Applied Economics Letters, 1997, 4, 173–176

Duration and mobility of young male workers in a segmented labour market JEFFREY WADDOUPS and DJETO ASSANE Department of Economics, University of Nevada, Las Vegas, Las Vegas, NV 89014, USA Received 22 March 1996

This study analyses the notion that the length of spells in secondary segment employment inhibits mobility to primary segment employment. Findings for young male workers aged 20–25 detect the existence of such an effect with secondary segment hazard rates exhibiting negative duration dependence after three years of duration.

I.

INTRODUCTION

According to labour market segmentation (LMS) theory, the labour market is composed of ranked sets of qualitatively distinct submarkets identified by both occupational and industrial characteristics (Gordon, Edwards and Reich, 1982). Dual labour market analysis, a simplified two segment version of the LMS approach, suggests the existence of a primary labour segment, consisting of stable occupations offering relatively good pay, benefits, and working conditions. This is in contrast to the secondary segment characterized by a relative lack of these attributes. Because of extremely low levels of compensation and relatively unstable employment, location in secondary segment occupations is connected with a high incidence of poverty and unemployment. Previous research in the LMS tradition has focused attention on the incidence and determinants of mobility through a segmented labour market structure (e.g. Waddoups, Daneshvary and Assane, 1995). Among important characteristics found to restrict upward mobility through the segment structure are race, low educational attainment, and slack labour markets. An important determinant of mobility that is consistent with the literature in LMS, but has not been empirically tested, is the effect of time spent in the secondary segment on the probability of mobility to higher level segments (duration dependence). Vietorsisz and Harrison (1973), for example, articulate the idea that secondary segment employment may be characterized by a selfreinforcing cycle of stagnation in wages and low levels of human capital investment by firms and workers, which may progressively leave secondary workers incompatible with primary segment work. According to this logic, the longer a worker remains in the secondary segment, the less likely he or 1350–5851 © 1997 Routledge

she is to exit into the more stable and preferred employment of the primary segment. The purpose of this study is to fill a gap in the present literature by testing for duration dependence in secondary to primary segment transitions. In other words, we are testing the hypothesis that duration in the secondary segment is an important determinant of subsequent mobility to primary segment employment. We will proceed with the analysis as follows: first, empirical methodology and data are discussed; second, results are presented and analysed; and third, conclusions are drawn.

II.

EMPIRICAL IMPLEMENTATION

The analysis of secondary segment durations begins with modelling of the conditional probability of exit from that sector. The probability that a secondary sector spell will end at time t, given that it has persisted until t is described by the hazard rate, …t† ˆ lim P…t dt!0

T

ˆ

t ‡ dtjT

t†=dt

…1†

f …t† S…t†

where T is the random variable describing the duration of the current secondary segment spell, f …t† is the density of completed spells at time t, S…t† ˆ 1 ¡ F…t† is the probability of surviving until time t with F…t† being the cumulative distribution function. The survival and density functions are derived in terms of the hazard function as: 173

174

J. Waddoups and D. Assane S…t† ˆ 1 ¡ F…t† ˆ exp ¡

Z

…u†du

…2a†

and f …t† ˆ …t†S…t† ˆ …t† exp‰¡

Z

…u†duŠ

…2b†

The corresponding likelihood function utilizes data on both completed and uncompleted secondary segment spells. It takes the form Lˆ

N1 Y iˆ1

…ti † exp…¡

Z

…u†dui †

N2 Y iˆ1

exp ¡

Z

…ui †dui

…3†

where N1 and N2 represent the number of completed and uncompleted spells in the secondary sector, respectively. The shape of the hazard function reveals the nature of the impact of duration on the probability of exit from secondary to primary segment employment. Among the more widely used parametric hazard functions are the exponential, Weibull, lognormal and log-logistic. These distributions assume various restrictions concerning the shape of the hazard function.1 The exponential assumes a constant hazard rate, suggesting that the probability of exit from the secondary sector may be independent of duration in that sector. If a constant hazard rate best defines the data, the hypothesis of duration dependence will be rejected. The Weibull, on the other hand, allows for either a constant, monotonically increasing or monotonically decreasing hazard. An increasing hazard rate indicates a positive correlation between the conditional probability of secondary segment exit probabilities and duration (implying positive duration dependence). A decreasing hazard, on the other hand, indicates a negative relationship between the conditional exit probabilities and duration in the secondary segment (implying negative duration dependence). Though the Weibull parameterization is commonly used in labour market applications of duration models, it is somewhat restrictive in the sense that it allows only for monotonically increasing or decreasing hazards. The log-logistic and the lognormal parametric forms are not subject to these restrictions. These distributions allow for nonmonotonic hazards, such as Ushaped or inverted U-shaped functions. Data used to fit the empirical models originate from the Panel Study of Income Dynamics (PSID). Durations in the secondary sector are observed for a group of young male PSID household heads, who report some experience in secondary employment during the period 1981–87. 2 Young workers are defined as respondents aged 20–25 at the start of their secondary segment

1

spell. During any 12 month period, it is possible for a worker to experience one or more spells of low status employment without it being observed by the researcher. Accordingly, if a respondent is located in the secondary sector in the year 1981, he is assumed to have been located there for 12 months. Similarly, if secondary location is reported all seven periods, 84 months in that sector is assumed. In addition, if any spell of secondary employment is interrupted by nonemployment, it is counted as a continuation in the secondary segment. The only event that closes a secondary spell, therefore, is a transition to a primary job. Multiple spells in the secondary segment are allowed and are treated as independent. 3 Table 1 contains proportions of observed durations in the secondary segment disaggregated by race. Black respondents are less likely to have spells lasting 1 to 12 months than their white counterparts, and are substantially more likely to experience a spell lasting the entire 7 year period. In general, the results in Table 2 suggest that for young white secondary segment workers, employment in this segment is a transitory part of their work history. This pattern does not hold for black workers, with a substantially greater proportion of this group remaining in the secondary segment throughout the 7 year period and fewer remaining in the segment for short periods of time. Table 2 contains sample characteristics of secondary segment workers observed at the beginning of their spell of secondary segment employment.4

III.

ESTIMATION RESULTS

Table 3 reports maximum likelihood estimates of secondary segment durations allowing for the distributional assumptions presented above. Magnitudes and signs of estimates are largely similar across equations. Consistent with expectations and

Table 1.

Duration of secondary segment employment spells

Duration

1 to 12 months 13 to 24 months 25 to 36 months 37 to 48 months 49 to 60 months 61 to 72 months 73 to 84 months Number of observations

See Kiefer (1988) for a detailed treatment of hazard models. See Waddoups, Daneshvary and Assane (1995) for a description of how occupations are allocated into segments. 3 Forty six per cent of workers in the sample experienced multiple spells of secondary segment employment. 4 The unemployment variable is calculated as an average of the yearly county unemployment rates over the duration of the spell. 2

Percentage of secondary segment durations White 34.8 18.3 15.9 7.9 6.1 3.7 13.4 164

Black 16.5 13.9 20.1 6.1 9.6 7.0 26.1 115

175

Mobility of young male workers Table 2.

Sample characteristics of secondary segment workers

Years of education Age Marital status (1 = single) Union Residence outside SMSA County unemployment Race (1 = black) Disability Number in sample

11.85 23.24 0.129 0.280 0.10 8.30 0.414 0.065 279.0

Table 3. Maximum likelihood estimates of secondary segment durations Variable

Exponential

Weibull

Normal

Logistic

Intercept

5.036 (3.861) ¡0:089 (1.478) ¡0:0001 (0.001) ¡0:433 (1.777) 0.474 (2.670) 4.394 (1.277) ¡0:026 (0.830) 0.475 (2.929) ¡0:141 (0.416) 0.019 1.000 0.000 37.405

5.024 (6.599) ¡0:082 (2.265) 0.003 (0.103) ¡0:403 (2.792) 0.394 (3.503) ¡0:438 (2.346) ¡0:023 (1.254) 0.402 (3.992) ¡0:146 (0.726) 0.019 1.408 0.115 40.538

5.107 (5.462) ¡0:075 (1.960) ¡0:026 (0.792) ¡0:410 (2.388) 0.378 (3.253) ¡0:236 (1.061) ¡0:023 (1.029) 0.398 (3.617) ¡0:075 (0.332) 0.027 1.211 0.081 37.088

5.350 (5.765) ¡0:082 (2.185) ¡0:033 (0.991) ¡0:438 (2.629) 0.385 (3.342) ¡0:227 (1.114) ¡0:025 (1.272) 0.403 (3.703) ¡0:070 (0.317) 0.027 2.009 0.145 36.423

¡336:484 279

¡318:249 279

¡323:223 279

Education Age Marital status Union SMSA

previous research, formal education is negatively correlated to secondary segment duration across specifications.5 The parameter estimate on the age variable, which acts as a proxy for labour market experience, is not statistically significant. This result is not surprising given the low variation in the age across respondents. The marital status of respondents is a uniformly negative predictor of duration, suggesting that single household heads experience shorter secondary segment durations than their married counterparts. The race, union coverage and county unemployment estimates are found to be positively correlated with durations in the secondary sector, with the former two being statistically significant at conventional levels. These results are also consistent with previous research in this area (e.g. Waddoups, Daneshvary and Assane, 1995). The estimated hazard functions are presented graphically in Fig. 1. Results from the Weibull model suggest that the data exhibit a monotonically increasing hazard of secondary segment exit, which indicates that secondary segment duration is positively correlated with exit probabilities. This result is inconsistent with the hypothesis that duration in the secondary segment reduces the probability of exit. Hazard functions estimated according to the log-normal and log-logistic distribution allow for an inverted U shaped hazard, indicating that the data may still be consistent with the duration dependence hypothesis. The negative duration dependence, however, does not emerge until after approximately the third year of secondary sector employment. Since the exponential, Weibull and log-normal are members of the same distributional family, classical log-likelihood tests may be used to compare the relative fits of the various models (Heckman and Walker, 1987). 6 Results of log-likelihood tests demonstrate that the normal provides the best fit to the data, further supporting the assertion that secondary segment hazards take on an inverted U shape.

5

County unemployment Race Disability

Standard error Median (months) Log-likelihood Number

¡351:742 279

Source: Estimates generated from PSID data. Note: Absolute t statistics are in parentheses.

IV.

CONCLUSION

Our findings add to the existing labour market segmentation literature by offering empirical content to the notion that duration in secondary employment affects the probability of upward mobility to the primary segment. In particular, for young male workers, we find that the hazard function most likely takes on an inverted U shape, which suggests positive duration dependence in the short run (about three years) followed by negative duration dependence. This research also confirms previous findings that factors treated as exogenous such as low levels of educational attainment and race remain important barriers inhibiting secondary-to-primary segment mobility, even after controlling for duration in the secondary segment. Our results are also consistent with earlier studies outside the LMS literature, which have found that previous labour market experiences do matter in the determination of present outcomes (i.e. Lynch, 1985).

A positive (negative) correlation between the covariate and duration implies a negative (positive) relationship with the variable and the probability of exit. The log-logistic distribution does not originate from the same distributional family, making its log-likelihood figure not directly comparable to the others. As Fig. 1 indicates the log-logistic and normal parameterizations take on strikingly similar shapes. 6

176

J. Waddoups and D. Assane 0.030

Logistic

Hazard (Probability of Exit)

0.025

Log-normal

Exponential

0.020

0.015 Weibull 0.010

0.005

0.000 0

12

24

36

48

60

72

84

Duration (Months) Fig.1. Hazard functions

ACKNOWLEDGEMENTS The authors acknowledge financial support from the First Interstate Bank Institute for Business Leadership

REFERENCES Gordon, D.M., Edwards, R. and Reich, M. (1982) Segmented Work, Divided Workers: The Historical Transformation of Work in the United States, Cambridge University Press, Cambridge Heckman, J.J. and Walker J.R. (1987) Using goodness of fit and other criteria to choose among competing duration models: a case study

of Hutterite data, in C. Clogg (ed.), Sociological Methodology, American Sociological Association, Washington DC Kiefer, N.M. (1988) Economic duration data and hazard functions, Journal of Economic Literature, 24, 646–79. Lynch, L.M. (1985) State dependency in youth unemployment: a lost generation?, Journal of Econometrics, 28, 71–84. Vietorisz, T. and Harrison, B. (1973) Labor market segmentation: positive feedback and divergent development, American Economic Review, 43, 366–76 Waddoups, J., Daneshvary, N. and Assane, D. (1995) An analysis of occupational upgrading differentials between black and white males, Applied Economics, 27, 841–47.