Applied Economics, 2000, 32, 583 ± 591
T emporary workers and productivity: the case of Spain  N C H E Z * and L U I S T OH A R I A { R OS A R I O S A Universidad de V alencia, Edi® cio Departamental Oriental, Campus de los Naranjos, Avda. de los Naranjos s/n. V alencia 46022, Spain and { Universidad de Alcala de Henares, Madrid
The primary focus of this paper is on e ciency wages and their testable implications. In particular the nature of the relationship between e ciency wages, productivity and the make up of the labour force is analysed, modelled and subjected to an empirical test. This theory is consistent with the views of many managers and personal administrators , who tend to ascribe primary importance in wage setting to indirect control of the ® rm’s workforce. Here we test a labour augmenting production function where e ort depends not only on wages but also on the proportion of temporary workers. I . I N T R OD U C T I ON The high rate of unemployment (around 23% in 1994)1 is one of , if not the biggest, weaknesses of the Spanish economy. This has overwhelming consequences both in terms of misallocation of resources and in terms of equity and social integration of the unemployed. Involuntary unemployment appears to be a persistent feature of the Spanish labour market. The presence of such unemployment raises the question of why wages do not fall to clear the labour market. In this paper we show how e ciency wage theories could contribute to explaining the unemployment rate as an equilibrium phenomenon. Traditional microeconomic theory assigns to wages the single role of attracting workers to ® rms and thereby allocating labour. Managers and personal administrators , however, often see this role of wages as secondary. Over the last 15 years many economists have helped to develop a new branch of labour economics and microeconomic theory which bridges the gap between traditional theory and the views of many wage administrators. This is now widely known as the e ciency wage theory. Of course, standard economic theory does describe the nature of contracts when there are many possible standards
of performance. According to the theory when a ® rm hires a worker, there is an understanding by both parties that certain minimum standards of performance must be met. The neoclassical ® rm can purchase all the labour services it wishes if it pays a wage at least as high as the market wage. If the ® rm chooses a wage below the market-clearing level, it receives no labour. Unlike the traditional microeconomic theory (which assigns to wages only an allocate role) , e ciency wage endogenizes the real or relative wage as a pro® t maximizing or cost minimizing choice of the ® rm. Many ® rms may ® nd it optimal to pay a wage above the opportunity wage of their worker. If a wage premium induces a lower turnover , less shirking, selective recruitment of better workers and improved morale,2 then the resulting savings in costs, or gains in productivity , may outweigh the costs of the higher wage up to a point. In this paper we focus on the predictions of the shirking model. The novelty that our studies incorporates is to estimate the e ect of the proportion of temporary workers over e ort and its implications on the level of employment inside an e ciency wage framework. First we analyse the impact of a number of variables on the likelihood of a ® rm paying a wage above the average of their sector. We have found that the higher the proportion of temporary workers a ® rm has, the less likelihood
* To whom correspondence should be addressed. E-mail:
[email protected]. 1 Eurostat de® nition. 2 See e.g. Pencavel (1972) , Stiglitz (1974 ), Yellen (1984 ), and for many further references, Stiglitz (1987 ). Applied Economics ISSN 0003± 6846 print/ISSN 1466± 4283 online # 2000 Taylor & Francis Ltd http://www.tandf.co.uk/journals/tf/00036846.htm l
583
584 there is of a ® rm paying a wage above the average of its group. The traditional shirking model is based on the imperfect information that ® rms have concerning the e ort of their employees. Monitoring individual performance is assumed to be impossible, or very costly, and the punishment for low performance is limited by legal constraints. As an incentive for workers to work instead of shirking, ® rms may ® nd it pro® table to raise wages. This attitude increases the cost of job loss for the worker in two ways. First, because this wage is higher than the wage paid by other ® rms. Secondly, it is above the market clearing level and will thus generate involuntary unemployment and also diminish his expected outside earnings. In the simplest version, according to Shapiro and Stiglitz (1984 ) workers and ® rms are assumed to be homogeneous and, as ® rms are identical, they would all ® nd it pro® table to raise wages. Relative wages will stay constant, and only unemployment will act as a discipline device, increasing the cost of job loss. A whole family of models has been built around this basic framework by varying what is assumed measurable, at what cost, and the payment schedules that are feasible. Krueger and Summers (1986 ) estimate standard wage equations using cross-section data on individuals. They use data from the US Current Population Surveys for 1974, 1979 and 1984. The industry and occupation variables were found to be relatively important explanatory variables for variation in earnings. Murphy and Topel (1987 ) also use longitudinal data, and produce di erent results from those obtained using cross-section data. Gibbons and Katz (1989 ) point out that the previous evidence based on longitudinal data only deals with the objection that interindustry wage di erential proxy for unobserved ability di erences if each worker’s productive ability is valued equally in di erent industries. In the Gibbons and Katz model, information about ability is imperfect ex ante, but improves ex post, i.e. the market observes signals about each worker’s ability at the time of hiring, but more is learnt through actual subsequent productivity. The workers who are subsequently discovered to belong to the wrong industry then move on. Wadhwani and Wall (1991 ) , using UK ® nancial data show that ® rms that increase their real wages improve their productivity. Levine (1992) ® nds that business units that increase the relative wages for their workers, have productivity gains approximately large enough to pay for the wage increases. In Sa nchez (1994 ) it is tested that sector productivity level increases when either relative wage or the level of unemployment rises. In Sa nchez et al. (1995) , it is shown that those 3 4
R. Sa nchez and L . T oharia Spanish ® rms paying above the average also obtain a level of productivity above the average. In Spain, after the labour market reform of 1984, a very extended use of temporary contracts was observed in almost all sectors of the economy and types of jobs during the expansionary period of the second half of the 1980s (98% of all new contracts were temporary contracts). After the reform, any unemployed person could be hired on a temporary contract without the requirement of a speci® c cause. This implies that for any job related to the permanent activity of the ® rm, the ® rm could either choose a permanent or a temporary contract. However, even if job creation increased temporary contracts, no permanent employment was created because renewal rates within permanent contracts were very low. Indeed, the share of workers with a temporary contract increased to approximately one-third by the early 1990s, more than three times the European average of just fewer than 10%. For given wages, temporary contracts are cheaper and so there is a rationale for this extended use. However, this does not satisfactorily explain why they are even used in jobs where long labour relations are an important source of e ciency. For instance, this is the case for jobs requiring speci® c human capital investment for which the investment decision depends crucially on the expected duration of the job. Another quite extended example is jobs where the provision of incentives is important. E ciency wage models can help to explain the consequences of this kind of hiring policy used by these ® rms. If a ® rm chooses this contract policy it would a ect its labour costs. Thus, our ® rst approximation inside the e ciency wage framework is checking the e ect that having temporary workers has on wages. The high share of temporary workers may have reduced the incentive to job search, because of the less attractive features of temporary work, which has also contributed to reduce the replacement ratio for certain job categories (overall these categories where the proportion of temporary workers have increased ). In medium-term perspectives, and to a lesser extent even in the short term, the lower incentive, both for companies and for their temporary workers to invest in their training, is expected to bring about serious negative e ects for productivity. Apart from incentive e ects, temporary workers may be intrinsically less e ective also because they are trained less and have less in¯ uence. 3 Also, these kinds of policies a ect the average level of e ort of the labour force. Even if the temporary workers are more productive in order to obtain a permanent contract, the probability of the collaboration of the permanent workers is low because they know that the likelihood of renewal of temporary contract into permanent contract is small. 4 Also,
For more complete information about the quali® cation of temporary workers in Spain, see Segura et al. (1991). GuÈell-Rotllan (1997 ) shows how in Spain the optimal renewal rate of temporary contracts into permanent is less than one.
585
T emporary workers and productivity the higher the proportion of temporary workers, the worse the signal is the permanent workers perceived about the type of policy used by the ® rm. We assume that if the industrial sectors of the Spanish economy pay e ciency wages, they will have a positive labour augmenting production function. The factor of augmentation in this context, is the e ort. And one of the determinants of this e ort is the proportion of temporary workers hired by the ® rm during the period analysed. The paper is organized as follows: in Section II a more detailed description of temporary contracts and labour costs is provided. Then in Section III we test the e ciency wage hypothesis through a panel data analysis. Finally in Section IV we introduce the concluding remarks.
I I . L A B OU R C OS T A N D T E MPOR A R Y WOR K E R S : A C R OS S - S E C T I ON A NA LY SIS The 1984 labour market reform introduced a compromise between the previous rigid labour market legislation and a more ¯ exible labour market of hiring and laying-o practices. As a consequence an emergent dualism appeared in the Spanish labour market, with one segment of the workforce ¯ exible and the other, formed for permanent workers, rigid, in view of the high severance pay and administrative procedures involved to dismiss those workers. An interesting issue is whether the introduction of temporary contracts has contributed to an expansion of employment or whether only a substitution e ect between temporary and permanent workers has taken place. 5 A ® rst possibility to assess the impact of temporary contracts is to see if the labour demand function has changed since 1984. The results showed in the Reports and Studies of the Economics and Financial Situation in Spain (1994 ) , suggest that the introduction of temporary contracts neither changed the level or the sensitivity of labour demand to the real wage. Based on this report, there is, however, another possible channel for temporary contracts to make an impact on employment; for example, the process of wage formation. Empirical analysis suggests that the introduction of temporary contracts has had an impact on wage formation, reducing both nominal and real wage costs. This is in line with what has been frequently reported that temporary workers are paid less than permanent workers with identical quali® cations. In addition to the impact in terms of wage formation, the increment on temporary contracts has another e ect in terms of short and long-term productivity. In the short 5
term there are two factors that could increase productivity. On the one hand, the higher risk for a temporary worker to be dismissed if he or she is found shirking should increase his or her e ort. On the other hand, most workers do not expect to be rehired once the temporary contract expires, when they would have to become permanent workers. This acts as a strong disincentive for e ort and productivity. Overall, it is more likely that a higher share of temporary workers reduces labour e ort and productivity. One of the key features of the Spanish labour market is the behaviour and the process of wage formation. After peaking in 1977, nominal wage growth decelerated markedly until 1987, thus contributing to a real wage moderation process that lasted for most of the 1980s. However, in the early 1990s wage growth rose again both in nominal and real terms, having slowed down only at moderate pace in 1993 and 1994. This is particularly striking given the high level of unemployment in Spain which rose, moreover, uninterruptedly between the third quarter of 1991 and the ® rst quarter of 1994. In Spain the minimum wage has continuously risen by less than the average wage. The number of workers paid the minimum wage is surprisingly low (0.9% of wage earners in the second quarter of 1992 according to the `Encuesta de Coyuntura Laboral’ of the Ministry of Labour ). This suggests that the statutory minimum wage is probably more important indirectly on employment promotion, the statutory minimum wage stands as a default reference for apprentice pay. Hence, relatively low minimum wages can help to improve employment prospects of low quali® ed workers. With respect to the regulations and employment policies of unemployment bene® ts, a package of measures adopted in March 1992 tightened the access duration and amount of unemployment bene® ts. T he Data The data source is published in the Spanish Industrial Inquiry on management’s strategies ESEE (Encuesta sobre Estrategias Empresariales). The Ministry of Industry and Fundacion Empresa Pu blica carries out this Inquiry. The precise data de® nitions are discussed in the Data Appendix. The ESEE is a new Spanish Inquiry that started in 1990. This inquiry has three di erent characteristics that make it quite interesting in the study of the behaviour of the Spanish Industrial ® rms. First, the ESEE is sensibly representative of the distribution of ® rms by employment size. Secondly, this Inquiry is specially directed to that research based on a strategic type. Thirdly, it is supplied through a panel of ® rms. Each year
See GuÈell-Rotllan (1997 ) for a good theoretical explanation about how the introduction of temporary contracts can lead to a few permanent contracts in Spain and that could reinforce a segmentation in the labour market and not necessarily higher employment levels.
586 the inquiry is repeated for the same sample of ® rms. The observation over time of the behaviour of the same number of ® rms is specially important for the analysis of some kinds of economic facts, for example, the control of the cyclical e ects as well as the unobservable heterogeneity among ® rms. T he cross-section results Our basic results are reported in Table 1. We have estimated a Probit model for every year between 1990± 94. We show how the likelihood of having a ® rm with a wage above the average wage of its industry,6 depends on a set of regressors that represent characteristics of the make up of the labour force, the industrial activity and the size of the ® rm. The probability of having a ® rm with a wage above the average of its industry decreases as the percentage of temporary workers (PEV EN) increases. The results are reported in Table 1. The value of the coe cient of PEV EN is negative as we expect and signi® cantly di erent from zero for each one of the ® ve years that we have analysed. These coe cients are signi® cantly di erent from zero for a degree of signi® cance of a 1% level. The coe cient of POBR (percentage of blue-collar workers) picks up the negative e ect of having a high number of blue-collar workers on the probability of a ® rm of having wages above the average of its group. This coe cient appears as negative and signi® cantly di erent from zero at a 1% level of signi® cance. Another important result is the in¯ uence of the quality of the labour force on the e ect on wages. As we can see in Table 1, the coe cient of the variable (PIL ) , which is the proportion of engineers and graduates, is positive and signi® cantly di erent from zero at a 1% level of signi® cance for 1990, 1991 and 1994. PL CT is the share of labour in the total cost and obviously it increases as the wage bill does. The coe cient of this variable is positive and signi® cantly di erent from zero at a 1% level of signi® cance. The variable PBSL (productivity for worker) is picking up the relationship between wages and productivity. Here we have to comment that there exists a simultaneity bias between these two variables and for this reason, it is important to check on which direction the causality works. As was expected, this coe cient appears as positive and signi® cantly di erent from zero for each one of the ® ve years. The variable CAPL (stock of capital by worker) appears positive and signi® cantly di erent from zero at a 1% level of signi® cance for 1992, 1993, 1994 and it is signi® cantly di erent from zero at a 10% level of signi® cance for 1990. 6
R. Sa nchez and L . T oharia Also it was expected that the higher the level of capital per worker the higher the level of wages that the ® rm is willing to pay. Next, we have in Table 1 a set of industrial dummy variables. The ® rms in the sample are classi® ed on their industrial activity following the `nace-clio’ classi® cation system. As we can see, the coe cients are negatives and signi® cantly di erent from zero at a 1% level of signi® cance. What this means is that if the ® rm belongs to any other industrial sector di erent from `Wood and derived’ then the probability of that ® rm paying a wage above the average decreases. Finally, what we have are ® ve dummy variables that take into account the in¯ uence of the ® rm’ s size on the probability of paying a wage above the average. Here the category of reference is a ® rm with a size of between (101 ± 201 ) workers. As we can see, a size smaller than (101 ± 201 ) lowers the likelihood of having higher wages, while a bigger size increases this probability. In general we can say that the overall estimation is signi® cantly di erent from zero as we can check through the À2 that is reported at the end of Table 1. What we have obtained here is that the likelihood of a ® rm paying a wage above the average wage paid by the group of ® rms, that belongs to the same industrial sector of activity, decreases as the percentage of temporary workers or blue collar workers gets higher. Belonging to an industrial sector of activity di erent from `wood and derived’ , which is the reference’ s category, diminish also this probability, given everything equal. Instead of this, the ® rms that have a higher proportion of workers with a high degree (engineers and graduates) , or higher average product of labour or a higher ratio of capital by worker increases the probability of the ® rm paying more than other ® rms of the same group of activity. The correlation obtained here could be explained by a high number of theories like for example the insider± outsider theory. Now to be sure that these results could be explained by e ciency wage theories, we have to check in which direction the causality works. In this case higher wages are the explanation of higher e ort and not the other way round. The fact that high relative wages are associated with higher levels of productivity may just re¯ ect the existence of either unobserved human capital or rent sharing. If workers share rents, as, for example in insider± outsider models of wage determination, then high productivity in this ® rm will cause high relative wages. In the next section, we attempt to control for the existence of unobserved human capital by allowing for a ® rm-speci® c ® xed e ect. Also, we attempt to control the simultaneity bias, between productivity and wage, by using a standard instrumental variable estimator.
Here we use the industrial group NACE-CLIO that appear in Table A1 to obtain the average of the industry in which the ® rms we are studying belong.
587
T emporary workers and productivity Table 1. W age di erential: a probit model estimation Variable CONST . PEV EN POBR PIL PL CT PBSL CAPL
1990 0.23071
(0.538 )
70.01736
1991 0.98608
(2.259 )*
70.01733
0.02221
(0.059 )
70.02291
1993 1.1577
(2.847 )*
70.02660
1994 2.2390
(5.169 )*
70.03351
(76.738)*
(76.482 )*
(77.707 )*
(78.471 )*
(79.754 )*
(73.830 )*
(74.967 )*
(74.906 )*
(75.099 )*
(75.367 )*
(2.492 )*
(2.501 )*
(0.754 )
(0.997 )
(3.757 )*
(7.900 )*
(7.375 )*
(7.159 )*
(5.271 )*
(2.409 )*
(9.569 )*
(9.196 )*
(9.833 )*
(7.960 )*
(4.617 )*
(1.769 )
(1.452 )
(2.770 )*
(2.516 )*
(4.946 )*
70.01378 0.03670 3.2515
0.00825 0.00002
70.018253 0.03846 3.0365
0.00755 0.0004
Industrial sector of activity. Reference category: Wood and derived 71.2400 71.9944 Mineral and Iron industry (74.437 )* (74.287 )* 70.80927 70.61070 Minerals and products no metallic (72.975 )* (72.240 )* 72.3705 72.2242 Chemical products (77.045 )* (76.450 )* 70.63561 70.54143 Metallic products (72.607 )* (72.179 )* 70.96114 71.0906 Machinery (73.124 )* (73.753 )* 71.3598 72.8559 O ce machinery and others (72.315 )* (73.513 )* 71.4754 71.4212 Electric materials (75.304 )* (75.047 )* 71.5382 71.4481 Cars and engine (74.833 )* (74.599 )* 71.2994 71.1926 Other material of transport (73.278 )* (73.189 )* 71.5048 71.1602 Meat and manufacturing of meat (73.858 )* (72.904 )* 71.2940 71.4454 Food industry and tobacco (74.956 )* (75.353 )* 71.9497 72.4481 Drinks (74.658 )* (75.785 )* 70.77218 70.85190 Textile and clothing (73.178 )* (73.486 )* Leather, shoes and derived 0.12571 0.34940 (0.396 ) (1.107 ) 71.4475 71.3110 Paper and derived (74.856 )* (74.567 )* 70.68346 70.84322 Couch and plastic (72.337 )* (73.024 )* 70.43714 70.39984 Other manufacturer products (71.226 ) (71.111 ) Size of the ® rm. Reference category from 101 to 200 employees 70.85275 Less than 20 workers (73.741 )* 70.59377 From 21 to 50 employees (72.671 )* 70.04530 From 51 to 100 employees (70.180 ) From 201 to 500 employees 0.59001 (2.671 )* More than 500 employees 0.98371 (3.706 )* N 962 À2 (28) Log-likelihood
1992
505.85 7412.36
71.1742
70.01726 0.01012 2.9786
0.007413 0.00003
72.0039
70.01831 0.01451 2.1686
0.00571 0.00004
71.7050
70.01895 0.0570 1.0405
0.00268 0.00006
72.1316
(74.591 )*
(73.994 )*
(74.699 )*
(71.571 )
(72.591 )*
(74.072 )*
(76.244 )*
(76.462 )*
(77.322 )*
(72.266 )*
(71.837 )**
(72.520 )*
(73.320 )*
(73.350 )*
(73.917 )*
(73.383 )*
(73.497 )*
(73.162 )*
(74.747 )*
(75.762 )*
(75.870 )*
(74.572 )*
(74.142 )*
(74.954 )*
(71.541 )
(73.262 )*
(71.821 )**
(72.778 )*
(72.407 )*
(72.628 )*
(74.052 )*
(74.004 )*
(75.391 )*
(74.933 )*
(74.844 )*
(76.541 )*
(71.753 )**
(72.023 )*
(72.256 )*
70.41492 72.0252
70.55921 70.95113 72.3366
71.3080 71.3978
70.58105 71.1507 71.0479 72.0681
70.41624 70.80098
(72.068 )* 71.0990
70.71837 72.1554
70.46122 70.98648 73.1881 71.6362 71.2990 71.2014 71.0311 71.0835 71.9988
70.49579 0.62916
(1.876 )**
71.4019
71.1825
72.6094
70.63915 71.2171
72.6875 71.7259
71.6268
70.75195 71.0553 71.5159 73.0192
70.55654 0.10513
(0.319 )
71.3667
(73.884 )*
(74.654 )*
(74.511 )*
(71.477 )
(72.029 )*
(72.491 )*
(70.319 )
(70.647 )
(72.199 )*
70.41081 70.11004
70.39505
70.57370 70.22222
70.99108
70.73181 70.79727
71.3031
(75.233 )*
(73.087 )*
(75.522 )*
(76.736 )*
(74.336 )*
(72.324 )*
(74.686 )*
(75.141 )*
(71.743 )
(70.520 )
(71.220 )
(70.646 )
70.97472 70.43934 0.06403
70.22311 70.02310 0.71930
70.88272 70.2208 0.1795
71.0039
70.15124 0.15079
(0.295 )
(5.535 )*
(0.973 )
(0.784 )
(2.575 )*
(5.928 )*
(1.845 )**
(2.231 )*
0.70603
1.1783
0.42143
0.56097
962
962
962
962
488.89 7420.29
473.75 7429.84
483.51 7415.69
554.55 7385.86
Note: t-student between brackets. (*) (**) Statistically signi® cant at a 1% and 5% level of signi® cance respectively. DW : Dependent variable. Takes value 1 if the labour cost for employee is higher than average and zero otherwise. PEV EN: Proportion of temporary workers. POBR: Proportion of blue-collar workers. PIL : Proportion of Graduates and Engineering over labour force. PL CT : Share of labour cost in total cost. CAPL : Amount of stock of capital by worker. PBSL : Level of production by worker.
R. Sa nchez and L . T oharia
588 Table 2. L abour augmenting production function (1990 ± 94) Model (1)
Variables Average wage of the ® rm (w)* Average wage of the industry (Awr) Permanent workers
0.2441
(2.003 )
70.3179
(712.03 )
All workers Proportion ( peven) of temporary workers Stock of capital (k ) Size (more than 200) Wald test Sargan test 1st order serial correlation 2nd serial correlation
0.6003
(5.360 ) ±
70.0136
(73.177)
0.2768
(11.43 )
0.1120
(7.962 ) 9464.40 (6) 0.0788 (4) 3.326 0.267
Model (2) 0.2090
(2.172)
70.2985
(711.34 ) ±
0.6120
(10.20 )
70.0019
(74.031 )
0.1639
(9.110)
0.0935
(6.876 ) 9405.90 (6 ) 0.9542 (4) 1.997 0.335
Note: t-student between brackets. The dependent variable is the level of production. (*) This variable has been taken as endogenous. The instrument is the wage lagged one period. Y : Dependent variable is the level of production of the ® rm, in logs. W : This is the log of real wage at the level of the ® rm. W ar: This is the average real wage at the level of the industry, in logs. This variable has been created using the classi® caiton NACECLIO. For each ® rm we have calculated the average wage for all the ® rms included in the same group NACE-CLIO without taking into account the ® rm of reference. Permanent workers: Number of workers that have long-term ® xed inside the ® rm in logs.
I I I . T E S T I N G T H E E F F I C I E N C Y WA G E H Y POT H E S I S : A PA N E L D A T A A NA LY SIS T he theoretical approach The basic framework of a standard e ciency wage model starts assuming wages a ect worker’s e ciency e. In the implementation of this hypothesis worker e ciency is assumed to be labour augmenting, and enters in the production function as an additional input. In order to test the e ciency wage hypothesis, it can be assumed a Cobb± Douglas production function with only two inputs: e ective labour E and capital K. 7 yit ˆ Ai Kit¬ Eit exp …uit †
… 1†
Eit ˆ eit L it
… 2†
eit ˆ W it¯1 …AW R †¯it2 exp …peven †¯it3 exp …size†¯it4
… 3†
where Y it is the level of output of a ® rm i in the year t; Ai is a sector speci® c ® xed e ect; K is the stock of capital; E is 7
the e ectiveness of the labour force; uit is an error term which is assumed to be normally distributed with zero mean and ¼2 variance ; eit is the average e ort of the labour force in ® rm i in year t, L is the level of employment of the ® rm; W is the average wage paid by the ® rm; AW R is the average wage paid by other ® rms that belong to the same sector of activity; peven is the percentage of temporary workers on the whole labour force of the ® rm; and size is a variable that picks up the e ect of having more than 200 workers. Substituting Equations 2 and 3 in Equation 1 and taking logarithms we obtain the simplest equation that we seek to estimate: yit ˆ ai ‡ ¬k ‡ l ‡ ¯1 wit ‡ ¯2 awr ‡ ¯3 peven ‡
…4 †
¯4 size ‡ uit
Here lower case letters denote logarithms. Equation 4 could be written as follows: yit ˆ ai ‡ ¬k ‡ l ‡ ¿1 wit
‡
¿2 awr ‡ ¿3 peven ‡ ¿4 size ‡ uit …5 †
Thus, Equation 5 is more general and allows us to check whether the results obtained with this equation, in the panel data regression, are in line with our previous results in the cross-section analysis. T he results The panel data estimation refers to cross-section data that have been pooled over time. Our panel is composed of 1098 ® rms of the ESEE for the period of 1990 to 1994. In our case we can follow the same ® rm through this ® ve years. When we move from the single cross-section to panel data, the information improves. In Table 2, we report the results obtained by the regression through panel data. We use the model in ® rst di erences to eliminate the ® xed e ect component. The Wald and Sargan tests also appear in this table because the Dynamic Panel Data Program (DPD) due to Arellano and Bond (1988) computes them automatically. The Wald test is of joint signi® cance for all the variables entered in X (a test of the null hypothesis that their estimated coe cients are all zero) , it is asymptoticall y distributed as a À2 with the degrees of freedom provided. DPD reports tests for the lack of ® rst order and second order serial correlation on the residuals. If the model has been transformed to ® rst di erences, ® rst order serial correlation is to be expected but not second order. These tests are based on the standardized residual autocovariances , which are asymptotically distributed as an N ¹ …0;1 † under the null hypothesis of no autocorrelation. More generally, Sargan
The manner in which the e ciency wage variable enters the production function is analysed in Wadhwani and Wall (1988 ).
589
T emporary workers and productivity tests of overidentifying restrictions are also reported. This statistic is asymptotically distributed as a À2 with as many degrees of freedom as overidentifying restrictions, under the null hypothesis of the validity of the instruments. Again, Arellano and Bond (1988 ) provide a complete discussion of these procedures. In Table 2, we report two models. The di erences between them are in the de® nition of the number of workers of the ® rm. In model (1) we have used for estimation the number of permanent workers, while in the second model it has been changed by total employment. Notice that the coe cients of the variables do not make a signi® cant change when we modi® ed the de® nition of the labour force used for estimation. The values of the coe cients of both labour and capital, are positive and signi® cantly di erent from zero at a 1% level. The sum of the coe cients on employment and capital is about 0.87, which implies mild decreasing returns to scale. The e ect of the variables, included on the e ort function, on productivity is nontrivial. Given the coe cient of 0.2441 the elasticity of e ort with respect to the ® rm’ s own wage is, actually, about 0.41 in the case of model (1) and quite similar for model (2). These values are smaller than one. However, it may be more plausible to have production function with lower e ort± wage elasticity (see, e.g., Akerlof and Yellen 1986 , pp. 14± 15). What this means is that there exist in the Spanish economy di erent types of wage formation that deviate the ® rm from the equilibrium wage, which imply that the elasticity of e ort with respect to the wage is equal to one (the Solow’s condition). The results about the elasticity of e ort with respect to each one of their components is reported on Table 3. The coe cient of the industry average wage has the correct sign and is signi® cantly di erent from zero at a 1% level of signi® cance. Dividing this coe cient by (coe cient of labour) , we obtain the value of the elasticity of the e ort with respect to the average wage of the industry and it is 70.53. Each time the average wage of the industry rises, the average level of e ort of the ® rm diminishes by a smaller proportion (0.53). Hierarchy of wages seems to be more important than the own wage. Also, the proportion of temporary workers a ects negatively the level of e ort of the ® rm. The coe cient of peven is negative and signi® cantly di erent from zero at a 1% level. The elasticity of e ort with respect to the proportion of temporary workers Table 3. T he value of the elasticity of e ort Variables W -e ort elasticity Awr-e ort elasticity Peven-e ort elasticity Size-e ort elasticity
Model (1) 0.41 70.53 70.02 0.18
Model (2) 0.34 70.49 70.003 0.15
is 70.02. An increase on the proportion of these types of workers has a negative e ect on the average level of e ort of the ® rm. As we have analysed in the previous Probit model, the higher the proportion of temporary workers of the ® rm the smaller the likelihood of that ® rm having a wage above the average of its industry. Then, what we have here is an indirect e ect on the wages of the ® rm and, in consequence , on e ort. Thus, if the proportion of temporary workers inside the ® rm is high, then the probability of having a ® rm that pays an average wage above the average in its group is small and that a ects the e ort of the labour force of the ® rm. First, because this kind of policy a ects the morale of all types of workers. Secondly, because his policy breaks the stability of relations between the workers and makes them care less about the training of new entrants. In the end, it also a ects the level of e ort. Size is also another variable that has a positive coe cient that is signi® cantly di erent from zero, at a 1% level of signi® cance. As we have obtained with the Probit model, when the ® rm has more than 200 workers the likelihood of that ® rm paying a wage above the average increases and that positively a ects the average e ort of the workers. The higher the number of workers, the less chance of monitoring them and then it is cheaper for the ® rm to pay more, and doing so they increase the opportunity cost for the workers of being ® red. Also, in this case we pick up an indirect e ect on wages with this variable.
I V . C ON C L U D I N G R E MA R K S The fundamental predictions of e ciency wage models are supported by these results. We have found cross-section evidence for the existence of di erences in labour costs for ® rms belonging to the same industrial sector of the Spanish economy. Furthermore, we have analysed the negative in¯ uence of temporary contracts on wages. Empirical analysis suggests that the introduction of temporary contracts has had an impact on wage formation reducing the real wage cost. It has been made empirical estimation throughout a Probit model for the period of 1990 to 1994 in order to show that the likelihood of having a ® rm paying wages above the average of its industry decreases as the ratio of temporary workers increases. The estimation of the e ciency wage model has been carried out using a panel of 1098 Spanish manufacturing ® rms. When we made the regression with panel data, the basic results obtained in the cross-section estimation remained. Here, hierarchy of wages inside the industry is more important than the wage paid by the ® rm to increase e ort. This result is important if we think that any bargain made by one ® rm will a ect the wage of the industry. Thus, if relative wages are important for productivity , any shock
R. Sa nchez and L . T oharia
590 that can a ect the high wage sector will produce an employment adjustment instead of a wage adjustment. A C K N OW LE D G E ME N TS We are grateful to Fundacion Empresa Publica for providing the data source, especially J. Jamandreu for his help in data managing. R E F E R E N C ES Akerlof, G. A. (1982 ) Labour contracts as partial gift exchange, Quarterly Journal of Economics, 97, 543± 96. Akerlof, G. A. and Yellen, J. (1986 ) E ciency W ages Models of the L abour Market, Cambridge University Press, Cambridge. Arellano, M. and Bond, S. (1988 ) Dynamic Panel Data Estimation Using DPD: a Guide for Users, Institute of Economics and Statistics, London. Arellano, M. and Bond, S. (1991) Some T ests of Speci® cation for Panel Data: Monte Carlo Evidence an Application to Employment Equations, Review of Economic Studies, 58, 277 ± 97. European Commission (1994) , T he Economic and Financial Situation in Spain, Reports and Studies. Directorate General for Economic and Financial A airs, no 7. Gibbons, R. and Katz, L. (1987) Does unmeasured ability explain inter-industry wage di erences? Working Paper no. 3182, NBER. GuÈell Rotllan, M. (1997 ) Hiring decisions in a two-tier system within an e ciency wage context: an application to Spain, Mimeo, London School of Economics. Jimeno, J. F. and Toharia, L. (1993 ) T he Productivity E ects of Fixed-Term Contracts: Are T emporary W orkers L ess Productive than Permanent W orkers? Fedea, Madrid, D.T. no 93± 104. Krueger, A. and Summers, L. H. (1986 ) E ciency wages and inter-industry wage structure, Econometrica , 56, 259 ± 94. Levine, D. (1992 ) Can wage increases pay for themselves? Test with a production function, T he Economic Journal, 102, 1102 ± 15. Murphy, K. and Topel, R. (1987 ) Unemployment Risk and Earnings: T esting for Equalising W age Di erences in the L abour Market, Basil Blackwell, New York. Pencavel, J. (1972 ) Wages, speci® c training, and labour turnover in US manufacturing industries, International Economic Review, 13, 53± 64. Ramaswany, R. and Rowthorn, R. E. (1991 ) E ciency wages and wage dispersion, Economica, 58, 501± 14. Sa nchez, R. (1994 ) Can the Previous Y ear Unemployment Rate A ect Productivity? A DPD Contrast. Valencian’s Institute of Economics Research (IVIE), Valencia, WP-EC 94-10. Sa nchez, R., Urbano, A. and Ortõ , A. (1995 ) Wage premium in the industrial sector of the Spanish economy: empirical evidence, L abour, Review in L abour Economics and Industrial Relations, 9 (2) 253 ± 74. Segura, J., Duran, F., Toharia, L. and Bentolila, S. (1991 ) Ana lisis de la Contratacio n T emporal en EspanÄa, Ministerio de Trabajo y Seguridad Social, Madrid. Shapiro, C. and Stiglitz, J. (1984 ) Equilibrium unemployment as a worker discipline device, American Economic Review, 74, 433 ± 44.
Stiglitz, J. (1974 ) Alternative theories of wage determination and unemployment in LDC’s: the layout turnover model, Quarterly Journal of Economics, 88 (2), 194 ± 227. Stiglitz, J. (1987 ) The causes and consequences of the dependence of the quality on price, Journal of Economic L iterature, XXV (1), 1± 48. Straka, J. (1989 ) E ciency wages and collective bargain: theory and evidence, PhD dissertation, Cornell University. Wadhwani, S. and Wall, M. (1988) A direct test of the e ciency wage model using UK micro-data, Discussion paper, no 313, June, Centre for Labour Economics, London School of Economics, UK. Wadhwani, S. and Wall, M. (1991) A direct test of the e ciency wage model using UK micro-data, Oxford Economics Papers, 43, 529 ± 48. Yellen, J. (1984 ) E ciency wage models of unemployment, American Economic Review, 74 (2), 200 ± 5.
A PPE N D I X Industrial sector classi® cation on the NACE-CL IO system of classi® cation NACE-CLIO (1): Mineral and iron industry. NACE-CLIO (2): Minerals and products no metallic. NACE-CLIO (3): Chemical products. NACE-CLIO (4): Metallic products. NACE-CLIO (5): Machinery. NACE-CLIO (6 ): O ce machinery and others. NACE-CLIO (7): Electric materials. NACE-CLIO (8): Cars and engineering. NACE-CLIO (9): Other material of transport. NACE-CLIO (10): Meat and manufacturing of meat. NACE-CLIO (11): Food industry and tobacco. NACE-CLIO (12): Drinks. NACE-CLIO (13): Textile, clothing and shoes. NACE-CLIO (14): Leather, shoes and derived. NACE-CLIO (15): Wood and derived. (This is the category of reference. ) NACE-CLIO (16 ): Paper and derived. NACE-CLIO (17): Couch and plastic. NACE-CLIO (18): Other manufacturing products.
Size of the ® rm by the number of workers employed Less than 20 employees. From 21 to 50 employees. From 51 to 100 employees. From 101 to 200 employees. (This is the category of reference. ) From 201 to 500 employees. More than 500 employees.
591
T emporary workers and productivity Table A1. Average proportion of temporal workers by industrial sector of activity NACE-CLIO
1990
1991
1992
1993
1994
Mineral and iron industry Minerals and products no metallic Chemical products Metallic products Machinery O ce machinery and others Electric materials Cars and engine Other material of transport Meat and manufacturing of meat Food industry and tobacco Drinks Textile and clothing Leather, shoes and derived Wood and derived Paper and derived Couch and plastic Other manufactures products
18.48 25.94 13.51 22.52 23.81 12.82 23.31 12.82 23.31 18.06 21.42 32.30 27.98 14.34 26.68 29.60 30.76 19.65
17.97 24.79 13.77 22.54 18.67 14.56 22.68 18.67 14.56 22.68 13.64 17.85 32.98 28.36 14.29 24.04 33.77 32.81
17.01 20.57 12.63 17.58 15.92 15.00 17.93 10.83 12.35 31.17 30.33 15.89 21.63 31.20 26.76 19.88 21.02 18.76
12.27 18.02 13.92 17.29 14.43 15.26 16.60 9.26 11.42 31.61 29.56 16.77 22.15 32.24 23.76 17.04 18.69 20.13
10.06 19.58 12.44 18.13 13.73 17.89 18.91 11.22 16.10 31.98 28.30 14.21 22.58 27.29 27.57 15.78 22.44 19.04
Total
22.97
19.57
20.61
19.68
20.05
