Experts Discuss Big Data’s Effect on Hiring, Bias Claims

The HR & Payroll Resource Center is your integrated, comprehensive source for HR and Payroll information that merges news, analysis, and guidance – including custom answers,...

By Lydell C. Bridgeford

August 3— “Big data” analytics will increasingly affect talent acquisition decisions and possibly employment discrimination claims under Title VII of the 1964 Civil Rights Act, a panelist said July 30 at the National Industry Liaison Group's annual conference in New York.

Talent acquisition, a key business imperative, is driving “big data” analytics, and “the need for talent will only get stronger in a global, competitive marketplace,” said attorney Heather Morgan, a partner at Paul Hastings in Los Angeles.

Morgan participated in an interdisciplinary panel examining big data analytics’ growing influence on talent acquisition, their potential impact on disparate treatment and impact claims under Title VII, and the Labor Department Office of Federal Contract Compliance Programs’ Internet applicant rule.

Some employers are venturing into big data analytics to assist them in making hiring and promotion decisions. In addition, advances in online technology mean employers want to take advantage of data from large applicant pools.

“We are in the early stages of using big data analytics, and it’s exciting. Yet, there are many unknowns, which is kind of scary. We don’t have a lot of the answers, and the law is not out there yet. Court cases will help us answer questions that will emerge,” Morgan said.

Rick Holt, a principal at ERS Group, prefers one of the original definitions of the term “big data”—data that originated from many different sources, contained a wide variety of information and arrived at a fast pace.

“Over time, the notion of big data has morphed to include the analytic tools and methods that are used to examine the data,” Holt, a labor economist, said. It may be time to “retire the term big data and come up with something that’s a little more precise, because the term has become abused and comically broad,” he added.

Computers Select Employees

The human resources technology vendors are offering software products that contain “big data analytics” to assist companies with assessing the quality of their sourcing and selection practices and strategies on talent acquisition, according to Morgan. The software vendors sell proprietary computer algorithms designed to predict the best candidates for job openings and promotions.

The data points selected by the vendors to build the predictive analytic programs include publicly available information from social media profiles and activity, and other information available through the Internet, as well as the employer’s information about its workforce, such as performance evaluations and test results.

For example, software has been created to sift through social media information to find data on certain individuals, Morgan said. “That information becomes part of the algorithm used to find job candidates,” she added. Employers are looking for the most effective, efficient and accurate way of finding qualified people for their organizations, she said.

Applicant tracking systems are not new, but they are being used in innovative ways with respect to big data analytics, Morgan said. For instance, some systems allow employers new and robust ways of doing workplace reporting, such as linking data and information from the written affirmative action program to the company’s diversity objectives reports, she said.

The idea is to use the information from the applicant tracking system to get a better sense of how the organization can obtain top-notch talent, Morgan added. Some employers may have an applicant tracking system that feeds into its human resources information systems (HRIS), which provides the company with the “whole picture from start to end,” she said.

An industrial psychologist on the panel, however, advised employers that just because “you can use” big data analytics “doesn’t mean you should.”

Before an employer decides to use predictive analytics to identify potential job candidates, it first has to establish “a theory underpinning what it’s looking for regarding the position and how that connects to the person’s ability to perform the job,” said Kathleen Lundquist, president and chief executive officer at APT Metrics.

Holt agreed with Lundquist that employers need to develop a clear theory about their hiring objectives before embarking on big data solutions. He also urged employers to make sure they have “clean and good” data that have been vetted for accuracy. “Integrating data from several sources into one unified data base is a huge challenge,” he added.

Big data analytics are “just a tool and vehicle for gathering information about job applicants. As a vehicle, it is only as good as the questions you ask and the way you implement it,” said Lundquist. “The more structured and standardized the tool the better, because you will get reliability, validity and possibly defensibility,” she added.

The panelists reminded federal contractors that the assessments, outcomes and products generated from the big data analytics must meet the validation standards on job relatedness and business necessity under the Uniform Guidelines on Employee Selection Procedures (UGESP), which the OFCCP codified at 41 C.F.R. § 60-3.

The OFCCP’s Internet applicant rule “has failed to keep up with the times, given that employers have to react quickly on talent acquisition,” said Jon Geier, a partner with Paul Hasting in Chicago.

In 2005, the agency issued the Internet applicant rule (41 C.F.R. §§60-1.3 and 1.12), addressing recordkeeping requirements for individuals who submit resumes or applications via the Internet to federal contractors.

Under the rule, job seekers must meet a four-prong definition of “Internet applicant.” Those criteria include the following: the individual submits an expression of interest in employment through the Internet or related technologies; the contractor considers the individual for a particular position; the individual's expression of interest indicates he or she has basic qualifications for the position; and the individual does not remove himself or herself from the further consideration or otherwise indicate he or she is no longer interested in the position.

The agency created the Internet applicant rule to try to meet the technological advances that had made it difficult for the OFCCP to pursue violations based on applicant data, Geier said. The agency also realized, going into 2006, that the definition of an applicant had become outdated.

Back then, LinkedIn was embryonic and other online and social network sites had not yet gained in popularity among job seekers, Geier said. “For the better or worse, the speed within technology has changed” even more rapidly since the 2005 Internet applicant rule.

“Individuals can now rapidly apply for jobs on their mobile devices,” while employers can examine large data sets of potential job applicants’ qualifications profiled on public websites, he said. According to some federal contractors, the 2005 Internet applicant rule fails to meet the needs and demands of businesses today. “It impedes legitimate business needs and imposes unrealistic recordkeeping obligations,” Geier said.

Analytics based on big data may reduce adverse impact and the recordkeeping obligations under the OFCCP’s Internet applicant rule, said Nathaniel Glasser, a partner at Epstein, Becker & Green in Washington.

Employers may want to take advantage of big data during the sourcing stage for talent in which individuals aren’t officially “applicants,” thus avoiding the Internet applicant rule and reducing adverse impact risk, according to Glasser. Recruiting practices aren’t selection procedures under UGESP, he said.

The results and tools from big data research can be adopted early in the hiring process to quickly weed out job candidates who would not qualify for the positions due to the basic job requirements, he said.

“You can reduce the number of candidates that you are originally screening. Big numbers are bad numbers in terms of statistical analyses for adverse impact. When you have a large data set to work with, it’s a lot easier to find statistical significance,” Glasser said.

During the selection process, the predictive analysis and models can be utilized to limit the subjectivity in the selection process, he said. This makes it a level playing field for all candidates, which “can lead to some good results and protect you against claims of unintentional discrimination,” Glasser said. Still, he advised employers to review predictive analytics and models for encoded biases.

Using big data, however, may trigger disparate treatment claims when the analytics and selection procedures are used for one protected group but not others, Glasser noted. Relying on the predictive criterion also may produce fewer candidates in a protected group, he said.

Entry-Level Bias Claims

Meanwhile, an analysis of OFCCP data for fiscal year 2014 shows discrimination in hiring for entry-level jobs remains the top allegation in agency conciliation agreements reached with federal contractors, consultant David Cohen said July 30.

Speaking at the NILG conference, Cohen said despite the OFCCP's emphasis on reviewing contractors' compensation practices, settlements alleging pay discrimination still represent only a small portion of the OFCCP's enforcement record, both in terms of conciliation agreements reached and monetary remedies obtained.

During fiscal 2014, the OFCCP settled 34 hiring discrimination cases for a total of about $9.4 million while settling 11 compensation cases for a total of about $1.2 million, said Cohen, president of DCI Consulting Inc. in Washington, which advises federal contractors on compliance with Executive Order 11,246, Section 503 of the Rehabilitation Act and the Vietnam Era Veterans' Readjustment Assistance Act.

So far in fiscal 2015, the OFCCP has reached 17 settlements with contractors containing financial remedies, including three conciliation agreements involving alleged pay discrimination, Cohen said.

It's striking that the agency has closed only 1,606 cases so far in fiscal 2015 and settled relatively few, as the OFCCP historically reaches 60 to 90 conciliation agreements per year, Cohen said. The agency will have to pick up its pace to reach that total in fiscal 2015, which ends Sept. 30, he said.

Compliance officers may be “overwhelmed” with implementing the amended regulations under the Rehabilitation Act and VEVRAA, as well as the new scheduling letter, especially Item 19 on compensation, Cohen said. The agency now is getting a lot more “raw data” on compensation from contractors and it might be a case of “be careful what you ask for,” he continued, adding, “I don't think [the OFCCP] knows what to do” with the information collected under Item 19.

The OFCCP's settlement data for fiscal years 2007 through 2014 show about 90 percent of settlement dollars come from cases alleging hiring discrimination, Cohen said. Such cases are relatively easy to pursue because of the wealth of applicant flow data, the “discrete” event of failure to hire being assessed and the large numbers often permitting the OFCCP to find statistical significance when comparing the employment of different races, genders and ethnic groups, he said.

In contrast, building a pay discrimination case is complicated, as hundreds of factors may contribute to a contractor's compensation decisions, data may not be so readily available and comparisons between employees aren't as easy to make, Cohen said.

The OFCCP in recent years has ramped up its focus on compensation during compliance reviews, spearheading the Obama administration's drive to eliminate a gender pay gap that has women on average earning about 77 cents for every dollar earned by male workers. The comparative figures for black and Hispanic women compared to white male workers are even worse.

To contact the reporters on this story: Lydell C. Bridgeford in Washington at and Kevin McGowan in New York at

To contact the editors responsible for this story: Heather Bodell at and Susan J. McGolrick at