The British low-wage sector and the employment prospects of the unemployed

ABSTRACT

Are low wages a way for the unemployed to switch to higher-paying jobs? Using data from the British Household Panel Survey, the labour market dynamics of unemployed, low-paid and higher-paid employed men are analysed. Moreover, the respective (un)employment duration and occupational skill level are accounted for. Results show that in general low wages significantly reduce the risk of future unemployment and increase the chances of ascending the salary ladder, especially in the case of long-term unemployment (>360 days). Furthermore, the occupational skill level has a substantial influence on the upward mobility of low-paid jobs: individuals working in the initial period in a low-paid and higher-skilled occupation have on average an 11 percentage points higher probability of entering higher pay compared to when working in a low-paid and low-skilled occupation.

KEYWORDS:

• Unemployment dynamics
• low-pay dynamics
• initial conditions problem
• random-effects probit models
• maximum simulated likelihood

JEL CLASSIFICATION:

• J64
• J62
• J31

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ Low pay and low wage are used as synonyms in this study.

² This might go along with the differentiation of ‘good’ and ‘bad’ jobs (Acemoglu 2001).

³ In the literature, this effect is also called the ‘stepping stone effect’ (see, for instance, Uhlendorff 2006) or ‘springboard effect’ (see, for instance, Knabe and Plum 2013).

⁴ To control for this effect, in a robustness estimation, the sample is restricted to individuals without post-secondary education (see the section on Job-related characteristics).

⁵ To identify self-employed, we use the information whether the individual is an employee or self-employed at the current job (WAVE-JBSEMP).

⁶ According to the ILO definition, the second restriction for the differentiation between being unemployed and inactive is whether the person is ready to begin work within the next two weeks. The BHPS does not have any information concerning this issue; therefore, the differentiation is solely based on the searching scheme.

⁷ Moreover, we differentiate the currently employed according to the length of their employment spell into the following three categories: employment spell <90 days, 90–360 days and >360 days. However, it is not possible to differentiate the employment spell into higher-paid and low-paid employed.

⁸ Some men receive a gross hourly wage that is slightly above or below the low-wage threshold. A small change in their pay or in their working hours could lead to a change in their labour market position, even though their gross hourly wage remains almost unaffected. When a large percentage of men change their labour market position and only experience a negligible change in the hourly wages, this could result in an overestimation of the transition probability between the two employment positions. In a robustness check, those workers who changed their labour market position and experienced a change in their wage of 10% (in absolute terms) or less were dropped. However, this did not affect the main findings of the study.

⁹ In a robustness estimation, to derive the occupational skill level, we used information based on the socio-economic group classification that distinguishes jobs into 19 categories. Personal service work, semi-skilled or unskilled manual work and agricultural work were classified as low-skilled occupations; otherwise, occupations were higher skilled. The findings hardly changed.

¹⁰ The wage distribution for the total sample can be found in the Appendix (see Figure A 1).

¹¹ Another potential approach is to model each combination of labour market position separately (e.g. unemployed, higher-paid employed in a higher-skilled occupation, higher-paid employed in a low-skilled occupation, low-paid employed in a higher-skilled occupation and low-paid employed in a low-skilled occupation). However, the multinomial random effects logit model as suggested by Uhlendorff (2006) did not converge. This issue may be picked up in the future research, for example, by using a different data set or estimation strategy.

¹² Due to the lack of information, it is not possible to decompose further the employment duration into higher-paid and low-paid spells. We assume that the effect of the employment duration is homogeneous between low-paid and higher-paid employed, and therefore, no additional interaction included.

¹³ A list of control variables can be found in Table A 1 and descriptive statistics in Table A 2.

¹⁴ ${\mathrm{H}}_0:{\hat{\eta }}_{21}={\hat{\eta }}_{22};{\chi }^2\left(1\right)=0.08\left[p-\mathrm{v}\mathrm{a}\mathrm{l}:0.7770\right]$.

¹⁵ ${\mathrm{H}}_0:{\hat{\gamma }}_{12}={\hat{\gamma }}_{13};{\chi }^2\left(1\right)=0.23\left[p-\mathrm{v}\mathrm{a}\mathrm{l}:0.6286\right]$; ${\mathrm{H}}_0:{\hat{\gamma }}_{22}={\hat{\gamma }}_{23};{\chi }^2\left(1\right)=0.61\left[p-\mathrm{v}\mathrm{a}\mathrm{l}:0.4356\right]$.

¹⁶ ${\mathrm{H}}_0:{\hat{\pi }}_{12}={\hat{\pi }}_{13};{\chi }^2\left(1\right)=2.09\left[p-\mathrm{v}\mathrm{a}\mathrm{l}:0.1479\right]$.

¹⁷ ${\mathrm{H}}_0:{\hat{\pi }}_{22}={\hat{\pi }}_{23};{\chi }^2\left(1\right)=19.63[p-\mathrm{v}\mathrm{a}\mathrm{l}:<;0.001]$.

¹⁸ The calculation of the APE can be found in the Appendix.

¹⁹ The estimation output can be obtained from the author upon request.