disentangling the minimum wage puzzle: an analysis … · 2009. 7. 16. · in this sense, our...

“zwu004060380” — 2006/7/6 — page 988 — #1

DISENTANGLING THE MINIMUM WAGEPUZZLE: AN ANALYSIS OF WORKERACCESSIONS AND SEPARATIONS

Pedro PortugalBanco de Portugal and UniversidadeNOVA de Lisboa

Ana Rute CardosoIZA Bonn, University of Minho,and CEPR

AbstractChanges in the legislation in the mid-1980s in Portugal provide remarkably good conditionsfor analysis of the employment effects of mandatory minimum wages, as the minimum wageincreased sharply for a very specific group of workers. Relying on a matched employer–employee panel data set, we model gross worker flows—accessions and separations—incontinuing firms, as well as in new firms and those going out of business, using a count regres-sion model applied to proportions. Employment trends for teenagers, the affected group, arecontrasted to those of older workers before and after the raise in the youth minimum wage. Themajor effect on teenagers of a rising minimum wage has been the reduction of separations fromthe employer, which, during the period under analysis, has compensated for the reduction ofaccessions to new and continuing firms. In this sense, our results can reconcile some of the pre-vious evidence in the empirical literature when analyzing the aggregate impact of the minimumwage on youth employment without decomposing it by type of worker flow. (JEL: D21, J23, J38)

1. Introduction

The impact of the minimum wage on employment remains a controversial issuein the economics literature. Using mostly aggregate data, a certain consensushad been built around the idea that minimum wages reduce employment (see the

Acknowledgments: Part of this work was carried out while the second author was visiting theEconomics Program, RSSS, Australian National University, whose financial support and stimulat-ing hospitality are gratefully acknowledged. The support while visiting the Bank of Portugal isacknowledged as well. We would like to thank Orley Ashenfelter, David Autor, Dan Black, OlivierBlanchard, Pedro Carneiro, Bruce Chapman, Denise Doiron, Dan Hamermesh, Francis Kramarz,José A. Machado, Alan Manning, José Mata, Ray Rees, and José Varejão for their comments. Wealso thank seminar participants at Princeton University, University of Pompeu Fabra, UniversidadeNova de Lisboa, Bank of Italy, IZA, Bank of Portugal, Australian National University, and NewcastleUniversity. Helpful comments from participants at the meetings of the European Association ofLabour Economists (Jivaskyla), Applied Econometric Association (Brussels), Sociedade Portuguesade Investigação em Economia (Aveiro), the International Conference on Comparative Analysis ofEnterprise (micro) Data (Aarhus), and the New Zealand Conference on Database Integration andLinked Employer–Employee Data (Wellington) are also acknowledged. We thank the Ministry ofEmployment, Statistics Department, and the National Statistical Office (INE) for having providedthe data used. We are indebted to Lucena Vieira for excellent research assistance. We thank thefinancial support from the Fundação para a Ciência e a Tecnologia. The usual disclaimer applies.E-mail addresses: Portugal: [email protected]; Cardoso: [email protected]

Journal of the European Economic Association September 2006 4(5):988–1013© 2006 by the European Economic Association

“zwu004060380” — 2006/7/6 — page 989 — #2

Portugal and Cardoso Worker Accessions and Separations 989

overview by Brown, Gilroy, and Kohen 1982). This evidence was challenged inthe 1990s by studies using mainly data on net employment changes at the firmlevel and taking advantage of the framework of quasi-experiments. Such studies,by Card and Krueger (1994, 1995) and Katz and Krueger (1992), have launchedan intense debate; this is illustrated, for example, in the Review Symposium in theIndustrial and Labor Relations Review of July 1995, or by Burkhauser, Couch,and Wittenburg (2000), Card and Krueger (2000), Deere, Murphy, and Welch(1995), Dolado et al. (1996), Kennan (1995), Machin and Manning (1996), andNeumark and Wascher (2000).

In this paper we analyze the dynamics of employment at the firm level follow-ing a sharp increase in the minimum wage. Several novel aspects are introducedin the empirical analysis of the topic.

First, the change in the legislation being examined provides remarkably goodconditions for economic analysis, because the increase in the mandatory mini-mum wage was large and unpredictable, and it affected only a specific group ofworkers. Indeed, in 1987 the minimum wage for workers aged 17 increased inPortugal by 50%—due strictly to changes in the legislation—and for workersaged 18 or 19 it was raised from 75% to the full minimum wage rate, an increaseof 33%.

Second, the analysis relies on a panel of linked employer–employee data thatcovers, for each year, nearly all of the wage earners in the private sector (morethan 2 million workers)—an outstanding data set. Every firm with wage earnersis legally obliged to reply to this inquiry by the Ministry of Employment, and theresponse rate is thus extremely high.

Third, the study handles issues that have not previously been explored inthe literature. By explaining employer behavior regarding both accessions andseparations and by focusing not just on firms remaining in business but alsoon new firms and those closing down, this study adds to the previous literaturethat has explained net job flows at the firm level.1 We can therefore address theminimum wage puzzle by identifying exactly where and how the minimum wagebites: Do employers dismiss mainly youngsters following a rise in the youthminimum wage? Or do youngsters become underrepresented among the newlyhired workers? Do firms that are about to be set up, which are free to choose theirlabor force from the available pool of workers, hire relatively fewer youngstersthan comparable firms did before the minimum wage was raised? Does a risingminimum wage place unbearable constraints on firms and so contribute to firmclosure? Identification of the precise source for changing employment levels canreconcile some of the evidence that has previously been presented in the literatureas contradictory.

1. See, however, Card and Krueger (1994) for a study of the impact of changes in minimum wageson entries and exits of McDonald’s franchises.

“zwu004060380” — 2006/7/6 — page 990 — #3

990 Journal of the European Economic Association

We will also rely on micro data on workers to follow a line of analysispreviously used by Abowd et al. (2000), Currie and Fallick (1996), and Zavodny(2000). At the individual level, we can provide a direct answer to the followingquestion: Are those workers directly affected by a rise in the minimum wage lesslikely to keep their jobs?

Therefore, in the current study we are able to relax two types of constraintstraditionally faced by labor economists in the empirical analysis of the minimumwage. First, the impact of the minimum wage is usually hard to isolate from that ofother economic forces. Ideally, changes in the minimum wage would affect only aspecific group of workers, subject to conditions otherwise similar to other workers,a situation that very rarely occurs in economics. Second, even when changes in theminimum wage provide good conditions for economic analysis, the study to beundertaken still depends crucially on the data available, in particular its quality anddetail.2 On both of these fronts, our study relies on unusually favorable conditions,enabling the analysis to progress in directions left unexplored by previous studies.

Section 2 describes the institutional framework in Portugal and the majorchanges that have taken place with respect to the mandatory minimum wage.Sections 3 and 4 provide an aggregate perspective on changes in the wage distri-bution and in employment, showing a high degree of compliance with the newlegislation. Section 5.1 models job accessions and job separations in continuingfirms, and Section 5.2 performs a similar exercise for new firms and for firmsclosing down. Worker turnover is the subject of Section 6. Section 7 discussesthe results, highlighting an extension of the theoretical model of Burdett andMortensen (1998) because it provides clear predictions that fit well the empiricalevidence disclosed for Portugal. The last part of the paper presents concludingcomments.

2. Institutional Framework

In Portugal, a mandatory minimum wage was set for the first time in 1974,which covered workers aged 20 or older and excluded agriculture and house-hold services. Ever since, it has been annually updated by the parliament undergovernment proposal.3 Decisions on the level of the minimum wage are madeon a discretionary basis, usually taking into account past and predicted inflationand in consultation with representatives of employers and workers. It is definedas monthly earnings.4 Throughout the period under analysis, the minimum wage

2. Meyer and Wise (1983a, 1983b) have developed a methodology for estimating the impact of theminimum wage on employment and the wage distribution using just one cross-section of data. Theirmethodology is discussed in Dickens et al. (1998).3. The only exceptions were 1982, when it was not updated, and 1989, when it was updated twice.4. For the period under analysis (1986–1989) the national minimum wage was set at PTE 22,500;25,200; 27,200; and 30,000, respectively.

“zwu004060380” — 2006/7/6 — page 991 — #4


Table 1. Share of the general minimum wage enforced, according to the age of the worker,in percentage.

Age group

Year 15 16 17 18 19 20–25

1979 to 1986 50 50 50 75 75 1001987 50 50 75 100* 100* 100*1988 75 75 75 100* 100* 100*

Source: Portugal, Diário da República, several issues.*Note: 80 if apprentice.

was defined as the base monthly payment, excluding overtime and other nonbasiccomponents of pay.

Since it was first enacted the minimum wage has undergone several changes,5

and the major one concerns age coverage. Indeed, the ages subject to exemption(as well as the percentage of exemption) have changed several times, and a clearerdescription can be provided by a table in which the major changes are highlightedin bold (see Table 1).

Three changes in the legislation have yielded remarkable wage increases forthe ages involved:6

• In 1987 the minimum wage for workers aged 17 increased by 50% due tochanges in the legislation ceteris paribus, since it was raised from 50% to75% of the full minimum wage.

• Also in 1987, the minimum wage for workers aged 18 or 19 was raised from75% of the minimum wage to the full minimum, thereby increasing 33%.

• In 1988 the minimum wage for workers aged 16 or younger was raised from50% to 75% of the full minimum, thus increasing by 50%.

Exceptions could, however, apply to the second change described if the workerwas formally classified as an apprentice, who would be entitled to 80% of thegeneral minimum wage instead of the full rate. This exemption could underminethe impact of the rise in the minimum wage that we intend to study if employerscould change the occupational category of the worker to an apprentice in orderto avoid paying the full minimum wage—an issue to be addressed in the nextsection.

This study concentrates on the two changes that took place in 1987, becausethey provide excellent conditions for economic analysis.

3. Aggregate Changes in Wages and Employment

Evidence on the impact of the rise in the minimum wage, based on aggregatefigures (Table 2 and Figure 1), indicates that compliance with the minimum wage

5. Fully described in Appendix B.6. Each year, changes in the legislation were enforced from January onward.

“zwu004060380” — 2006/7/6 — page 992 — #5


Table 2. Growth in the average wage of full-time wage earners, by age group, 1987–1989.

Wage growth factor

Age 1987 1988 1989

17 1.21 1.14 1.1518 1.21 1.12 1.1319 1.18 1.12 1.1320 1.15 1.11 1.1321 1.14 1.12 1.1222 1.14 1.12 1.1323 1.14 1.12 1.1424 1.14 1.12 1.1425–29 1.14 1.10 1.1330–34 1.15 1.10 1.1535–39 1.14 1.09 1.1340–65 1.16 1.11 1.14

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1986–1989).Note: Wages refer to the gross base monthly wage, and the change is computed as wa,t /wa,t−1, where w stands for the

average wage of the age group a and the subscript t denotes time.

legislation was high. Note the impact of the legal changes on the average wage ofthe different age groups, as shown by three cross-sections of data. As expected,wages increased most sharply precisely for the groups of workers affected by therise in the minimum wage (see highlights in bold in Table 2). For example, between1986 and 1987 the average wage for workers aged 17, 18, or 19 increased the most,whereas all the other age categories had a much lower and more homogeneouswage growth. The rise in the legal minimum wage thus seems to have had arelevant impact on the wages of eligible workers, indicating that sub-minimum

Figure 1. Wage distribution, teenagers, 1986 versus 1987. Source: Computations based on Portugal,MTSS, Quadros de Pessoal (1986–1987). Notes: Reports the gross base monthly wage for teenagersaged 17 to 19. The first vertical line indicates 75% of the national minimum wage in 1986; the secondline indicates the full national minimum wage in 1987.

“zwu004060380” — 2006/7/6 — page 993 — #6


Table 3. Share of apprentices, by age group, 1985–1989.

Age 1986 1987 1988 1989

16 0.838 0.847 0.837 0.83517 0.779 0.774 0.761 0.76118 0.625 0.617 0.597 0.58319 0.493 0.503 0.483 0.45920 0.349 0.367 0.372 0.35721 0.250 0.281 0.283 0.28622 0.197 0.213 0.230 0.22423 0.153 0.175 0.182 0.18424 0.120 0.138 0.150 0.14725–29 0.058 0.073 0.082 0.09030–34 0.025 0.034 0.034 0.00035–39 0.016 0.022 0.022 0.00040–65 0.010 0.014 0.012 0.000

Total 0.104 0.116 0.122 0.127

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1986–1989).

wages had been widespread among teenagers7 and that registering a worker asan apprentice did not increase during this period (see Table 3).

Visual inspection of the wage distribution in Figure 1 stresses the relevance ofthe changes that took place. In 1986, the wage distribution for teenagers showeda peak at 75% of the national minimum wage—that is, the sub-minimum allowedfor workers aged 18 or 19—and another one at the full minimum wage. By 1987,the peak at the sub-minimum wage level had almost vanished, and the share ofteens at the minimum wage had increased sharply. The kernel plot (produced witha common bandwidth) therefore indicates that the rise in the applicable minimumwage shifted sub-minimum teenager wages to the new minimum wage level, andthus compliance to the new regulation was widespread.

Consider now the yearly changes in employment between 1986 and 1989, aperiod marked by sharp economic growth (see the growth in total employment,Table 4). During that period, teenage employment increased at a fast pace. Indeed,whereas total employment increased by 3.3% between 1986 and 1987, the numberof jobs taken by workers aged 17 to 19 increased more sharply, by 4.3% to6.1%. Nevertheless, the group aged 20 to 23 presented even higher employmentgrowth rates. The following year, total employment increased by 1.2%, whereasthe youngest workers had a higher employment growth, which compensated forthe decline in the employment of prime-age workers. Hence this level of analysisdoes not suggest the existence of an aggregate negative impact on employmentresulting from the sharp rise in the minimum wage for specific age groups.

The same indication is given in Table 5, which was produced from a different(and independent) data source: the Labor Force Household Survey run by the

7. In 1986, the shares of workers aged 17, 18, and 19 earning sub-minimum wages were 84%,63%, and 46%, respectively.

“zwu004060380” — 2006/7/6 — page 994 — #7


Table 4. Growth in number of full-time wage earners by age groups, 1987–1989.

Employment growth factor

Age 1987 1988 1989

17 1.052 1.094 1.11918 1.043 1.090 1.10619 1.061 1.066 1.10820 1.086 1.042 1.10821 1.082 1.051 1.08822 1.109 1.054 1.09923 1.063 1.083 1.09524 1.055 1.041 1.13625–29 1.014 0.995 1.07130–34 1.023 1.002 1.03235–39 1.029 0.983 1.00140–65 1.027 0.996 1.049

total 1.033 1.012 1.077

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1986–1989).Note: Employment growth was computed as La,t /La,t−1, where La stands for employment of age group a and the

subscript t denotes time.

Portuguese Statistical Institute.8 This is the survey that feeds the OECD laborforce statistics. Here it is worth noting the sharp decrease from 1986 to 1989 in theunemployment rate for youngsters aged 17, 18, and 19. The overall unemploymentrate declined as well, reflecting cyclical conditions, but at a much slower pace(from 9.6% to 7% over the same period). This evidence is hard to reconcilewith the view that minimum wage hikes always generate significant employmentlosses.

It is worth referring to the previous literature on this issue for Portugal. Theoverall employment trends reported for teenagers in the affected age group—bothby the Ministry of Labor, covering the population of firms employing wage earnersin the private sector, and by the National Statistical Office, in its Labor Force Sur-vey—are at odds with the ones offered in the study by Pereira (2000, pp. 40–41;2003, p. 242, Table 5) on the impact of changes in the minimum wage in Portugal.

Table 5. Teenage (17, 18, and 19) employment and unemployment.

Employment Full-time employmentin the private sector in the private sector Unemployment rate

(thousands) (thousands) (%)

1986 173.7 170.0 20.21987 180.9 176.9 16.81988 177.9 173.1 13.71989 188.1 184.0 11.6

Source: Portugal, INE, Labor Force Survey (1986–1989).

8. We thank the INE for making available to us the disaggregated data.

“zwu004060380” — 2006/7/6 — page 995 — #8


Her data set is a nonrandom sample from the Ministry of Labor data; this sampleis not representative of the population of firms (that we use) and, in particular, isseverely biased with respect to the actual trend in employment for the affectedgroup of workers.9

4. Worker Flows

Using the unique firm identifiers in our data set (see Appendix A), it is possible topinpoint newly started and closing firms. And because there are unique individualidentifiers, one can trace hirings and separations. Combining the two sources ofinformation, we compute the four components of the net employment flows:hirings, separations, jobs created by the entry of new firms, and jobs destroyedbecause firms were shut down. Worker flows are exhibited in Table 6 for all wageearners and separately for teenage wage earners.

It is clear from Table 6 that hirings and separations increased significantlyover the 1986–1989 period. The employment resulting from new firms alsoincreased, whereas employed lost on account of plant closings remained more orless constant. The gross flows of teenage workers followed the general (cyclical)pattern.

However, it is interesting that a closer look reveals apparently distinct patternsfor the evolution of the teenage share in the gross worker flows (see Table 7). Theteenage contribution to the employment created by new firms decreased, whereasthe proportion of teenagers in the workforce of firm closures increased. The

Table 6. Gross worker flows.

Worker flows (thousands)

1986 1987 1988 1989

Total Teens Total Teens Total Teens Total Teens

Employment 1,424.1 87.3 1,473.2 91.7 1,488.6 99.3 1,569.1 109.8Hirings 191.8 27.9 237.7 35.3 285.8 42.0 323.2 46.8Separations 176.4 32.2 205.4 34.6 299.7 39.3 290.7 42.0New firms 53.2 6.7 63.1 8.1 77.7 9.5 91.0 10.1Firm closings 46.3 4.4 48.4 4.7 42.9 4.4 50.7 5.3

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1985–1989).Note: Teenagers aged 17, 18, and 19. The stock of employment refers to year t indicated in the heading of the column;

the flows refer to the period from t − 1 to t .

9. We tried hard to replicate Pereira’s computations using a data set with the same restrictions (ontime, age, industry, and region) employed in her paper. In this endeavor, one is faced at the outsetwith the sharp contrast between her figures on the evolution of aggregate teenage employment andthe figures obtained from our counterfactual data set (which is not a sample, but does conform to thenotion of the population). Not unexpectedly, we were unable to obtain results that would resemblethe ones produced in her study. In other words, we always uncover evidence of teenager employmentincrements following the change in the minimum wage legislation.

“zwu004060380” — 2006/7/6 — page 996 — #9


Table 7. Teenage share in gross worker flows.

Teenage worker flow share

1986 1987 1988 1989

Hirings 0.146 0.149 0.147 0.145Separations 0.182 0.168 0.131 0.144New firms 0.126 0.128 0.123 0.112Firm closings 0.095 0.097 0.101 0.105

Source: Table 6.

fraction of teenagers in hirings does not seem to have changed significantly. Theremarkable result appears to be a sharp reduction in the proportion of teenagersamong those workers that separate from their firms. Relating the change in theminimum wage to gross worker flows is a novel feature of our investigation, andwe shall deepen the analysis of these outcomes in what follows.

5. Worker Accessions versus Worker Separations

Firms may change the composition of their labor force following an increase in theminimum wage for a particular group of workers. In particular, they might dismissworkers whose marginal productivity falls below the new minimum wage or avoidhiring those workers. This section models worker flows, distinguishing betweenhirings by continuing firms, hirings by companies newly set up, separations fromcontinuing firms, and layoffs from companies that close down.

We will concentrate on the share of teenagers in total job accessions/sepa-rations involving workers aged up to 35 years.10 Whereas overall figures for jobaccessions and separations may be influenced by the business cycle, concentratingon the share of teenagers in those flows can highlight changes in employer policiesfollowing the rise in the youth minimum wage. In fact, considering just the flowsfor workers aged up to 35 means that we are comparing similar groups, which areexpected to be affected in a similar way by trends in the business cycle.11 Insightinto the impact of minimum wages on worker flows was obtained via regressionanalysis. A Poisson regression model applied to grouped data was employed. Thesize of the flow of teenage workers was modeled as

yi ∼ P [µi = λi], (1)

withλi = exp {x′

iβ} Ni; (2)

10. Other reference groups were also tried, with similar results.11. Another reason we did not include older workers is because we want to minimize the contami-nation of our results generated by early retirement decisions, a phenomenon that was fairly frequentduring this period.

“zwu004060380” — 2006/7/6 — page 997 — #10


in a more convenient reparameterization,

log(λi) = x′iβ + log(Ni), (3)

where µ is the expected value of the number of events (in this case, the number ofteenage workers), x is the vector of explanatory variables, and β is the vector ofcorresponding regression coefficients. Here Ni denotes the size of the exposureset in the same unit in which the events occur (in this case, the firm’s flow of work-ers). Note that this approach automatically accommodates the need to properlyweight each observation by the size of the total flow of workers (aged less than35 years).

When modeling trends in continuing firms, one must account for the possi-bility that there are repeated observations. Firms will be present as long as thereare hirings (or separations) in a given year. A natural way to accommodate thelongitudinal nature of this data is to specify a model with firm-specific randomeffects:

yit ∼ P [µit = αiλit], (4)

whereλit = exp {x′

itβ} Ni (5)

and where αi is a random error term that is assumed to be gamma distributed withunit mean and variance equal to 1/δ. Integration with respect to αi leads to theunconditional density (Cameron and Trivedi 1998):

P [yi1, . . . , yiT ] =(∏

t

λyit

it

yit !

) (δ∑

t λit + δ

)δ

×⎡⎣(∑

t

λit + δ

)− ∑t yit

⎤⎦ [

�(∑

t yit + δ)

�(δ)

]. (6)

Maximum likelihood estimation of the β and δ parameters is straightforward.The results of both specifications, with and without random effects, are presentedfor comparison.12

5.1. Gross Job Flows in Continuing Firms

Consider first the recruitment and dismissal policy in continuing firms. The regres-sions in Table 8 compare the recruitment policy in 1986 (before the change in

12. Negative binomial models, with and without firm random effects, were also estimated; theresults were identical to the ones presented.

“zwu004060380” — 2006/7/6 — page 998 — #11


Table 8. Share of teenagers hired by continuing firms, Poisson regression model.

No random effects With random effects

Variables Coef. Std. Coef. Std.

Year 1988 (dummy) −0.057 0.009 −0.036 0.010Year 1989 (dummy) −0.065 0.009 −0.043 0.010Firm size −0.099 0.003 −0.106 0.004Firm hiring rate −0.474 0.021 −0.455 0.027Market concentration (Herfindahl index) −0.970 0.045 −0.977 0.062

1/δ 0.228Log–likelihood −96,177 −92,433N 99,608 99,608

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1986–1989).Notes: Controlling for the industry (7 dummies) and for public or foreign ownership of the company. Firm size is in

logs.

the minimum wage) with that of 1988 (14 months after the minimum wage hike),and with that of 1989, thereby allowing for a longer lag in the impact of theminimum wage on recruitment policies.13 Thus, year dummy variables aimed atcapturing the impact of the policy change were included in the pooled regression.The affected group and the control group (workers aged 20 to 35) were kept inthe sample.

The results of the models with and without random effects point in the samedirection, though the magnitude of the coefficients changes slightly. A likelihoodratio test compares the random effects estimator with the pooled Poisson estimator,revealing that the panel estimator is more adequate. Therefore, the comments thatwe shall present refer just to the model that includes random effects.

After the rise in the youth minimum wage, companies decreased the share ofteenagers among their newly hired workforce (Table 8). Note that for both 1988and 1989, the dummy variable coefficient estimate is negative and statisticallysignificant. For both 1988 or 1989, the share of teenagers in overall job accessionsto continuing firms was about 4% lower than in 1986. In that sense, results on thebehavior of employers concerning their hiring policy following an increase in theminimum wage lend support to the competitive view of the labor market, becausethere seems to have been a shift away from teenagers. Larger firms are less likelyto recruit teenagers; in a company 10% larger, the share of teenagers recruited is1% lower. The inclusion of the hiring rate among the regressors aims at furthercontrolling for the impact of business cycle conditions. Firms with a higher hiringrate are less likely to recruit teenagers; a 1 percentage-point increase in the hiringrate decreases by 0.5% the share of teenagers hired. The concentration index ismeant to proxy for the profitability of the industry: a 1 percentage-point increasein the concentration index in the industry decreases the share of teenagers hiredby about 1%.

13. The data were gathered in March each year, and the changes in legislation took effect in January.

“zwu004060380” — 2006/7/6 — page 999 — #12


Table 9. Share of teenagers separated from continuing firms, Poisson regression model.

No random effects With random effects

Variables Coef. Std. Coef. Std.

Year 1988 (dummy) −0.132 0.010 −0.150 0.010Year 1989 (dummy) −0.111 0.009 −0.140 0.010Firm size −0.112 0.003 −0.113 0.005Firm hiring rate −0.123 0.016 −0.121 0.020Market concentration (Herfindahl index) −0.184 0.054 −1.140 0.075

1/δ 0.379Log–likelihood −97,770 −93,045N 125,397 125,397


logs.

However, the share of teenagers in job separations also decreased followingthe rise in the minimum wage (Table 9). Indeed, it was 15% (resp. 14%) lower in1988 (resp. 1989) than in 1986. These results contrast with those of the previousTable, indicating that workers affected by a sharp rise in the minimum wage arenot overrepresented among those who afterwards separate from their employers.

5.2. Firm Turnover

We go on to inspect the impact of the rise in the minimum wage on firm closureand firm creation, a topic earlier analyzed by Portugal and Guimarães (1998). Thisanalysis provides a more complete picture of the effect of the minimum wage inan economy with high levels of firm creation and firm destruction (Blanchard andPortugal 2001).

Firms set up in 1988 or in 1989, a year or two after the policy change,recruited a 4% lower share of teenagers than those set up in 1986 (Table 10).As with accessions to continuing firms, these results lend support to the idea thatrising minimum wages reduce the demand for the affected workers, since firmsset up in 1988 or 1989 increased their tendency to employ older workers when

Table 10. Share of teenagers hired by new firms, Poisson regression model.

Variables Coef. Std.

Year 1988 (dummy) −0.042 0.018Year 1989 (dummy) −0.041 0.018Firm size −0.113 0.005Market concentration (Herfindahl index) −2.880 0.174

Log–likelihood −29,936N 38,138


logs.

“zwu004060380” — 2006/7/6 — page 1000 — #13


Table 11. Share of teenagers dismissed from firms closing down, Poisson regression model.

Variables Coef. Std.

Year 1988 (dummy) 0.050 0.023Year 1989 (dummy) 0.025 0.023Firm size −0.106 0.007Market concentration (Herfindahl index) −2.941 0.257

Log–likelihood −15,023N 19,203


logs.

compared to similar firms that had been set up in 1986. Also in line with traditionaltheory, the results in Table 11 indicate that teenage workers are overrepresentedin the labor force of firms closing down in 1988 when compared to the situationbefore the rise in the youth minimum wage. Some support is therefore found forthe hypothesis that rising minimum wages may have placed unbearable costs onsome firms, leading to teenagers being overrepresented in firms closing down.

In synthesis, the results gathered so far indicate that the short-run (one ortwo years) impact of the rise in the youth minimum wage can be decomposedinto flows operating in opposite directions. On one hand, teenagers make up asmaller share of job accessions and a larger share of the workforce in firms goingout of business, and either fact points toward a reduction in relative demand forteenagers. On the other, however, they make up a smaller share of the workforceseparating from continuing firms. The balance between those flows operating inopposite directions led, in Portugal during the period under analysis, to a rise inoverall teenage employment.

A simple back-of-the-envelope exercise can be performed, using the regres-sion estimates and the information in Tables 6 and 7, to check the relativeimportance of the different teenage worker flows. The contribution of each teenageworker flow to overall employment was computed as

teen flowt−1,t

employmentt−1= total,flowt−1,t

employmentt−1× share, (7)

where the flow can be hirings, separations, recruitments into new firms, or dis-missals from firms closing down and where share refers to the share of teenagersin each worker flow. To simulate the impact of the minimum wage, teenage shareswere computed using three alternative scenarios: the actual share of teenagers in1986; the share that would prevail in 1988 and 1989 given the impact of the min-imum wage, ceteris paribus (computed using the 1986 shares and the regressioncoefficients); and the actual shares in 1988 and 1989. Table 12 reports the results.

Overall, the aggregate impact of teenage worker flows during 1988 repre-sented 0.53% of total employment in the previous year; for 1989, that value was0.64% (see the last line in columns 1 and 2 of Table 12). In 1988 (resp. 1989), the

“zwu004060380” — 2006/7/6 — page 1001 — #14


Table 12. Contribution of teenage worker flows to overall employment: simulations usingalternative procedures to compute teenage shares.

Contribution of teenage worker flow, using: Impact min w,

Shares 88, 89 Shares 86 Shares min w ceteris paribus

1988 1989 1988 1989 1988 1989 1988 1989

(1) (2) (3) (4) (5) (6) (5)–(3) (6)–(4)

Entry 0.64 0.68 0.66 0.77 0.64 0.74 −0.03 −0.03Exit −0.30 −0.36 −0.28 −0.32 −0.29 −0.33 −0.01 −0.01Accessions 2.85 3.14 2.82 3.16 2.72 3.02 −0.10 −0.14Separations −2.67 −2.82 −3.71 −3.56 −3.16 −3.07 0.56 0.50

Total 0.53 0.64 −0.50 0.04 −0.09 0.36 0.41 0.32

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1987–1989).

change in the minimum wage legislation was responsible for an increase in totalemployment of 0.41 (resp. 0.32) percentage points, a fairly large portion of thetotal observed increase in teenage employment. This outcome was generated bythe two forces operating in opposite directions previously identified: a decrease inteenage employment in 1988 (1989) of 0.14 (0.18) percentage points, attributableto fewer hirings by both new and continuing firms as well as to an increase inthe share of teenagers in the workforce of the firms that closed; and an increasein teenage employment in 1988 (1989) of 0.56 (0.50) percentage points that wasproduced by the sharp decrease in separations. Therefore, the decline in separa-tions has clearly driven the rising teenage employment level during the periodunder analysis in Portugal.14

6. Worker Flows: Do Rising Minimum Wages Increase Job Attachment?

A longitudinal perspective from a worker point of view can shed further light onthis issue. Following workers over time, we can directly check whether youngsterswho were earning sub-minimum wages tended to lose their jobs following thesharp increase in their legal minimum wage. Results reinforce the initial hint,as the rise in the minimum wage seems to have generated rising employmentattachment by teenagers already holding a job.

Workers aged 16 to 18 and earning sub-minimum wages in 1986 were ayear later eligible for a sharp wage rise. Therefore, if the labor market werecompetitive and if these workers were earning the value of their marginal product,then the rise in the minimum wage would price them out of employment. Instead,Table 13 shows that, in the age groups affected, the share of workers who remainemployed is greater for those initially earning sub-minimum wages than for thoseat the full minimum. Whereas 66% of the youngsters aged 16 to 18 employed at

14. In 1998 the Portuguese law extended the minimum wage to workers under 18 years of age.A study by the Ministry of Labor claims that this policy change was associated with a modest (1.2%)employment increase (Portugal, MSST, 2002).

“zwu004060380” — 2006/7/6 — page 1002 — #15


Table 13. Employment status in 1987 of full-time wage-earners in 1986, by age and wagecategory.

Situation in 1987

Age in 1986 Wage in 1986 Full-time Part-time Jobless

16–18 Below minimum 65.96 4.24 29.80At minimum 59.91 3.66 36.42Between min86 and min87 75.65 4.41 19.94At or barely above min87* 67.87 4.10 28.03Above min 87 × 1.1 71.72 4.71 23.57

19–25 Below minimum 56.55 5.12 38.33At minimum 57.63 3.26 39.11Between min86 and min87 67.66 4.79 27.54At or barely above min87* 65.91 4.32 29.77Above min87 × 1.1 71.66 3.70 24.64



Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1986–1987).Note: *“Barely above” is defined as not exceeding the minimum plus 10%.

sub-minimum wages in 1986 were still employed full-time a year later, just 60%of those earning the full minimum kept their employment status.

If, instead of taking teenagers at the minimum wage as the comparison group,we consider older workers on sub-minimum wages, a similar trend is detected.The share of sub-minimum wage workers remaining in employment is higher inthe ages directly affected by the rise in the minimum wage than in older groups.

This section models retention rates from a worker standpoint. We will modelthe probability that a worker will remain with the same employer.15 Let usfirst briefly discuss the different alternatives considered for the definition of thetreatment group and then report on net wage changes for those groups.

Three alternative specifications of the treatment group have been defined, asfollows.

Treatment 1. A broad specification that considers simply whether a workeris a teenager (aged 16 to 18) in measuring the employment impact of a rise in theyouth minimum wage.

15. The probability of losing a job and becoming unemployed has been more frequently debatedin the literature, so we also considered the probability of remaining employed (whether or not withthe same employer); the results are in line with those reported here.

“zwu004060380” — 2006/7/6 — page 1003 — #16


Treatment 2. An alternative specification considers whether or not theteenager is eligible for a wage hike in the initial period (that is, earned a sub-minimum wage, disregarding how far the sub-minimum wage deviates from theminimum wage).

Treatment 3. This specification takes into explicit account the changes inlegislation when computing the distance between the worker’s previous wage andthe new minimum wage (wage gap). This measure has been used in the literatureas an independent variable to capture the impact of changes in the minimum wage(Card and Krueger 1994; Currie and Fallick 1996; Zavodny 2000). According tostandard competitive reasoning, the probability of remaining employed should belower the greater is the change required in the worker’s wage in order to adjust tothe new minimum wage. The wage gap was computed as

wage gapit = min wt − w86,i

w86,i

(8)

for period t and worker i, a teenager initially earning sub-minimum wage w; it iszero otherwise.

Attention was restricted to workers aged 16 to 35 (teenagers and youngadults), since the affected and the comparison groups should be similar (Meyer1995). Results are robust with respect to the definition of alternative controlgroups. In particular, if we consider only workers at or slightly above the mini-mum wage (or only workers aged 25 or less) as the comparison groups, then thesame results are obtained.

The evidence reported in the previous sections suggested that the workersaffected by the change in the legislation actually had a sharp wage increase.Complementing the data reported in Table 2 and Figure 1 on the evolution of grosswages, we now report, in Table 14, the net wage growth of each of the treatmentgroups after controlling for factors such as the industry, size, and ownership(foreign or national, public or private) of the company as well as the schooling,

Table 14. Wage growth, 1986 to 1988.

Regression estimates

Coef. Std. Coef. Std. Coef. Std.

Treatment 1 0.132 0.002Treatment 2 0.214 0.002Treatment 3 0.382 0.004

N 356,973 356,973 356,973R2 0.085 0.100 0.103

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1986–1988).Note: All three regressions control for: industry (7 dummies), foreign or public ownership of the company and its size,

worker schooling (7 dummies), age, gender, tenure, a dummy for newcomers into the firm (tenure less than one year), anda dummy if apprentice.

“zwu004060380” — 2006/7/6 — page 1004 — #17


gender, and tenure of the worker. It is clear that the treatment group was entitled toconsiderably higher wage growth than the control group, in line with the evidenceon gross wages previously reported. Moreover, as we adopt a finer definition ofthe treatment group—from just teenagers, to sub-minimum wages, to the wagegap—larger wage increases are noticed.

We now turn to modeling worker retentions. Workers employed in 1986, ayear before the rise in the minimum wage, were kept for analysis in order toexplain their probability of remaining employed in 1988, 14 months after the risein the minimum wage.16

The three specifications in Table 15 provide a consistent view of the impactof the minimum wage.17 After controlling for age and tenure effects, teenagersin general were more likely to keep their jobs, from 1986 to 1988, than wereolder workers. The impact of the change in the minimum wage is more telling incolumns 4 and 6, which provide a more rigorous account of the quasi-experimentand which also control for being a teenager. As their minimum wage was raised,

Table 15. Probability of remaining with the same employer from 1986 to 1988 (logitmodel).



Treatment 1 0.326 0.011 0.093 0.018 0.226 0.014Treatment 2 0.325 0.020Treatment 3 0.273 0.025Male −0.132 0.005 −0.137 0.006 −0.137 0.006Age 0.011 0.001 0.010 0.001 0.010 0.001Tenure 0.062 0.001 0.062 0.001 0.062 0.001Tenure less than 1 −0.379 0.007 −0.378 0.007 −0.380 0.007Apprentice 0.065 0.008 0.044 0.008 0.057 0.008Wage86 0.607 0.010 0.640 0.011 0.636 0.011Firm size 0.082 0.002 0.082 0.002 0.082 0.002Constant −7.266 0.104 −7.571 0.106 −7.534 0.107

N 695,751 695,751 695,751χ2 77,255.7 77,522.7 77,373.3

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1986–1988).Note: The regression controls for schooling (7 dummies), industry (7 dummies), and foreign or public ownership of

the company.

16. The probability of remaining employed in 1987, two months after the rise in the minimumwage, should not be considered because this time lag is too short for the change in legislation to havehad an effect. For example, Neumark (1999) claims that the effect of changes in the minimum wageis felt most with a one-year lag. Analyzing the separations from 1986 to 1989 would be stronglyinfluenced by the military draft for men aged 20–21 in 1989 and therefore we have not consideredthat longer lag. Note, however, that 18-year-olds in 1986 were already affected by the military servicein 1988.17. The wage of the worker was included among the regressors. Following Card and Krueger(1995) and Abowd et al. (2000), it is intended to capture heterogeneity in labor force attachmentacross the wage distribution.

“zwu004060380” — 2006/7/6 — page 1005 — #18


compressing the bottom part of the wage distribution, the expected benefits fromsearching for a new job decline and so there is now less incentive to quit the job.Teenagers earning sub-minimum wages in 1986 are much more likely to remainwith their employers than other teenagers. The change in legislation led to a wageraise that may have increased job attachment for the affected group of workers.The regression coefficient on the wage-gap variable (treatment 3) is also positiveand statistically significant, revealing that teenagers initially earning the lowestwages—and therefore subject to the highest wage raise—are the workers mostlikely to stick with their employers.18

The magnitude of the effects is, however, larger than what would be expectedfrom the worker flows regressions. The regression coefficients in Table 15 implyan increase in retention rates of about 38%, a value significantly larger than the13–15% decrease in the share of teenagers among separations in Table 8. Furtherinquiry is needed to reconcile these results.

What if sub-minimum wages had not changed? Would teenagers have shownthe same pattern of behavior anyway? To check if the differences in behaviorthat we are capturing are indeed due to changes in the minimum wage legislationand are not due to persistent contrasts across the groups of workers, different timeperiods were considered as well—when no change in the legislation had occurred(1984 to 1986 and 1989 to 1991). Thus, worker retentions from 1984 to 1986 andfrom 1989 to 1991 were compared with worker retention from 1986 to 1989 bypooling the data for the three periods.19 The impact of the change in the minimumwage was captured by interacting a dummy variable for the 1986–1988 period,after the change in the legislation, with the treatment variables.

Table 16 sheds additional light on the previous results. The relevant coeffi-cients are now given by the interaction of the treatment variables with the yeardummy representing the employment experience of 1986 to 1988. When con-trasted with the job-holding experience of 1984 to 1986 and 1989 to 1991, theimpact of the change in sub-minimum wages is now smaller than before, but isstill sizable and statistically significant. Indeed, these estimation results revealthat, conditional on all the covariates (including age and tenure), teenagers showhigher retention rates than do their older counterparts regardless of the periodbeing considered. But it happens that retention rates are highest for the 1986 to1988 period. It is worth mentioning that stronger effects are, again, obtained forthe measures that best identify the group of workers subjected to the change in theminimum wage legislation. Whereas in the first specification (column 2) there is

18. Nevertheless, observe that higher wage raises are also likely to impose a heavier burden onemployers, increasing the chances of a dismissal. Our results seem to suggest that the decrease inquit rates more than offsets, in the aggregate, the increase in displacement rates.19. The choice of these periods was dictated by data availability considerations. Before 1984 thereare some miscoding problems regarding the registry of the date of birth, and after 1991 changes inmandatory schooling affect labor supply.

“zwu004060380” — 2006/7/6 — page 1006 — #19


Table 16. Probability of remaining with the same employer, from 1984 to 1986, from 1986to 1988, and from 1989 to 1991, pooled data (logit model).



Treatment 1 0.319 0.007 0.182 0.010 0.215 0.009Treatment 2 0.227 0.011Treatment 3 0.202 0.012Year*treatment 1 0.044 0.010 −0.057 0.019 0.214 0.009Year*treatment 2 0.103 0.022Year*treatment 3 0.093 0.027Male −0.152 0.003 −0.156 0.003 −0.156 0.003Age 0.015 0.0004 0.014 0.0004 0.015 0.0004Tenure 0.059 0.0004 0.059 0.0004 0.059 0.0004Tenure less than 1 −0.413 0.004 −0.412 0.004 −0.413 0.004Apprentice 0.088 0.005 0.070 0.005 0.080 0.005Wage 0.610 0.006 0.640 0.006 0.640 0.0062Firm size 0.111 0.001 0.111 0.001 0.111 0.001Constant −7.299 0.059 −7.578 0.060 −7.587 0.061

N 2,200,197 2,200,197 2,200,197χ2 287,225.2 287,882.4 287,626.8

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1984–1988).Note: The regression controls for schooling (7 dummies), industry (7 dummies), foreign or public ownership of the

company, and two time dummies.

an indication that retention rates are 4% higher for the target group, in column 4the corresponding effect is about 11%. A similar picture is drawn by the resultsof the third specification. The estimation results appear now to be more in linethan before with those reported for the worker flows.

Further checks were pursued with the use of different definitions of the con-trol group. Table 17 summarizes those attempts. To obtain a comparison groupcloser to the treatment group, the sample was first restricted to workers earningwages below 1.25 times the minimum wage. In a similar vein, the sample wassubsequently restricted to workers aged less than 25 years. In the final exercise,the comparison group was restricted to workers aged between 25 and 34 to avoidthe contamination of results by spillover effects on individuals aged close to theaffected group.

The main thrust of these results is, again, that separation rates declined withthe change in minimum wage legislation, irrespective of the comparison group.This indication is particularly clear for the treatment 2 and 3 variables, as hintedbefore. In this admittedly crude exercise, there is no clear pattern that wouldsuggest the presence of spillover effects.

These results contrast with those obtained in several previous studies. Stewart(2002) found no impact on employment for the affected group after the introduc-tion of the minimum wage in the United Kingdom. In the work by Currie andFallick (1996) for the United States, the affected group of workers was foundto have a lower probability of remaining employed than low-wage workers who

“zwu004060380” — 2006/7/6 — page 1007 — #20


Table 17. Probability of remaining with the same employer from 1984 to 1986, from 1986to 1988, and from 1989 to 1991, with alternative control groups, pooled data (logit model).



Sample: Wage less than 1.25 of the minimum wageYear*treatment 1 0.020 0.011 −0.074 0.021 0.048 0.017Year*treatment 2 0.100 0.023Year*treatment 3 0.071 0.023N 1180946 1180946 1180946

Sample: Comparison group aged less than 25

Year*treatment 1 0.023 0.011 −0.076 0.020 0.047 0.016Year*treatment 2 0.099 0.022Year*treatment 3 0.054 0.027N 1055402 1055402 1055402

Sample: Comparison group aged between 25 and 34

year*treatment 1 0.060 0.011 −0.030 0.020 0.045 0.012year*treatment 2 0.095 0.022year*treatment 3 0.090 0.027N 1509450 1509450 1509450

Source: Computations based on Portugal, MTSS, Quadros de Pessoal (1984–1991).Note: The regression controls are the same as those employed in Table 16.

were not affected by the rise in the minimum wage. Zavodny (2000) found that theprobability of remaining employed in the United States following a rise in the min-imum wage was lower for low-wage teens than for high-wage teens. Abowd et al.(2000) also found that job-loss effects are more pronounced for workers directlyaffected by the rise in the minimum wage than for those initially marginally abovethe new minimum wage.

Neumark and Wascher (1995a, 1995b) explicitly considered supply-sideeffects. They concluded that, following a rise in the minimum wage, high-skillteenagers may leave school to start work and thus replace low-skilled teens whosemarginal revenue falls below the new minimum wage. According to Neumark andWascher, the weak youth disemployment effects detected by some studies couldsimply result from the fact that teenagers are encouraged to leave school, driv-ing out of work less-skilled teens previously employed. Hence, there would bestrong disemployment effects on teenagers already working, but they would notbe captured by the analysis simply because there would be rising demand (andsupply) for more-skilled youngsters. Explicitly modeling the retention rates, wedo not find support for that hypothesis because the retention rate is higher forteens earning sub-minimum wages than for the unaffected groups of workers.Moreover, the fact that, following the rise in the youth minimum wage, there isa decrease in the opportunities for job accessions by teenagers relative to olderworkers (as shown in Section 5) reduces the extent to which the minimum wagemight be pulling teenagers out of school.

“zwu004060380” — 2006/7/6 — page 1008 — #21


7. Discussing the Results

Could our results be driven by the business cycle, with economic expansionexplaining the rise in teenage employment? We believe that this was not the case.One reason is that the periods pre- and post-rise in the minimum wage did notdiffer in terms of economic growth, as the economy was already growing at afast pace from 1985 to 1986, before the rise in the minimum wage. Second, theunemployment rate was around 9% in 1986, so there was a large pool of olderworkers seeking jobs. Third, our methodology partially controls for business cycleeffects that might have influenced the group of workers affected by the change inthe legislation and similar workers that were included in the comparison group.In fact, what we are explaining is the share of teenagers in the flow of hiringsor dismissals. The total flow is certainly influenced by the business cycle, butthere are no strong reasons to believe that the share of teenagers in that flowshould be. Even if it were, the use of hiring and separation rates at the firmlevel as explanatory variables would, in our view, attenuate the business cyclecontamination of the results. Fourth, and most important, if business cycle effectswere driving the rise in youth employment, then job accessions by teenagerswould have increased relative to older workers. On the contrary, we find that itdecreased, and youngsters’ rising employment was driven mainly by a decline injob separations.

Burdett and Mortensen (1998) have developed an equilibrium wage disper-sion search model that nicely matches the patterns observed during this period inPortugal. It assumes that workers can search on the job or while unemployedand that they are heterogeneous with respect to the utility while unemployed,which determines their reservation wage. Firm-level worker flows (crucial in ourempirical analysis) are derived, and in particular the model predicts that the vol-untary quit rate decreases with the wage offered: Better-paying employers facelower quit rates, a theoretical result of great importance in explaining the detectedempirical patterns.

Addressing specifically the impact of the minimum wage, the model estab-lishes that inefficient unemployment is generated when the wage offered is belowthe worker’s reservation wage, even though his productivity may be larger. Sucha situation would result from the existence of frictions in the labor market, asso-ciated with employers’ monopsony power. Raising the minimum wage wouldshift the distribution of wage offers and reduce inefficient unemployment. Themodel therefore explicitly predicts that increases in the minimum wage can reduceunemployment and raise employment.

Another line of reasoning would consider adjustment costs. New firms, whichare unconstrained in their recruitment policy, would react to the change in factorprices and so recruit a lower share of teenagers. In contrast, firms that werealready in business were partially constrained by employment firing costs and

“zwu004060380” — 2006/7/6 — page 1009 — #22


by sunk costs incurred in hiring and training of teenagers, so it may not be aprofit-maximizing option to dismiss a teenager who has been with the companyfor one or two years in order to recruit an older worker at the same wage. Thereis thus an “inaction” band in which the firm decides not to react to a change inthe factor prices. As a result, the substitution effect could have been partially heldback. A scale effect could nevertheless operate, and some firms might reduce theirscale of operation, which would reduce job accessions by teenagers.

8. Conclusion

This study goes beyond the explanation of net employment changes by workertypes at the firm level following a rise in the minimum wage. It studies grossworker flows—both accessions and separations—in continuing firms, as well asin new firms and those going out of business, relying on a matched employer–employee panel data set. Decomposition of the changes in employment by itssources can help reconcile some of the evidence that has previously been presentedin the literature as contradictory, helping to disentangle the minimum wage puzzle.

Our results show that the share of teenagers among newly hired workers,both in continuing firms and in new firms, decreased following the increase in theyouth minimum wage. These flows would thus point toward the reduction of therelative demand for teenagers. On the other hand, the share of teenagers in jobseparations in continuing firms decreased sharply following the rise in their min-imum wage. From a worker perspective, we find that teenagers subject to a highwage increase resulting from the change in the minimum wage are more prone tokeep their job than comparable groups of workers. This result points to the rele-vance of supply-side factors, as job attachment for low-wage youngsters may risefollowing an increase in their minimum wage, reducing the high job turnover thatis characteristic of low-wage workers. In synthesis, the main short-term impactof the 1987 minimum wage change in Portugal was the reduction of separationsfrom the employer, which compensated for the reduction of job accessions.

Having said that, a number of shortcomings should be mentioned. First, wehave looked solely at short-term effects; second, our quasi-experiment is difficultto compare with other studies in that it essentially corresponds to the introductionof minimum wages at the margin; third, we could not distinguish between quitsand layoffs; fourth, the policy change occurred in a labor market with one of themost stringent employment protection legislations among OECD countries.

The overall employment impact of a rise in the minimum wage in an economywill be dictated by the relative magnitude of the different flows that have beenanalyzed. In the Portuguese economy during the period under analysis, decliningjob separations have had a decisive impact raising the share of employment heldby teenagers. This impact was amplified during the expansionary period under

“zwu004060380” — 2006/7/6 — page 1010 — #23


analysis, generating an increase in the level of youth employment. Thus, in ourview it is possible that, under different economic environments, minimum wagepolicies may lead to distinct employment outcomes depending on the relativeweight of worker flows and, importantly, on the balance between the reduction inquits and the increase in layoffs.

Appendix A: Data Set and Concepts Used

Data Set. Quadros de Pessoal (QP) is a longitudinal data set matching firmsand workers in the Portuguese economy. The data are gathered every year by theMinistry of Employment and Social Security, based on an inquiry to which everyestablishment with wage-earners must, by law, respond. Reported data cover allpersonnel working for the establishment in a reference week, in March each year.Personnel on short leave (such as sickness, maternity, strike, or holidays) areincluded, whereas personnel on long-term leave (such as military service) are notreported. Civil servants and domestic service are not covered, and the coverage ofagriculture is low given its low share of wage earners. Reported data includethe firm’s location, industry, employment, sales, ownership, and legal setting; theworker’s gender, age, skill, occupation, schooling, admission date, earnings, andduration of work; and the mechanism of wage bargaining.

The mandatory nature of the survey leads to an extremely high response rate,and in fact the population of private-sector firms with wage earners in manufac-turing and services is covered. In light of the nature of the data set, which coversnot just every company with wage earners but also all of its workers, problemscommonly faced by panel data sets—such as under- or oversampling of certaingroups and panel attrition—are reduced.20 Also, employer-reported wage infor-mation is known to be subject to less measurement error than worker-reporteddata. Moreover, the legal requirement that the reported data be made available toevery worker in a public space in the establishment contributes to its reliability,reducing measurement errors.

Each firm entering the database is assigned a unique identifying number,enabling it to be followed over time. The ministry implements several checksto ensure that a firm that has already reported to the database is not assigned adifferent identification number. The worker code results from a transformation ofthe social security number, and it was considered to be eligible for matching acrossyears if it contained the number of digits declared valid by the Social Security

20. A detailed analysis of the representativeness of this data set has been performed by Cardoso(1997), focusing in particular on manufacturing and relying on the Census of Manufacturing, whoseexhaustive nature is stressed by the data producer, the National Statistical Office. Comparison of thecoverage of QP and the census indicated that QP provides a wider coverage of the population ofwage earners than the explicit attempt on the part of the National Statistical Office to exhaustivelycapture every firm in the economy when making interviews for the census.

“zwu004060380” — 2006/7/6 — page 1011 — #24


Office and if it was not repeated within a year. In the period under analysis, 87–92% of the workers had a code that could be matched across years. Moreover,we were able to overcome the lack of information on this code for some workerssimply by using information on date of birth, gender, date of admission into thefirm, and firm identification number.

Wages. The database reports gross monthly earnings, which are split into thefollowing components: base wage, regularly paid subsidies, overtime work, andirregularly paid subsidies. Normal and overtime hours of work are reportedas well.

Given the legal definition of the minimum wage and the concepts used bythe data source, the analysis focuses on the gross base monthly wage for full-time wage earners. This concept corresponds exactly to the legal definition of theminimum wage enforced throughout the period under analysis.

New Firms, Continuing Firms, and Those Gone Out of Business. A firm isconsidered a new one if it had never previously reported to QP. A firm is reportedas having gone out of business the first year it fails to report if it never returnsto the database. A firm is considered a continuing one if it reported both inthe previous and the current periods. The regressions on the share of teenagersrecruited by continuing firms include all continuing firms that recruited workersin the 16–35 age bracket. Similarly, the regressions for dismissals by continuingfirms include all the continuing firms that dismissed workers in that same agebracket.

Appendix B: Changes in the Minimum Wage Legislation

Apart from the changes in the age coverage of the minimum wage as describedin the text, the minimum wage legislation has undergone other changes, whichcan be summarized as follows.21

Industry Coverage. In 1977 it was extended to cover workers in agriculture,though their wage rate was lower; the following year, household service wasalso covered, but at a rate lower than agriculture. In 1991 the agricultural andgeneral minimum wages were harmonized, and in 1998 the same was applied tohousehold service.

Raising the minimum wage for agriculture and for household work to the gen-eral level took place gradually, so the full harmonization was highly predictable.As such, these changes in the legislation do not provide adequate conditions foreconomic analysis.

21. Throughout the period, minimum wage reductions applied to handicapped workers.

“zwu004060380” — 2006/7/6 — page 1012 — #25


Firm Size Coverage. Small firms outside agriculture with fewer than five work-ers were allowed to pay the lowest minimum (that set for agriculture), a possibilityintroduced in 1978 and revoked in 1991. Before 1978, several regimes had existed:in 1974, firms with fewer than five workers were exempted; in 1975 and 1976,those with fewer than 10 workers were also exempted; in 1977, firms with fewerthan 10 workers outside agriculture were allowed to pay the minimum set foragriculture by proving they were in a difficult financial situation.

On request, firms claiming that application of the new minimum wage wouldimpose an unbearable rise in labor costs were allowed to pay the minimum wageset for agriculture. This possibility was introduced in 1978. In 1987, only firmswith fewer than 50 workers were eligible for this reduction, and the firm sizebenchmark was lowered to 30 workers in 1988, and 20 workers in 1989; theprovision was revoked altogether in 1990.

Because the general minimum wage and that for agriculture had been grad-ually harmonized, the abolition of these exemptions did not lead to a remarkablerise in the applicable minimum wage.

References

Abowd, John M., Francis Kramarz, David N. Margolis, and Thomas Philippon (2000). “TheTail of Two Countries: Minimum Wages and Employment in France and the United States.”Working Paper, Cornell University.

Blanchard, Olivier, and Pedro Portugal (2001). “What Hides Behind an Unemployment Rate:Comparing Portuguese and U.S. Labor Markets.” American Economic Review, 91, 187–207.

Brown, Charles, Curtis Gilroy, and Andrew Kohen (1982). “The Effect of the Minimum Wageon Employment and Unemployment.” Journal of Economic Literature, 20, 487–528.

Burdett, Kenneth, and Dale T. Mortensen (1998). “Wage Differentials, Employer Size andUnemployment.” International Economic Review, 39, 257–273.

Burkhauser, Richard V., Kenneth A. Couch, and David C. Wittenburg (2000). “A Reassessmentof the New Economics of the Minimum Wage Literature with Monthly Data from the CurrentPopulation Survey.” Journal of Labor Economics, 18, 653–701.

Cameron, A. Colin, and Pravin K. Trivedi (1998). Regression Analysis of Count Data.Cambridge University Press.

Card, David, and Alan B. Krueger (1994). “Minimum Wages and Employment: A Case Studyof the Fast-food Industry in New Jersey and Pennsylvania.” American Economic Review,84, 772–793.

Card, David, and Alan B. Krueger (1995). Myth and Measurement: The New Economics of theMinimum Wage. Princeton University Press.

Card, David, and Alan B. Krueger (2000). “Minimum Wages and Employment: A Case Studyof the Fast-food Industry in New Jersey and Pennsylvania: Reply.” American EconomicReview, 90, 1397–1420.

Cardoso, Ana Rute (1997). “Earnings Inequality in Portugal: The Relevance and the Dynamicsof Employer Behaviour.” Ph.D. thesis, European University Institute, Florence, Italy.

Currie, Janet, and Bruce C. Fallick (1996). “The Minimum Wage and the Employment ofYouth: Evidence from the NLSY.” Journal of Human Resources, 31, 404–428.

Deere, Donald, Kevin M. Murphy, and Finis Welch (1995). “Employment and the 1990–1991Minimum-wage Hike.” American Economic Review, 85, 232–237.

“zwu004060380” — 2006/7/6 — page 1013 — #26


Dickens, Richard, Stephen Machin, and Alan Manning (1998). “Estimating the Effect ofMinimum Wages on Employment from the Distribution of Wages: A Critical View.” LabourEconomics, 5, 109–134.

Dolado, Juan, Francis Kramarz, Stephen Machin, Alan Manning, David Margolis, and CoenTeulings (1996). “The Economic Impact of Minimum Wages in Europe.” Economic Policy,23, 318–357.

Katz, Lawrence F., and Alan B. Krueger (1992). “The Effect of the Minimum Wage on theFast-food Industry.” Industrial and Labor Relations Review, 46, 6–21.

Kennan, John (1995). “The Elusive Effects of Minimum Wages.” Journal of EconomicLiterature, 33, 1950–1965.

Machin, Stephen, and Alan Manning (1996). “Employment and the Introduction of a MinimumWage in Britain.” Economic Journal, 106, 667–676.

Meyer, Bruce (1995). “Natural and Quasi-experiments in Economics.” Journal of Business andEconomic Statistics, 13, 151–161.

Meyer, Robert H., and David A. Wise (1983a). “Discontinuous Distributions and MissingPersons: The Minimum Wage and Unemployed Youth.” Econometrica, 51, 1677–1698.

Meyer, Robert H., and David A. Wise (1983b). “The Effects of the Minimum Wage on theEmployment and Earnings of Youth.” Journal of Labor Economics, 1, 66–100.

Neumark, David (1999). “The Employment Effects of Recent Minimum Wage Increases:Evidence from a Pre-specified Research Design.” NBER Working Paper No. 7171.

Neumark, David, and William Wascher (1995a). “Minimum Wage Effects on Employment andSchool Enrollment.” Journal of Business and Economic Statistics, 13, 199–206.

Neumark, David, and William Wascher (1995b). “Minimum Wage Effects on School and WorkTransitions of Teenagers.” American Economic Review, 85, 244–249.

Neumark, David, and William Wascher (2000). “Minimum Wages and Employment: A CaseStudy of the Fast-food Industry in New Jersey and Pennsylvania: Comment.” AmericanEconomic Review, 90, 1362–1396.

Pereira, Sonia C. (2000). “The Impact of Minimum Wages on Youth Employment in Portugal.”Research Memorandum No. 4, Research Centre for Economic Policy, Rotterdam.

Pereira, Sonia C. (2003). “The Impact of Minimum Wages on Youth Employment in Portugal.”European Economic Review, 47, 229–244.

Portugal. Instituto Nacional de Estatística (1986–1989). Labor Force Survey.Portugal. Ministério da Segurança Social e do Trabalho (2002). “Salário Mínimo e Emprego

dos Jovens em Portugal.” Working Paper.Portugal. Ministério do Trabalho e da Segurança Social (1984–1991). Quadros de Pessoal.

Data in magnetic medium.Portugal (1979–1988). Diário da República, several issues.Portugal, Pedro, and Paulo Guimarães (1998). “The Influence of Minimum Wages and Local

Labor Market Conditions on the Entry and Exit of Plants.” Working Paper, Bank of Portugal.Stewart, Mark B. (2002). “The Impact of the Introduction of the UK Minimum Wage on the

Employment Probabilities of Low Wage Workers.” Journal of the European EconomicAssociation, 2, 67–97.

Zavodny, Madeline (2000). “The Effect of the Minimum Wage on Employment and Hours.”Labour Economics, 7, 729–750.

disentangling the minimum wage puzzle: an analysis … · 2009. 7. 16. · in this sense, our...

Documents