MANA 4328 Dr. Jeanne Michalski michalski@uta.edu Combining Test Data MANA 4328 Dr. Jeanne Michalski michalski@uta.edu 1
Selection Decisions First, how to deal with multiple predictors? Second, how to make a final decision?
Developing a Hiring System OK, Enough Assessing: Who Do We Hire??!!
Interpreting Test Scores Norm-referenced scores Test scores are compared to applicants or comparison group. Raw scores should be converted to Z scores or percentiles Use “rank ordering” Criterion-referenced scores Test scores indicate a degree of competency NOT compared to other applicants Typically scored as “qualified” vs. “not qualified” Use “cut-off scores”
Setting Cutoff Scores Based on the percentage of applicants you need to hire (yield ratio). “Thorndike’s predicted yield” You need 5 warehouse clerks and expect 50 to apply. 5 / 50 = .10 (10%) means 90% of applicants rejected Cutoff Score set at 90th percentile Z score 1 = 84th percentile Based on a minimum proficiency score Based on validation study linked to job analysis Incorporates SEM (validity and reliability)
Selection Outcomes PERFORMANCE No Pass Pass PREDICTION Regression Line Cut Score PERFORMANCE 90% Percentile No Pass Pass PREDICTION
Selection Outcomes PERFORMANCE Type 1 Error False Negative True Positive High Performer Type 2 Error False Positive True Negative Low Performer No Hire Hire PREDICTION
Selection Outcomes PERFORMANCE High Performer Low Performer Prediction Line Cut Score High Performer Low Performer Unqualified Qualified PREDICTION
Dealing With Multiple Predictors “Mechanical” techniques superior to judgment Combine predictors Compensatory or “test assessment approach” Judge each independently Multiple Hurdles / Multiple Cutoff Profile Matching Hybrid selection systems
Compensatory Methods Unit weighting Rational weighting Ranking P1 + P2 + P3 + P4 = Score Rational weighting (.10) P1 + (.30) P2 + (.40) P3 + (.20) P4 = Score Ranking RankP1 + RankP2 +RankP3 + RankP4 = Score Profile Matching D2 = Σ (P(ideal) – P(applicant))2
Multiple Regression Approach Predicted Job perf = a + b1x1 + b2x2 + b3x3 x = predictors; b = optimal weight Issues: Compensatory: assumes high scores on one predictor compensate for low scores on another Assumes linear relationship between predictor scores and job performance (i.e., “more is better”)
Multiple Cutoff Approach Sets minimum scores on each predictor Issues Assumes non-linear relationship between predictors and job performance Assumes predictors are non-compensatory How do you set the cutoff scores?
Multiple Cutoff Approach Sets minimum scores on each predictor Issues Assumes non-linear relationship between predictors and job performance Assumes predictors are non-compensatory How do you set the cutoff scores? If applicant fails first cutoff, why continue?
Multiple Hurdle Model Finalist Decision Background Test 1 Test 2 Interview Pass Pass Pass Pass Fail Fail Fail Fail Reject
Profile Matching Approach Emphasizes “ideal” level of KSA e.g., too little attention to detail may produce sloppy work; too much may represent compulsiveness Issues Non-compensatory Small errors in profile can add up to big mistake in overall score Little evidence that it works better
Making Finalist Decisions Top-Down Strategy Maximizes efficiency, but may need to look at adverse impact issues Banding Strategy Creates “bands” of scores that are statistically equivalent (based on reliability) Then hire from within bands either randomly or based on other factors (inc. diversity)
Banding Grouping like test scores together Standard Error of Measure Function of test reliability Standard Error of Measure Band of + or – 2 SEM 95% Confidence interval If the top score on a test is 95 and SEM is 2, then scores between 95 and 91 should be banded together.
Combined Selection Model Selection Stage Selection Test Decision Model Applicants Candidates Application Blank Minimum Qualification Hurdle Candidates Finalists Four Ability Tests Work Sample Rational Weighting Finalists Offers Structured Interview Unit Weighting Rank Order Offers Hires Drug Screen Final Interview
Alternative Approach Rate each attribute on each tool Desirable Acceptable Unacceptable Develop a composite rating for each attribute Combining scores from multiple assessors Combining scores across different tools A “judgmental synthesis” of data Use composite ratings to make final decisions
Categorical Decision Approach Eliminate applicants with unacceptable qualifications Then hire candidates with as many desirable ratings as possible Finally, hire as needed from applicants with “acceptable” ratings Optional: “weight” attributes by importance
Sample Decision Table
More Positions than Applicants
More Applicants than Positions