The Role of Two Statistical Approaches in EEO Cases
by Richard E. Biddle
Richard E. Biddle is the president of Biddle and Associates, Inc., a Sacramento, California based EEO consulting firm. He has published numerous article on various aspects of disparate impact theory. He has served as an expert/consultant on more than 100 EEO cases. The firm of Biddle & Associates, Inc., specializes in litigation support, test development and validation, affirmative action development and software development.
Copyrightę 1995 Biddle & Associates, Inc.
In the EEO (Equal Employment Opportunity) field, there are two theories of discrimination: discriminatory treatment and discriminatory impact. Both can use two types of statistics: hypergeometric and binomial under different circumstances. Both theories of discrimination can be used to prove individual or class claims, although individual cases are usually taken with the treatment theory, and class cases with the impact theory. To prove charges of discriminatory treatment, intentional discrimination must be shown against a protected person or group. To prove charges of disparate impact, facially neutral practices, procedures or tests are challenged to reveal that although they may not be intended to be discriminatory, they are discriminatory in effect.
Statistics have a role in both types of cases. The "stark pattern of discrimination" referred to in disparate treatment theory relies on the same statistical approaches as the initial burden requirement in disparate impact theory. This paper will discuss the role of hypergeometric and binomial statistics in EEO cases, what data should be used for each, along with an overview of disparate impact and treatment burdens so the reader can see the context for each statistical approach.
Burdens of Proof
Prima Facie: The plaintiff has the initial burden of proving a prima facie (i.e., inference) case by showing that (1) the plaintiff is a member of a protected group (based upon race, color, religion, sex, or national origin, (2) the plaintiff was qualified for the position and applied for a job for which the employer was seeking applicants, (3) despite the plaintiff's qualifications he or she was rejected, and (4) after that rejection, the job remained open and the employer continued to seek applicants with the same qualifications as the plaintiff's and selected an applicant who was not a member of the plaintiff's protected group. This last point of the prima facie argument demonstrates that the plaintiff was treated differently than non-protected persons under similar circumstances.
The procedures for establishing proof are flexible. Evidence is presented and interpreted. Experts can be used to present and interpret statistical evidence, and statistics may be used to indicate discrimination against an entire protected group. Significant statistical disparities, supported by specific instances of discrimination, may be used to prove a prima facie case of discrimination. Conversely, statistics may also be used to prove that a protected group was not treated differently than a non-protected group.
If a prima facie case is made by the plaintiff, there is a presumption that the challenged acts -- if not explained or rebutted by the employer -- are more likely than not to be based upon consideration of impermissible factors, such as race.
Rebuttal: After the plaintiff establishes a prima facie case, as defined above, the burden of going forward with producing evidence to rebut the prima facie case shifts to the defendant. This production rebuttal burden can be met with the articulation of a legitimate, nondiscriminatory reason for the challenged conduct. The rebuttal reasoning must allow for a rational conclusion that the employment decision was not motivated by discrimination.
The central issue concerns the alleged intentionally discriminatory motive of the employer in terms of the selection process. If there is evidence of a discriminatory motive, this allegation can be rebutted by showing that the same legitimate decision would have been made absent the discriminatory motive.
To prove charges of discriminatory treatment, a plaintiff must show that intentional discrimination was directed against a protected person or group. To prove charges of disparate impact, a plaintiff would challenge facially neutral practices, procedures or tests to reveal that although they may not be intended to be discriminatory, they are discriminatory in effect. Statistics have a role in both types of cases.
Once the employer rebuts the prima facie case of disparate treatment, the presumption of discrimination arising from the prima facie case is dismantled. However, the question of whether or not there has been intentional discrimination still remains to be resolved.
Pretext: After the employer has rebutted the plaintiff's prima facie case, the plaintiff is then entitled to try to show that employer's asserted reason was nothing more than a mere pretext to cover the discriminatory motive. A reason cannot be proved to be a pretext for discrimination unless the reason was false; therefore, discrimination was the reason.
To show disparate treatment based solely on statistics, the plaintiff must show "a stark pattern of discrimination unexplainable on grounds other than [race]."
In disparate impact, there are three burdens of proof -- alternately maintained by the plaintiff and the defense: (1) the plaintiff has the first burden to prove adverse impact of a selection device or system (i.e., prima facie statistical case of disparate impact where a group protected by law is shown to be disproportionately impacted); (2) if the plaintiff is successful in proving adverse impact, then the burden shifts to the employer to prove that the selection device or system is job related and justified by business necessity; and (3) if the employer is successful with its burden, the plaintiff must prove the alternate employment requirement in order to win the case (i.e., that another selection device or system, or another way of using the selection device or system would also serve the employer's legitimate purpose without the undesirable adverse impact).
Because the "stark pattern of discrimination unexplainable on grounds other than [race]" under the disparate treatment theory and the initial burden of the disparate impact theory (to show adverse impact) share the similar statistical approaches, we will consider adverse impact analyses in depth to cover the role of statistics in both kinds of cases.
Adverse Impact - First Burden
In the first burden of a disparate impact analysis, up to three tests are performed to determine adverse impact:
1. the "threshold" analysis (also called the initial inquiry) to see if the sex and racial composition (i.e., percentages) of the at-issue job is underutilized compared to the composition of the qualified population in the relevant labor market
2. a "barriers" analysis to see if there are barriers or practices which disproportionately deter sex or racial group members from applying, and then
3. the "selection" analysis to see if a practice, procedure or test is disproportionately impacting a sex or racial group, unless the practices, procedures or tests are not capable of separation for analysis, then the entire decision-making process can be evaluated as one practice.
If a practice, procedure, or test is found to be a "barrier" as defined above, an adverse impact finding could be expected on the practice, procedure, or test causing the barrier. However, even if the practice, procedure or test causing the "barrier" to an at-issue job is not involved in the action, it still can be a "barrier" for statistical purposes. If a barrier is found, a binomial statistical test will be needed in the "selection" analysis and a "proxy" group will be needed in the "selection" analysis. If a barrier is not found (i.e., applicant flow is very similar to availability), then actual applicants can be used in the "selection" analysis and a hypergeometric statistic is used.
Two kinds of statistical distributions are used to measure the differences between observed data and data expected by chance in the above analyses. The binomial distribution can be used in all three of the tests, while the hypergeometric distribution can be used in the "selection" test, if a barrier does not exist.
Synonyms used for each are:
|Guidelines Method||Hazelwood Method|
|two sample approach||one sample approach|
|rates analysis||pools analysis|
The "Guidelines Method" is a label used for the hypergeometric approach because of the reference to Section 4D of the Uniform Guidelines on Employee Selection Procedures, Federal Register, 43, 38,290-38,315. "Two sample" refers to the two samples, or groups, used in the formula (i.e., men versus women; whites versus blacks, etc.). "Rates analysis" is used as a label because the rate of one group is compared to the rate of another group (i.e., hire rate, selection rate, promotion rate, etc.). Therefore, the hypergeometric distribution is appropriate in situations comparing applicants versus hires, those who take a test versus those who pass it, those who start a process versus those who complete it, and so on.
The "Hazelwood Method" is an alternative label for the binomial approach, so-called after its U.S. Supreme Court source. "One sample approach," describes how a single sample, or group, is evaluated from the pool (i.e., Hispanics selected versus Hispanics available). "Pools analysis" refers to the comparison between one group in a pool and the same group in another pool (i.e., the percent of Hispanics in the selected pool compared to the percent of Hispanics in the available pool). The binomial distribution is appropriate in situations comparing hires versus availability, promotions versus availability, applicants versus availability, etc.
Minimum Values Needed with the Hypergeometric and Binomial
The hypergeometric and binomial are slightly different statistical approaches used to evaluate the probability of the discrepancy between the data observed in a situation and data that is expected by chance. When the probability of a difference between what is observed in a data set and what is expected from a similar data set, randomly selected, is so great that no more than 5 percent of the time will the differences be attributed to chance (i.e., a 5 percent probability), then a legal inference can be drawn of statistical significance.
Two problems occur with the 5 percent probability. First, how is it derived? It can be derived from a direct calculation, or it can be estimated after first calculating a standard deviation statistic and then translating the statistic to a probability. The resulting answers will differ slightly but will be close. The direct calculation takes much longer than the estimated way for a computer, and some canned statistical programs do not calculate the probability for both tails of the distribution. Secondly, there are one-tailed probabilities and two-tailed probabilities that yield different standard deviations. (One-tailed probabilities address directional hypotheses, such as do women score higher on a certain test compared to men. The EEO field compares group rates and pool percentages using two-tail probabilities. Two-tailed probabilities address the nondirectional hypotheses, such as is there a difference in the passing rates between men and women on a certain test, regardless of which group is higher.) A one-tailed probability of 5% equals 1.64 standard deviations while a two-tailed probability of 5% equals 1.96 standard deviations.
The United States Supreme Court has concluded that the EEO field will consider 2-3 standard deviations as the range for statistical significance in disparate impact cases. Therefore, as a general rule, if the difference between an expected value and the observed number is greater than 2-3 standard deviations, the difference is greater than that which can be attributed to chance. Standard deviations of 5 or more have been used in "treatment" cases to infer intentional discriminatory treatment
In sampling theory, there is a population and a sample. The population is the pre-selection pool of people; it is the full universe of individuals having some common, observable characteristic. A sample is a subset of the population, or the post-selection pool of people. The sample size could vary from 1 person to all persons in the entire population.
When we take repeated turns at sampling from a limited population without replacing each sample back into the population, the sampling processes are no longer independent of one another. Since the population has a limited sample size and, therefore, a limited number of "successes" to offer in total, the success of one sampling turn will affect the probability of success with the next sampling turn. The hypergeometric distribution (which does not involve replacement of the sample) takes into account the established number of "successes" that exist in the population. The hypergeometric approach is often referred to as "distribution-free," because it does not assume the scores in the analysis were drawn from a population distributed in a certain way.
When the sample selected from one sampling turn is replaced into the population before the next sampling turn, the chance for success with each sampling turn is identical. The binomial distribution (which involves replacement of the sample) assumes that the chance for success in one sampling turn is independent of any other sampling turn, while the population remains constant. The binomial approach approximates a normal distribution.
Obviously, with the replacement method or binomial approach, it is possible to select the same person twice. However, when sample sizes are much smaller than the population, this risk is considered slim. In addition, only minimal error will occur by not returning the person sampled back into large populations when sample sizes are small.
Probability theory has several criteria: (1) the way an event can occur is equally likely, (2) events are mutually independent, and (3) probabilities are additive for alternative outcomes.
The probabilities calculated with either the binomial or hypergeometric approach range from zero (no chance) to 1.0 (absolute chance) that the observed sample departs from a random sample (a sample expected by chance and chance alone from the population). Judging from the size of the probability, people in the EEO field make inferences on whether the events that caused the observed sample can be treated as a random sample. If the probability is high (i.e., more than .05), the events that caused the observed differences are inferred as chance factors. If the probability is low (i.e., .05 or less), the inference is made that factors other than chance caused the observed differences in the samples. Probability measures the risk of inference error.
While the specific level of statistical significance is usually placed at the 5% probability level, the level of statistical significance increases as the probability decreases. In other words, a 1% probability is more statistically significant than a 5% probability. A 1% probability means that 1 time in a hundred, with data sample sets like the one for which the formula was applied, chance caused the observed differences. A 5% probability means 5 times out of a hundred chance caused the observed differences. There are fewer risks of inference error with the 1% probability compared to the 5% probability, hence the statistical significance is more with the 1% probability compared to the 5% probability.
Hypergeometric versus Binomial Calculations
Slightly different numbers are used for the hypergeometric formula and the binomial formula. An example is provided below to illustrate the numbers used in each formula. The numbers used for the hypergeometric formula comparing whites and blacks in this example are: whites 5214 (number of whites in the population, such as applicants) and 858 (number of whites successful after the sampling turns, such as those hired), and blacks 355 (number of blacks in the population) and 37 (number of successful blacks after the sampling turns). These are the numbers used to calculate rates within the hypergeometric formula. The success opportunities and success rates for Hispanics, Asians, and American Indians were not included in the hypergeometric calculations for whites versus blacks.
The numbers used for blacks with the binomial formula are 5% (the percent blacks make up of the total population, such as census requisite skills availability), 37 (the number of successful blacks, such as those hired), and 1077 (total number of successful sample considering all groups, such as all groups hired). Dividing the number of successful blacks (37) by the total number of successful sample (1077), results in the percentage blacks make up of the successful pool.
|BEGINNING POPULATION POOL||SUCCESSFUL SAMPLE|
|Total||7059||100 %||1077||100 %||15.3%|
The hypergeometric standard deviations will usually be higher than the binomial standard deviations. Hence, plaintiffs prefer the hypergeometric and defendants prefer the binomial. However, when the successful sample total is small relative to the population size, (or as the population reaches infinity) the hypergeometric standard deviations will approach the binomial standard deviations.
The following standard deviations include both types of distribution, using the data presented above:
Note: Some calculations are not applicable, because the data shows a group to be advantaged compared to other groups.
Support for Each Distribution
Uniform Guidelines on Employee Selection Procedures, (1978), Federal Register, 43, 38,290-38,315.
Adoption of Questions and Answers to Clarify and Provide a Common Interpretation of the Uniform Guidelines on Employee Selection Procedures, (1979), Federal Register 43, 11,996-12,009, and amended April 1, 1980.
Contreras v. City of Los Angeles, 656 F.2d 1267 (9th Cir. 1981).
Bouman v. Block, 940 F.2d 1211 (9th Cir. 1991).
Castenada v. Partida, 97 S.Ct. 1272 (1977).
Hazelwood School District v. United States, 97 S.Ct. 2736 (1977).
Teamsters v. United States, 97 S.Ct. 1843 (1977).
Proof of statistical significance is not quite the penultimate step in an adverse impact finding; evidence of practical significance must follow. Practical significance needs to be evaluated only when there is statistical significance per the Uniform Guidelines on Employee Selection Procedures, Section 4D. Disparate impact requires a finding of statistical significance and practical significance.
Practical significance focuses on the practical effect of differences found in statistical significance tests. Theories and formulae for statistical significance are taught in the social sciences, but the mechanics of practical significance are learned by reviewing the application of statistics in litigation. Practical significance is concerned with the effect small number changes have on statistical conclusions, and with an extensive, contextual and practical assessment of the utility of those conclusions. It offers a means to ask: What happens when the results are just barely significant? What is the impact of adding a few people in a hypothetical way?
In the U.S. v. Commonwealth of Virginia, when adding two people from an unfavorable status (e.g., not passing a test) in the plaintiff group to a favorable status (e.g., passing a test) changed a finding of statistical significance to one of no statistical significance, the results were not found to be practically significant. In Waisome v. Port Authority of New York, the same events led to the same conclusion.
In Contreras v. City of Los Angeles, two practical significance tests were discussed. One dealt with the effect of adding three people to the plaintiff group in a favorable way that eliminated the 80 Percent Rule of Thumb conclusion. When the 80 Percent Rule of Thumb conclusion can be changed by adding only three people, this shows that the sample is considered unreliably small, and of no practical significance.
Also from Contreras, when adding four people to the plaintiff group in a favorable way so that it brings the selection rates very close, the sample is too small to rely on for statistical purposes. Being "close" in this situation means selection rates are within 2.1% of each other after adding four to the plaintiff group's passing numbers. 10
The Fifth Circuit has twice discussed the topic of practical significance with a type of practical significance test, similar to the second Contreras test of comparing selection rates. In Frazier v. Garrison I.S.D., the Court stated that the selection rates were within 4 1/2% of each other -- insufficient for a finding of disparate impact, even with statistically significant rate differences. In Moore v. Southwestern Bell, the court found that a 7.1% difference in selection rates was not enough for a finding of disparate impact, even though the standard deviations were above 3.0.
Adverse Impact Burden Summary
The "threshold" analysis has been specified by Wards Cove Packing Co. v. Atonio for the initial inquiry in disparate impact cases. In that case the United States Supreme Court stated that the initial inquiry compares the racial composition of those in the at-issue jobs to the racial composition of the qualified population in the relevant labor market (i.e., availability). Regardless of the "threshold" analysis results, a "barriers" analysis must be run to determine whether to use actual qualified applicants in the "selection" analysis with the hypergeometric formula or a "proxy group" such as outside availability of the qualified population in the relevant labor market with the binomial formula.
Wards Cove instructs that: "[A]s long as there are no barriers or practices deterring qualified" protected group members from applying, then using the actual applicants is acceptable in the "selection" analysis. A "barriers" analysis compares those who apply to those qualified to apply in the relevant labor market. If the applicant pool is not significantly different from the sex or racial composition of the qualified population in the relevant labor market (i.e., availability), then the selection analysis should be conducted with the actual qualified applicants. If the "barriers" analysis shows significantly less qualified applicants than availability, then the "selection" analysis should use a "proxy group" or availability.
As long as there are no barriers or practices deterring qualified protected group members from applying, Wards Cove then requires a comparison between the percentage of selected applicants who are from a protected group to the percentage of qualified applicants from that group. The 1991 Civil Rights Act tells us to perform this analysis on each challenged practice, procedure or test unless it can be demonstrated that the elements of the selection process are not capable of separate analysis. Under these circumstances, the overall process can be analyzed as one practice.
If the "barriers" analysis shows no difference in percentages by sex and race/ethnic groups between the qualified applicants and availability, the hypergeometric test can be used to compare applicants to those selected. If the "barriers" analysis shows statistically lower percentages of women or a protected race/ethnic group, then the binomial should be used with a proxy group to compare availability to those selected.
If the "barriers" analysis or the "selection" analysis (using either the actual qualified applicants or the "proxy group") shows statistical and practical significance for a practice, procedure, or test, then the employer must prove it is job-related.
Adverse impact comprises only the first of three burdens shared between plaintiffs and defense in a disparate impact case. A discussions of the two remaining burdens follows.
Job Relatedness (Business Necessity) - Second Burden
After adverse impact has been proven by the plaintiff, the employer must prove that the practice, procedure, or test causing the adverse impact is job-related and consistent with business necessity. If the different practices, procedures, and tests within the selection procedure cannot be evaluated separately, then the entire selection process must be evaluated as one.
The 1991 Civil Rights Act defines the concepts of "job related" and "business necessity" as described in Griggs v. Duke Power Co. and other Supreme Court Decisions prior to Wards Cove. In Griggs, the Supreme Court stated the requirement with these words:
The Act proscribes not only overt discrimination but also practices that are fair in form, but discriminatory in operation. The touchstone is business necessity. If an employment practice which operates to exclude Negroes cannot be shown to be related to job performance, the practice is prohibited....Congress has placed on the employer the burden of showing that any given requirement must have a manifest relationship to the employment in question.
In 1978 the Uniform Guidelines on Employee Selection Procedures were published. The Guidelines "are intended to establish a uniform Federal position in the area of prohibiting discrimination in employment practices on grounds of race, color, religion, sex, or national origin."
The Uniform Guidelines melds the concepts of "job relatedness" and "business necessity" into a validation requirement. "If adverse impact exists, it must be justified on grounds of business necessity. Normally, this means by validation which demonstrates the relation between the selection procedure and performance on the job." The Uniform Guidelines provide three methods of validation: criterion-related, content, and construct. Criterion-related validity is demonstrated "by empirical data showing that the selection procedure is predictive of or significantly correlated with important elements of work behavior." Content validity is demonstrated "by data showing that the content of a selection procedure is representative of important aspects of performance on the job." Construct validity is demonstrated "by data showing that the selection procedure measures the degree to which candidates have identifiable characteristics which have been determined as significant for successful job performance."
Content validity is the technique used most frequently by employers attempting to defend their practices as job related under the Guidelines. Decisions that are reported serve to be very helpful in identifying the points courts want covered to successfully defend a disparate impact challenge. See Contreras v. City of Los Angeles. However, unreported decisions also address important points courts want covered to successfully defend a disparate impact challenge. See Martinez v. City of Salinas, Parks v. City of Long Beach, and Simmons v. Kansas City, Kansas. The points repeated in these decisions are a need for a job analysis that identifies the knowledge, skills, and abilities (KSAs) needed for important or critical work behaviors, a link-up made of the KSAs to important or critical work behaviors, and the selection tool measuring only important or critical KSAs. Different types of content validity systems have been used successfully. The Contreras case reports a system developed and used by the City of Los Angeles. The Martinez, Parks, and Simmons cases all used the GOJA (Guidelines Oriented Job Analysis) system.
Alternate Employment Practice - Third Burden
After the employer proves job relatedness and business necessity through a criterion-related, content, or construct validity study, the plaintiff has an opportunity to prove that an alternate employment practice was available to the employer. If the plaintiff can prove an "alternate employment practice" was available to the employer, and the employer did not use it, the plaintiff can prevail. Otherwise, the plaintiff cannot prevail under these circumstances.
Uniform Guidelines Alternate Employment Practice Requirement
The Uniform Guidelines in Section 3B requires:
Where two or more selection procedures are available which serve the user's legitimate interest in efficient and trustworthy workmanship, and which are substantially equally valid for a given purpose, the user should use the procedure which has been demonstrated to have the lesser adverse impact. Accordingly, whenever a validity study is called for by these guidelines, the user should include, as a part of the validity study, an investigation of suitable alternative selection procedures and suitable alternative methods of using the selection procedures which have as little adverse impact as possible, to determine the appropriateness of using or validating them in accord with these guidelines. If a user has made a reasonable effort to become aware of such alternative procedures and validity has been demonstrated in accord with these guidelines, the use of the test or other selection procedure may continue until such time as it should reasonably be reviewed for currency. Whenever the user is shown an alternative selection procedure with evidence of less adverse impact and substantial evidence of validity for the same job in similar circumstances, the user should investigate it to determine the appropriateness of using or validating it in accord with these guidelines. This subsection is not intended to preclude the combination of procedures into a significantly more valid procedure, if the use of such a combination has been shown to be in compliance with the guidelines.
Suitable alternative selection procedures and suitable alternative methods of using the selection procedures (which serve the user's legitimate interest in efficient and trustworthy workmanship, and which are substantially equally valid for a given purpose) are the two critical features of the alternate employment practice requirement, according to the Uniform Guidelines.
The 1991 Civil Rights Act Alternate Employment Practice Requirement
The 1991 Civil Rights Act states that evidence necessary for the alternate employment practice requirement "...shall be in accordance with the law as it existed on June 4, 1989, with respect to the concept of 'alternative employment practice.'" Wards Cove, was published on June 5, 1989. It appears that Congress was specifically refuting the Wards Cove interpretation of the "alternate employment practice." Therefore, it is important to know what Wards Cove mandates, and what Congress may have refuted via the 1991 Civil Rights Act, in order to know what not to rely upon for an interpretation. Wards Cove says in relevant part:
Of course, any alternative practices which respondents offer up in this respect must be equally effective as petitioners' chosen hiring procedures in achieving petitioners' legitimate employment goals. Moreover, "[f]actors such as the cost or other burdens of proposed alternative selection devices are relevant in determining whether they would be equally as effective as the challenged practice in serving the employer's legitimate business goals." Watson v. Fort Worth Bank and Trust, 487 U.S.--- (1988) at ---(O'CONNOR, J.). "Courts are generally less competent than employers to restructure business practices," Furnco Construction Corp. v. Waters, 438 U.S. 567, 578 (1978); consequently, the judiciary should proceed with care before mandating that an employer must adopt a plaintiff's alternative selection or hiring practice in response to a Title VII suit.
The 1991 Civil Rights Act has told EEO practitioners not to include the June 5, 1989 Wards Cove interpretation of the "alternate employment requirement" in our work. Rather, we are instructed by Section 105(b)of the 1991 Civil Rights Act to rely upon part of the Congressional Record:
No statements other than the interpretive memorandum appearing at Vol. 137 Congressional Record S 15276 (daily ed. Oct. 25, 1991) should be considered legislative history of, or relied upon in any way as legislative history in construing or applying, any provision of this Act that relates to Wards Cove -- Business necessity/cumulation/ alternative business practice.
The entire Interpretive Memorandum from the Congressional Record (S 15276), October 25, 1991 states:
The final compromise on S. 1745 agreed to by several Senate sponsors, including Senators DANFORTH, KENNEDY, and DOLE, and the Administration states that with respect to Wards Cove--Business necessity/cumulation/ alternative business practice--the exclusive legislative history is as follows:
The terms "business necessity" and "job related" are intended to reflect the concepts enunciated by the Supreme Court in Griggs v. Duke Power Co., 401 U.S. 424 (1971), and in the other Supreme Court decisions prior to Wards Cove Packing Co. v. Atonio, 490 U.S. 642 (1989)
When a decision-making process includes particular, functionally-integrated practices which are components of the same criterion, standard, method of administration, or test, such as the height and weight requirements designed to measure strength in Dothard v. Rawlinson, 433 U.S. 321 (1977) the particular, functionally-integrated practices may be analyzed as one employment practice.--Interpretive Memorandum, Cong. Rec. (S 15276), October 25, 1991.
Elsewhere in the Congressional Record it states:
Our intention with respect to the "alternative practices" issue is reflected at S15276 of the Congressional Record on October 25, 1991.--Sponsors' Interpretive Memorandum, Cong. Rec. (S 15484), October 30, 1991.
The alternative business practice did not receive much amplification in these interpretive memoranda. However, Kennedy did put into the Congressional Record some interpretive words on alternative practices:
Even if the respondent proves business necessity, the challenged practice or process is unlawful if the complaining party demonstrates that a different employment practice with less disparate impact exists, and the respondent fails to adopt the alternative employment practice. The bill restores the law regarding the demonstration of alternative business practices to its status before June 4, 1989.
Once the employer fails to adopt such an alternative practice, the employer cannot escape liability under this "third-prong" by adopting the practice at a later time, such as during the trial of the disparate impact claim.--Cong. Rec. (S 15234), October 25, 1991).
The Wards Cove decision, published on June 5, 1989, places emphasis on "equally effective" and "cost or other burdens" on the proposed alternative employment devices. The 1991 Civil Rights Act puts the alternative employment practices burden back "as it existed on June 4, 1989, with respect to the concept of 'alternate employment practice.'" In addition to ignoring the Wards Cove words of interpretation, the 1991 Civil Rights Act further states that we can rely on the Congressional Record in relevant part which states that the burden is showing "a different employment practice with less disparate impact..." The difference in emphasis is clear. It is much easier for a plaintiff to present an alternative employment practice with less disparate impact than it is for a plaintiff to present an alternative employment practice that is equally effective and is cost effective as well.
Not many cases have made it to the Alternate Employment Practice burden in court. Not many studies have been conducted comparing the degree of adverse impact for similar tests. In the early 1980's, the City of Salinas presented a comparison at trial of different reading tests with varying degrees of adverse impact. The one selected by the City of Salinas had the least amount of adverse impact and was found to be valid for entry-level firefighters. In another case decided by a three judge panel, the judges found that the state had proven the testing practices were a business necessity in certifying public school teachers, and the only proposed alternative (graduation from an approved program) for certification purposes was found inadequate to achieve the state's purpose in certifying minimally competent teachers. Perhaps this will be a new frontier in EEO.
Summary of Disparate Impact
In disparate impact analyses, two kinds of statistical distributions are applied: the hypergeometric and the binomial. Each distribution produces a standard deviation and a probability used to determine statistical significance. After statistical significance is shown, practical significance needs to be shown.
Plaintiffs usually attempt to aggregate data for analysis purposes (i.e., combining race/ethnic groups, combining years of data, combining similar jobs, etc.) to increase the chance of finding statistically and practically significant results.
The statistical distributions are applied to three disparate impact analyses: the "threshold" analysis, the "barriers" analysis, and the "selection" analysis. The "threshold" analysis is calculated to see if a protected group is underutilized on a job or group of jobs. The "barriers" analysis is calculated to see if there have been any barriers in the application process. If a barrier is detected, then a finding of disparate impact can be expected in regard to the practice, procedure, or test causing the barrier. In addition, a "proxy group" (usually availability) is used in the "selection" analysis rather than the actual applicants. The "selection" analysis is conducted on practices, procedures, or tests to see if there is an adverse impact on a protected group. In situations where the practices, procedures, or tests cannot be analyzed separately, the entire selection process is analyzed as one.
After adverse impact is shown, the employer has an opportunity to prove validity (i.e., job relatedness and business necessity), usually through a validation study. If the employer can prove validity, the plaintiff has an opportunity to demonstrate that another practice, procedure, or test is available that will suit the employer's legitimate needs -- or that another use of the practice, procedure or test will be substantially equally valid for the employer with less adverse impact.
*Richard Biddle is President of Biddle & Associates, Inc., a Sacramento-based EEO consulting and software firm. Mr. Biddle's practice concentrates on litigation support in the areas of statistical analyses, job analysis, validation of practices, procedures, and tests, affirmative action, and expert witness work. He has been involved in over 100 cases.