Appendix 6: Discriminant Analyses

[next chapter] [previous chapter] [contents]

Several different discriminant analyses were run on the questionnaire data using Statistical Package for the Social Sciences. The following variables were used: mode (on-campus or distance education); gender; whether student had studied in previous 12 months; family situation; level of prior education at entry; and whether students enjoyed studying at university.

In the first run using variables of mode, enjoyment and gender, 63.6 per cent of cases could be correctly classified as being likely to drop out. When whether the student had studied in the previous 12 months was added, 65.8 per cent could be correctly classified. This increased to nearly 70 per cent when the family situation and previous level of education were also added.

The variables below were coded in the following way:

Standardised canonised discriminant function coefficients were developed for the variables in each of the four runs of the data with above variables entered.

RUN1

Variables entered. Mode, enjoy, gender.

n=77. Stayed=45, dropped out=32.

P=.049, Canonical R=.318

The standardised canonical discriminant function coefficients were:

mode .67

enjoy .63

gender .42

The number of cases that could be correctly classified was 63.6 per cent.

RUN2

Variables entered. Mode,enjoy,gender,do1x.

n=76. Stayed=44, dropped out=32.

P<.05, Canonical R=.356

The standardised canonical discriminant function coefficients were:

mode .56

enjoy .53

do1x -.52

gender .36

The number of cases that could be correctly classified was 65.8 per cent.

RUN3

Variables entered. Mode,enjoy,gender,do1x,famx.

n=74. Stayed=42, dropped out=32.

P<.05, Canonical R=.37

The standardised canonical discriminant function coefficients were:

do1x .50

enjoy .47

mode .41

gender .39

famx .33

The number of cases that could be correctly classified was 68.9 per cent.

RUN4

Variables entered. Mode,enjoy,gender,do1x,famx,edx.

n=63. Stayed=36, dropped out=27.

P=.044, Canonical R=.448

The standardised canonical discriminant function coefficients were

gender .63

enjoy .46

do1x -.45

famx .37

mode .31

edx .06

The number of cases that could be correctly classified was 68.3 per cent.

This discriminant analysis technique led to the following conclusions.

1. Education level is not useful as a predictor to determine if indigenous students will drop out.

2. The most useful predictors are gender, enjoyment level, mode of study, whether the student had been studying in the previous 12 months prior to entry, and their family situation.

3. Indigenous students are more likely to drop out if:

• they are male;
• they live alone;
• they are on-campus rather than external students;
• they have not studied in the previous 12 months prior to entry;
• they had less enjoyment of the situation of studying.

4. It can be stated that there is about a 40 per cent chance of a student dropping out. If we have no information about a particular student and we predict that they will drop out, we will be correct about 40 per cent of the time. However if we know the values of the five variables mentioned in part (2) above for a particular student, we can increase the correctness of our prediction to about 70 per cent.

5. The number of students staying in analyses dropped from 77 in Run1 to 63 in Run4 due to certain categories not being included for some of the variables. This contributed to the fact that the relative importance of any particular variable as reflected by its standardised canonical discriminant function coefficient was not consistent over the different runs. For example, mode had a higher coefficient than enjoyment or gender in Run1, but a lower coefficient in Run4.

Most of the students eliminated from Run4 were in the TAFE etc category of the education variable. It may be useful at some other time, to examine these students more closely as their absence or presence in the analyses seems to affect the importance of some of the other variables, particularly gender and mode.