Abstract
The effectiveness of vocational education and training (VET) depends on the quality of interactions between the actors from the education and employment systems, which ensure the correspondence of skills supply and demand. This paper develops an instrument to measure education-employment linkage (EEL) by capturing EEL in each sub-process where these actors can and should interact. Surveying VET experts from 18 countries suggests that countries with dual VET have the highest EEL, while the included Asian countries score lowest in terms of EEL. The analysis further reveals that the three most important sub-processes are employer involvement in the definition of qualification standards; employer involvement in deciding the timing of curriculum updates; and the combination of workplace training with classroom education.
Acknowledgements
We thank the Center on International Education Benchmarking of the National Center on Education and the Economy (NCEE) for funding the study. We further thank participants of the LH-KOF-Retreat 2016 and the JVET Conference 2017 for helpful comments and suggestions.
Notes
1. A few features are missing for all experts in a country due to the low number of country-specific experts. We address this issue by aggregating these features before the estimation, using equal weights for each feature. This approach of aggregating features into a single variable allows to increase the number of observations to a reasonable amount, but has the drawback that the data-driven weighting within the aggregated variable remains unknown. Therefore, we only aggregate variables where necessary to ensure a reasonable sample size.
2. If our empirical framework were to suffer from substantial omitted variable bias because it fails to capture the most relevant features of EEL, the subjective and semi-objective methodologies would yield substantially different results. Comparing the two lets us evaluate the validity of our empirical framework to some extent.
3. Though our baseline methodology consists of estimating Equation (3) on the level of countries, we could calculate Equation (4) based on feature scores x ik of individual experts and aggregate the results to the country level. The problem of this approach is that missing values due to item non-response or filtering of features results in an implicit change in the weighting scheme. To illustrate this, consider two experts evaluating two features with equal weight that have the score seven and one, respectively. If the second expert has a missing value for the second feature, calculating EEL followed by averaging yields an EEL of 5.5. Calculating the average feature score before calculating EEL according to Equation (4) on the other hand yields the true score of four. In order to minimise this type of error, we use the country average, x c , in Equation (4). However, the KOF EEL Index based on expert-level feature scores has a correlation of 0.99 with the baseline index discussed in the paper.
4. However, some countries have missing values in some features. In this case, we adjust the weights of Equation (4) to sum to one.
5. An alternative approach is to aggregate features into sub-processes using equal weights of features and regressing Aggregate EEL cp on the resulting sub-process scores. Though the relative weights of sub-processes differ in some cases, the resulting KOF EEL Index has a correlation of 0.9 with the baseline strategy.
6. In analogy to the calculation of averages for feature scores, ϕ p represent the average of country averages. However, the correlation of the KOF EEL Index based on averages across experts yields a correlation of 0.99.
7. As an exception, based on the recommendation of the Singaporean government, we evaluate EEL of the Institutes of Technical Education (ITE) in Singapore, which are located at the post-secondary level.
8. Differences to the values reported in Renold et al. (Citation2016) arise because of averaging features across experts before calculating the KOF EELI rather than the other way around.
9. Note that the average of the KOF EEL Index across experts is not numerically identical to the KOF EEL Index based on the average of each feature across experts.