Work Sample

Types of Work Sample Tests


Work-Sample Tests of Trainability These are tests through a period of instruction when the applicant is expected to learn tasks involved in a work sample. The work-sample tests of trainability are suitable for untrained applicants with no previous job experience. The predictive validity of this technique is low relative to other techniques and there is evidence the validity of the instrument may attenuate over time.

Simulation of an Event These tests present the candidate with a picture of an incident along with quotations from those involved. The candidates then respond to a series of questions in which they write down the decisions they would make. The test is scored by subject matter experts.


Low Fidelity Simulations These tests present applicants with descriptions of work situations and five alternative responses for each situation. Applicants choose the responses they would most likely and least likely make in each situation.

Work-samples Applicants perform observable, job-related behaviors as predictors of criterion performance. It is not feasible to adapt certain work behaviors for testing. Work samples often are not conducive to group administration and, therefore, were dropped from consideration because of concerns regarding test security.

Validating Work Sample Tests


Content Validity The most direct relationship between the test and job would be shown through content validation. The tasks and duties performed on the test would be compared to the tasks and duties performed on the job. The test should encompass significant (in quantity or in importance) tasks/duties of the job.

Criterion Validity To measure this validity, you must first determine what criteria will be used. Two common forms of criteria are:

* Supervisory ratings of the incumbent's job performance. The disadvantage of using supervisory ratings as criteria is that they typically lack sufficient reliability to be used for statistical analysis. The reliability of these measures is attenuated by rater errors such as 'halo' or 'leniency'. These ratings alto tend to lack the variability necessary to show a correlation between predictor and criterion

* Production measures such as quantity or quality of work. Production measures are not available for some jobs.