Appendix C – Sensitivity checks

In the previous section we saw how we determined the number of classes in the Latent Class Analysis. In this section we will do some informal test of the sensitivity of the model with 8 groups (latent classes). We do so by estimating a model with 6 groups and compare it to the model with 8 groups. Our results are robust if we do not see any major differences between the model with 8 groups and the model with 6 groups.

Figure 6.2 presents the model comprising six distinct groups. The figure visually represents the identified groups, depicting the prevalence of each barrier within each group, the respective group sizes, and the designated names we have assigned to them. From the figure we see a great overlap compared to the model with 8 classes, which was presented in chapter 4.

First, in the LCA with 6 groups we also identify a group where everyone faces the barrier never worked, whereas we in the LCA with 8 groups identified two groups who has never worked. In the LCA with 6 groups it consists of 11 pct. of the individuals with no or weak labour market attachment, whereas the size of the two groups in the LCA with 8 groups who has never worked is 6 pct. and 5 pct. respectively. This suggests that the group who has never worked in the LCA with 6 groups is divided into two groups in the LCA with 8 groups; one consisting of individuals whose primary barrier is the fact that they have never worked, and one with individuals who have never worked but who also lack education.

Figure 6.2 Latent Class Analysis estimated with 6 latent classes

Second, in both models we identify a group of individuals who struggle with health issues and have low contact with PES and in both models these groups consist of 4 pct. of the individuals with no or weak labour market attachment in the Nordic countries. Similarly, we find a group of approximately the same size in both models consisting of individuals whose primary barriers is lack of education and lack of skills. This is the general finding between the two models i.e., groups identified in one model is possible to find in the other model.

In general, this suggests that our preferred LCA with 8 groups is relatively stable when it comes to changing the number of classes. It will per definition change something in the output when you vary the number of groups, but as just demonstrated above the differences are minor between the two models.