Regression models for applications, acceptances, and enrollment at University of California-Berkeley.

Our Markov chain model (see Fig 1) requires specification of the parameters δ, α, and ϵ, which are probabilities derived from counts of applicants, acceptances, and enrollments. These counts must be specified for each year and for each racial/ethnic group. We assess the fit of linear, log-linear, and optimal Box-Cox models on historical data. We choose the preferred model, specified in the table above, and use it to forecast future values. Fig 3 shows various models for application count, corresponding to the top section of the table. The column labeled Λ is an exponent used in the Box-Cox transformation, and thus is relevant only to those fits.