Public sector pay and private sector wage
premiums: Testing alternative models
of wage determination
Faculty of Economics
University of Cambridge
The paper focuses on a labour market dominated by a public sector where the links between pay and effort are weak, as in many developing countries. This feature, „a culture of shirking‟ is incorporated in an extension of the Shapiro-Stiglitz model of shirking to
explain why firms in the private sector do not bid down wages but offer a premium over the reservation wage of the marginal worker. A unique feature of this paper is that data on the premium over the reservation wage can be exploited to test for the existence of efficiency wages. Panel data on households and firms in Ethiopia are used to test the robustness of the premium, controlling for the heterogeneity of workers, and the dispersion in wage offers and reservation wages of workers relative to the marginal worker. The premium appears to be driven by efficiency wage considerations rather than alternatives such as bargaining models and specific investments in workers.
Keywords: public sector, reservation wages, efficiency wages, panel data JEL classification: J41 J45
Address for correspondence:
Faculty of Economics and Politics E-mail:Pramila.Krishnan@econ.cam.ac.uk
University of Cambridge Telephone: 44-1223-335236
Austin Robinson Building Telefax: 44-1223-335475
Public sector pay and private sector wage premiums: Testing alternative
models of wage determination
The consensus on why wage levels in urban labour markets are often high is that it reflects government wage policy, which in many countries sets urban wages at an artificially high level. In some countries and in some periods, trade unions might play a role - but the blame for “too high” wages is usually firmly assigned to public sector hiring and wage policies combined with urban bias in pricing of basic consumption goods, that inflate wages well above productivity and labour costs.
However, there is little reason to believe that government pay policies by themselves should have any direct effect on wage determination in a competitive private sector, in the absence of attempts to influence private sector pay. Of course, government employment policy might crowd out particular categories of skilled labour and hence affect wage setting for such workers. In fact, as long labour demand in the private sector is determined by the value of marginal product, the effect of government pay policies is to affect the supply of labour to the private sector. There is no reason, however, in these circumstances, to believe that the private sector will pay a premium over the reservation wage of the marginal worker, even if the public sector does do so (see Lindauer, 1991). Therefore, if the private sector is observed paying a premium over the reservation wage in the absence of trade union or government pressure, an alternative explanation for the payment of premiums must be offered. This paper focuses on the case of Ethiopia, which has a dominant public sector offering relatively high levels of pay (Taye, 1998, 1999). The private sector in Ethiopia is still rather small in comparison, but there is little evidence that it is over-regulated or that pressure is applied to keep wages high. Nevertheless, the descriptive statistics indicate that despite rates of unemployment as high as 30% and a period of inflation, real wages in the private sector did not budge and were significantly above reported reservation wages. In the circumstances, the existence of such premiums deserves investigation.
Why might wage premiums be observed in the private sector? In the absence of regulation or union pressure on wages, there are a number of different models consistent with the existence of such premiums. First, in a standard competitive model, wage premiums are likely to be observed if workers differ in their reservation wages or in ability (observable only to the employer), thus creating dispersion in the offered wage. In any empirical analysis, introducing controls for such differences should therefore eliminate evidence of a premium. The relatively higher risk aversion of workers might also induce premiums, if firms keep wages stable despite changing labour market conditions in order to protect
*I would like to thank Stefan Dercon, Stephen Nickell, Margaret Stevens and Francis Teal for their detailed comments. I would also like to thank the Department of Economics at Addis Ababa University and Arne Bigsten of the University of Gothenberg for providing the Urban Household Survey and Taye Mengistae for the data on manufacturing enterprises in Ethiopia.
workers against fluctuations in income.
Alternative models of wage determination that rely on the bargaining power of workers or the threat of union formation would also predict wage premiums. The existence of specific investments and turnover costs might afford rents to continuing in employment and might be the object of a bargain between employer and employee (Malcolmson, 1999). In such circumstances, rents should only be available to those with some experience on the job - first-time workers should not receive them. (This is also likely to be true for search and matching models that assume that it is costly to search and obtain the best employer-employee match since, if matches can only be assessed after a period, first-time workers are unlikely to be rewarded in advance.)
The previous models are concerned mainly with recruitment and retention of workers. Models concerned with the motivation of workers to exert themselves on the job also predict premiums but for different reasons. Models of efficiency wages such as that of Shapiro and Stiglitz (1984) predict that firms are likely to pay a premium over the reservation wage since they are unable to verify effort on the job without costly monitoring. Workers exert effort since if they do not and are found out, will lose that premium. The existence of wage premiums above the reservation wage of the marginal worker also implies that the labour market does not clear and the consequent unemployment acts to discipline workers. Other models of efficiency wages such as Akerlof‟s gift-exchange model (1982)
provide a justification for wage premiums, suggesting that a rise in relative wages results in workers reciprocating with increased effort if wage premiums engender loyalty and gratitude.
While there are many explanations for the existence of wage premiums, explaining persistently high rates of unemployment together with unyieldingly high levels of wages is more difficult. Arguably, explanations based on efficiency wage payments offer a coherent and persuasive account of why this might be so. In particular, they may matter in labour 1, where there is more than a suspicion that with markets dominated by a large public sector
a relatively large, rather inefficient public sector, the link between effort and pay (in wages or in other forms of rewards, such as privileges) is weak, which may affect private sector labour market functioning. Of course, while a large public sector might be unable to monitor the effort of its employees, it does not follow that the private sector must be similarly afflicted - but it is plausible that the problem of shirking might be infectious. In other words, a `culture of shirking? in the public sector made evident by absenteeism, the use of government offices for transacting private affairs and low levels of effort in general might pose a negative externality to private sector employers. In what follows, I lay out a standard market-clearing model of labour allocation, extended to account for a public sector
1The public sector remains the largest urban employer in most African countries even if its importance has diminished over the last decade: African central governments employed about 1.1% of the population in the 1990s compared with over 1.8% in the 1980s. However, less than 20% of the population is urban, so this means a larger share as a percentage of the urban labour force. The fiscal weight of the central government wage bill is one of the highest for Sub-Saharan Africa at 6.7% of GDP compared to 5.4% worldwide. When measured as a multiple of per-capita GDP, the wage is 5.7 times per-capita GDP. (See Stevenson (1992) and Schiavo-Campo et al (1997)).
where wages are set institutionally and a private sector where wage premiums are driven by the inability of employers to observe effort as in the Shapiro-Stiglitz model of shirking. In the absence of such incentives, effort in the private sector is assumed to fall to the low levels observed in the public sector. This is contrasted with a simple model of rent sharing where wages are determined by some bargain between workers and employers.
I use data on a sample of urban households in Ethiopia, surveyed twice in 1994 and 1997. 2 need not necessarily suggest that the The high rates of unemployment observed here
private sector pays a wage premium or that unemployment is largely involuntary. In particular, the hiring policies of the public sector until 1992 might have served to increase „wait‟ unemployment: hiring was largely from the ranks of the unemployed. Over 80% of the unemployed claim to be supported by their family - only 15% admit taking up an occasional job in order to support themselves. This would suggest that they are able to endure some length of time in unemployment, while queuing for work. It also sits comfortably in a world where graduates of tertiary education were guaranteed employment in the public sector and secondary-school leavers given some assurance that work would eventually be found for them. However, this policy was dramatically changed in 1993 when the new government undertook a programme of structural adjustment - and expectations about employment ought to have changed with it.
One would expect the unemployed to enter some form of informal employment (as in other developing countries) in such circumstances. However, recent evidence (both anecdotal and derived from urban surveys and the Census of 1994), suggests that petty self-employment is increasing but the unemployed do not appear to be taking this up: the new entrants into this sector are mainly students and housewives. In a country that is one of the poorest in the world and where the incentives to finding some work (even in relatively better-off households) would appear to be strong, it is odd to see such high rates of urban unemployment. It is also inconsonant with wages in the private sector being maintained at par with public sector wages - one would expect wages in the private sector to be bid down by the unemployed.
In what follows, I allow that unemployment might still be motivated by expectations of employment in the public sector, compounded by the inability (or reluctance) of the unemployed to enter informal or self-employment. Wage premiums in the private sector 3labour market will in part explain high unemployment without ruling out other explanations
2Unemployment rates have remained high over the last decade. The reforms of the economy undertaken since a new government took power in 1992 have had little effect on unemployment. Some of this persistence in unemployment might be attributed to the fact that the private sector (and self-employment) was repressed under the previous government of the Derg and the main source of employment lay in the public sector. It is unlikely that the high rates of unemployment are caused by in-migration for Ethiopia has one of the lowest rates of rural-urban migration in Africa; in part, due to the land reforms undertaken in rural areas and the fact that there are only user rights in land, so that migration would mean abandoning the claim to land.
3I assume that other factor markets clear, or more simply, abstract from problems in labour demand caused by spillovers from other factor markets. This may be a strong assumption: investment in private sector activities appears very slow and sluggish, despite recent measures to improve incentives to invest with structural adjustment. However, given non-convexities in production possibilities, private sector labour market adjustment may be too slow to absorb the increased number of unemployed in particular skill categories, resulting in high levels of involuntary unemployment.
- and in what follows I explore the reasons for the existence of such premiums. It ought to be noted immediately that the test is weighted against finding evidence of a premium in the private sector: if the high urban unemployment is caused mainly by a combination of queuing for public sector work and the inability to enter informal employment, it ought to be a sufficiently stern disciplining device and there should be little reason for the private sector to pay a premium.
The empirical approach taken here differs from other empirical tests in that it focuses primarily on data on households rather than on firms. In particular, using household-level data allows the exploitation of the unique feature of the data on individuals available here: the ability to examine the behaviour of accepted (actual) wages relative to the reported reservation wage. Data on firms has always seemed the more natural testing ground; however, the tests are bedevilled by the problem of distinguishing whether more profitable firms pay their employees high wages as a form of rent-sharing or whether more productive 4. However, the test on workers, who are paid to exert themselves, raise the profits of firms
household-level data is backed up by data using a sample of 220 private and public firms in the manufacturing sector. Firms were surveyed over the same period, 1994 and 1995 and were asked about the manner in which they attempted to encourage effort from workers. There is also evidence from a survey of workers in each firm surveyed. I deploy evidence from both firms and households in order to investigate the determinants of wage determination in the private sector.
The next section provides the theoretical framework used to organise the empirical tests followed by a summary of the data in Section 3. Section 4 sets out the econometric model and the results are in Section 5. Section 6 concludes the paper.
2. Unemployment and job allocation
I describe a model of labour force allocation, taking into account the existence of a public sector that pegs wages at a relatively high level, while the private sector offers market wages. This is modelled in two stages, in order to make explicit the difference in empirical prediction of a standard market clearing model of wage determination and that in which a wage premium is offered. The model is based on that of Boadway et al (1990).
The simple model of allocation describes a world in which offers of employment are made to the unemployed and those engaged in job search while in the private sector. This is extended to account for the possibility that the private sector pays an efficiency wage to encourage workers to exert themselves on the job - and the predictions here are compared with that of a simple rent-sharing model. I assume that public sector workers do not look for other employment. As will be demonstrated in the next section where the aspirations of the unemployed are described, this is consonant with behaviour in the labour market in Ethiopia.
4Manning and Thomas (1997), point out that most of the tests thus far are unpersuasive precisely because it is difficult to discriminate between efficiency wage explanations and alternatives such as rent-sharing or competitive compensating wage differential models. They also argue that there are few direct tests of efficiency wage models and most tests seem to suggest that incentives matter in wage setting but not much more.
To model the problem at hand, I begin by describing a traditional model of job allocation,
workers and a public sector employing L with two sectors: a private sector employing L12
workers, and the remainder unemployed out of a labour force N. The workers are assumed to be identical as are the firms in the private sector. The model is set in continuous time, denoted by t. Instantaneous utility is assumed linear in income, i.e. u(y(t))=y(t). I assume that the worker seeks to maximise the expected present value of instantaneous utility, discounted over an infinite horizon, at a rate r, i.e.
； (1) V = E y(t) exp(-rt) dt！0
Income y can be derived from different sources. Individuals receive an income of b while unemployed. Job offers in each sector i, are denoted by the wage rate, w, i = 1,2; where i
sector 1 is the private sector and sector 2 the public sector. Offers arrive with constant 5probability in each period.
The model describes the situation of the unemployed in an unchanging environment. Job offers arrive at the rate of Π from the private sector and Π from the public sector while 12
those in the private sector receive offers from the public sector at rate Π. Workers also p
leave work for unemployment at an exogenously given rate of q for private sector workers 1
and q for public sector workers. To derive optimal solutions to the decision problem for 2
being in each sector, consider the present values over a short interval of time [0,t]. Let V, U
V, and V denote respectively, the present value of someone currently unemployed, a 12
private sector worker and a public sector worker. Since exp(-rt)，1 - rt and since over this
interval there is a probability qt (qt) that private (public) sector workers would lose their 12
jobs, it follows that:
= t + (1-rt)[t+ (1-t - t)+ t ] (2) qqVwVVVΠΠpp11U1211
= bt + (1-rt)[t + t + (1-t-t)] (3) VVVVΠΠΠΠ1212U12U
= t + (1-rt)[t+ (1-t)] (4) qqVwVV22U222
Note that (4) assumes that workers take the first job offered (as in Boadway et al. (1990)). This follows if the expected lifetime utility V above is larger than that gained from u6accepting jobs only in either the public or private sectors. Note also that (4) assumes that
only worthwhile offers are given to the unemployed (or alternatively, are used in their
5Typically, further assumptions are made on the nature of offer distribution. In particular, it is usually assumed that offers are received according to a Poisson process (Devine and Kiefer, 1996). This assumption ensures that the model has a simple stationary structure. 6The possibility of self-employment could be introduced in a simple fashion by assuming that the unemployed can always take up self-employment - and will do so if the expected lifetime utility V of being self-3
employed is higher than V but this will not affect the analysis here. However, given that entry into self-employment u
might be determined by quite different considerations from that of wage work, and that the data reveal that that there is no movement between self-employment and wage-work, I confine myself to examining the behaviour of those looking for wage work.
as given and taking limits as t ？ 0, equation (2), (3), (4) for V V valuation). Taking VU 1,2
and V are solved as: U
= + ( - ) + ( - ) (5) qrVwVVVVΠp11U1211
(6) = b + ( - )+ ( - )rVVVVVΠΠU11U22U
= + ( - ) (7) qrVwVV22U22
These equations can be interpreted as asset equilibrium equations: the return from being in a particular state („the interest rate times the asset‟) equals the income earned from the state in each t and the expected capital gain or loss. They must be valid for each individual.
Workers are only observed in jobs if it is worth their while and so the participation constraint has to be satisfied. This means that for someone observed working, V1 ？ V U
(private sector) or V ？ V (public sector) has to hold. I treat b and w as given and 2U2
assume that the private sector labour market clears. Consequently, equilibrium in the private sector labour market will imply that for the (marginal) worker V = V. Equilibrium wages 1U
w can then be obtained as follows. First, solving (5), (6) and (7) for V, and then 1U
substituting this in (5), the following expression is obtained.
- ( - b) b ( r + )qwwΠΠ2p222 = + (8) w1r + + r + + qqΠΠ2222
In short, the equilibrium wage in the private sector must be equal to the discounted value of income while unemployed and the discounted expected value of the government‟s wage offer, where the discount takes into account the probability of obtaining wage offers in the government sector. If the government were to pay reservation wages, so would the private sector. The private sector‟s wages are distorted to the extent that the government pegs wages above the level of discounted income available to the unemployed i.e. b. The last
term is the discounted value of government pay, less the excess of government pay over the unemployment support, multiplied by the probability of getting a government job offer while in the private sector. In fact, if public sector wages are pegged at very high levels and accession rates into this sector are high (so that expected returns to being in private employment are high), firms might even charge the unemployed a fee rather than pay a wage.
Efficiency wages in the private sector and a `culture of shirking? in the public sector
I now extend the model by introducing the possibility of workers in the private sector shirking on the job. Private sector workers are now assumed to be able to shirk on the job; the presence of a large public sector in which employees are well paid and in which efficiency is not of prime concern to employers may affect the motivation of workers in the private sector. In short, low levels of effort in the public sector pose a negative externality to private sector employers. Workers might choose to shirk on the job since it is the norm and private sector employers will wish to prevent them from doing so. I assume that monitoring by employers is imperfect so another mechanism will have to be used by employers as well. Workers can choose to shirk or not. Let superscript S (N) denote that a
be the probability of losing the job through worker is shirking (not shirking). Let mp
monitoring of shirking by the firm. To introduce the consequences of expending effort (or not) on utility, utility is assumed to be linear in wages and effort, u(w,e)=w-e. To allow for a `culture of shirking?, effort expended by a worker who shirks in the private sector by workers is assumed to be the same as that of a worker in the public sector and set at zero for simplicity (i.e. their productivity is zero). The „asset equilibrium‟ equations for a worker when shirking or not shirking in the private sector can then be written as:
NNN = ( - e) + ( - ) + ( - ) (9) qrVwVVVVΠp11U1211
SSS = + ( + )( - ) + ( - ) (10) qrVwmVVVVΠp11pU1211
Given that if the worker shirks, his productivity is zero, the firm will try to ensure that non-shirking behaviour is the norm, i.e.
NS ？ (11) VV11
so that the value to the worker of not shirking is not lower than shirking. Using (9) and (10), (11) implies that:
e (r+++)qmΠpp1？ + (r+). - (12) wVVΠΠ1pUp2mp
If the non-shirking condition (12) is satisfied, workers will not shirk on the job. The „asset equilibrium‟ equations for public sector employment and unemployment are
straightforwardly defined by (13), and (14).
= + ( - ) (13) qrVwVV22U22
= b + ( - )+ ( - ) (14) rVVVVVΠΠ12U1U2U
The non-shirking condition (12) has important consequences. In equilibrium, firms will pay wages that just induce workers never to shirk. Assuming that a worker indifferent between shirking and non-shirking chooses not to shirk, (12) can be restated using the „asset
equilibrium‟ conditions for V and V as before, to derive the optimal wage policy of firms. 2U
In the constrained private sector labour market equilibrium, wages will then satisfy:
- ( - b)) (r + + )b (r + )qqewwΠΠΠΠ2p1p2222 = + + [ (r + + + ) + ](15) qwmΠp1p1r + + r + + r + + qqqmΠΠΠ222p222
Comparing this solution with (8), the equilibrium wage condition without the non-shirking condition, it is now clear that the wage offered will be higher than the equivalent wage in the previous models - it exceeds it by the last term in the equation above. The premium paid will be increasing by the extra effort e needed in the private sector, the rate of public sector job offers to workers in the private sector (Π), the exogenous quit rate in the private sector p
q and the rate at which unemployed get private sector job offers (Π) it decreases with 11;
better monitoring technology, which increases the shirking detection rate (m) and with the p
7). rate of job offers for the public sector to the unemployed (Π2
However, while offering clear and testable predictions, the model is difficult to test, for the predictions cannot be distinguished from those of alternative models. To see this, the case of bargaining over wages by workers and employers is presented below.
Wages in private sector set by bargaining between employers and workers
Suppose, instead of the firm setting an efficient wage as above, wages and effort are determined by a bargain between employer and employee. In order to avoid being pinned to a specific model (which is unlikely to be amenable to an empirical test), the basic prediction of any such model, that V is above reservation utility V is used, (see Machin and Manning 1u
(1992)). It is assumed that
= θ (16) VV1u
where θ >1, is assumed to be constant, (though it might well be a function of exogenous variables that determine bargaining power). If so, the following expression for private sector wages in equilibrium can be derived as before:
b(r + ) + qwΠwΠp2222 = A  - (17) w1r + r + qq22
r (r + + ) θ + (θ - 1) + (r + ) (θ - 1)qqqqΠΠpp2212A = (18) r (r + + ) - (r + ) (θ - 1)qqΠΠ2121
where A is larger than 1. Note that if θ=1, (17) reduces to (8).
The contrast between (8) on the one hand and (15) and (17) on the other forms the basis of the test proposed here: if firms pay efficiency wages (or share rents), the distribution of accepted wages will be truncated below, not by the reservation wage alone (as in a standard job allocation model), but by the reservation wage plus a premium. Furthermore, in the case of the shirking model, the premium is higher, the higher the quit rate from the private sector, the higher the rate of accession into the public sector from the private sector, the worse the ability of the firm to monitor its workers and the higher the likelihood of the unemployed getting into the private sector. The bargaining model suggests that the premium is higher, the higher the quit rate from the private sector, the higher the rate of accession into the public sector from the private sector and the higher the likelihood of the unemployed getting into the private sector but the most important factor is the size of θ and whether it is substantially different from 1. Hence, in the absence of specific information about monitoring technologies or the factors driving the bargain, both kinds of models offer similar predictions of a premium over the reservation wage and an empirical test that finds
7The rates, Π and Π, taken as given to the worker thus far, can be related to the unemployment rate and to 12
levels of employment L and L in a steady state equilibrium. In the steady state, flows into unemployment must 12
equal flows into employment in the different sectors. This is turn, allows us to examine the response of the premium to the unemployment rate and it is easy to show that the premium varies inversely with the unemployment rate, rises with increased employment in the private sector and falls with rising employment in the public sector.
evidence of a premium is unlikely to be able to come out strongly in favour of the one over the other.
The challenge is therefore to discover other proximate evidence that might persuasively distinguish these theories in the empirical analysis. The next section discusses the data available to do so and possible ways in which these theories can be distinguished. In particular, a combination of data on firms and households is to be deployed in order to both offer different perspectives on the evidence in favour of the various possibilities and more important, assess the consistency of the evidence.
3. The empirical tests and the data
Using household data on premiums
There are two possible approaches to distinguishing between the various models of wage determination. As the previous section demonstrated, the existence of premiums alone cannot do so - but a detailed examination of the data might do better. The first step is to establish that wage premiums actually exist and are not driven by heterogeneity in the labour market. If wage premiums do not disappear once controls for worker and firm heterogeneity are introduced (by using fixed effects estimation), the next step is to ask what might cause them. The primary distinction is between efficiency wage models and models where insider effects determine wages. If insider effects are because of specific investments in workers, which allow workers to bargain over the rents available thus, a possible test is to check whether new and first-time workers receive the wage premiums. If they do, this is inconsistent with such effects for they are generated to those continuing in work (Malcolmson, 1999). If union effects matter, comparing workers in union and non-union firms should allow the effects to be distinguished (but this is unhelpful if the threat of union formation drives the sharing of rents).
Efficiency wage models ought to generate premiums to all workers but might also generate different premiums by occupational or education group as in the shirking model discussed earlier if unemployment rates differ across groups or types of workers - and ought to be responsive to movement in unemployment rates over time. A difficulty with using data on households is the paucity of controls available for the match between employer and employee. The only data available are controls for firm size, occupation and sector - but given such controls, if wage premiums persist and are determined primarily by worker type, the argument must be on the side of efficiency wages. However, the main prediction of this class of models is that firms offer wages above the going rate because it raises productivity and this is not readily testable using household-level data and so I turn now to a discussion of what might be achieved by examining data on the firm.
Using firm-level data
An alternative approach to the problem considered above is to contrast the predictions of efficiency wage models with models of bargaining using firm-level data. Wadhwani and Wall (1991) provide an example of this approach. It begins by specifying a standard Cobb-