4. Latent class analysis

In this chapter, we present the results of a latent class analysis (LCA) based on a pooled sample of individuals with no or weak labour market attachment in the five Nordic countries. The use of a pooled model serves multiple purposes: it increases the sample size, enhances comparability between countries, and simplifies the interpretation of results by avoiding unnecessary complexity associated with estimating separate models for each country.

Combining data from multiple countries in an LCA presents potential challenges, such as cultural and contextual variations that may inappropriately influence the latent classes. While acknowledging this, it is important to note that the Nordic countries share a relative similarity in terms of culture and labour markets, reducing the magnitude of this challenge. Further, combining data from different countries can also affect the model in negative ways due to measurement invariance across countries. In the analysis, we use the harmonised EU-SILC. Therefore, the data and the measurements we use to identify employment barriers are as similar as possible across the Nordic countries, meaning that measurement invariance does not possess a great challenge in the analysis below.

The chapter begins by presenting the key findings and interpretations from the LCA. Subsequent sections provide a detailed exploration of the outputs that led to these conclusions, allowing for a comprehensive understanding of the results.

In this section, we will present the main conclusions and our interpretation of the results from the latent class analysis (LCA). Table 4.1 summarises the output of the LCA. First, it reveals the identification of 8 data-driven target groups, which represent subgroups of individuals sharing a similar set of employment barriers. The primary barrier and – if any – secondary barrier refer to the most important barrier(s) for the data-driven target group. The table also contains information about the share of individuals with no or weak labour market attachment assigned to the group. Lastly, we provide a short description of the group.

The table demonstrates several interesting findings from our analysis. First, it shows that there is a somewhat clear distinction between groups with a relatively high share of individuals with no recent experience in the labour market and groups with a relatively low share of individuals with recent experience. The first 5 groups, to a large degree, all face an employment barrier related to no recent experience, whereas the remaining 3 groups, to a lesser degree, face barriers related to recent experience.

Another interesting finding relates to Group 8, which comprises approximately one-third of the individuals with no or weak labour market attachment, making it the largest identified data-driven target group. This group consists of individuals who are relatively close to the labour market and can be considered temporarily out of work, as a small percentage of individuals in this group face the barrier of no recent experience in the labour market. In fact, this group does not have any dominant barriers, and their barrier set appears to be simpler compared to the barriers sets of the other 7 data-driven target groups. On average, they face only one barrier, which is significantly low compared to some of the other identified groups.

Lastly, it is also important to be aware of the existence of individuals with more complex combination of barriers, such as those who have never worked (Group 1) or those who lack both work experience and education (Group 2). On average, individuals in these groups face three and four employment barriers, respectively.

Consideration of these 8 identified data-driven target groups is paramount when aiming to place a heightened emphasis on addressing employment barriers within future employment policies in the Nordic countries. These data-driven target groups could be used to facilitate discussions of the strengths and limitations of different policy interventions for concrete groups of policy clients. They could also be used to help inform decisions on whether to channel additional efforts towards specific priority groups.

In this section, we will delve deeper into the 8 identified groups and highlight the most and least prevalent barriers within each group, which forms the basis for naming them. To briefly recap, LCA is a method that exploits the interrelations of the employment barriers to identify population subgroups sharing the same employment barriers. In other words, the only input to the statistical model is the employment barriers each individual faces. Hence, the model completely disregards the traditional demographic target groups.

Figure 4.1 shows how prevalent the 10 barriers are in each identified group. Each column represents the prevalence of the barrier in the group. For instance, the first group of individuals all face an employment barrier related to having never worked, and this is the primary barrier that identifies this group of individuals. In comparison, the second group of individuals are quite like the first one in many ways since they have also never worked, but this group of individuals also lack education. These two groups consist of 11 pct. of all the individuals with no or weak labour market attachment in the Nordic countries.

The largest group is Group 8, which consists of 33 pct. of the individuals with no or weak labour market attachment. The group consists of individuals who are relatively close to the labour market, hence can be considered to be temporarily out of work, since a low share of the individuals face the barrier related to no recent experience. In fact, this group of individuals do not have any ruling barriers in general, and their barrier set is simpler compared to the barrier sets of the other 7 data-driven target groups.

The second-largest group is Group 3, which consists of 21 pct. of the individuals with no or weak labour market attachment in the Nordic countries. We have labelled this group health issues and lack of education since 98 pct. of the individuals in the group have either physical or mental health issues that limit the individuals in daily activities. In addition, 42 pct. of the individuals do not have any competence-giving education, which is relatively high compared to most of the other groups.

In addition to variations in the most common barriers among the 8 data-driven target groups, there are also differences in the average number of employment barriers in each group. Figure 4.2 illustrates these differences, showing that the average number of barriers ranges from 1 to 4. Group 2 is the group facing the highest number of employment barriers on average, while Group 8 experiences the fewest barriers. This further emphasises that individuals in Group 8 are closer to the labour market and may require less comprehensive support to overcome their employment barriers, unlike Group 2, where individuals face a more complex set of barriers. In addition to variations in the most common barriers among the 8 data-driven target groups, there are differences in the average number of employment barriers with each group.

In this section, we look further into the demographic characteristics of the individuals in each data-driven target group. It is possible to provide these statistics since everyone in our sample is assigned to one of the data-driven target groups. Hence, ex-post, it is possible to study the individuals in, e.g., the first group in terms of demographic characteristics. If each traditional demographic target group is uniquely identified in any of the data-driven target groups, it shows that the traditional demographic individuals within a target group face the same employment barriers. If each traditional demographic target group, on the other hand, is grouped into various data-driven target groups, this shows that each of the traditional demographic target groups is highly heterogeneous in terms of the employment barriers faced by the group. Therefore, tailoring policies to only the most prominent real or assumed barriers facing these groups may not be sufficient for increasing their employment chances.

Figure 4.3 shows how each of the data-driven target groups consists of the traditional demographic target groups. For example, the first data-driven target group consists of 28 pct. young people, 4 pct. seniors, 39 pct. immigrants, 13 pct. individuals with severe physical/mental health issues, and 16 pct. other individuals with no or weak labour market attachment. Starting with the discussion from the introduction to this section, the figure shows that none of the traditional demographic target groups completely dominate any of the data-driven target groups. For example, none of the data-driven groups consist of more than 50 pct. of any of the traditional target groups, and the general picture is that each of the data-driven target groups, broadly speaking, consists of a diverse composition of the traditional demographic target groups. However, there are exceptions to the rule since certain groups, to a significant degree, comprise some of the traditional target groups. For example, Group 6 consists of 45 pct. immigrants, Group 7 consists of 38 pct. young people, and Group 4 consists of 39 pct. persons with health issues.

In Table 4.2, we look further into the composition of gender, age, and immigrant status in the data-driven target groups. On the one hand, the table further supports the picture that none of the data-driven target groups are dominated by any specific demographic characteristics. On the other hand, some characteristics are relatively more prevalent in each of the data-driven target groups. For example, Group 4 consists of 86 pct. individuals who are older than 50 years, and Group 5 consists of 74 pct. women. In that sense, the traditional demographic way to define target groups cannot be completely disregarded since, according to the LCA, they share similar set of barriers in some instances and to some degree.

Table 4.2 also shows that women, in general, are overrepresented in the population of individuals with no or weak labour market attachment in the Nordic countries. More specifically, women constitute 57 pct. of the individuals with no or weak labour market attachment in the Nordic countries, which points to women, in general, being more likely to have potential labour market difficulties. Concerning the number of barriers between men and women with no or weak labour market attachment, we also see that women are slightly more challenged compared to men. Women with no or weak labour market attachment face 1.96 employment barriers on average, whereas men with no or weak labour market attachment face 1.86 employment barriers on average.

Notice that Figure 4.3 and Table 4.2 are not completely comparable since it is required in Figure 4.3 that each individual be assigned to only one of the traditional target groups. If any overlap occurred (e.g., both young and immigrant or both health issues and immigrant), we made use of the following hierarchy: persons with disabilities, immigrants, young people/seniors. This means that individuals are assigned on the basis of health issues before immigrant status and on the basis of immigrant status before being either young or senior. Therefore, some discrepancy can be found between Figure 4.3 and Table 4.2. For example, Group 2 consists of 32 pct. immigrants in Figure 4.3, while it consists of 50 pct. immigrants in Table 4.2.

As alluded to in the introduction of this chapter, we have used a pooled sample of individuals with no or weak labour market attachment in the five Nordic countries. Therefore, it is interesting to examine the groups to which the populations of these five Nordic countries are allocated. This is presented in Figure 4.4, which shows the fraction of, e.g., Swedes allocated to Group 1, Group 2, Group 3, etc. The figure shows that the populations of the five Nordic countries are more or less evenly distributed in the 8 data-driven target groups. For example, Group 3 consists of 15 pct. of the Swedes, 25 pct. of the Danes, 20 pct. of the Finns, 29 pct. of the Norwegians, and 20 pct. of the Icelanders with no or weak labour market attachment. This demonstrates that no Nordic country stands out significantly in terms of the barriers faced by their individuals with no or weak labour market attachment.

These findings indicate two important points. First, the cultural and contextual differences among the five Nordic countries are relatively similar, justifying our use of a pooled sample. Second, individuals with limited labour market attachment in the Nordic countries generally encounter similar employment barriers, suggesting the potential for cross-Nordic learning and collaboration. However, it is worth noting that this result is naturally influenced by the fact that we have been able to operationalise only 10 employment barriers related to 9 out of the 24 employment barriers from our framework.

The descriptive statistics in the previous chapter reveal that the majority of the individuals with no or weak labour market attachment face two or more employment barriers, while a not-trivial minority face four or more barriers. The simultaneous presence of several barriers is confirmed in the analysis using LCA, which also shows that each of the traditional target groups is very heterogeneous in terms of employment barriers. None of the data-driven groups estimated by LCA consist of more than 50 pct. of any of the traditional target groups, and the general picture is that each of the data-driven target groups, broadly speaking, consists of a diverse composition of the traditional demographic target groups. An implication of this is, for example, that a group of immigrants may have more in common with a group of young people than with other immigrants in terms of composition of employment barriers.

Young people, seniors, immigrants, or persons with disabilities are often used as proxy groupings in policy discussions, assuming that these categories effectively capture distinct sets of employment barriers that can inform policy formulation and implementation (Fernandez et al., 2016). However, it is important to note that being, for example, young or a senior, in and of itself, does not constitute an employment barrier, and the empirical findings in this report demonstrate that relying on broad demographic categories as a shorthand for the unique challenges experienced by individuals with no or weak labour market attachment has its limitations. Regardless of how employment policy initiatives are organised, it is crucial to recognise that experiences about effective interventions for one of the traditional groups, for example seniors, are also considered for other groups, for instance immigrants, with the same combination of barriers. Nevertheless, it is equally essential to acknowledge that the traditional demographic target groups cannot be entirely disregarded, especially when addressing specific barriers. For instance, policy actions aimed at addressing the lack of education among young people will differ from those aimed at addressing the lack of education among seniors. Moreover, specific barriers associated with particular traditional target groups do exist. For example, lack of language skills is a barrier specifically affecting immigrants.

In the next phase of this project, we will assess how current activation and labour support policies in the Nordic countries suit the individuals with weak labour market attachment. Here, a starting point is the findings in the previous report, which identify 24 employment barriers, along with the empirical findings in this report of the prevalence and combination of barriers among individuals with no or weak labour market attachment. The evaluation will be achieved through a combination of extensive literature reviews on the effect of existing labour market measures and qualitative research, including interviews with relevant experts and practitioners.