Department of Education Oversight, Gatekeeping

In the last two years, schools reported a higher proportion of part-time staff. For the first three years for which we have data, the ratio of full- to part-time staff was about 3 to 1. In 1993 and 1994, it was approximately 2 to 1. The number of full-time equivalent staff, however, has not declined. The average number of full-time equivalents in 1994 was 16, the highest in the five years, and the ratio of students to faculty was 24.9, the lowest in five years.

School Performance by Enrollment and Program Length

The tables in this section present a direct comparison of the performance of school with differing full-time enrollments and program lengths. These tables repeat analyses conducted for the first time with the 1993 data. The 1994 results are very similar to found for 1993.

For the enrollment comparison, we divided the schools into four groups, and for the program length comparison, we divided the schools into five groups. We then calculated the average outcomes for schools in each of these groups. The results are presented in Tables 2.2 and 2.3.

School with full-time enrollment of 300 or less clearly have more success retaining and graduating their students than larger schools. The differences are most pronounced between the smallest and largest schools, but there are even differences of 4 to 7 percentage points between the 300 or

less and the 301 to 600 schools. The smallest schools also place more of their students in related employment than the larger schools.

Program length has an even stronger influence on graduation and withdrawal than size of enrollment. Table 2.3 presents schools outcomes for five categories of average program length. Recall that average program length is based on the length of each program offered by a school weighted by the number of full-time students enrolled in each program. Schools in the shortest category have graduation rates 29 percentage points higher and withdrawal rates 9 percentage points lower than schools in the longest category. These differences are slightly less than those found in 1993, but the overall pattern is identical.

On training-related placement there is a difference of 7 percentage points between the schools with the shortest and longest average programs. There is no difference in related placement among the three program lengths between the shortest and longest.

On three of the four measures, longer programs have less desirable outcomes. As in 1993, however, schools with the longest average programs have the lowest default rates. We think the two possible explanations we advanced in 1993 are still valid. The first is that students who complete longer programs may obtain higher paying jobs and be better able to repay their loans. The second is that longer programs are more expensive and student loans pay a smaller

proportion of total costs. Consequently, students in longer programs may be from families with higher incomes which are less likely to default on loans.

CHAPTER 3

MULTIPLE REGRESSION ANALYSIS OF ANNUAL TOTAL DATA

This chapter presents the results of multiple regression analyses of the annual total variables defined in Chapter 1. Five years of these data are available and this chapter summarizes those variables that have been found to have statistically significant associations with schools outcomes in at least three of those five years. The full regression results showing the results for all the variables used in the analyses are presented in Appendix Tables 3.5 to 3.8. The results for the 1994 annual total data are compared to those found for the cohort data in Chapter 4.

A multiple regression coefficient (R) reflects the degree of association between an outcome variable and the best possible combination of the explanatory variables. The square of this coefficient (R2) indicates the proportion of variability in the outcome that can be attributed to the explanatory variables. The closer the R approaches a value of 1.00, the better the independent variables explain variations in the outcome variable. The R must be over .70, however, before half the variability is explained. The adjusted R' controls for spuriously high Rs based on a small number of observations. Since many observations were used in these analyses, the adjustment reduces the R2 very little.

Before discussing the multiple regression results, however, we summarize the one-to-one correlations between the explanatory variables (school characteristics) and the outcome variables (school performance) for full-time enrollment'. One-to-one correlation coefficients (7) are interpreted in much the same manner as multiple Rs: the closer the coefficient comes to 1.00, the highest possible correlation, the more similar are the rates of variations in the two variables. It is not necessary that the measures of the two variables be similar, but changes in one variable must be accompanied by similar changes in the same direction in the other variable if there is to be a positive correlation.

The school characteristics that usually have one-to-one correlations of .20 or higher with school performance are presented in Table 3.1. We used the .20 level as a cutoff because is it highly significant statistically, and also begins to have practical significance as a school characteristic that should be given attention. (The full tables listing all the correlations of the school characteristics with the outcome variables are presented in Appendix Tables 3.1 to 3.4.)

It is important to note that correlation does not necessarily mean causation. Similar rates of variation in two variables may or may not reflect the effect of one of the variables on the other. To repeat an analogy used in previous reports: If we were to correlate the shoe size of men with their height, we would find a significant correlation. Taller men tend to have larger feet than shorter men. This does not mean that large feet cause men to grow taller or that height causes large feet. What causes both of these characteristics are the genetic components of

1 The correlations for part-time enrollments tend to be in same directions, but usually lower than the correlations for full-time enrollments. In general, school characteristics are less related to the outcomes of part-time students than they are for full-time students.

individuals as these components interact with the nutrition available in the environment. Both shoe size and height are only reflections of basic causes. In a similar manner, many of the variables used in this analysis are only reflections of more basic relationships between school and student characteristics and school outcomes.

All of the correlations with graduation rates are negative indicating that as programs become longer, the percentage of students receiving Pell grants increases, and the number of full-time students increases, graduation rates decrease.

TABLE 3.1

CORRELATIONS OF 20 OR MORE BETWEEN SCHOOL OUTCOMES
AND SCHOOL CHARACTERISTICS FOR FULL-TIME ENROLLMENTS,
SCHOOL YEARS 1990 TO 1994

Note: All correlations but two significant, p < .01; -.07 p < .02, - .05 p> .10

The first four correlation with withdrawal rates indicate that as these characteristics of schools increase, withdrawal rates increase also. The coefficient for the Ability-To-Benefit (ATB) variable has been declining, especially in the last two years. We noted in Chapter 2 that the percentage of ATB students has also been declining. If there is more careful screening of ATB students, this could explain the drop in this coefficient.

-.21

-.29

.26