Your Data Skills Test Should Look Like the Real Job

The SQL test nobody checked

Many Oklahoma employers now ask data candidates to take a test. It might be a timed SQL quiz, a Power BI exercise or a take home case study. Done well, a test shows who can actually do the work.

Often the test was written years ago by someone who has since left. Nobody has compared it to the job in a while. Nobody knows who passes it and who does not.

What the evidence says

Title VII of the Civil Rights Act allows employment tests. But if a test screens out people of one race, sex, religion or national origin more than others, the employer must show that it is job related for the position and consistent with business necessity. Even then, the employer can be liable if a less harmful option would work just as well and it refuses to use it. Title VII covers employers with 15 or more employees.

The federal Uniform Guidelines on Employee Selection Procedures give a rule of thumb. If one group's pass rate is less than four fifths of the rate for the group that passes most often, federal agencies generally treat that as evidence of adverse impact. For example, if 50 percent of one group passes and 30 percent of another passes, the lower rate is only 60 percent of the higher one.

The same guidelines say the closer a test is to the actual work, the stronger the case that it is valid. Time limits should match how fast the work must be done on the job. A test vendor's reputation or sales material is not proof that a test works, and the EEOC says the employer is still responsible for its tests.

The EEOC has shown what happens when a test drifts from the job. In its case against Dial, women made up 46 percent of hires before a strength test and 15 percent after. Expert testimony showed the test was much harder than the job, and the court found it violated Title VII. In another case, Ford paid $8.55 million and agreed to replace a written test for an apprenticeship program.

The Americans with Disabilities Act adds two duties. A test must measure the skill it claims to measure, not a person's disability. And employers must make reasonable accommodations in how tests are given, unless that would cause undue hardship.

Oklahoma law reaches further down. The Oklahoma Anti Discrimination Act defines an employer as any organization that pays one or more people. It bans classifying applicants in a way that tends to deprive them of opportunities because of race, color, religion, sex, national origin, age, genetic information or disability.

In April 2025, Executive Order 14281 told federal agencies to put less priority on enforcing disparate impact rules, including under Title VII. An order sets agency priorities. It does not rewrite the statute, and priorities can change with each administration.

Why data skills tests go wrong

The test is harder than the job

A puzzle full of rare SQL tricks may look rigorous. If the analyst will mostly join a few tables and build monthly reports, the puzzle measures something the job does not need.

The clock is tighter than the work

Real reports are built over hours or days, with time to check the numbers. A 20 minute timer rewards speed that the job may never ask for.

It tests what a new hire learns in a week

Questions about your own table names, your report layout or one tool version test whether someone has worked at your company before. The federal guidelines warn against basing decisions on skills that are normally learned in a brief orientation.

Not everyone gets the same test

If one manager gives a hard case study and another skips the test for a referral, the results cannot be compared. The guidelines expect tests to be given and scored under the same conditions for everyone.

How to build a fair data skills test

1. 1Write down the five or six tasks the person will do most in the first six months.
2. 2Build the test from those tasks, using sample data that looks like yours but holds no real customer records.
3. 3Set the time limit by timing someone already doing the job, then add room.
4. 4Give every candidate for the role the same test, the same instructions and the same scoring guide.
5. 5Write the scoring guide before anyone takes the test, and have two people score the work when you can.
6. 6Tell candidates up front how to ask for an accommodation, and grant reasonable requests.
7. 7Keep records of who took the test and who passed, and check the pass rates at least once a year.
8. 8Give people a fair chance to retest, and update the test when the job changes.

What this means if you hire data people in Oklahoma

A good test helps you and the candidate. It shows the real work, it gives everyone the same shot and it makes your choice easier to explain. A test that nobody has checked can turn away the people you most want to hire.

At Oklahoma Data Partners, we recruit only for data roles and only in Oklahoma. We have been data hiring managers ourselves, so we help employers build skills checks that match the real work, then bring them carefully vetted candidates who can do it. If you want a fresh look at your hiring test, book a consultation.