The Effect of Dismissals Protection on Employment: More on a Vexed Theme.

• Author: Addison, John T.

John T. Addison [*]

Paulino Teixeira, [+]

Jean-Luc Grossot [++]

This paper presents new results on the relationship between severance pay and labor market performance for a sample of 21 OECD countries, 1956-1984. Specifically, it evaluates Lazear's empirical argument that severance pay reduces employment and elevates joblessness. His findings are shown not to survive correction for errors in the data and the application of correct estimation procedures. Furthermore, adverse labor market consequences of severance pay are not detected in a dynamic characterization of the Lazear model. Limitations of the approach followed here are also addressed and contextualized.

1. Introduction

   Concern over the adverse employment consequences of employment protection legislation is a recurring theme in labor market analysis. Recent, and conflicting, applied treatments include Scarpetta (1996) and Nickell (1997). The unfolding empirical analysis of the effects of employment protection shares certain similarities with investigation of the covariation of collective bargaining structures and macroeconomic outcomes (see, e.g., Calmfors and Driffill 1988; OECD 1997). In both cases, the models are typically reduced form and the empirical evidence rather mixed.

   The present paper offers a replication and critique of Lazear's (1990) famous empirical model of employment protection, which is the sole extant empirical treatment to allow for changes in a measure of employment protection over an extended time interval. Our concern is not so much with theoretical issues as with the sensitivity of Lazear's results to data problems. Suffice it to say here that the theoretical analysis of the long-run effects of employment protection produces ambiguous results (e.g., Bentolila and Bertola 1990; Bertola 1991; Hopenhayn and Rogerson 1993; Bentolila and Saint-Paul 1994; Saint-Paul 1995). Accordingly, there is a premium on empirical analysis. [1]

   It should perhaps come as no surprise to learn that Lazear (1990) did not deny that employment protection could be benign--in perfectly functioning markets, the parties would efficiently negotiate around a severance pay mandate via appropriate side payments from worker to firm--or argue that outcomes were independent of the stage of the cycle (see Hamermesh 1993) or, for that matter, assert that negative effects would be observed across all outcome indicators (adverse effects being more clear-cut for employment than for unemployment, where discouragement could even generate a reduction in unemployment). Rather, his position was that, in the presence of constraints on efficient contracts, employment protection rules might be expected to bind in regular (and especially European) markets. His tests were designed accordingly.

   Our criticism of Lazear is that, quite apart from errors in his data, he did not adequately investigate the statistical problems stemming from his use of pooled cross-sectional and timeseries data, even if recognition of the problems emphasized in the present treatment is apparent in his narrative. We refer in particular to the problems of country heterogeneity and autocorrelation. We shall report that almost all of the statistical significance attaching to Lazear's key employment protection measure evaporates once the appropriate econometric procedures are employed. This is not the end of the story, however, because of the limitations of the employment protection variable and the parsimonious nature of the estimating equations used here. Both issues will be addressed in the context of the developing employment protection research literature.

2. Data and Methodology

   The data and variables used in this inquiry in principle follow those of Lazear (1990), to whom we are indebted for supplying us with a diskette containing the raw data (and programs) used in his study. The data have been corrected for the errors of omission and commission identified by Addison and Grosso (1996). Appendix A illustrates the main issues. We note that the data errors do not overturn Lazear's principal findings, taken at face value, even if they do serve substantially to alter the point estimates.

   In estimating the employment effects of statutory job protection, Lazear initially uses data on 20 OECD countries for the sample period 1956-1984. His data are not complete for all variables and years. Lazear actually collected data on 22 countries but subsequently dropped two nations (Canada and Hong Kong). Our sample of countries and time frame are the same as Lazear's, with the exceptions noted in Appendix A. Suffice it to say that sample differences were not material to any of the results reported below.

   Lazear examines the determinants of four outcome indicators: the employment-population ratio (EMPPOP), the unemployment rate (UNRATE), the labor force particiaption rate (LFPR), and the average hours worked by production workers (HOURS). Values of each of the first three dependent variables differ slightly from those used by Lazear because we were able to obtain updated estimates of the size of the population, civilian labor force, and employment used in their construction. Rather more important changes were introduced into the HOURS variable, chiefly because of the need to provide a consistent time series and also to substitute a correct measure of weekly hours for Italy to replace the erroneous daily hours measure inadvertently used by Lazear (1990, table I, column 3).

   Turning to the independent variables, the crucial employment protection measure is severance pay (SEV). This is defined as the statutory entitlement in months of pay due to a blue-collar worker with 10 years of service on termination for reasons unconnected with his/her behavior. The measure thus pertains to no-fault individual dismissals for economic reasons. Comment on the efficacy of this measure is provided below.

   The remaining covariates are a quadratic time trend, represented by YEAR and [YEAR.sup.2]; a demographic control for the population of working age (WRKAGE); [2] and the growth in per capita gross domestic product (GROWTH), which accommodates the notion that a growing economy vitiates at least in part the probabilistic costs of a severance pay mandate (see Gavin 1986). Unusually, in his fitted regressions, Lazear only enters the latter variable in interaction with severance pay. As is more conventional, we include both GROWTH and its interaction with severance pay (GROWTH.SEV).

   The majority of Lazear's estimates are from outcome equations that include just the time trend variables YEAR and [YEAR.sup.2] and the key dismissals protection indicator. This has perhaps served to amplify the principal criticisms of the model, namely, that it abstracts from many variables that may be expected to affect structural unemployment and employment rates, while its key independent variable, severance pay, is at best a partial indicator of the legal regulations applying in a particular country. (There is also the neglected issue of other constraints operating through the collective bargaining system.) But we note parenthetically that Lazear (1990) does at least use his expanded set of variables to explain changes in unemployment rates in a specification that combines cross-section and time-series variation. He concludes that, in some countries, more generous severance pay "can go a long way in explaining the changes in employment over time" (Lazear 1990, p. 720). Singled out as cases in point are France, Portugal, Italy, and Israel.

   Although our interest is primarily in reassessing Lazear's model per se, both criticisms have to be addressed at this point. Consider first the dismissals protection variable. Recent work in the employment protection area has sought to widen the definition of dismissals protection. Perhaps the best known measure has been constructed by Grubb and Wells (1993), who identify three elements of a system of employment protection: restrictions on dismissals, restrictions on temporary forms of employment contract (so-called atypical work), and restrictions on working hours.

   The first element covers not only severance pay, as in Lazear, but also procedural delays and unfair dismissal provisions. (Note, however, that there is no recognition of regulations concerning collective dismissals.) [3] The second element encompasses restrictions on the use of fixed-term contracts and temporary agency work, such as the permissible grounds for their use, the maximum number of successive contracts, and their maximum cumulated duration. Finally, hours restrictions cover such things as the length of the normal working week, annual overtime limits, minimum rest periods, and restrictions on night work.

   Using simple unweighted averages of rankings, Grubb and Wells (1993) provide summary indexes for each component together with a grand ranking for the overall strictness of the regulatory climate that is reproduced in column 3 of Table 1. An analogous procedure is employed by the OECD (1994) to (average) rank countries by the severity of their legal restrictions on regular and atypical work, the results of which are reported in column 4 of Table 1. In each case, the ranking is from least to most regulated, and the data describe the situation in the late 1980s. We note that the OECD ranking in column 4 has commonly been used in most of the recent employment protection studies (see below). [4]

   Given that legislative rules may be only part of the story, other researchers have exploited less ambitious but potentially more encompassing reputation indexes based on surveys of employers. One such index is provided in column 5 of Table 1. It uses data from a survey conducted by the International Organization of Employers (IOE) (1985) as distributed to European and non-European employer federations. The survey seeks to identify the importance of obstacles to the termination of regular employment and the deployment of atypical workers. The entries in...