Female labour force intermittency and current earnings: switching regression model with unknown sample selection

Abstract

Using the Health and Retirement Survey from the USA, this paper finds a 16% selectivity-corrected wage penalty among women who engage in intermittent labour market activity. This penalty is experienced at a low level of intermittent activity, but appears to not play an important role in a woman's decision to undertake such activity. In addition, employer preferences appear to play a larger role than human capital atrophy in the determination of the wage penalty.

Acknowledgements

The authors would like to thank Brian Armour, Eric French, Joyce Jacobsen, Robert E. Moore, Yongsheng Xu, and participants of the Federal Reserve System Microeconomics Conference, 2–3 June 2003, Chicago, for comments and suggestions. The views expressed here are those of the authors and do not necessarily represent the views of the Federal Reserve Bank of Atlanta or the Federal Reserve System.

Notes

See, for example Stratton (1995), Jacobsen and Levin (1995), and Baum (2002). This penalty for intermittency is often offered as one source of the lower wages observed for women relative to men. The role of intermittent activity in determining this gender wage differential, however, is not the subject of this paper. In addition, the word penalty is used to describe the loss in potential wages dues to intermittent activity and is not intended in a pejorative manner.

Based on the authors' calculations from the March Current Population Survey, 1988–2002.

The appeal of a single index in order to combine these multiple dimensions into one quantifiable measure is similar to the long-standing search by development economists for a single index to quantify the multi-dimensions of social development of a country (for an example, see Cahill and Sanchez, 2001).

This index construction is analogous to the commonly used index of female labour market attachment; see Maret-Havens (1977). As the index is constructed here, number of periods of absence and duration of absences receive the same weight. It is possible that these components are weighted differently by employers in assessing to which sector a woman should belong. Exploring further variations in this index is the subject of future work. In the actual construction of the index, the number of periods a woman experiences, N_i , is scaled by the maximum number of periods observed in the data set; this ensures that each component of the index ranges between zero and one. The data used for estimation provides no information to quantify the characteristics of a woman's previous employment (such as hours of work or occupation). This index builds upon an earlier version which did not include a measure of length of time since last spell (Hotchkiss and Pitts, 2003). Use of that index also indicated a wage penalty at very small values.

A woman with no spell of absence would have a value of 100% of work life since last spell.

This index does not account for delays in entrance into the labour force, only the penalty associated with intermittent attachment once the individual has chosen to enter the labour force.

See Baum (2002) for an analysis of the impact of maternity leave on wages exclusively.

The Quandt (1958, 1960) switching regression estimation strategy has been applied more recently by Moschos (1989), Hotchkiss (1992), Hotchkiss et al. (1994) and Averett and Hotchkiss (1996).

Also see Lahiri and Song (2000).

Values for

larger than 2.72 resulted in numerical errors and were clearly on the declining slope of the likelihood function.

The results in Table A.1 correspond to the optimal index value which determines intermittency. This value will be discussed below.

See footnote 10.

The Chi-square test statistic comparing all parameter coefficients except the selection terms was 52.45. This value indicates a difference in parameter estimates that is significant at a 99% confidence level.

This result is reminiscent of the dual labor market theory (Dickens and Lang 1985) which predicts lower returns to education in the secondary sector.

Notably, Reimers (1983), Cotton (1988), Idson and Feaster (1990), and Oaxaca and Ransom (1994) all address this issue.

Recall that neither selectivity term was significantly different from zero. However, the fact that educational attainment across sectors works to reduce the wage differential across sectors lends credence to this supposition. This is consistent with Neal (2002) who finds the selectivity-corrected wage differential between black and white women is larger than the observed wage differential because of positive selection by white women out of the labour market.

The results reported below are essentially unchanged when average experience and tenure values from either the intermittent or continuous sectors are used for both sectors in calculating the wage differential (i.e., the intermittent wage penalty).

A theoretical life-cycle model of labour supply (available from authors upon request) shows that this coefficient being greater than or equal to zero implies that labour supply across multiple periods of a woman's life is substitutable rather than complementary.