METHODS FOR SMOOTHING EXPECTANCY TABLES APPLIED TO THE PREDICTION OF SUCCESS IN COLLEGE

Six methods for smoothing expectancy tables were compared, using data for entering students at 86 colleges and universities. Linear regression analyses were applied to ACT scores and high school grades to obtain predicted first-term grade point averages (FGPA's) for students entering each institution in 1969-70. Expectancy tables were constructed for each institution using the relative frequency with which students in each predicted FGPA interval obtained FGPA's of "B or better" and "C or better." The methods used to smooth the 1969-70 expectancy tables were compared on how closely the smoothed tables corresponded to the 1971-72 and 1972-73 observed data. The smoothed tables were substantially more accurate than the 1969-70 observed data on which they were based. Two maximum likelihood methods provided the most accurate tables. Estimated relative frequencies were more accurate for the "B or better" level than for the "C or better" level.