中国经济管理大学 《心理学与人力资源管理》 （MBA研究生课程班）Chapter 8 – Fairness in Employment Decisions
      韦恩.F.卡西欧（Wayne F. Cascio）美国科罗拉多大学（丹佛校区）商学院Robert H. Reynolds全球领导与管理讲座教授，拥有罗切斯特大学工业与组织心理学博士学位。担任美国心理学会工业／组织心理学分会、美国管理学会、全美人力资源学会等多个组织会士。
　　赫尔曼·阿吉斯（Herman Aguinis）美国科罗拉多大学（丹佛校区）商学院Mehalchin管理讲座教授。曾担任美国管理学会研究方法分会会长。

Chapter 8 – Fairness in Employment Decisions

Overview 
 Chapter 8 Fairness in Employment Decisions presents considerations beyond the reliability and validity concepts discussed in Chapters 6 and 7.  Employment fairness includes both individuals’ perceptions of and the further statistical analyses of employment decisions.   Perceptions influence how willing employees are to accept and to demonstrate organizational human resource practices. These perceptions also influence the organization’s image beyond its workforce.  Perceptions lead to organizational profit or to organizational loss.  To statistically analyze HRM fairness, discriminant and predictive validity or predictor and criterion variables must be considered.  While some existing research exists, the methodologies behind these analyses present both procedural and practical challenges and limitations.  However, even with the limitations, a thorough understanding of employment fairness can lead to positive societal impact. 
Annotated Outline
I. At a Glance
A.  When it is possible, selection measures need to consider individual and group differences by
 1.  Race
 2.  Ethnicity
 3.  Gender
B.  These differences typically analyzed through the examination of differential validity.
 1.  Subgroup validity coefficients
C.  More enlightened approaches consider differential prediction or differential bias.
 1.  Standard errors of estimate
 2.  Intercepts of subgroup regression lines
D.  One or both may occur at the same time, however, current research indicates that both occur at the same time on an infrequent basis.
E.  While measures may predict differences in performance per group, job performance needs consideration as the predictions may cause unfair discrimination when job performance information would not. 
F.  When measures do not cause discrimination, adverse impact for protected groups must be considered to minimize legal challenges and to improve greater societal benefit. 
II. Introduction 
A.  To this point in our study of HRM, appropriate HRM decisions have been influenced by & dependent upon

  1.  Laws
 2.  Decision systems utility (cost & benefit)
 3.  Processes
 4.  Test results and characteristics such as
  a.  Reliability
  b.  Validity
B.  What does fairness mean?
 1.  Treating all people alike, justly, equitably 
 2.  Having no adverse impact on any group of individuals
C.  How do you determine fairness?
   1.  By analyzing the differential validity and predictive bias among groups
 2.  Must keep in mind that HRM decisions are based on individual differences  measures.
 3.  Therefore, HRM decisions will have some discriminatory effects.
 4.  Fairness in employment decisions means then that HRM decisions make  justifiable and wise discriminatory decisions.
D.  Resources for guiding HRM fairness

 1.  Uniform Guidelines on Employee Selection Procedures (1978)
 
 2.  Standards for Educational and Psychological Testing (1999)

 3.  Principles for the Validation and Use of Personnel Selection Procedures  (2003)
 4.  Computer program to explore decision making scenarios 
 www.cudenver.edu/~haguinis/mmr
E.  Legal precedence guiding HRM fairness

 1.  Ninth Circuit Court of Appeals
 Officers for Justice v. Civil Service Commission of the City and County
 of San Francisco, 1992

 2.  Seventh Circuit Court of Appeals
 Chicago Firefighters Local 2 v.
 City of Chicago, 2001 
F.  Fairness challenges
 1.  number subjects per group 
 2.  unbiased criterion
 3.  comprehension of differences between
  a.  differential validity &
  b.  differential prediction
 4.  value systems
 5.  societal costs
G.  Historically, fairness research focuses include

 1. Efficacy of selection decisions analysis of differential validity within  subgroups
 
 2. Accuracy of performance predictions analysis of mean job performances   and differential validity
H.  Critical Definitions  
 
 1.  Adverse impact
 
  a.  when HRM selections for members of subgroups are less than
   4/5 or  80% of group with highest selection rate

  b.  may exist fairly, may exist unfairly
 2.  Differential Validity   
  a.  when significant difference exists between two subgroups’ validity    coefficients
  b.  & when correlations in one or both groups are significant
 3. Single Group Validity 
  a.  when no significant difference exists between two subgroups’    validity coefficients 
  b.  & when significant difference does exist for one group’s predictor –    criterion relationship  
III.  Assessing Differential Validity - Basic Fairness Procedure
 1.  Divide data by group & subgroup
  Example:  Managerial Jobs by
   a.  Age
   b.  Race
   c.  Ethnicity
   d.  Gender
  
 2.  Determine predictor & criterion correlation 
  a.  For all managerial jobs using
   i.   Predictor = Test Score
   ii.  Criterion  = Performance Rating
  b.  Plot the relationship by gender
 3.  Analyze fairness implications
  a.  Positive validity  
   i.  Predictor – criterion relationship is the same for both      subgroups and elliptical in shape

   ii.  Conclude fairness, validity, and legality supported
  b.  Zero validity
   i.  Predictor – criterion relationship is the same for both      subgroups but circular in shape
   ii.  Conclude no differential validity, no point to consider      predictor
  c.  Positive validity but adverse impact

   i.  Predictor – criterion relationship shows differences per     subgroups and elliptical in shape
 
   ii.  Conclude valid and legal adverse impact but only if criterion     necessity proven
  d.  Positive validity combined groups, invalid for separate groups  
   i.  Predictor – criterion relationship is high for entire group but     low or  zero for either subgroup and elliptical in shape
 
   ii.  Conclude unfair, invalid, illegal, and discriminatory 
  e.  Equal validity, unequal predictor  means 
   i.  Predictor – criterion relationship is similar for both subgroups,    elliptical in shape, but predictor means differ 
   ii.  Conclude with successful performance as foundation the use     of different cut scores for decisions is fair, valid, and legal most     but not all of the time
  f.   Equal validity, unequal criterion  means
   i.  Predictor – criterion relationship is similar for both subgroups,    elliptical in shape, but criterion means differ
 
   ii.  Conclude fairness questionable, validity questionable, but no     adverse impact
  g.  Equal predictor means, valid for nonminority only
   i.  Predictor – criterion relationship differs for both subgroups,     shapes differ, but valid for nonminority only
 
   ii.  Conclude fairness questionable, validity limited, no adverse     impact, but definite social implications
  h.  Unequal criterion means and validity only for nonminority
   i.  Predictor – criterion relationship differs for both subgroups,     shapes differ, but valid for nonminority only
 
   ii.  Conclude fairness questionable, validity limited, no adverse     impact, but definite social implications
 4.  Basic Fairness Summary
 
  a.  Perfect fairness may not be possible when HRM decisions applied to    heterogeneous groups.   
  b.  Implementing different HRM decision systems may be empirically    more fair but may be perceived with suspicion and lose any credibility. 
  c.  Additional Differential Validity Issues 

        i.   Very few well-controlled studies 
     ii.  Samples sizes existing research too small
     iii. Predictors not always relevant to criterion
   iv. Lack of unbiased, relevant, reliable criteria
     v.  Limited number of cross-validated studies  
IV.  Assessing Differential Prediction & Moderator Variables 

A.  To completely study and understand fairness, differential predictions for subgroups must be considered.
 
B.  Differential predictions focus on the slope of the differential validity coefficients.
 
C.  Slopes are best understood by considering the regression line (line of best fit) between the predictor and criterion variances.  
D.  Regression line accuracy can be improved by considering the sub-groupings as additional variables or moderators.
 
E.  Considering multiple moderators brings in the concept of Moderated Multiple Regressions (MMR) or R².
F.  Interesting evidence for MMR research

 1.  Looking at differences over time and predictions of job performance  

  a.  Cognitive Differences
   i.   Some research showed minorities tended to do less well on     job than test scores predicted for Dutch, African-American,     Hispanics  
  b.  Physical Ability differences
   i.  Gender differences existed but varied by occupation      considered
  c.  Personality differences
   i.  Gender differences found by occupation
 
 2.  This evidence indicates that for HRM, selection decisions would tend to hire  more minorities rather than fewer
 3.  Assessing differential prediction & moderator variables problem to consider
 
  a.  small sample sizes for minority groups

  b.  increase chance that procedure deemed unfair when procedure is    fair
 
  c.  decrease statistical power 
 4.  To avoid low MMR statistical power, carefully plan a validation study to  include technical feasibility & credible results
V.  Further Considerations - Adverse Impact, Differential Validity, & Differential Prediction
A.  To reduce adverse impact
 
 1. Improve minority recruiting strategy
 2. Use cognitive abilities in combination with noncognitive predictors
 3. Use specific cognitive abilities measures
 4. Use differential weighting for the various criterion facets
 5. Use alternate modes of presenting test  stimuli
 6. Enhance face validity
 7. Implement test-score banding to select among applicants
  a.  considers distributive justice for appropriateness of HRM testing    decisions
 
  b.  must remember that HRM tries to maximize profitability
   i.   maximizing profits may lead to adverse impact
 
   ii.  values based HRM may lead to decreased profitability
  c.  Test-score banding methods
 
   i.  Sliding-band method – considers range of test scores      as equivalent given imperfect reliabilities for tests
     
    1)  maximizes both utility and social objectives
 
      ii.  Criterion-referenced banding method - considers range of test    scores (predictors) and range of performance scores (criteria)
 
    1)  also maximizes utility and social objectives  
     iii.  Strengths

    1) Use of validity evidence  
    2) Bandwidths are wider
    3) Inclusion relevant criterion data
    4) Use of reliability information
   iv.  Weaknesses

    1) Possible legal issues
    2) Possible violation scientific values 
    3) Possible violation intellectual values
    4) Emotions associated w/Affirmative Action       Programs 
    5) Conflict between goals of research and        organizations
    6) Measurement objections
VI. Fairness - the Social and Interpersonal Context of Employment Testing
A.  Fairness requires professionalism, courtesy, compassion, & respect
 
B.  Perceived unfairness may lead to
 1.  negative organization impression
 2.  litigation challenges
C.  Fairness perceptions include 
 1.  distributive justice - outcomes
  
 2.  procedural justice – processes to reach decisions
VII.  Fair Employment and Public Policy
A.  While not always popular, tests and measurements serve public in several ways
 1.  diagnostic – to implement remedial programs 
 2.  assessing candidate qualifications   
 3.  protection from false credentials
B.  Each generation must reconcile the meaning of equal employment opportunities
 
C.  Policies are not for or against tests and measurements.
D.  Policies are about how tests & measurements are used.
Discussion Questions
1. Why is the assessment of differential prediction more informative regarding test fairness than an assessment of differential validity?  Pages 176 – 180
 Within HRM, test fairness includes individuals’ perceptions, statistical characteristics, and justifiably legal employment decisions. Test fairness means that all people who take an employment test perceive that the test measures everyone in the same manner, alike, justly, and equitably. Fair tests accurately predict employment performance, are accepted by those affected by the test decisions, and withstand legal challenges. Test fairness also means that test measurements result in no adverse impact on any group of individuals. 
 Historically, attention to test fairness focused on the differential validity (predictive accuracy) of employment decisions based on test measurements.  However, while conceptually appropriate and important, differential validity analysis does not completely explain many aspects of employment bias that may result in employment discrimination.  To improve selection (test) measures, more thorough analysis of the standard error of estimates and the intensity of the errors (slopes and intercepts) is necessary. Differential prediction assessment  provides this information.  Standard error of estimates and the intensity of the errors may be understood through the use of regressions analyses (line of best fit regression lines) and the considerations of multiple or moderator variables (multiple correlation analyses).
2. Summarize the available evidence on differential validity and its relationship with adverse impact. What advice on this issue would you give to an employer? Pages 169 – 174
 Differential validity research demonstrates that there can be various relationships between predictors and criteria. Because multiple relationships can exist, I would recommend that the employer first conduct a basic differential validity study. The basic steps to complete a differential validity study follows.
 1.  Divide data by group & subgroup
  Example:  Managerial Jobs by
   a.  Age
   b.  Race
   c.  Ethnicity
   d.  Gender
  
 2.  Determine predictor & criterion correlation 
  a.  For all managerial jobs using
   i.   Predictor = Test Score
   ii.  Criterion  = Performance Rating
  b.  Plot the relationship by gender, age, race ethnicity
 Once the differential study is complete, I would recommend that the relationship plots be analyzed for implications and that employment policies be adjusted or confirmed based on the results. Additionally, I would recommend that the study be periodically updated to keep the policies as fair as is possible.  Possible plot relationships and corresponding conclusions follow. 
  a.  Positive validity  
   i.  Predictor – criterion relationship is the same for both     subgroups and elliptical in shape

   ii.  Conclude fairness, validity, and legality supported
  b.  Zero validity
   i.  Predictor – criterion relationship is the same for both     subgroups but circular in shape
   ii.  Conclude no differential validity, no point to consider       predictor
  c.  Positive validity but adverse impact

   i.  Predictor – criterion relationship shows differences per     subgroups and elliptical in shape
 
   ii.  Conclude valid and legal adverse impact but only if criterion     necessity proven
  d.  Positive validity combined groups, invalid for separate groups  
   i.  Predictor – criterion relationship is high for entire group but     low or  zero for either subgroup and elliptical in shape
 
   ii.  Conclude unfair, invalid, illegal, and discriminatory 
  e.  Equal validity, unequal predictor means 
   i.  Predictor – criterion relationship is similar for both subgroups,    elliptical in shape, but predictor means differ 
   ii.  Conclude with successful performance as foundation the use     of different cut scores for decisions is fair, valid, and legal most     but not all of the time
  f.   Equal validity, unequal criterion means
   i.  Predictor – criterion relationship is similar for both subgroups,    elliptical in shape, but criterion means differ
 
   ii.  Conclude fairness questionable, validity questionable, but no     adverse impact
  g.  Equal predictor means, valid for nonminority only
   i.  Predictor – criterion relationship differs for both subgroups,     shapes differ, but valid for nonminority only
 
   ii.  Conclude fairness questionable, validity limited, no adverse     impact, but definite social implications
  h.  Unequal criterion means and validity only for nonminority
   i.  Predictor – criterion relationship differs for both subgroups,     shapes differ, but valid for nonminority only
 
   ii.  Conclude fairness questionable, validity limited, no adverse     impact, but definite social implications
 With selection measure discriminant validity demonstrated (or not) and with employment policies reflecting the efforts to establish and maintain fairness, I would caution the individuals responsible for this aspect HRM to recognize that perfect fairness may not be possible when HRM decisions are applied to heterogeneous groups.  Additional managerial challenges include the realty that implementing different HRM decision systems may be empirically  more fair but may be perceived with suspicion and lose any credibility. 
3. Discuss some of the difficulties and suggested solutions for conducting a differential-prediction analysis. Pages 177 - 180
 To completely study and understand fairness, differential predictions for subgroups must be considered.  These differential prediction analyses present several procedural and practical problems that researchers must attempt to address and to consider as they evaluate the results of the analyses. For example, there may be small sample sizes for minority group and for other subgroups.  This reality may lead to the perception that a selection measure is unfair when the procedure is indeed fair. Small sample sizes also lead to decreased statistical power.
 To avoid low differential prediction statistical power, HRM researchers must carefully plan the validation study to include technical feasibility & credible results and when weaknesses exist to acknowledge the limitations and strive to improve the data over time.  Which strategy to use will vary based on practical considerations such as time, availability of participants, and access to statistical software.  However, assuming that the practical constraints can be resolved, some of the available strategies include the following.
        1.   Improve the research design with larger sample sizes
2.   Adjust statistically for small sample sizes
3.  Implement a synthetic validity approach
4.  Adjust the level of significance test that can be absorbed
5.  Make reasonable predictions with regard to effect sizes
6.  Use the power of random sampling
7.  Use an extreme group design
8.  Develop and use reliable measures
9.  Use continuous criterion scales
10. Equalize sample sizes across subgroups by oversampling
11. Strengthen theory behind relationship of predictor to criterion
12. Monitor compliance with the research assumptions
4. Describe strategies available to reduce adverse impact. Pages 183 - 189
 Adverse impact exists when HRM selections for members of subgroups are less than 4/5 or 80% of the group with the highest selection rate. This adverse impact may exist fairly or unfairly. Adverse impact may exist fairly when there are genuine or bonafide differences among subgroups. Unfair adverse impact may exist when differences are not valid and lack predictive performance validity. To reduce adverse impact several procedures can be implemented. These procedures include the following.
 
 1. Improve minority recruiting strategy
 2. Use cognitive abilities in combination with noncognitive predictors
 3. Use specific cognitive abilities measures
 4. Use differential weighting for the various criterion facets
 5. Use alternate modes of presenting test  stimuli
 6. Enhance face validity
 7. Implement test-score banding to select among applicants
  a.  considers distributive justice for appropriateness of HRM testing    decisions
 
  b.  must remember that HRM tries to maximize profitability
   i.   maximizing profits may lead to adverse impact
 
   ii.  values based HRM may lead to decreased profitability
5. When is a measure of individual differences unfairly discriminatory? Pages 189, 190
 Individual differences measures are by definition discriminatory. These measures exist to make the best decisions about who can do the best job for the organization making the decisions. Measures are unfairly discriminatory “when persons with equal probabilities of success on the job have unequal probabilities of being hired for the job” (Guion, 1966).
6. Provide arguments in favor of and against the use of test-score banding. Pages 184 - 189
 Test-score banding methods include the sliding-band method and the criterion-referenced banding method. The sliding-band method considers a range of test scores as equivalent given the imperfect reliabilities for test scores. This method maximizes both the utility of the decisions and the social objectives that the organization supports.  Criterion-referenced banding methods consider a range of test scores (predictors) and a corresponding range of performance scores (criteria). Decisions are made within the context of the corresponding ranges. This method also maximizes utility and social objectives for the organization.   
 Strengths of the test-score banding methods include the use of validity evidence, the options to change the bandwidths, the inclusion of relevant criterion data, and the inclusion of reliability information.  Test-score banding is consistent with psychometric theory. Additionally, court decisions have accepted the use of test-score banding to justify employment decisions.
 Weaknesses or arguments against the use of test-score banding include possible legal issues not yet determined by the courts, possible violation of scientific values, possible violation intellectual values, negative emotions associated with Affirmative Action Programs, conflicts between the goals of research and organizations, and finally, measurement objections by those affected by the decisions.
7. What are the advantages and disadvantages of implementing a criterion-referenced banding approach as compared to a predictor-referenced approach? Page 181, 185 - 187
 The advantages and disadvantages of implementing a criterion-referenced banding approach relative to a predictor-referenced approach can be considered along four dimensions. These dimensions are the use of validity evidence, bandwidth, inclusion of criterion information, and the use of reliability information.
 With regard to the use of validity evidence, criterion-referenced banding allows the inclusion of comparisons based on performance data (criterion data) resulting in greater predictive accuracy. Sliding test bands alone do not. However, criterion data is not always available so sliding test bands may be the only banding approach available.  Criterion-based bandwidths are generally wider than those produced by test-bands.  Thus, criterion-based decisions appear to be more flexible and fair but may not be as accurate.  The inclusion of criterion data may produce inaccurate results when the criteria are imperfect or inadequate to accurately reflect performance.  Because reliability information can be included with criterion-referenced bandwidths, the corresponding validity coefficients are strengthened because both the predictor and the criterion reliability coefficients are considered. 
8. What are some strategies available to improve fairness perceptions regarding testing?       Page 189 - 191
 Perceived fairness for employment testing can be influenced by making sure that research surrounding the test includes strong scientific methodologies with the maximization of well-controlled studies and making sure that sample size is considered during the analysis (and, where possible is as large as is methodologically needed). Another strategy is to make sure that the predictors are relevant to the performance criterion.  Attempts should be made to make sure that the criteria not biased, are relevant, and reliable.  And lastly, fairness perceptions can be improved when efforts are on-going to make sure that the tests and test measures are cross-validated.
9. Discuss some of the public policy issues that surround testing. Pages 190, 191
 While not always popular, tests and measurements serve the public in several ways. These ways include                                                                                                                                    1.  diagnostics  – to implement remedial programs                                                                  2.  assessing candidate qualifications                                                                                           3.  protection from false credentials.
 Public policy will reflect each generation’s reconciliation of and management of equal employment opportunities. General public opinions will shift and change over time and based on factors such as the general economy, the need for greater or fewer employees, and changes in the workforce demographics. Regardless of these factors, though, employers and employees need to remember that policies are not for or against tests and measurements. Policies are about how tests are used and how measurements are used.

中国经济管理大学 MBA微课程：免费畅享MBA