Statistical discrimination has been tested in a series of experiments. Before discussing the experiment of Fryer Jr. et al (2005) in detail, some earlier experiments are summarised based on the review of Anderson et al (2005). Davis (1987) studied an experimental labour market in which worker groups were of different size. If more observations can be drawn from one group, then it is more likely to produce a higher maximum observation. If employers focus on this higher draw, this may result in a bias towards the larger group. In the experiment, the employer group was in the first period confronted with 80 % of draws from the majority population, in the second they chose themselves how intensively the respective groups should be sampled. Still, 60 % of the employers sampled the majority group, pointing to a mechanism with which a bias towards one group might arise simply induced by population properties (Anderson et al 2005; p.105).
In the experiment ofAnderson and Haupert(1999), workers were divided into green and yellow groups. The productivity of each worker in each group was assigned exogenously. Before making a decision, employers could interview the workers at a certain cost. Anderson and Haupert (1999) observed that in markets with lower average productivity of one colour, employers tended to hire fewer workers of that group. They claim that in the absence of an interview, employers focus on the population average. This is supported by the fact that employment levels rose after the cost of interviewing was reduced (Anderson et al 2005; p.106)
Whereas in the previous experiments differences were exogenous,Fryer Jr. et al (2005) conducted a classroom experiment where productivity and hir-
ing decision could evolve simultaneously as in the model ofCoate and Loury (1993).
The experiment is set up as follows: Half of the players are employers, the other workers. Half of the workers are green, the other purple. Workers are told that their investment cost is drawn from an interval between $0 and $1, and that costs are independent and vary randomly. Workers make their investment decision after observing their cost. After the decision, a test result is generated. If a worker invests and gets hired, he gets a wage of $3.00. If he is not hired, he gets a low-skill job at a wage of $1.50. The net gain for an investing worker is the wage minus his investment cost. Two draws of the test are made to determine the final test result. Test results are represented as marbles in an urn. A blue marble (B) represents a positive test result, a red one (R) a negative result. The probability that a result is good is 0.5 if the worker invested, or 0.2 if not. A test result of BB thus means that the chance that a worker invested is high, a result of RR means he probably did not invest, whereas in the event of BR (or RB), the result is unclear. An employer only knows the worker’s colour and test result. An employer earns $4.00 if a worker who invested is hired; $0.00 if a worker who did not invest was hired, and $2.00 if the worker was not hired. To both workers and employers, the hiring rates of each colour are presented, i.e. information about the market outcome is public.
Two treatments are presented: In the first treatment, investment costs are drawn for both worker groups over 20 periods from the interval [$0.00,
$1.00]. The second treatment was conducted to ‘investigate the effects of historical discrimination’ Fryer Jr. et al (2005; p.166). In this treatment, for the first five periods investment costs for purple workers were drawn from the interval [$0.5, $1.00], whereas for green workers from [$0.00,$0.5], so that green workers had higher incentives to invest. For the remainder 15
rounds, the cost distributions were equal.
In general, Fryer Jr. et al (2005) observe that discrimination emerges only in some experiments. Of these they present two instances.
In the first experiment discrimination against purple workers emerged quickly. Around 80-90% of green workers were hired most of the time, whereas purple workers were hired at around 40-50%. Hiring rates remained almost constant for green, and slightly improved for purple workers. Invest- ment rates for both groups increased for some periods, after which they fell again. Investment costs in the first two rounds was (by chance) higher for purple workers. This ‘may have been a factor that kept investment rates much higher for green workers in most periods’ (Fryer Jr. et al(2005; p.165)). Employers hired always when the test result was BB. Employers were more liberal with green workers: If the test result was unclear, they were hired invariably, but only 78% of purple workers. If the result was RR, employers still hired 64% of green, but only 15% of purple workers. In the following discussion, it emerged that beliefs that purple workers would not invest formed quickly, as well as the corresponding belief that this group is unlikely to get hired. This lead most workers of that colour to decrease their efforts. Moreover, the consistent liberal treatment of green workers encouraged most of them in their investment behaviour, while some players stopped investing because they expected to get hired anyway. Thus, invest- ment rates for both groups declined in the second half of the game, but for different reasons.
In the second experiment, it emerged that investment rates of green and purple workers were similar, although the costs for purple workers were much higher. They were hired at an only slightly lower rate than green workers. After step 5, the cost distributions became equal again. Pur-
ple workers continued to invest at similar rates, while investment rates for greens dropped quickly, resulting in higher employment for purples (raising from about 60% to 90%), and lower employment for greens (decreasing from about 65% to 50%).
Summarising, these results highlight some driving factors in experimen- tal environments:
– Negative stereotypes can form quickly and are persistent. It might only take some random perturbations (here, initial cost asymmetries) to generate these stereotypes.
– Decisions are not independent. The belief that one group is more pro- ductive leads to the belief that this still holds if bad or mixed outcomes occur, while the opposite is true for the disadvantaged group.
– The height of the cost does not necessarily have a large impact on the investment decisions, as long as the return to investment is positive.