What is it?
How do you calculate it?
Kyle E. Brink
Jeffrey L. Crenshaw
Personnel Board of Jefferson County
www.AdverseImpact.org
Sex
National Origin
Total
Discrimination
Chargesa
30,510
2,880
24,826
9,396
67,612
$ 135,400,000
$ 22,800,000
$ 232,300,000
aIncludes
Source: http://www.eeoc.gov/types/index.html
Defined discrimination
Appears you only have to violate one or the other to claim AI exists
AI Analysis Considerations
Span covered
Comparison group
Test/analysis type
Descriptive statistics
Practical significance*
Statistical significance*
Decision/outcome in question
Hires
Selection
Rate
White
80
48
Black
40
12
48/80 =.6
(60%)
12/40 = .3
(30%)
.3/.6 = .5
(50%)
Impact Ratio
The term adverse impact does not appear in APA Standards for
Testing or SIOP Principles
Statistical Significance:
Decisions & Errors
Null hypothesis: There is no difference (no AI);
any difference is due to chance.
Truth (unknown)
No AI
No AI
Decision
AI
AI
Correct
acceptance Type II error ()
(1- )
Correct
Type I error
rejection
()
(Power; 1-)
Statistical Significance
Impact ratio is much more powerful than significance
test, but at the expense of Type I error
Tests of statistical significance can control Type I error
Error variance
Greater power when less error variance
Selection rate
Greater power when high selection rate (e.g., 50%)
Widen timeframe
Combine geographic areas
Combine events from several jobs, job groups
or divisions
Combine selection procedures
Combine different ethnic groups
Group
Dallas
White
Ft. Worth
Combined
Hispanic
15
30
50.0%
White
40
300
13.3% 1.00
Hispanic
30
225
13.3%
White
140
500
28.0% 0.63
Hispanic
45
255
17.6%
level
Higher level results in greater power
Court prefers = .05
Tails
1 tail (directional) has greater power than 2 tail (nondirectional)
Court prefers 2 tails
2 X 2 contingency table
E.g., association between pass vs. fail X male vs.
female
Statistical Tests: ZD
Z-test of the difference in selection rates
Comparison Problem
When comparing test results, we are
comparing apples and oranges
ZD vs. ZIR
Both
ZIR
OFCCP
ZIR
Chi-square or ZD
Practical Tests
N of 1 (flip-flop) rule
Practical Tests
Shortfall analysis
Conclusions
If IR < .8 and a statistical test is not significant
Consider magnitude of IR
Confidence intervals may show promise
References
Biddle, D. (2005) Adverse Impact and Test Validation: A
practitioners Guide to Valid and Defensible Employment Testing.
England: Gower.
Collins, M. W. & Morris, S. B. (2008). Testing for adverse impact
when sample size is small. Journal of Applied Psychology, 93, 463471.
Equal Employment Opportunity Commission, Civil Service
Commission, Department of Labor, & Department of Justice. (1978).
Uniform guidelines on employee selection procedures. Federal
Register, 43, 38290-38315.
Hays, W. L. (1994) Statistics (5th ed.). New York: Holt, Rinehart &
Winston.
Lawshe, C. H. (1987). Adverse Impact: Is it a viable concept?
Professional Psychology Research and Practice, 18, 492-497.
Moore, D. S. & McCabe, G. P. (1993). Introduction to the Practice
of Statistics (2nd ed.). New York: W.H. Freeman & Company.
References
Morris, S. B. (2001). Sample size required for adverse impact
analysis. Applied HRM Research, 6, 13-32.
Morris, S. B. & Lobsenz, R. E. (2000). Significance tests and
confidence intervals for the adverse impact ratio. Personnel
Psychology, 53, 89-111.
Office of Federal Contract Compliance Programs (1993). Federal
contract compliance manual. Washington, D.C.: Department of
Labor, Employment Standards Administration, Office of Federal
Contract Compliance Programs (SUDOC# L 36.8: C 76/1993).
Roth, P. L., Bobko, Pl, & Switzer III, F. S. (2006), Modeling the
behavior of the 4/5ths rule for determining adverse impact: Reasons
for caution. Journal of Applied Psychology, 91, 507-522.
www.AdverseImpact.org