Discrimination in Hiring

1
Discrimination in Hiring

• Special Topic

2
Readings

• READING FOR NEXT CLASS
• http//money.cnn.com/2005/04/08/news/funny/beautif
  ul_money/

3
Opinion Poll

• In their hiring decisions, do firms discriminate
  on the basis of gender and race?
• A. No firms do this
• B. Few firms do this
• C. Many firms do this
• D. Almost all firms do this

4
Clip

• http//www.youtube.com/watch?vaJDD0kERpdc

5
DISCRIMINATION THEORIES

• What does it mean to say firms discriminate?
• Preference-based Discrimination
• (Becker, 1951)
• Employers dont like hiring minorities
• Statistical Discrimination
• (Arrow, Phelps)
• Uncertainty over workers productivity leads to
  discrimination

6
Preference-based Discrimination

• d coef of discrimination
• All workers equally productive
• U pF(Na Nb) waNa wbNb - dNb
• MBpF
• MCa wa MCb wb d
• Optimal
• wa lt wb d Hire only from group a
• wb d lt wa Hire only from group b
• Depending on size of d, firm hires all group a or
  all group b.

7
Preference-based Discrimination (Contd)

• At price p, markets clear when
• demand for a workers supply of a workers
• demand for b workers supply of b workers
• No wage differential if there are enough firms
  with d0
• But if enough firms have large d then
• demand for b workers lt supply at wa wb
• Too many b workers to all get jobs at
  non-discriminatory firms
• Creates wage gap

8
Preference-based Discrimination (Contd)

• BIG POINT 1 Discriminating firms do not max p.
  They earn lower profits than non-discrim firms.
• pF(Nb) wb dMRgt True marginal cost
• Discrimination should not prevail
• Competitive market means non-discriminating firms
  can price output lower, and force discriminating
  firms out of business.

9
Notes

• Subtlety Could take entrepreneurs compensation
  in the form of satisfying a preference
• Normal profits could be lower.
• Result is different if firms have market power,
  not subject to Law of One Price.

10
Variants

• Employee Discrimination
• Employers dont care but employees do
• Hire a minority ? must pay other workers a
  compensating wage differential
• Implications
• Segregation - No wage differential
• Customer Discrimination
• Only customers care
• Implications
• Hide minority workers
• Segregation by region - Little expected wage dif
• Seems fragile. This should be going away, but
  wage differential by race is not...

11
Statistical Discrimination 1

• Aigner Cain, 1977
• Setup
• W and B workers
• MPi known to worker, not firm.
• Test scores noisy signal of true MP
• W and B have same average prod, but.
• Test score noisier predictor of Bs ability than
  As

12
Aigner Cain - Graphically

• If test were perfect, everyone would be paid
  there prod
• 45 degrees
• W types
• Test noisy, not a perfect predictor of prod
• B types
• Test noisier, but same average MP
• Upshot High ability W workers paid more than
  high ability B workers

wi
W
B
Ti
13
Aigner Cain - Graphically

• Are low ability W workers better off than low
  ability B workers?
• A. Yes
• B. No
• C. Cant tell

wi
W
B
Ti
14
Aigner Cain - Graphically

• Ws have greater variance of pay, but same mean
• Discrimination?

wi
W
B
Ti
15
Opinion Poll

• Suppose there are 2 midterms.
• Suppose midterm 1 counts for 1/10 of your grade
• Suppose midterm 2 counts for 1/3 of your grade
• Which do you study harder for?
• A. Midterm 1
• B. Midterm 2
• C. The same amount for each

16
Statistical Discrimination 2

• Lundberg and Startz, 1983
• Differences in signal-to-noise ratio may yield
  endogenously different investments in H.C.
• Add HC investment to Aigner-Cain
• For group b, investment in schooling raises wage
  less than for group a.
• So equilibrium choice of HC is lower.

17
Graphically

• Dif in signal to noiseratio ?
• Dif incentive to invest?
• Dif in Test score?
• Dif w in equilibrium
• Discrimination?
• LS say YES
• Groups have same average initial ability but
  different w

18
Problems

• For Statistical Discrim to be correct, it must
  be that return to education for minority workers
  is less than that for other workers
• Doesnt appear to be true!
• (Ashenfelter and Rouse, 2000)
• Why is noisy measure noisy?
• Noisy to whom?
• Why not fix it?

19
Discrimination Theory - Recap

• What does it mean to say discrimination causes
  wage difs?
• Preference-based Discrimination
• Implies lower profits for discriminators
• Not expected to persist in the long-run (in
  perfect competition.)
• Statistical Discrimination
• Firms profit-max, but uncertainty about MP leads
  to wage difs
• Aigner Cain Noisier signal of MP for group B
  means high-scoring group Bs get lower wages than
  group Ws with identical scores. Reverse true for
  low-scoring group Bs. Same average wage
• Lundberg and Startz Add HC investment to Aigner
  Cain. Noisier signal of MP for group B means
  group B workers will choose to get less HC than
  Group A workers. Theyll receive less of a reward
  for acquiring HC, so they invest less, in
  equilibrium. End up with lower average wages

20
Detecting Discrimination

• Empirical Evidence

21
Opinion Poll

• Suppose, on average, people born in San Francisco
  had more education than people from Bakersfield
• Suppose people born in San Francisco, on average,
  earned higher wages than people from Bakersfield
• Would you be inclined to say that Bakersfieldians
  are victims of discrimination?
• A. Yes
• B. No

22
Oaxaca Decomposition

• Difs in observables
• Difs in unobservables
• Sometimes called Discrimination
• Example Gender wage gap
• Part of ave wage dif explained by dif in
  schooling
• Part unexplained
• Hypothetical Set Up Plot wage against schooling
  for men and women
• Suppose Men have higher wage and higher schooling

M
W
S
23
Question

• Which point indicates the average wage women
  would have earned if they had the same average
  education as men?

M
B
D
W
C
A
E
S
24
Oaxaca Decomposition

• A) shows total average wage dif between M and W
• B) shows wage dif explained by differences in ed
• C) shows total unexplained wage dif between M and
  W
• Shows how large dif in ave wage would be if women
  had as much ed as men, but were still rewarded
  for ed according to womens pay scale
• Assumes womens wage increase for schooling
  (slope) is correct

M
W
S
25
Oaxaca Decomposition

• A) shows total average wage dif between M and W
• B) shows wage dif explained by differences in ed
• C) shows total unexplained wage dif between M and
  W
• Shows how large dif in ave wage would be if men
  had as little ed as women, but were still
  rewarded for ed by mens pay scale
• Assumes mens wage increase for schooling (slope)
  is correct

M
W
S
26
Oaxaca Decomposition

• Distance C often interpreted as due to
  discrimination
• Usually, many more explanatory variables are used
  besides schooling
• Age, experience, parental ed, etc.
• Wage dif unexplained by all these factors is
  called discrimination
• Sometimes also called the residual

M
W
S
27
Problem 1

• Assumes all important factors are in explained
  portion
• This is rarely true
• Usually there are important factors we cant
  observe (unobservables).
• Example In Malaysia studies showed large wage
  dif between Chinese and Malays
• Not explained by education
• But ed measure treated all college degrees
  equally
• Differences in majors
• Male maternity?

M
W
S
28
Problem 2

• Ignores discrimination that may have caused
  unequal average characteristics in the first
  place
• Example Why did women have lower average ed than
  men?
• This itself may have been result of barriers to
  ed for women!

M
W
S
29
Evidence from Oaxaca Decompositions

• Most empirical studies show significant
  unexplained wage gaps by gender and race
• These are what are usually reported in the press
• Example Women earn 70 cents for every dollar
  men earn, for equal work.
• Very hard to draw confident or meaningful
  conclusions from these studies

30
Other Approaches Audit Studies

• Send 2 workers to a job interview, one worker
  from each group (e.g., by race, by gender, by
  sexual orientation, etc.)
• Give them identical resumes
• See which one is more likely to get the job, on
  average
• Many audit studies find applicants treated
  differently based on race or gender

31
Problem 1

• Studies are not double blind
• The actors in audit studies
• may not be identical
• may have an agenda

32
Extension

• Use names as proxy for group (Bertrand and
  Mullanathan)
• Tabulate name data from state-level birth
  certificates
• Determine distinctively white names and
  distinctively African-American names
• Distinctively black (frequency black)/(frequency
  white)
• Send out at least 2 fictitious resumes in
  response to each of a large number of job
  advertisements
• Make resumes virtually identical
• Randomly assign white-sounding or black-sounding
  name
• See if applicants are treated differently

33
Bertrand and Mullanathan - Results

• Call back rates
• Whites 9.7, or about 1 in 10
• African-Americans 6.5, or about 1 in 15
• Equivalent to about 8 yrs experience on resume
• Strengths
• 1. Solves Double-blind problem.
• 2. Solves Factors observed by employers at the
  interview problem

34
Problem 1

• Does name capture SES?

35
Additional evidence

• Levitt and Fryer, 2003, find names do not predict
  long-run outcomes, after controlling for SES

36
Other Problems

• Market discrimination vs individual
  discrimination
• In BM 88 of firms treat both applicants equally
• Audit studies may find firms that discriminate,
  but this may have no effect on wages if there are
  enough firms that do not
• Marginal firm is what matters
• (Helps explain differences in audit study results
  and wage regressions)

37
Differences in Variance

• Example Men, Women and math scores
• If cut-off high then non-dscriminating firm
  will hire more men
• If cut-off low then non-dscriminating firm will
  hire more women
• Sidebar Summers

prod
prod
38
Conclusions

• Very difficult question to answer
• Lots of interesting work being done