L7: Fairness Flashcards
what is meant by “adverse impact”?
if with regards to employment or promotion, one group shows sig higher % of rejection (lower selection ratio) than another group
- does not necessarily mean the selection procedure is biased but should investigate
- but still negative on a societal level (they may have had differential opportunities to learn a particular required skill)
- not biased if shows actual differences between groups in job related skills and abilities (but this is still adverse impact)
- ex: min height requirements for a job have adverse impact on women
what is the adverse rejection rate, or four fifths rule?
- a rule of thumb in determining adverse impact
- selection rate for a group is less than 4/5 (80%) of selection rate for group w highest selection rate
- not a legal def of discrimination, but used to mointor severe discrimination practices
when is adverse impact unfair?
as soon as different selection ratios are not justifyable due to different performance on the job
what is a test bias?
When the test systematically overestimates or underestimates the performance of a group (e.g. , based on gender, race) if you use one regression line for both groups OR test estimates better for one group than another
ex: on average, members of different groups have an equal probability of being successful on the job but an unequal probability of being hired due to a biased selecion test. So using it will unfairly affect the chances of Person from disadvantaged group in getting the job
but if differences between groups in mean test scores are associated w a corresponding difference in mean job performance = no evidence of bias
how can u detect test biases?
- are there group differences in validity (r) aka differential validity?
- are there group differences in regression lines aka differential prediction?
- are there group differences in average predictor scores aka adverse impact?
what is differential validity?
difference between subgroups in the correlation (r) between a predictor and a criterion
- bias exists if there is differential valiidty between 2 groups -> problem if validity lower for minority
- this gives an indication if the test is valid within groups: does the test predict job performance for all groups
what is differential prediction?
Same chance of success but different selection ratesbias exists if the same regression line is used to predict criterion scores when regression lines differ across groups
- predictor use is fair if a common regression line can be used to describe the predictor criterion relationship for all subgroups of interest
when is adverse impact justifiable?
same regression slope & intercept across groups
cause this indicates no differential prediction or predictive bias
eg cognitive ability tests
evaluate cognitive ability tests on biases & adverse impact
- no sig differential prediction found which means that they predict job performance of minority & majority groups in the same way
- BUT large ethnic group differences found on cog ability tests (so gropu differences in averages of predictor scores) -> adverse impact
why is there a large gap in ethnic groups scores on cog ability tests?
- in immigrant groups may be due to language bias (so tests w substantial verbal components may underestimate g)
other explanatory vairables: income, maternal education, maternal verbal ability/knowledge, learning materials, physical home env, birth weight, birth order
evaluate work sample tests on biases & adverse impacts
- smaller average score differences between ethnic groups than cog abiltiy tests -> lower adverse impact
- similar levels of predictive validity
Since work samples have similar levels of predictive validity (slightly more) & less adverse impact than cognitive ability tests -> why not always replace ‘g’ tests with work samples?
- work samples may be less suitable for assessment of applicants w/o job experience
- expensive
- validity of work samples decreases over time more so than other seleciton procedures
- g is fundamental for acquiring the knowledge & mastering skills found in most jobs (esp those w large training component)
evaluate personality inventories on biases & adverse impacts
- no strong indication of differential prediction in different ethnic groups
- studies show negligible adverse impact
- but in nl theres evidence of differences in average scale scores on some personality measures (less favorable for immigrants) -> adverse impact
why do personality inventories show adverse impact in nl?
could be due to language problems, socioeconomic background, cultural differences
evaluate interviews on biases & adverse impacts
- show considerably less adverse impact than cog ability tests (particularly highly structured interviews)
suggestions: supplement cog ability tests w other selection procedures (less adv impact, high predictive validity)
what are 6 strategies to reduce adverse impact?
- improve recruiting for minorities -> get top applicants (try to increase pool of qualified applicants from minority group)
- use g tests in comb w non g predictors
- use measures of specific, not general g abilities
- use alternate modes of presenting test stimuli
- enhance face validity (ensure that applicants preceive test as valid & are motivated to do well,decreases stereotype threat)
- implement test score banding to select applicants (instead of strict top down selection, remember that test is th not perfect, search for minority applicants within the top scorers band; all scores within a band are regarded as equivalent)
define affirmative action
Policy that exceeds equal employment opportunity by requiring to correct past discriminatory practices by
increasing the numbers of minorities (e.g. women, minority ethnic groups) in specific positions
idea is to level the playing field in the short term & to affect a long term transformation whereby society eventually no longer needs affirmative action
what are the 2 challenges to affirmative action?
- individual hired under AA feel prejudged as inferior performers & can be viewed as tokens
- perceptions of reverse discrimination
- preferences toward one protected class may create conflicts between other minority groups
- criticzed for being inefficient in assigning the best available candidates irrespective of gender, to a particular job when several candidates compete for it
BUT BENEFITS OUTWEIGH THE COSTS
