Binary data are of interest in many scientific researches. The most common model for examining these data is the logistic model, which is in the family of generalized linear models. Optimality criteria for finding optimal designs are usually based on the covariance matrix and the asymptotic properties of this matrix. The asymptotic properties of this matrix, including the unbiasedness of the elements, hold as long as the sample is large. Therefore, in small samples, the issue of parameter estimation bias arises, and in this case, the use of the mentioned criteria is not sufficient. Robinson and Khoury (2003) studied the investigation and comparison of designs for logistic models for small samples. In this paper, using the graphical and intuitive method of Robinson and Khoury (2003), we show that the prediction performance of minimax optimal designs is better than that of local optimal designs and weaker than that of Bayesian optimal designs.