Statistical Discrimination

1
Statistical Discrimination

• David L. Dickinson
• Appalachian State University
• April 2006 GATE.

2
Maximal Quality Selection and Discrimination in
Employment. Douglas D. Davis, 1987, JEBO.

• Statistical Discrimination
• Non-prejudiced based. Motivations for racial
  (e.g.) ability perceptions are not abundant.
• Ability perceptions may be lower for minority
  workers if employers interview fewer minorities.
• That is, the sample maximum increases in draws.
• External validity is with upper-credential
  hiring environments..we want the best!
• The benefit of experiments is that prejudice can
  be lab-controlled (difficult to control in the
  field, however).

3
Experimental Design

• Critical features
• Costly search
• Incentive to select only best candidates
• Opportunity to reflect perceptions of minority
  inferiority
• Controlled payoff distributions to searcher
  (employer), and proportion of minority
  observations controlled.
• Context neutralno reference to race, gender, etc.

4
Design Specifics

• Subject sees Box g and Box h.
• Each box is a prize distributions, and subjects
  may sample each distribution for 1-3 cents per
  observation (80 of sample purchased comes from
  majority Box, which subjects do not know).
• Payoffs are chosen prize minus total sampling
  costs.
• Distributions of Boxes changes every six decision
  periods, and only on 6th decision period can
  subjects choose Box from which sample is draws
  occur.
• Main hyp is supported if subjects choose from
  majority Box more frequently

5
Results

• Each subject generates 2 observations on
  experimental parameter cells (Table 1) (N80).
• Proportion of minority Box choices.36lt.50.
• Modal choice proportion, however, is .50.
• Minority superiority also a regular choice (order
  statistics go both ways).
• TAB sheet recording may worsen the situation
  (though not statistically significant)
• A prejudiced MARKET TIP significantly biases
  selections away from minority Box (Table 4).
• Evidence in Table 5 suggests that minority Box
  (population) must be significantly better, in
  terms of mean prize, than majority Box to create
  perception of equality.

6
Conclusions

• Greatest perception of minority inferiority
  appear with TAB or MARKET TIP.
• Presumed equality only occurs with a higher mean
  outcome from less frequently sampled (minority)
  group.
• Practical applicability?
• EEO interview quotas, productivity distribution
  equality?