New Forms of Dualization? Labour Market Segmentation Patterns in the UK from the Late 90s Until the Post-crisis in the Late 2000s

Abstract

There has been an increase in literature that examines the patterns of dualization in labour markets across different welfare states. However, rarely do these studies empirically explore how labour markets are divided. Rather they assume a certain type of division to exist in a market, and apply this assumption to measure the extent to which this division can be observed. This paper aims to overcome this limitation by examining the labour market dualization patterns of the UK’s employed population over the past decade through a latent class analysis model. Our analysis shows that the UK labour market could be characterised by a three group system during the period between 1999 and 2010. This divide supports the theoretical literature on labour market divisions in that there are clear distinctions between those who are insiders and those who are not. However, what is interesting is that rather than having a dichotomised pattern of division of insiders and outsiders, we find a third group which can be characterised as a “future insecure” group. What is more, the main characteristics that divide the groups are not contract types (involuntary part-time or temporary employment), but rather income levels (low pay), occupational profile (low-skilled occupations) and social security benefits stemming from employment (occupational pension coverage). From the results, we conclude that the patterns and characteristics of labour market divisions may not be generalised and further empirical investigations are needed to understand the cross-national variations.

Appendix

1.1 Specification of the Model

The basic latent class model can be specified as follows:

Let j = 1,…, J observed variable, and observed variable j has r _j  = 1,…, R _j response categories. L is categorical latent variable with c = 1,…, C latent classes. We model the probabilities of belonging to class c as γ _c and the item-response probabilities of objected variable j is referred to as ρ _j .

As the latent classes are mutually exclusive and exhaustive, each individual is a member of one and only one latent class. Thus,

$$\mathop \sum \limits_{{{\text{c}} = 1}}^{\text{C}} \gamma_{c} = 1$$

(1)

Also, each individual provides one and only one response alternative to variable j, the vector of item-response probability ρ _j ,γ _j |c (the probability of response γ _j to observed variable j, conditional on membership in latent class c) always sums to 1. Therefore,

$$\mathop \sum \limits_{{r_{j} }}^{{{\text{R}}_{\text{j}} }} \rho_{j,} \gamma_{j} |c = 1$$

(2)

Suppose ${\mathcal{Y}}_{j}$ representing element j of a response pattern y. We model an indicator function I (${\mathcal{Y}}_{j} = \gamma_{j}$) that equals 1 when the response of variable $j=\gamma_j$, and otherwise equals 0. Then equation of how the probability of observing a particular vector of response is a function of the probabilities of membership in each latent class (the γ) and the probabilities of observing each response conditional on latent class membership:

$$P\left( {\text{Y} = \text{y}} \right) = \mathop \sum \limits_{c = 1}^{C} \gamma_{c} \mathop \prod \limits_{j = 1}^{J} \mathop \prod \limits_{{\gamma_{j = 1} }}^{{R_{j} }} \rho_{{j} ,\gamma_{j} |c}^{{I\left( {{\mathcal{Y}}_{j} = \gamma_{j} } \right)}}$$

(3)

The equation of the probability of a particular observed response pattern y conditional on membership in a particular latent class c is,

$$P\left( {\text{Y} = \text{y}|\text{L} = \text{c}} \right) = \mathop \sum \limits_{{c} = 1}^{C} \gamma_{c} \mathop \prod \limits_{j = 1}^{J} \mathop \prod \limits_{{\gamma_{j = 1} }}^{{R_{j} }} \rho_{{j} ,\gamma_{j} |c}^{{I\left( {{\mathcal{Y}}_{j} = \gamma_{j} } \right)}}$$

(4)

The next step, after computing the conditional response probabilities for each latent class c, is to obtain the unconditional joint probabilities of latent class c and response pattern y, P (Y = y, L = c), for each of the C latent classes. The equation for this is,

$$P\left( {\text{Y} = \text{y}, \text{L} = \text{c}} \right) = P\left( {\text{L} = \text{c}} \right)P\left( {\text{Y} = \text{y} |\text{L} = \text{c}} \right) = \gamma_{c} \mathop \prod \limits_{j = 1}^{J} \mathop \prod \limits_{{\gamma_{j = 1} }}^{{R_{j} }} \rho_{j} ,\gamma_{j} |c^{{I\left( {{\mathcal{Y}}_{j} = \gamma_{j} } \right)}}$$

(5)

In terms of conducting the multiple group latent class models, we introduce a group variable V with q = 1,…,Q groups. In our paper, V represents cohorts and there are Q = 11 groups: total dependent employees in BHPS from 1999 to 2010. Each response pattern y corresponding to cohort q is related with a probability of occurrence P (Y = y|V = q), and within each time point 1, $\sum P(\text{Y} = \text{y}|V = q) = 1$. Thus, the equation of conditional membership in latent class c and group q is,

$$P\left( {\text{Y} = \text{y}| \text{V} = \text{q}} \right) = \mathop \sum \limits_{c = 1}^{C} \gamma_{c|q} \mathop \prod \limits_{j = 1}^{J} \mathop \prod \limits_{{\gamma_{j = 1} }}^{{R_{j} }} \rho_{j} ,\gamma_{j} |c,q^{{I\left( {{\mathcal{Y}}_{j} = \gamma_{j} } \right)}}$$

(6)

See Table 5.

Table 5 Classification of occupation, based on ILO’s ISCO-08 (2012) and Schwander and Häusermann (2013)