2.1 What is law?

The notion of a legal rule dates back at least as far as 2000 BC, with the Babylonians following Hammurabi’s code and the Mesopotamians following the code of Urukagina. While some of these ancient edicts imply curious beliefs about causality,¹² it is clear that the purpose of many of these rules is to modify or constrain human behavior. For as long as law has existed, its purpose has been to facilitate the efficient functioning of civil society.

The aforementioned codes even had room for what we would now call labor law. Hammurabi’s code includes, for example:

These are remarkable examples of how legal rules are created to regulate the employment relationship. That minimum wage rules persist to this day suggests that there are robust reasons for the existence of such rules. Recognizing this possibility, the law-and-economics approach seeks to explain the rules as solutions to well-defined market imperfections. Ostrom (2000), for example, has shown that many societies have developed efficient systems of rules and adjudication for regulating the use of common-pool resources, thereby avoiding the tragedy of the commons (Hardin, 1968). Ostrom observes that all successful commons-governance regimes consist of a set of rules that have the following features:

All organizations, including firms and families, have rule systems with these features. The economic analysis of such rule systems typically entails asking what set of rules would achieve an efficient allocation.¹³ In the common-pool-resource problem, for example, one seeks a set of rules ensuring that each person with access does not overuse the resource. Organizational economics tries to determine which systems of rules and compensation in firms or public entities ensure that agents reveal useful information and choose efficient levels of effort.

What distinguishes the law’s rule system from a family’s or organization’s is not the existence of binding rules, but the sources of enforcement and adjudication. When we speak of a legal system, we mean the set of rules and associated penalties that are enforced by the state.¹⁴ As discussed above, however, the jurisdiction associated with a dispute is often poorly defined. Even when the jurisdiction is well-defined, disputes can often be decided by multiple judicial bodies (including private mediation and arbitration). Many countries, such as Italy and the United Kingdom, maintain a separate system of courts for ruling on employment disputes. In the United States, disputes regarding a collective-bargaining agreement may be brought before the National Labor Relations Board, employment disputes before a state court, and Title VII discrimination suits before a federal court.

To illustrate what we mean by a legal system (and employment law in particular), let us consider the proverbial worker-firm relationship. In the standard neoclassical model of employment, the worker agrees to supply L hours of labor for a wage w. This can be viewed as a supply contract where the hours are, say, consulting time. In that case, the relationship would be governed by contract law. The worker must supply L hours, and the employer must pay w L dollars.

Should the worker supply less than L hours, the worker has breached the contract. Suppose that the firm has paid P₀ in advance, a common practice if the worker is, for example, a lawyer. Some sort of binding agreement is needed; otherwise, the worker would simply take the P₀ and try to find employment elsewhere. The question, then, is: What incentives does the worker have not to breach the agreement?

One possibility is the use of an informal enforcement mechanism. This would include firms’ telling each other that the individual has breached, and hence should not be dealt with.¹⁵ Another alternative is to use physical violence against the individual, a common technique in the illegal drug trade and other black markets.¹⁶ While both enforcement systems are still widely used, societies have evolved more legalistic systems of adjudication for the simple reason that parties sometimes fail to perform even if they act in good faith. Enforcement systems that trigger punishment regardless of the reason for breach, such as violence among drug dealers, are simply not always efficient.¹⁷

In contrast, if the contract is viewed as a legally binding agreement, then the breach of contract by the worker gives the firm the right to seek damages in court. If the firm prevails, the court can order the worker to pay damages to the firm; if the worker refuses to pay, the court can still enforce the decision by ordering the seizure of the worker’s assets.

The fact that contract breach leads to the right to file suit—as opposed to an automatic penalty—is a feature that distinguishes legal enforcement from other forms of rule systems, such as rewards within a firm. In the economics literature, it is common to view any agreement between parties that links future rewards to actions as a “contract”. Jensen and Meckling (1976), for example, famously proposed that one should conceptualize a firm as a “nexus of contracts”. If all contracts are enforceable at negligible cost, a reward system that promotes an employee for good performance and an agreement with an outside supplier to pay a bonus for sufficient quality are assumed to be equally enforceable. If employment contracts are enforceable at no cost, subject only to information constraints, then explaining the contract form requires only that one carefully specify the environment and then use principal-agent theory to work out the optimal contract.

The difficulty, as Williamson (1991) observes, is that the law uses forbearance for transactions within a firm. Even if a firm promises a promotion, that does not confer a right upon the worker to sue the firm should the promotion not be offered. All dispute resolutions of this sort occur strictly within the firm, with no appeal to an outside legal authority available. A full discussion of the role of courts in such disputes must wait until the next subsection, but for now the relevant point is that for any contract between two legal persons (in our example, firm and consultant), both parties always have the right to seek damages in court should there be a breach of contract.¹⁸

That the decision to sue is discretionary implies that the same rule may have different effects depending on the characteristics of the parties of the contract. The characteristics of the parties might even be the most important factor in whether a lawsuit is filed. For example, suppose that a company has illegal discriminatory hiring practices. If the market is thick, with plenty of employment alternatives, potential employees may not find it worthwhile to bring suit against the company. In a less friendly employment market, if an individual believes he has been the victim of illegal discrimination and cannot find another job, he may bring suit. This point is illustrative of the fact that we will always find differences between a legal rule on paper and the same rule in practice.¹⁹

A second source of uncertainty is how the court will decide a case. In the case of the breaching consultant, the court has to make two decisions, each of which is prone to statistical error: (1) deciding whether a breach has occurred, and (2) if so, the damages to be paid. For damages, the general rule is that courts order expectations damages— namely, the losses associated with the worker’s breach. An example of a formula that the court might use is:

Here w₁ is the wage for the replacement worker. The worker has to pay the costs associated with finding a replacement worker rather than the value of the work done, due to contract law’s requirement that injured parties make every effort to mitigate losses arising from breach. The mitigation rule is mandatory, meaning that parties cannot contract around it. Employment law has many other mandatory rules; slavery is prohibited, for example, as is discrimination on the basis of race, age, or sex.

Expectation is not the only way to calculate damages from contract breach. In a famous paper on contract damages, Fuller and Perdue (1936) identified restitution and reliance as other possible measures of damages.²⁰ Restitution damages, for one, are intended to put the firm into the same financial position—as if the contract had never been signed. In our consultant example, restitution damages would only require the worker to repay P₀. Matters are less clear if P₀ is paid to the worker so that he can buy passage to the job site and secure accommodation. Suppose upon arriving, the worker is injured and cannot perform his duties; the firm sues for breach. If the contract does not specify what happens in this contingency, a court may be asked to fill in the gaps in the agreement. Some other issues that might not be described in the contract (and which the court must adjudicate) include whether the injured worker may be fired or whether he must take out a loan to repay P₀.

Alternatively, the firm, in the expectation of being able to rely upon performance, may have paid for passage and accommodation. In that case, if the worker does not perform, then she may be asked to compensate the firm for these reliance expenditures.

Finally, the contract could have provided that a non-performing worker pay a fine P₀. In this case, non-performance would not be a breach of contract, and a breach would occur only if the worker failed to pay back P₀. Notice that in this case, the damages would simply be P₀, even under the expectation-damages rule. This example illustrates that for the same exchange relationship there is no unique notion of contract breach. Rather, the contract defines what constitutes breach, and then the courts must decide whether or not to enforce the terms set out in the agreement.²¹

Defining the conditions for breach is not a clear-cut exercise. For example, the contract could have specified liquidated damages in the amount P₀. In this case, breach would occur for non-performance, but then the contract would direct the court to set damages at P₀. This contract seems to be equivalent to the previous one (indeed, most economic theories of contract would see them as equivalent), but there is an important distinction. In the previous case, if the worker pays P₀, no breach has occurred, and the firm has no right to bring suit against the worker. In the liquidated-damages case, even if the worker offers to pay P₀, the firm still has a right to sue the worker because technically a breach has still occurred. Admittedly, the firm would face an uphill battle in court if the worker had offered to pay—and normally would have no reason not to accept the offer as part of a settlement agreement—but the right exists nonetheless. The firm might believe, for example, that the worker abused the liquidated-damages clause, accepting the consulting job only as a contingency in case another opportunity did not work out. If the firm’s belief is true, the worker has arguably violated the requirement of good faith and fair dealing—another mandatory rule in the common law of contracts. Arguing that the liquidated-damages clause no longer applies, the firm might ask for expectation damages larger than P₀.

Conversely, assume that the firm sets liquidated damages at three times P₀. Breach occurs, and the worker declares these damages to be unconscionably high. The worker may have a valid claim under contract law’s prohibition against penalty clauses. This doctrine—another mandatory rule—provides that liquidated damages far exceeding the losses to the injured party will not be enforced by courts.

The prohibition on penalty clauses, along with the other examples of mandatory rules discussed above, illustrate that the law allows for a significant degree of judicial intervention into private contracts. Among other things, a court can fill in missing terms and refuse to enforce unreasonable terms. The decision whether to intervene is often at the court’s discretion, but courts usually turn to the Uniform Commercial Code and other statutes, as well as previous court decisions, to justify these interventions. Thus the legal system is an adjudication process that modifies contracts in the face of a breach as a function of past experience and practice.

The extent to which courts should intervene into freely entered agreements has proven to be controversial. Early scholars, such as MacNeil (1974), proposed that courts rely on many sources of information, including industry custom and the history of the relationship between the plaintiff and the defendant, when making a decision. If parties have a longstanding relationship, Macneil argued, contract terms should be enforced within the context of the relationship. Macneil and many others believed that this sort of context-sensitive adjudication could help repair the parties’ relationship and facilitate the continuation of mutually beneficial exchange. Most economic theories of contract, in contrast, work from the assumption that parties have well-defined interests and can draft agreements efficiently, implying that contracts should be enforced as written. This presumption has led to a law-and-economics scholarship that mostly argues for the curtailment of judicial discretion, and for a more systematic dependence upon basic economic reasoning when ruling on a case (see Goetz and Scott (1980) and more recently Schwartz and Scott (2003)).

Regardless of one’s theoretical commitments, it remains the case that the law does not simply enforce a set of well-defined rules. The law does include a set of rules, but along with a system of adjudication that results in a context-sensitive application of these rules to individual cases. This context sensitivity includes, among other things, consideration for the idiosyncratic features of the parties. The basic principles of contract law apply to all agreements between two parties, but more specialized bodies of law have evolved to regulate specific classes of contracts. The insurance industry is regulated by a specialized area of contract rules (see for example Baker (2003)) as is employment law. It is to this latter body of law that we now turn.

2.2 Employment law

Employment law evolved from contract law and master-servant law to deal with the unique problems characterizing the modern employment relationship. The first task is to determine the difference between (1) a firm’s relationship with an outside contractor selling services, and (2) its relationship with an employee. The difference not only affects the area of law that regulates the relationship, but it also affects the relevant tax law. In the United States, the Internal Revenue Service will find that an employment relationships exists when “the person for whom services are performed has the right to control and direct the individual who performs the services, not only as to the result to be accomplished by the work but also as to the details and means by which that result is accomplished”.²²

As this tax regulation exemplifies, the obligation of the employee is to follow his employer’s directions, not to produce a specific service with particular characteristics. Simon (1951)’s model nicely captures this distinction between sales and employment. In a sales contract, says Simon, the seller agrees to supply a particular good or service x from the set of all possible goods and services X, and in exchange the buyer agrees to pay a sum P. An employment relationship, in contrast, is characterized by a subset of all possible goods and services, A ⊂ X, that represents the set of duties that the employer might ask the employee to carry out. A might include the service x defined in the aforementioned sales contract, but that single task would normally be just one component in a broad complex of obligations defining an employment relationship. In exchange for a promise to carry out these duties, the employer agrees to pay a wage w.

Simon’s simple model highlights an essential feature of the employment relationship, namely, the admissible scope of a person’s job as represented by the set A. The admissible set of tasks S ⊂ X −that is, the set of acts that an employer is allowed by law to command—has been subject to a plethora of regulation and litigation. For example, is it conscionable for a firm to require 50 hours a week? Can a manager ask her assistant to commit crimes?

An employment relationship often begins with little formal agreement about the tasks the employee will be asked to carry out. The longer the potential duration of the employment, the more incomplete the initial employment agreement. Given the informal nature of such agreements, when disputes do arise the courts will have little to rely upon when constructing the obligations of each party. With poorly defined obligations, determining whether a breach occurred presents a difficult task, as does choosing an appropriate remedy.

The combination of extreme contract incompleteness and daunting litigation costs have convinced many legal scholars that the appropriate default rule is at-will employment. The courts have converged to this default rule partly because they now view employment law as an extension of contract law. That view diverges significantly from the early case law on employment disputes, which was mostly governed by “master-servant law”.²³ That old body of law consisted of a set of legal default rules developed in England and the United States to deal with cases involving domestic servants. In the master-servant relationship, the customary period of employment lasted one year; courts held that neither party should sever the relationship before then.

In a widely cited work, Wood (1877) argued for replacing this law with the rule of at-will employment, where both parties can sever the relationship whenever they wish and face no liability beyond the requirement that the employer pay her employee the agreed-upon wage for work already completed. Wood’s argument was a pragmatic one, based on the bad experiences of many employers and employees with the inflexibility of master-servant law. As detailed in Feinman (1978), the new rule was quickly adopted by the New York courts and remains the default rule today. In California, the legislature adopted what is now Section 2922 of the California Labor code, which provides that “employment, having no specified term, may be terminated at the will of either party on notice to the other”.

The at-will-employment rule figures prominently in most economic models of the labor market. As these models have it, workers and firms enter into relationships that are preferred to the alternatives in the marketplace. Should a firm mistreat a worker, or have high standards for performance or number of hours worked, the firm will have to pay relatively high wages or else the worker will leave. Similarly, if a worker demands a higher wage or better working conditions, the firm is free to search for another worker who will abide by the current arrangements. In equilibrium, all firms and workers are satisfied with their lot relative to the alternatives.

The hypothesis of a perfectly competitive market can explain many broad features of wages and employment over time, but it cannot explain the emergence of the at-will-employment rule. This is an example of the model’s inability to explain the emergence of laws that seem to constitute reasonable responses to real economic issues. That failure indicates flaws in popular economic models—but not in economic reasoning generally. Consider the case of child labor. As societies have become more wealthy, they have gradually imposed stricter legal constraints on the minimum age and maximum hours for minors in the workplace. By the perfectly-competitive-market hypothesis, these restrictions would be unjustified because only those families for which child labor is efficient would put their children to work rather than in school. On the other hand, a more realistic economic inquiry recognizes the market imperfection imposed by liquidity constraints: children (and parents) cannot borrow against future income arising from education, so many families send their children to work for a short-term gain in income rather than invest in a long-term gain from education. Investing in education results in superior overall welfare, so the choice to put children to work is inefficient. Laws that regulate child labor, like the Fair Labor Standards Act, are justified because, by increasing the cost of child labor, they motivate families to substitute education for labor. By increasing investments in education, these laws increase social welfare.

The analysis in the previous paragraph indicates the potential insight to be gained from an evolutionary perspective when investigating the law and economics of employment law. Applying this perspective to our law’s historical origins, we observe that employment law adapts to the changing macroeconomic environment. One of the earliest labor statutes on record, the Ordinance of Labourers, addressed the problems of unharvested crops, rising wages, and poaching of workers faced by English landowners at the height of the Black Death. Similarly, in 1630, the Massachusetts General Court placed a wage cap of 2 shillings a day on skilled craftsmen, who were at that time taking advantage of limited supply. More recently, the Fair Labor Standards Act became law in the midst of the Great Depression, when workers lacked market power to ensure good working conditions. Mandatory overtime pay, meanwhile, incentivized firms to hire more workers at fewer hours each, thereby serving as an income- and risk-sharing function.

These legal adaptations to changes in the labor-market environment can all be conceived as forms of insurance, whether against the waste of unharvested crops, gouging by craftsmen, or unemployment. In the next subsection, we discuss recent work based on the hypothesis that workers are risk-averse, which might help explain some of the features of these laws. Certainly, both minimum-wage laws and unemployment insurance can be viewed as forms of imperfect insurance.²⁴

Once the issue of risk is put aside, the law-and-economics movement has tended to take the view that employment at will is the optimal default rule (see, e.g., Epstein (1984)). Within economics, a tradition including Friedman (1962) and Alchian and Demsetz (1972) has viewed labor services from the perspective of the buyer-seller contract—with no remedies for contract breach. Specifically, Friedman argued that a competitive market with free entry and exit is the most efficient market form, even when contracts cannot clearly specify quality. In his vision of the world, workers and firms trade freely within the context of the sales contract (quality x in exchange for price p); should performance be inadequate, the worker would gain a poor reputation and thereafter be excluded from the market.

There are two difficulties with this argument. The first, as MacLeod (2007a) discusses, is that the literature on relational contracts shows that reputational concerns are neither necessary nor sufficient to ensure efficient exchange. Second, the common law has developed many doctrines that limit the freedom of contract in the context of the simple buyer-seller model. Posner (2003) suggests that these developments tend to enhance contract performance. Chakravarty and MacLeod (2009) present evidence that this is indeed the case for a large class of contracts that are common in the construction industry.

As the Simon model highlights, the employment relationship is different because a performance obligation is created ex post. If the worker accepts a contract with scope A ⊂ X, then the performance obligation is created when the employer asks the employee to carry out x* ∈ A. If the relationship were governed by standard contract law, then if the employee chooses x^b ≠ x* the employer-cum-buyer could sue for damages B(x*) – B(x^b), where B(x) is the benefit to the employer of action x. Under employment at will the general rule would be that the employer has no right to sue, but she can freely dismiss the employee, even if performance is satisfactory.

Correspondingly, the employee has the right to leave whenever he wishes. For example, if the employer asks the employee to carry out an action outside the scope of his duties, then under at-will employment the employee has the right to refuse to carry out the task and find another job. In contrast, in the case of a construction contract, if the buyer were to ask for a modification to the building plan, then under US law the contractor would have an obligation to carry out the modification, but he would also have the right to sue for the additional cost of the change if the buyer/builder does not adequately compensate him for the changes.

The defining feature of at-will employment is that in each period parties are free to renegotiate the contract, with the outside options defined by each parties’ market opportunities. Consistent with the Coase theorem, we should expect at-will employment to give rise to arrangements that are ex post efficient. This observation has led some legal scholars (e.g., Epstein (1984)) to suggest that exceptions to employment at will are inefficient. Yet today in the United States, as in most other jurisdictions worldwide, the law of wrongful discharge is alive and well. Indeed, there are clear exceptions to the rule of employment at will. In the next section, we discuss three of the most important exceptions figuring in recent empirical work on employment law.

2.3 Exceptions to employment at will

This section discusses three exceptions to employment at will that have found broad support in US courts. These exceptions are judge-made laws, created in response to difficult cases; hence, they are good examples of how the common law evolves in response to the disputes that arise in practice. The three exceptions we consider are (1) the public policy exception, protecting from employer retaliation those workers that act in a way consistent with accepted state policy, (2) the implied contract exception, protecting workers who can show that the implicit contract with the employer entails just-cause dismissal, and (3) the good-faith exception, requiring employers and employers to behave in ways consistent with fair dealing.

US courts rarely order specific performance—that is, the losing employer typically still has the right to discharge the employee—and hence the issue is usually one of damages: How much should she have to pay for this right? In other jurisdictions, however, reinstatement is sometimes considered an acceptable remedy. One of the few US cases in which specific performance was granted in an employment dispute was Silva v. University of New Hampshire.²⁵

The question of damage awards is not straightforward, but economics can assist in organizing our thinking. If markets are perfectly competitive—and a worker’s compensation is equal to his best market alternative—dismissal does not entail any harm. However, as Mincer (1962) has shown, the second assumption can break down when the worker’s training costs were significant. If the worker paid for some of the training costs, he will be compensated for them through increased future compensation, and therefore his income may be in excess of the best market alternatives. More often, dismissal entails a costly job search and possibly relocation. When an employee has been wrongfully discharged, the court will award damages that reflect these costs.

An additional complication is whether the wrongful-discharge action comes under tort or contract law. The first exception to the at-will employment rule, a claim for wrongful discharge as violation of public policy, is considered a tort claim.²⁶ A tort claim, put briefly, is distinguished from contract disputes by there being no requirement for a prior contractual relationship. Standard examples include traffic accidents and medical malpractice. The practical implication of this distinction is that tort law allows for the recovery of both consequential damages and punitive damages, which may far exceed the direct economic harm suffered by the discharged employee. In contract law, consequential damages and punitive damages are in general not recoverable.

2.4 Discussion

The economic model of contract tends to view legal rules as constraints upon individual decision-making, either in terms of increasing transaction costs or imposing constraints upon the wages, hours, and other conditions of employment. In practice, the law is a complex adjudication system that is difficult to describe with an elegant model. Some of the distinctive features of a legal system that are not captured in the economic model of the employment contract include:

The fact that courts may overrule contract terms is well-recognized in the legal literature. One of the central issues of this literature is the question of whether or not there is anything we can reasonably call “the law” that allows one to consistently anticipate how courts will rule on a given dispute. There is a related debate regarding how best to think about judicial behavior.³⁶

Within economics, there is a small but growing literature that explores the role of the law in ensuring performance. Johnson et al. (2002) find that even if enforcement is imperfect, the existence of courts can help entrepreneurs enter into new supply contracts. Djankov et al. (2003) construct a database consisting of how costly it is to evict a tenant and collect on a bounced check across a large sample of countries, finding that the cost of collection in civil-law countries is significantly higher than in common-law countries. This result is consistent with subsequent work reported in Djankov et al. (2008) for the problem of debt collection. See La Porta et al. (2008) for a more comprehensive discussion of this literature.

For the most part, this work focuses on the costs of the legal system and assumes that variations in these costs across jurisdictions affect economic performance. Botero et al. (2004), and more recently Djankov and Ramalho (2009), explore the extent to which employment law and regulation affect labor market performance. This work uses crosscountry variation in measures of employment-law flexibility to identify the effects of the law upon labor-market performance. These papers suggest that the historical origin of the country’s legal regime—whether common-law or civil-law—is often the decisive factor in the evolution of the country’s employment rules. However, these papers do not explain this observation. One possible interpretation, perhaps in need of further research, is that laws, like organisms, are adapted not just to the environment but to other laws. The various laws in a legal system—of which employment law is a small part— persist at a steady-state equilibrium unless an overwhelming shock—whether political or economic—suffices to move enough laws to another equilibrium to pull the rest of the legal regime along with them. The rarity of such events—Russia’s transition to capitalism is a plausible example—might explain the durable influence of common-law and civil-law institutions on employment laws.

Regardless of this latter conjecture, what is clear from the analysis above is that the law is adaptive, yet existing work does not adequately explain why there is variation in the law. The plausible view taken here is that the law evolves in response to cases brought before the courts. New types of disputes breed new types of law. To understand why these cases arise, we need to understand what exactly is the role of the law in an employment contract. In the next section, we review the literature on the economics of the employment relationship, placing legal rules in the context of the full relationship.

3.1 Why do we need models?

The purpose of this section is to review the role that economic theory plays in understanding the significance of the law. As discussed in Section 2, even though economic concerns shape the development of the law, economic analysis as developed by the economics profession has played a relatively minor role in explicitly guiding court decisions.³⁸ While there may be no explicit accounting of economic effects, it is safe to say that employment rules established by courts have measurable effects on the economy. Accordingly, the goal of the theory discussed here is to structure empirical tests of the impact of employment policy on economic performance.

The recent empirical work in labor economics has been greatly influenced by the potential-outcomes framework, as beautifully exposited by Holland (1986).³⁹ I shall briefly review this approach when using economic theory to understand the effects of the law on labor-market performance. First, the framework provides guidance on how to best organize and represent data. Second, it provides guidance on how to estimate a causal effect. Holland emphasizes that it is impossible to establish a causal relationship without some additional hypotheses that themselves can rarely be tested; they must rely on a model of how the world works.

Formally, the model proceeds by supposing that we have a universe of units to be treated, denoted by u ∈ U. For the purposes of our discussion, let U denote all potential workers in the economy. In addition, we might also be interested in outcomes at a state or country level. In that case we let u^s ⊂ U be the subset of individuals living in state s ∈ S, where s could denote a US state among all states or one country among all countries.

This chapter’s main concern is labor-market performance, so we restrict our attention to the question of how policy might affect wages and employment. For individual u, let y^E ∈ Y^E = {0, 1} be employment status (with 1 meaning employed), and let her wage per period be given by y^w ∈ Y^W = [0, ∞). Suppose that these outcomes will be observed in the next period (t). Employment and wages are likely to be affected by employment policies in the next period, denoted by l_t.

Rubin’s model was developed in the context of a medical treatment where one asks if a particular drug has an effect. This question is typically answered by randomly dividing a group of individuals into a treatment and a control group. The causal effect of the drug is measured by comparing the outcomes in the two groups. The problem is that this procedure does not identify the effect of the treatment on a particular individual. In some illnesses, individuals become well in the absence of treatment. For others, the illness may be fatal regardless of the treatment. By chance, it is possible that all the former individuals (the false positives) would be assigned to the treatment group, while the latter individuals (the false negatives) would be assigned to the control group. In that case, the experiment would show that the drug had an effect, even though it did not.

The first issue is how to define a causal effect. In the context of our simple model, let y(u, l⁰, t) be the outcome under the status-quo law in the next period, and let y(u, l¹, t) be the outcome under the new rule, say an increase in the minimum wage. Let Δ = l¹ – l⁰ denote the policy change. Following Holland (1986), we say that the policy change Δ at date t causes the effect:

D(u, Δ, t) = y(u, l¹, t) – y(u, l⁰, t).

This definition is concrete: It is the difference in potential outcomes. In order to measure this “effect”, we would have to observe the same outcome for two different policies at the same time, something that is clearly impossible without time travel. Holland (1986) calls the impossibility of observing a causal effect the Fundamental Problem of Causal Inference. His analysis emphasizes the fact that measuring the causal impact of a treatment entails additional hypotheses.

Most solutions to the problem of causal inference rely upon versions of unit homogeneity or time homogeneity. By unit homogeneity we mean that there is a set of units U′ ⊂ U with the feature that the effect of the change Δ is the same for all u ∈ U, in which case the effect can be estimated by policy change to unit u¹ ∈ U′ but not to unit u⁰ ∈ U′, in which case for u ∈ U′ we have:

D(u, Δ, t) = y(u¹, l¹, t) – y(u⁰, l⁰, t).

By time homogeneity, we mean that the effect of the treatment in different periods is the same. Hence, if we can estimate the effect of a treatment on a unit u by comparing the effect over time:

D(u, Δ, t) = y(u, l¹, t + 1) – y(u, l⁰, t).

The challenge then becomes finding the homogeneous group. Regression discontinuity is an example of a recent popular technique that provides a way to create homogeneous groups that allow for the estimate of the effect of a treatment.⁴⁰ For example, DiNardo and Lee (2004) argue that firms in closely contested unionization drives are almost identical in most respects. Because the outcomes of union certification votes are very close, one can assume that for these firms union status is randomly assigned. Consequently, we can compare the change in firm value for those firms that were unionized to the change for those that were not, and thereby procure a robust measure of the effects of unionization on a firm’s productivity.

Lee and McCrary (2005) provide an example of time homogeneity. Specifically, they look at the effect on behavior of sanctions against crime. Their study exploits the fact that when a person turns 18, they suddenly become eligible to be tried in adult courts, where they will face more severe sanctions than a juvenile court would impose. On the supposition that a person’s characteristics just before and after they turn 18 are the same, observed changes in crime-related behavior can be ascribed to changes in criminal sanctions.

Notice that all the work is being done by the assumption of continuity over time with the same union, or across units with very similar characteristics. The great benefit of this approach is that, beyond the continuity assumption (which is a strong assumption), this approach is relatively model-free. The problem is that while it may provide a credible measure of the effect of a policy change, the approach says little if one moves away from the point at which the policy change or treatment is applied.

A formal model in this framework has two distinct goals. The first is that it may provide a concise representation of a set of facts about the world. It describes the set of measured characteristics that one needs to know in order to capture the effect of a treatment. For unit u ∈ U, let X (u, t) ∈ Θ be a set of characteristics. In practice, one may not be able to measure all dimensions of Θ, but let us suppose for the moment that we can. Suppose u is a worker and we are interested in explaining worker wages. Then, we would say that a model that specifies a wage f(X, l, t) for a worker with characteristics X is an unbiased representation of the data at date t if for all u ∈ U,

ϕ_ut = y(u, l_u, t) – f(X(u, t), l_u, t)

is an iid set of random variables with zero mean.

If our model is linear, we can let β(t, l) = ∂f/∂X, in which case we can write our model in the familiar regression form:

y_ut = β(t, l_ut) X_ut + ϕ_ut.

If our model is unbiased, then this is a well-specified model that can be estimated by ordinary least squares. However, even if the model is well-specified, as Holland (1986) emphasizes, the coefficients of the model cannot be assumed to represent a causal relationship. For example, one of the parameters might be the gender of a worker, say 0 is male and 1 is female. If y_ut is the wage, and the coefficient on gender is negative, we cannot say that gender causes a wage drop. This is because gender is not a treatment or something that one normally assumes can be varied within a person.

We can use the coefficient on gender to test various theories. For example, human capital theory predicts that a person’s wage is a function only of their productive characteristics, such as schooling, ability, and experience. One reason women might be paid less is that they spend more time out of the labor force in child rearing. This reasoning implies that once the full set of characteristics reflecting productivity is included in X, then the coefficient on gender should be zero. If it is not, then we can say there is discrimination in the labor force.

A good theory specifies the set of parameters X that provide all the information necessary to describe wages while preserving time independence:

y_ut = β(l_ut) X_ut + ϕ_ut.

Any variation in wages that occurs over time is explained via either changes in the parameters X_ut or by changes in the environment 1_ut. In practice, the econometrician may not have access to all the relevant information X_ut, which leads to the well-known omitted variable bias problem in econometrics. For the present discussion, let us suppose that the relevant data are available and ask how the model can help in measuring the causal impact of a change in law l.

In general, economic theories do not provide precise point predictions; more typically, they make predictions about the sign of an effect. In the context of measuring the effect of the law on outcomes, the variation in treatment typically occurs either across jurisdictions—namely, the experiment assumes that all individuals in a particular jurisdiction u ∈ U_s face the same legal environment l_st,, and it is the legal environment that varies across jurisdictions. For example, many countries can be characterized as civil-law or common-law legal systems. We can let U₁ be individuals in common-law countries and U₀ be individuals in civil-law countries, and set l_s = s.

In the example of civil- and common-law countries, one could estimate β_s = β(l_s) for each jurisdiction. In this case, the causal impact of the legal system depends on the distribution of characteristics of individuals in the economy. We would estimate the causal effect of changing from a civil-law system to a common-law system for regions that are currently under civil law in period t by:

In order to estimate the causal effect of a change in the legal system, one needs to use the characteristics of the jurisdiction where the change is to occur. This adjustment is a version of the well-known Oaxaca decomposition, which is widely used in studies of income inequality (see Altonji and Blank (1999)) and union wage differentials. As we discuss in more detail in Section 4, this is not the literature’s usual technique. The more common assumption is that the effect of a policy is linearly separable, where for u ∈ U_s we have:

(1)

Using data for a single period t, then, we can estimate the average effect of the legal system on the wages and employment of individuals by:

The goal of the theory discussed in this section can be summarized as follows. The theory makes predictions regarding the characteristics X that are needed to represent individual outsources. In particular, it will provide predictions regarding how variations in individual characteristics relate to variations in outcomes. Theory has predictive power if we can safely assume that the relationship between the Xs and the ys is stable over time.

A theory has more predictive power if one can represent outcomes using a smaller set of Xs. Given the difficulty of obtaining good measures of individual characteristics, theories with fewer Xs are inherently easier to test. On a related note, there is a line of inquiry in statistics that attempts to be model free. This is achieved by supposing that one has a rich set of X variables and that the environment is inherently continuous; as a result, good representations of the data can be used to make predictions on how changes in an individual’s Xs will affect outcomes. Breiman (2001) suggests that such an approach is sometimes more feasible given present computing resources and the large data sets we have in some domains.

However, representation is not causation. Many individual characteristics are not amenable to experimental treatment. Making causal statements requires that we assume we have a valid representation of the data that allows one to compare outcomes either: (1) across units with similar characteristics but in different treatments, or (2) the same units faced with different treatments over time. In these cases, the theory—in addition to specifying the relevant X variables—also specifies an explicit mechanism by which the law affects the actions of individuals, and hence how one can obtain a valid measure of the causal impact of a change.

3.2 Economics of the employment contract

This section discusses the literature that seeks to explain the form and function of employment contracts. Fundamentally, parties who enter into a contract have agreed to have their behavior constrained in the future. The most basic reason for a contract is to support inter-temporal exchange, something that cannot be avoided in the context of labor services. For example, a day laborer agrees to work eight hours in exchange for a wage at the end of the day. When it comes to paying, the employer may have an incentive to renege or attempt to reduce the agreed-upon wage. One role of the law is to enforce such agreements.

In economics, such simple exchanges are typically assumed to be enforceable. The literature has focused on explaining the form of observed contracts that address one of three more subtle issues:

What makes the study of employment contracts difficult is that every relationship has elements of these three ingredients. In particular, teasing out the empirical implications of these models has proven difficult. Nonetheless, much of the structure of the legal rules governing employment can be understood as an attempt to address risk, information asymmetries, or hold-up.

3.3 Implementing the efficient employment contract in a market

The literature on the employment contract has identified three broad economic motivations for an employment contract: insuring risk-averse employees, ensuring the revelation of relevant information for decision-making, and encouraging relationship-specific investments. Given these transaction costs, the next issue is: What sort of labor-market institutions ensure efficient matching and trade?

In principle, one could construct a model that includes all the ingredients that have been identified as relevant for understanding employment. At a purely abstract level, Rogerson (1992) and Aghion et al. (1994) have shown that under the appropriate conditions, one can construct an abstract mechanism that implements the efficient allocation in a variety of cases, some of which combine risk aversion and asymmetric information. However, as Tirole (1999) discusses, we still do not know how to relate these abstract results to observed institutions and contract forms. The literature on employment typically explores the implications of regulation for a simple model that has one or at most two transaction costs. The literature on employment protection has for the most part followed the lead of Bentolila and Bertola (1990) and Lazear (1990) in supposing that an increase in labor protection is parsimoniously modeled as an increase in turnover costs.

Lazear (1990) observes that if complete contracts are possible, mandated severance payments can always be undone via the labor contract. In that case, a law mandating severance payments would have no effect on employment but would lower starting wages. Lazear carries out a study of 22 countries over a 29-year period and concludes that increasing severance pay to 3 months’ salary for workers with 10 years’ experience leads to a 1 -percent reduction of the employment-to-population ratio. This is a reduced-form analysis that does not take into account the complex inter-temporal optimization problem faced by firms. This is the goal of Bentolila and Bertola (1990). They find that firing costs create complex inter-temporal incentives that depend upon the state of the business cycle. Specifically, firing costs reduce labor demand in good times but increase demand in bad times. Lower starting wages translate into lower firing costs, and hence firms have greater incentives to hire workers during downturns.

The literature has mostly followed the lead of this work and modeled employment protection as a turnover cost. The theoretical contributions have begun with one of the transaction costs (risk, asymmetric information, or holdup) and then explored the implications of employment protection modeled as a turnover cost. This allows one to explore the implications of treating a relationship with a particular policy choice. If we suppose that in different relationships one of the three transaction costs is more important, then this approach generates testable predictions of the effect of a law change for different relationships, which hopefully can be measured and hence form right-hand-side X variables.

I complete the section with a brief discussion of unions. From the perspective of transaction costs, one can view unions as an alternative to employment law. This provides a system for the implementation, enforcement, and arbitration of employment disputes between the firm and unionized employees.

3.4 Summary

In this section, we have reviewed the literature on the theory of the employment relationship from the perspective of transaction-costs economics and contract theory. From this research, we learn that an optimal employment contract is shaped by many factors in addition to the demand and supply of factors of production. The need to provide insurance to workers underlies many of the contributions, in part because this model is quite elegant and can in many situations deliver clear predictions. One prediction that it does not deliver is a theory of why employment relations can entail conflict, and why the allocation of bargaining power has important efficiency consequences.

Models of asymmetric information naturally deliver a theory of conflict, and can explain why parties for whom trade is efficient may fail to reach an agreement (see Crawford (1982)). The vast majority of work on employment focuses on the case in which the asymmetric information concerns the outside options for the worker and firm. Many of the employment disputes discussed in the previous section deal with disagreements concerning what happened within the relationship. The model in Simon (1951) is a useful starting point for thinking about these issues, but currently there is little work on the role of the courts in finding facts in employment cases.⁵¹ The main message here, consistent with the work of Milgrom (1988), is that the efficient contract for task assignments is to give the informed party decision rights. Chakravarty and MacLeod (2009) discuss how the law can achieve this goal in the context of construction contracts. The way the law achieves the formal allocation of authority within an employment relationship has not been explored in detail, however. Williamson (1991) makes the point that for the most part courts do not intervene in the day-to-day management of employees. The question of how various employment law doctrines affect this authority relationship is still an open question.

The provision of incentives to make relationship-specific investments gives the third motivation for entering into a binding contract. What is interesting is that the contractual instruments here—specifically, the allocation of authority and bargaining power—also play a central role in achieving efficient investment. There is a need for work that helps us understand which, if any, of these theories provide the most useful way to think about the employment contract. The theories by themselves are not typically framed in terms of making causal inferences. Rather, they make predictions regarding how variations in match characteristics (X variables) are related to observed features of the relationship (wages, employment, bonus or severance pay).

There is some recent work by Cahuc et al. (2006) and Postel-Vinay and Robin (2002) that uses the holdup model of wage determination developed in MacLeod and Malcomson (1993) to empirically estimate models of wage and employment determination. Cahuc et al. (2006) find that only highly skilled workers have significant bargaining power, while low-skilled workers have none. Postel-Vinay and Robin (2002) find in a panel with French data that personal characteristics tend to be more important for highly skilled workers. These are not causal exercises, but they do suggest that employment law is more likely to be important for highly skilled workers.

4.1 Employment law

In terms of Rubin’s model, the unit of analysis for employment laws are typically governments. The most common outcome variables are employment per unit of population, wages, the unemployment rate, and GDP growth. The issue then is how does a change in employment law change these outcome variables? This is a very difficult question because many events occur along with changes in the law that make it difficult to identify a causal effect. Surprisingly, the results tend to be relatively consistent. The majority of studies find either no effect or a negative effect of increasing employment protection. The results of this work are summarized in Table 2. We do not discuss all of these papers, but note that they can be divided into three broad classes: cross-sectional country studies, cross-sectional country studies with time, and the studies from the United States and India that use within-country state variation.

A good starting point are the cross-country studies. Botero et al. (2004) gather data from a several sources to construct measures of various types of employment regulation, including hours restrictions and dismissal procedures. They find that stricter employment protection is associated with countries whose law originated in the civil law tradition. For these countries, it is found that labor force participation is lower and unemployment is higher. Of course, these statements are not causal, but rather say that there is a covariation between legal origins and employment performance. Djankov et al. (2003) show that civil codes make it more costly to use the courts for contract enforcement, but even this evidence is not necessarily causal. For example, if judges in civil law countries are more corrupt, then the more bureaucratic rules may be a response to this corruption. These basic results have recently been replicated by Djankov and Ramalho (2009), and hence there appears to be a relationship. The question is, why?

A natural way to control for cross-country variations is to use variation over time in laws. Early studies along these lines include the influential work of the OECD (1993). They found that employment protection increased jobless duration. In subsequent work, the OECD (2003) finds that employment protection is often insignificant, but is associated with increased unemployment for prime-age males. These, like the cross-section studies, must construct measures of employment protection that are meaningful in different countries. The difficulty, as we see in Table 3, is that there are a large number of possible laws, each of which get implemented in an idiosyncratic fashion. One way to deal with this complexity is to have a more narrow study of a law change. A series of studies by Kugler (1999, 2004a, 2005) uses the 1990 market-based reforms in Columbia. Kugler finds that these reforms generally lead to more flexible labor markets and lower unemployment.

Given that these changes occurred at a single point in time, this implies a true causal effect under the hypothesis that other secular changes would not have produced this effect. One way to satisfy this assumption is to narrow the analysis to a set of units that are more similar, but face changes at different times. Besley and Burgess (2004) pioneered this approach using variations in employment law across states in India. Given that all the units (states) are in the same country, this controls for legal origin. Using data from 1958–1992 on employment law legislation, Besley and Burgess (2004) find that pro-worker legislation lowers economic growth. Similar results have been replicated by Aghion (2008) and Ahsan and Pages (2009). However, given the large cultural diversity in India, one might question the extent to which the law is exogenous to other events in society.

Possibly, the most convincing studies on the effect of the law exploit the fact that in the United States employment law is under state jurisdiction. One can then estimate the effect that state level changes in exceptions to employment at will have on the labor market. These law changes for the 1983–95 period are illustrated in Fig. 1. As we can see, a large number of states adopted the implied-contract and public-policy exceptions during this time period. In this case, the identifying assumption is that US states are sufficiently similar that one can assume that the impact of the law in each state is similar. One then uses a generalization of model (1):

where u denotes state. In most papers, the Current Population Survey (CPS) is used to measure employment and wages by state. Bertrand et al. (2004) show that one cannot assume that the error term ϕ_ui is iid, and one must allow for correlation over time. In practice this is achieved by adding state-specific time trends and computing standard errors with clustering of the errors at the state level (in effect allowing arbitrary covariance over time within states, but assuming independent errors across states).

Miles (2000) is an early study by a legal scholar who carefully ensured that employment law is correctly coded. This can be difficult in the United States because law is created by both the courts and the states. He finds very little effect on employment from any of the law variables, but finds that the implied contract exception leads to an increase in temporary employment, consistent with the later work of Autor (2003). Autor et al. (2004, 2006) further refine the law variables used by Miles (2000), finding that the implied-contract exception leads to a robust reduction in state employment, but that the public-policy and good-faith exceptions have no effect.

The theory discussed above predicts that the effect of the law depends upon the characteristics of the employment relationship. In particular, if hold-up is the transaction cost most responsible for the creation of employment law, then we would expect the law to have more effect on matches with higher levels of relationship-specific investment. This idea motivates the work of MacLeod and Nakavachara (2007). They match the CPS job training supplement that measures the amount of skill in a job with an occupation code to divide occupations in the low, middle, and high skill. They then use the data created by Autor et al. (2004) to explore the effect of relating skill requirement with the law. They find that the negative effects of the implied contract and good faith exceptions tend to be concentrated in low-skill (high-turnover) jobs. In fact, the good faith exception has a positive effect on employment for high-skill workers.

Though this appears to be at odds with the previous literature, it should be noted that the earlier literature focuses on the average effects, which obscures the effect on different subgroups. One of the messages from the recent research on contract theory is that optimal contract form should be sensitive to the characteristics of the job, particularly for high-performance jobs. Consequently, we cannot obtain a complete understanding of how the law works without taking into account its impact on different types of relationships. Both the theory and the early work of Freeman and Medoff (1984) predict that if transaction costs are a significant source of inefficiency in employment relationships, then there is a role for unions to enhance performance.

4.2 The effect of unions

Table 4 provides a list of studies that explore the effect of unions upon firm productivity. The early studies by Brown and Medoff (1978) and Clark (1980b) explicitly recognize that unions may enhance performance by increasing investments in firm-specific human capital, improving worker morale and other organizational parameters. Clark (1980b) also observes that union contracts have many terms that address issues in the workplace other than compensation. Brown and Medoff (1978) estimate a Cobb-Douglas production function using data from the May CPS merged with the 1972 Census of Manufacturers (COM). They find that labor productivity is consistently higher at unionized firms. They are careful to point out that this result can be explained by several factors, and while consistent with the hypothesis that unions enhance productivity, they do not establish a causal link.

Clark (1980b) refines this approach by using a time series on US cement plants, which allows one to explore the effect of unions upon plants in the same industry, with precise control on capital equipment. Like Brown and Medoff (1978), Clark finds that unions enhance productivity. One of the interesting findings is that if one compares only new plants, then unionized firms have 3.6% higher productivity, but a lower capital/labor ratio. This is consistent with the holdup model of Grout (1984), which predicts that firms will invest less in physical capital in response to union bargaining power. Abowd (1989) uses stock market data to conclude that unions and firms maximize total wealth.

Recently, developments have begun to focus on workplace organization.⁵² Black and Lynch (2001) find that there is an interaction between workplace practice and unionization. When firms have more consensual decision-making in the workplace, then unionization is associated with higher productivity. Conversely, unions are associated with lower productivity in firms that use “traditional” management practices. Doucouliagos and Laroche (2003) carry out a meta-analysis of a large number of studies on the productivity impact of unions. They find that the evidence is broadly consistent with a positive-productivity effect in the United States and a negative effect in the United Kingdom. This is an interesting observation given that the law is quite different in the two jurisdictions. In the United Kingdom, unions cannot commit to a binding agreement, whereas this is possible in the United States. This difference in productivities is consistent with the evidence, though we are far from having convincing causal evidence.

Even if unions enhance firm productivity, this does not necessarily translate into an increase in profits. This depends on the bargaining power of unions, and the extent to which firms earn rents.⁵³ A common strategy is to merge National Labor Relations Data on union certification with firm stock-market data to look at the impact of certification on profits. Ruback and Zimmerman (1984) find that unionization causes a 3.8% fall in equity value. Abowd (1989) finds that unionization can also reduce profits, but that this is an efficient redistribution from firms to workers. These and similar studies must deal with the fact that unionization is endogenous, and hence it is very difficult to estimate the causal impact of a union, independent from other factors. DiNardo and Lee (2004) use a regression discontinuity design in which they compare the outcomes in firms where the unions barely won certification, to ones where there was a close lost. Under the hypothesis that the groups have similar characteristics, then differences can be attributed to union status. Their finding is that the effect is essentially zero.

Lee and Mas (2009) use a similar approach, but obtain a better measure of abnormal stock-market returns. They find that unionization causes a 10% decline in abnormal returns. What is particularly interesting (especially in the light of the efficient-markets hypothesis) is that the negative effect is largest a year after the vote for unionization. This result is consistent with the body of research that looks at the impact of unions on firm productivity (see Doucouliagos and Laroche (2009)). Even if unions truly enhance productivity, if they lower profitability then we should expect firms to reallocate resources to jurisdictions with less union penetration. Kuhn (1998) suggests that there seems to be little evidence of this in Canada, while Machin (2000) finds that new firms in the United Kingdom tend not be unionized.

As Farber and Western (2001) documents for the United States, and Machin (2000) for the United Kingdom, there has a been a large decline in unionization that is consistent with the hypothesis of excessive rent extract by unions in the face of alternative, nonunion, investment opportunities for firms. Yet, there are still some areas with significant union presence. For example, workers in the entertainment industry are unionized. This is an industry where highly skilled actors, writers, and musicians must move from job to job, and where for each job there is a large number of potential candidates. In the absence of a union contract, the wage would be set at the talent’s opportunity cost, which is likely to be far below the return necessary to make it profitable to invest in his or her particular skill. In this case, a union has the potential to increase the talent pool, though this is a hypothesis that has not been carefully tested.

The other area with significant union presence is the public sector. Ehrenberg et al. (1983) find that unions do not significantly affect productivity for municipal librarians. Byrne et al. (1996) find that unionized police are less effective in dealing with crime. Eberts and Stone (1987) explore the effect of unions upon teacher productivity, finding that on average unionized teachers increase test scores by about 3%. However, their impact is more homogeneous, and they do not do as well with students with above-or below-average ability. In a recent study, Lovenheim (2009) finds that teacher unions raise costs by about 15% while having no impact on school performance.⁵⁴

In summary, the evidence on unions is consistent with the hypothesis that they do have some bargaining power with respect to the firm. As we discussed in the previous section, bargaining power may allow parties to implement more efficient arrangements. The extent to which this is possible in practice is controversial. It is clear that over the past century we have witnessed a large rise and fall in unionization rates in the United States. The transaction-cost perspective suggests that unions can be viewed as substitutes for legal enforcement of contracts and other forms of labor-market regulation. The fact that employment law in the United States has become stronger in the last 30 years—in the sense of providing more protection for disadvantaged groups⁵⁵ and stronger employment protection—suggests that private law may be providing a substitute for unionization.

In the case of the public sector, the skills acquired by workers are likely to be job-specific, and moreover the demand for these skills are stable. This suggests that the optimal contract is of a long duration, which may explain why public-sector unions are so prevalent. However, it is extremely difficult to measure public-sector productivity, and hence to evaluate properly the available labor institutions for these relationships. Moreover, it is difficult to argue that public-sector unions are purely rent-seeking organizations. As Blank (1994) documents, the public/private sector wage ratio has been falling in the United States, even while private-sector unionization has been falling. If unions were the main source of wage growth for workers, then we should observe the opposite. We do not have definite answers to any of these questions, and hence there is much room for further research.