Difference between revisions of "Adverse Impact Tests"
(Created page with "Four “statistical” tests are available for analysis of adverse impact data: 80% Rule, Expected / Difference, 2 Standard Deviation, and Fisher’s Exact. Statisticians genera...") |
|||
| Line 1: | Line 1: | ||
| − | Four “statistical” tests are available for analysis of adverse impact data: 80% Rule | + | Four “statistical” tests are available for analysis of adverse impact data: |
| + | * '''80% Rule''' | ||
| + | * '''Expected / Difference''' | ||
| + | * '''2 Standard Deviation''' | ||
| + | * '''Fisher’s Exact''' | ||
| − | + | Statisticians generally feel that in order for any of these tests to be valid you should have at least 30 employees in the group, 5 transactions, and 5 in the protected group. If you have less than this, you could argue that the test results are not reliable. | |
| + | |||
| + | ==80% Rule== | ||
* Applicant Flow/Promotions (positive activities)<br>If the rate being measured is greater than 80% of the rate for the base group for applicant flow and promotions the test passes. | * Applicant Flow/Promotions (positive activities)<br>If the rate being measured is greater than 80% of the rate for the base group for applicant flow and promotions the test passes. | ||
*Termination/Layoffs (negative activities)<br>If the rate being measured is less than 120% of the rate for the base group for terminations the test passes. | *Termination/Layoffs (negative activities)<br>If the rate being measured is less than 120% of the rate for the base group for terminations the test passes. | ||
* Applicant Flow Example<br>Assumes that males are the base group: If male applicants = 20 and male hires=10 then the male hire rate is 50%. If female applicants=30 and female hires=10 then the female hire rate is 33.33%. The 80% test indicates that a female hire rate of 40% (50% X 80%) is ‘passing’. The actual female hire rate being 33.33%, the 80% test fails in this example. | * Applicant Flow Example<br>Assumes that males are the base group: If male applicants = 20 and male hires=10 then the male hire rate is 50%. If female applicants=30 and female hires=10 then the female hire rate is 33.33%. The 80% test indicates that a female hire rate of 40% (50% X 80%) is ‘passing’. The actual female hire rate being 33.33%, the 80% test fails in this example. | ||
| − | See | + | See [[80%25_Or_Four-Fifths_Rule|80% or Four-Fifths Rule]] for more information. |
| − | + | ==Expected/Difference== | |
| + | (Calculated if the 80% test is selected.) | ||
If the 80% test fails then an expected value for the activity is calculated and compared to the actual value for that activity by subtracting the actual value from the expected value (disregard sign) to arrive at the difference. If the difference value is 0 the test passes. The expected value is determined by calculating the overall rate for the activity in test and then applying that rate to the individual subgroups. | If the 80% test fails then an expected value for the activity is calculated and compared to the actual value for that activity by subtracting the actual value from the expected value (disregard sign) to arrive at the difference. If the difference value is 0 the test passes. The expected value is determined by calculating the overall rate for the activity in test and then applying that rate to the individual subgroups. | ||
Continuing the example above: Since the 80% test failed we look to the expected/difference values. Since total applicants=50 and total hires=20 then the overall hire rate for this group is 40%. So, if all things were equal, we could reasonably expect 12 female hires (30 female applicants X 40%). Since we actually hired 10 females we have a difference of 2, therefore the expected/difference test fails and we will run either the 2 standard deviation test or the Fisher’s exact test as determined by group size and expected value. See report footnotes for these criteria. | Continuing the example above: Since the 80% test failed we look to the expected/difference values. Since total applicants=50 and total hires=20 then the overall hire rate for this group is 40%. So, if all things were equal, we could reasonably expect 12 female hires (30 female applicants X 40%). Since we actually hired 10 females we have a difference of 2, therefore the expected/difference test fails and we will run either the 2 standard deviation test or the Fisher’s exact test as determined by group size and expected value. See report footnotes for these criteria. | ||
| − | + | ==Two Standard Deviation Analysis== | |
The 2 Standard Deviation Analysis usually results in a lower showing of under representation than the Four-Fifths Rule, especially if the job group being tested is small. This analysis is available in the Utilization Summary, Availability Analysis, and the Adverse Impact Analysis. | The 2 Standard Deviation Analysis usually results in a lower showing of under representation than the Four-Fifths Rule, especially if the job group being tested is small. This analysis is available in the Utilization Summary, Availability Analysis, and the Adverse Impact Analysis. | ||
| − | See | + | See [[Two Standard Deviations]] for calculation example. |
| − | + | ==Fisher’s Exact== | |
The Fisher’s Exact Test is an analysis which was originally designed for smaller groups, but works well with any size group. | The Fisher’s Exact Test is an analysis which was originally designed for smaller groups, but works well with any size group. | ||
| − | See | + | See [[Fishers Exact Test]] for programming example. |
| + | |||
| + | ==See Also== | ||
| + | [[Adverse Impact]]<br> | ||
| + | [[Adverse Impact Data Requirements]] | ||
Revision as of 14:09, 18 March 2011
Four “statistical” tests are available for analysis of adverse impact data:
- 80% Rule
- Expected / Difference
- 2 Standard Deviation
- Fisher’s Exact
Statisticians generally feel that in order for any of these tests to be valid you should have at least 30 employees in the group, 5 transactions, and 5 in the protected group. If you have less than this, you could argue that the test results are not reliable.
Contents
80% Rule
- Applicant Flow/Promotions (positive activities)
If the rate being measured is greater than 80% of the rate for the base group for applicant flow and promotions the test passes. - Termination/Layoffs (negative activities)
If the rate being measured is less than 120% of the rate for the base group for terminations the test passes. - Applicant Flow Example
Assumes that males are the base group: If male applicants = 20 and male hires=10 then the male hire rate is 50%. If female applicants=30 and female hires=10 then the female hire rate is 33.33%. The 80% test indicates that a female hire rate of 40% (50% X 80%) is ‘passing’. The actual female hire rate being 33.33%, the 80% test fails in this example.
See 80% or Four-Fifths Rule for more information.
Expected/Difference
(Calculated if the 80% test is selected.) If the 80% test fails then an expected value for the activity is calculated and compared to the actual value for that activity by subtracting the actual value from the expected value (disregard sign) to arrive at the difference. If the difference value is 0 the test passes. The expected value is determined by calculating the overall rate for the activity in test and then applying that rate to the individual subgroups.
Continuing the example above: Since the 80% test failed we look to the expected/difference values. Since total applicants=50 and total hires=20 then the overall hire rate for this group is 40%. So, if all things were equal, we could reasonably expect 12 female hires (30 female applicants X 40%). Since we actually hired 10 females we have a difference of 2, therefore the expected/difference test fails and we will run either the 2 standard deviation test or the Fisher’s exact test as determined by group size and expected value. See report footnotes for these criteria.
Two Standard Deviation Analysis
The 2 Standard Deviation Analysis usually results in a lower showing of under representation than the Four-Fifths Rule, especially if the job group being tested is small. This analysis is available in the Utilization Summary, Availability Analysis, and the Adverse Impact Analysis. See Two Standard Deviations for calculation example.
Fisher’s Exact
The Fisher’s Exact Test is an analysis which was originally designed for smaller groups, but works well with any size group. See Fishers Exact Test for programming example.
