Chapter 2: Workers and Wages

Organization Design

Colin F. Camerer, Caltech

1/30/2007 12:24 AM

Chapter 2: Workers and wages

This chapter is about workers and wages: What determines the amount of money people earn, and the kind of work they do? What incentives get workers do what firms want them to?

The first concept introduced is what determines wages in a labor market. We start with a basic model in which workers are just paid the “marginal revenue product” (MRP) of their labor—how much they can earn for a company. In highly competitive markets, where workers and jobs are interchangeable, competition will drive wages to the point where they equal the marginal revenue that a worker can produce. But several considerations usually create a gap between wages and the MRP of unskilled labor: “human capital”; compensating differentials; discrimination; upward-sloping wage profiles, and wage compression.

The second concept is the idea of an “agency relationship”; a simple idea that revolutionized organizational economics in the 1980’s. An agent is anybody who takes actions on behalf of a principal. The agency problem is the principal’s challenge of getting the agent to do what she (the principal) wants, rather than something else the agent is tempted to do. We use a simple model to illustrate the core of the agency problem and some properties of an optimal contract that minimizes agency costs. We also mention some psychological reasons to question the core assumptions underlying the model, which are meant to be hints about what new and improved models might eventually look like.

A more sophisticated model includes multitasking—the agent performs more than one valuable activity. This is a very important step because it reminds us that when we incentivize one kind of activity, we usually get more of that activity but less of another. There is an old joke about producing nails in Russia. One factory was instructed to produce a large mass of nails—so workers made a small number of huge, heavy nails. Another factory was instructed to produce a large number of nails—so workers made lots of tiny nails.

Once we have a model in place to highlight what’s important, we turn to the empirical question of how well financial incentives work—and when they backfire. We also discuss the “hidden costs of incentives”. Psychologists have been much more hostile to the idea that paying money for performance is always good. Their idea is that financial pay may extinguish or “crowd out” intrinsic motivation, so that sometimes it is best to pay nothing extra and just try to pick workers who will do a good job on their own.

The final concept in this chapter is the process of hiring, promoting and firing workers. Many companies create an internal labor market. Internal labor markets have special properties that insulate the workers from the forces of market competition, and also try to keep the worker tied to the firm for many years. Creating long-term relationships gives workers an incentive to work hard, takes advantage of the special firm-specific skills the worker builds up, and uses what the firm has learned about the worker to make promotion decisions.

I. What determines wages?

The simplest model of wage determination is a bland base case in which all jobs and worker skills are the same. Think of a low-skill job like migrant farm workers picking strawberries: Not too much skill is involved (not that the work is easy!) and there are many strawberry farms. Putting aside unions and collusion among firms, unskilled labor markets are often competitive—which means that if one firm offers more money everybody flocks there (price increases produce excess supply) and if one worker offers to work for much less, every firm wants to hire him (price decreases—lower wages—produce excess demand). Importantly, because workers and jobs are interchangeable (by definition), there is no advantage to forming long-term work relationships.

In this case, workers should be paid the marginal value of what they can contribute to the firm (net of capital costs—like the cost of equipment they use). Their value is called “marginal revenue product” (MRP). If firms pay more than MRP they go out of business. If some firms pay less than MRP, in theory, they will not attract enough workers so they will have to raise wages to the point where wages equal MRP—in theory.

Anybody who doubts the power of this model should visit a rural area or poor country where people are hungry for work and competition is fierce. Workers are often very well-informed about job opportunities and eager to switch when a good deal arises. Firms also know where the poorest, skilled workers are.

For example, there are street corners in parts of Los Angeles where private busses show up every morning at a certain time (e.g., the Home Depot on Sunset Boulevard).[1] The same is true in poor South African townships in Capetown. Drivers arrive in a bus or van and announce a type of work and wage rate. Workers just show up at the corner and get on the bus if they are willing to work. The job interview is “Yes, or no?”. The bus takes them to a field where they pick crops, or to a work site where they do relatively unskilled labor. There are many such busses, and many workers. Because the same bus drivers and many of the same workers show up every day, there is probably some tendency to form long-term attachments—e.g. if two busses pull up at the same time, a worker may go on the one driven by a person who is more familiar to him, and a busdriver might let a familiar worker get on first if many workers are clamoring to board the bus. There is also some reputation from repeated play—a worker who doesn’t get enough done probably gets known by the drivers who hire labor, and is scorned. These personal relationships aside, the workers getting on busses is as close to a competitive market for unskilled labor as you can imagine outside a textbook.

Next we will discuss seven ways in which actual wages and work deviate from this simple model in which workers and firms match up and create a wage which is equal to the MRP. The complications are:

1.  Human capital: Different workers are differentially productive due to knowhow;

2.  Compensating differentials: Wages may go up or down if work is particularly unpleasant or pleasant;

3.  Discrimination: Controlling for human capital, workers of different types might be treated differently due to ethnicity, gender, religion or other observable factors;

4.  Upward sloping wage profiles: When workers have long-term relationships with companies, wages may go up even MRP goes down

5.  Wage compression: Workers who have widely different MRP’s have similar wages (i.e. wages are statistically “compressed”).

6.  Interindustry wage differentials: Controlling for skill, education and other variables, people are paid different amounts for the very same job depending on the industry they are in (e.g. legal secretaries at high-priced law firms earn more than government secretaries).

7.  Internal labor markets: Often markets for entry-level jobs are rather competitive, but once a worker is “inside” a firm they enter a cocoon in which competition is replaced by employment terms which depend on history and relationships. It is difficult for outsiders to enter these markets except through a low-level entry portal. These “internal labor markets” help explain three other common patterns mentioned above: Why wage profiles are upward-sloping, wages are compressed, and there are persistent interindustry wage differentials (industries with internal labor markets will have higher wages than those which are competitive).

A useful tool for exploring determinants of wages is a Mincerian (after the economist Jacob Mincer) “wage equation”. A “wage equation” relates wages for a worker i at time t (left hand side) to various observables you think should be correlated with MRP and hence could influence wages. If we could measure MRP precisely (as can be done in sports studies, say) then we would include MRP. But usually we can’t MRP that accurately. So the goal of the wage equation is to use as many observable variables that are thought to be correlated to MRP as we can to see how wages vary with observable variables like age, education and so on.

Here is an example, which includes age, education, grades, skill, safety and fun of the job, gender and race “dummy” variables[2], and “tenure” on the job (number of years of work at a particular job and firm).

(W1) Wit = a + bAgeit+c*Educationit+dGradesit+eSkillit+fDangerit+ gFunit+hBlacki+kFemalei+m(Job tenure)it+nIndustryi+eit

Studies use statistical regressions to estimate the values of coefficients on age, education, etc., which best-explain the observed wages. The residual term eit is the unexplained or surprising part of the wage—that is, is the worker paid an unusually high amount (eit >0) or low amount (eit <0) given their age, education, and so forth.

Many of the variables in the wage equation above are difficult to measure—such as skill, danger and fun. So in practice we might simply estimate a simpler form of the equation, such as

(W2) Wit = a + bAgeit+c*Educationit+hBlacki+kFemalei+eit

If the true wages are being generated by longer equation (W1), but we include only age, education, and race and gender dummies (as in the incomplete specification W2), then the measured value of the residual e’it will equal the “true” (unobserved) eit, plus the effects of the omitted variables-- dGradesit+eSkillit+fDangerit+gFunit+m(Job tenure)it+nIndustryi.. Remembering that important variables are probably omitted is extremely important in interpreting the coefficients for variables which are included, and labor econometricians spend a lot of time worrying about how to correct for omitted variables.

For example, a common argument against apparent discrimination is that a negative coefficient on the Female variable for example (k<0) is really just picking up the effect of some valuable skill variable which is correlated with Female but omitted. For example, if k=0 and e>0 (Skill increases wages and there is no wage discrimination among Females), but Skill is omitted from the regression, if Skill and Female are negatively correlated then the measured coefficient k’ will be negative. It will look like females are being underpaid, but in reality low-skill workers are paid less and femaleness is a proxy for low-skill. (Keep in mind, I am not saying that the former argument is either true or fair, it is simply illustrates how conclusions are so sensitive to omitted variables.)

1. Human capital

In the 1960’s, Gary Becker advanced the theory that any component of the wage which cannot be explained by observables like age and education must be a reflection of “human capital”— some skill the worker has, or knowhow about their job. Human capital became a useful metaphor for what education and on-the-job training create, which is valuable for firms and hence induces firms to pay workers more.

Think of education as creating “general human capital”. Then wage equations can be used to estimate how valuable education is, treating it as an investment just like a house or shares of stock. Most studies show that the return to college education is quite high, about 12%.[3] This high rate of return raises a puzzle: Why doesn’t everybody finish high school (at least) and go to college?

One possibility is that teenagers are not planning ahead when they decide to quit high school, or do so because of other pressures (a struggling family needs income) or life-events (having a child). Another possibility is that they don’t realize the returns are so high.

A third possibility is that the estimated returns to education are biased upwards by “selection”. This is a subtle point and will crop up again and again so let’s spend some time unpacking it. The dollar value of education is most crudely measured by running regressions and computing the value of the coefficient c on the education variable. This gives a rough estimate of how much a degree is worth in the marketplace. But comparing more- and less-educated people fuses together two distinct causes: The education may have a “treatment” effect, raising wages; or there may be a “selection” effect—more talented workers go to college, and they would have earned a higher wage even if they didn’t go to college.[4]

In general, it is difficult to disentangle selection and treatment effects. The ideal way to separate the two is an experiment—e.g. we take two similar people who got into a college, and prohibit one of them from going. The difference in how much they earn later in life tells us how much the education is worth. The best studies use natural experiments or quasi-experiments. The next best alternative is an “instrumental variable”, which is a variable that is correlated with the decision of whether to go to college but not with the ability to get into college. For example, suppose you compare students who all got into a college, but in one sample an unexpected illness or death in the family forced the accepted student to work rather than go to college. Unless having an unexpected illness influenced the student’s ability to get into the college, the unexpected illness is like a “random assignment” of equally-skilled students to “go to college” and “don’t go to college” categories. Even controlling for sample selection, it appears there is a large return to education from the “treatment effect”.

An important distinction between types of human capital is general and firm-specific human capital.