to assuming that every application is accepted. This baseline correctly predicts 85 percent of the applications since 85 percent of the applicants were accepted for a loan. The base equation in column 1 of Table 2 correctly predicts over 95 percent of the applications. Using this criterion, the number of misses if diminished by two thirds with the addition of our supplementary variables to the equation.
I don't understand "The evaluation of the percentage correctly predicted will depend upon the sample frequencies of the alternatives; thus, the most commonly assumed baseline for comparison is a logit with only a constant term, which in this model is equivalent to assuming that every application is accepted. "

Can anyone help me?

Many thanks in advance!