If OFCCP’s proposed revised scheduling letter is approved anytime in 2012, it will ask you to submit your 2011 AAP data. If your AAP year tracks the calendar year, this is the data you are reviewing now.
From the time you receive a scheduling letter, you have 30 days to submit your written affirmative action program along with the documents specified in the enclosed itemized listing. OFCCP has proposed few changes to the letter itself. The most important changes are to the itemized listing. Perhaps even more important are the ways in which the emphasis on this data will change the desk audit itself.
Under the current scheduling letter, hiring data is submitted by job group and by Minority/Non-Minority status and gender. At the desk audit, OFCCP enters this data into an Impact Ratio Analysis (IRA) spreadsheet which calculates disparities in selection rates. Selection rates that are less than 80% of the selection rate of the favored group (referred to as violating the 80% rule) and/or selection disparities where the difference in selection rates is statistically significant at a level of 2 or more standard deviations (SDs) are considered indicators of potential discrimination.
In most cases, OFCCP uses 2 SDs as the trigger prompting further examination of the selection procedure. If there is no indicator of statistically significant disparities in any job group, the agency typically does not further refine the data and the review generally is limited to technical compliance issues rather than possible discrimination issues.
The IRA calculation compares two groups at a time. It can compare, for example, men and women or minorities as a group with non-minorities as a group. Since the current screening calculations are by gender and minority group, they are well suited to the IRA.
The proposed revised scheduling letter requires the submission of job title and race/ethnicity level data. In the past, even if the agency had the race/ethnicity and title level data, it may not have reviewed it if your Minority/Non-Minority IRAs produced no indicators at the job group level. You can expect this to change. Specifically, requesting race/ethnicity data and job title data will have at least two predictable effects. It will supply OFCCP with more data to review where contractors may not have previously provided such data, and it will put significantly more pressure on the field offices to examine this data at the desk audit for all contractors, not just those that spring an indicator at the job group level by Minority/Non-Minority calculations. If the agency does not regularly use this data, it will be difficult to justify its routine collection.
One reason OFCCP wants job title data is to eliminate the masking of indicators that can occur with job group data. For example, in the chart below there is no indicator at the Laborer Job Group level. The numbers in the chart would not be considered a trigger for further review of the selection procedures under the current desk audit approach. The impact ratio (IR) does not violate the 80% rule and there is no statistically significant standard deviation. The shortfall would not be pursued in the absence of an indicator.
Laborer Job Group | Non-Minority | Minority | IR | SD | Shortfall |
Applicants | 200 | 200 | |||
Hires | 115 | 105 | |||
91 | 5 |
Under the proposed scheduling letter, the hiring information for the Laborer Job Group would also be submitted at the job title level. Suppose this same Laborer Job Group is comprised of a Laborer 1 and a Laborer 2 title. Examined by title, an indicator surfaces in the job title Laborer 1, as shown in the chart below.
Laborer 1 Title | Non-Minority | Minority | IR | SD | Shortfall |
Applicants | 100 | 100 | |||
Hires | 70 | 50 | |||
71% | 2.887 | 10 | |||
Laborer 2 Title | |||||
Applicants | 100 | 100 | |||
Hires | 45 | 55 | |||
1.22 |
By requiring the data by job title, this masking is eliminated.
Race/ethnicity data can also eliminate masking. For example, the fact that a contractor favors a particular minority group over another minority group would be masked in a Minority/Non-Minority IRA. Let’s revisit the Laborer 2 job title which did not spring any indicator at the job title level in the Non-Minority/Minority analysis in the last chart above.
Race/ethnicity data shows us that of the 100 minority applicants for Laborer 2, 50 were Hispanic and 50 Asian. From that pool 45 Asian applicants were hired compared to only 10 Hispanic applicants. Without the race/ethnicity data at the title level, this issue would not likely have been found or investigated by you or OFCCP. With this data, OFCCP indicators are triggered.
Laborer 2 Title | Asian | Hispanic | IR | SD | Shortfall |
Applicants | 50 | 50 | |||
Hires | 45 | 10 | |||
22% | 7.035 | 17 |
As is rather apparent from these calculations, running race/ethnicity and gender calculations at the job title level multiplies the number of equations that can be used to identify potential discrimination indicators to pursue.
The race/ethnicity/job title IRA can also be run to compare the selection rates of non-minorities and Asians in the Laborer 2 job title. This IRA will give you an indicator favoring Asians and disfavoring non-minorities.
Laborer 2 Title | Non-Minority | Asian | IR | SD | Shortfall |
Applicants | 100 | 50 | |||
Hires | 45 | 45 | |||
.50 | 5.303 | 15 |
Finally, if you do this IRA for Non-Minority and Hispanic applicants, you will get statistical significance favoring non-minorities.
Laborer 2 Title | Non-Minority | Hispanic | IR | SD | Shortfall |
Applicants | 100 | 50 | |||
Hires | 45 | 10 | |||
44% | 2.995 | 8 |
Any indicator over 2 SDs at the desk audit will trigger more in-depth investigation by OFCCP. Keep in mind these indicators are based on the IRA which has its limitations. More analysis is required to know what these numbers really mean.
It is important when reviewing your data that you understand the limitations of the IRA. As you can see, you can run a variety of IRAs for the Laborer 2 title but they have to be done in pairs (Asian/Hispanic, Non-Minority /Asian, Non-Minority/Hispanic), but it cannot run all three groups at the same time; for that, you need another formula. At OFCCP, a Chi-Square formula is used to calculate the fairness of the selection rates among more than two race/ethnic groups.
A Chi-Square can include all racial/ethnic groups in the equation. In our Laborer 2 title example, you would enter the applicant/hire figures for non-minorities, Asians and Hispanics. The Chi-Square will show the shortfall for each as well as whether the equation is statistically significant. If the hiring figures from the Laborer 2 title above are put in a Chi-Square, it shows that there were 19 more Asian applicants hired than would have been expected, 5 fewer Non-Minority hires than would have been expected and 15 fewer Hispanic hires than would have been expected.
Laborer 2 | Non-Minority | Asian | Hispanic | Total |
Applicants | 100 | 50 | 50 | 200 |
Hires | 45 | 45 | 10 | 100 |
Selection rate % applicants | 45% | 90% | 20% | |
Expected Hires | 50 | 25 | 25 | 100 |
Expected Non-Hires | 50 | 25 | 25 | 100 |
Shortfall | 5 | -19 | 15 | |
Degrees of Freedom 2 | ||||
Chi-Square Value 51 | ||||
p-value 0 (this is statistically significant) |
This is a very different picture than you might get simply by running a series of IRAs. Where the Non-Minority/Hispanic IRA suggests discrimination against Hispanic applicants in favor of Non-Minority applicants, the Chi-Square shows that the only group hired above expected levels are the Asian applicants. Both OFCCP compliance officers and your compliance staff will have to be careful to properly analyze the actual hiring problem. While it is true that race/ethnicity data at the title level can eliminate some masking problems, because of the variety of equations and interpretations, it can also increase the opportunities for false positives or misleading calculations.
There may be other ways to properly calculate the expected distribution of jobs when there are multiple race and ethnicity groups in a single selection process that would work just as well or better. What does not work is simply adding up shortfalls from a series of IRAs or drawing conclusions from an individual IRA that ignores the fair distribution expected of the entire pool.
Job title information can also disclose steering. If you hire laborers from the same pool and do not make actual job title assignments until after hire, job group level data will only show if expected levels of hires were made at the job group level. Job title information can show if, for example, women are being steered to different and lesser jobs as a result of being assigned to a particular job title. If you hire by a generic job group and assign specific roles after hire, you will need to examine your 2011 data to ensure that this process did not put any protected group at a disadvantage in terms of earning or promotion potential or any other desirable job characteristic.
It is also important to know what to do about applicants who identify as more than one race/ethnic group. Initially, OFCCP will include those who identify as two or more races/ethnicities in the minority figures of a Non-Minority/Minority IRA. When refining by specific races/ethnicities, individuals who identify as two or more races, but do not specify which races/ethnicities, are put to the side and not included in the race specific calculations. I have seen this group handled as a separate category in the Chi-Square.
Individuals who identify as both Non-Minority and a specified minority group are usually included in the specific minority group. If they identified membership in more than one minority group and both are disfavored in the hiring process, they may be counted in either or both minority groups but will only receive a single remedy.
If there is no race/ethnicity data for certain applicants, they also may be left out of the IRA or included as a separate group in the Chi-Square for purposes of the analysis, but generally not for purposes of liability or remedy. If there are substantial numbers of unidentified applicants, the agency may attempt to calculate the racial distribution either using zip code demographics or other proxies. None of these proxies have proven particularly satisfactory; however, you should be aware that failing to obtain race/ethnicity data does not necessarily mean that the agency will not attempt to obtain this data or a reasonable proxy.
On the upside, the proposed new scheduling letter’s request for race/ethnicity and gender data at the job title level has the potential to unmask discrimination indicators that may not have surfaced under the current desk audit procedures. On the challenging side, you may find it extremely difficult to review your 2011 data through this lens if you do not already have systems in place to do so. You may have thousands of job titles and very diverse applicant pools and if your hiring picked up across those pools, you have your work cut out for you. As you go about it, make sure that you understand what your data is really saying and whether the right analysis is being used. I also suggest that you track how long it really is taking you to conduct the analysis at this level. It will probably be more than the predicted burden hours in the proposed scheduling letter. The proper cost benefit analysis for this approach can only be meaningfully measured if accurate data is maintained comparing the increase in and reliability of indicators with the increase in effort and burden hours.