Training Opportunities: Exploring what happens two months later

Model specifications

Using multinomial logit analysis, it is possible to calculate the contribution of each variable to the labour market outcomes. The dependent variable in this case is the category of outcome two months after each placement, and the independent or explanatory variables are individual and programme related factors collected for administrative purposes.

Logits were calculated for all variables in the model with respect to their contribution to a negative (other) outcome over other outcomes. This makes it possible to calculate an odds ratio for each value of the variable for each possible outcome.  For example, for the year variable, the logit shows how much more or less likely learners leaving placements in one year are likely to attain one type of labour market outcome compared with an other outcome. An other  describes a non-positive event, such as unemployment or out of the labour force status for the learner two months after leaving a placement, and is the base category for which all odds ratios are calculated in this analysis.

The advantage of this approach is that it enables control of all the other variables within the model so that the contribution of each variable to the attainment of each labour market outcome can be accurately assessed. Simply put, we can show how powerful each variable is in predicting outcomes. We can also calculate the odds of a positive outcome over a ‘negative’ outcome for any single variable value. For example, we can assess the likelihood of an outcome for a person in their third Training Opportunities placement, controlling for the fact that they are also male, European, and for their previous qualifications and employment experience. This is important because any number of possible combinations of these variables are possible and various interactions between them may change the final result. For instance, young learners who are males in Auckland may have different outcomes from young people who are female in the Southern region. If we calculate the odds for each iteration of each variable, while controlling for all other possible iterations of other variables, we can be more sure of what’s driving the placement outcome attainment than by just looking at the observed results.

The model is limited by the data that is available to it, so its explanatory power may be relatively low.³ However, this is common where analyses of education programmes are concerned. It is simply not possible to account for every variable that may have an effect on the outcome of Training Opportunities. But, the administrative dataset is a rich source of information, and by performing this analysis we are able to show with some confidence what some of the more powerful predictors of each outcome may be.

It is also not possible to include all of the administrative variables within the model, primarily because of collinearity issues. For example, trainee weeks and credit attainment are highly correlated, as are learner age and educational attainment. Where collinearity is indicated, one or more correlated variables are excluded to increase the precision of model estimates.

Footnote

3. A pseudo R Square statistic for the model was not produced.