How the Kind of Data You Collect Shapes Job Classification and What You Do With It

In organizational psychology, grouping separate jobs into broader categories is a routine step. Those categories then feed decisions about hiring, pay, and development. Patrick D. Converse and Frederick L. Oswald looked at what happens before that grouping is even possible: the choices an analyst has to make along the way.

Two of those choices stand out. First, what kind of data do you gather about each job? Second, once you have it, which quantitative method do you use to sort jobs into clusters? Neither decision is trivial, and both get made early enough that they quietly set limits on everything downstream.

To see how much the first choice matters, Converse and Oswald worked with a single dataset that described jobs through three different kinds of ability information. Same jobs, three lenses on the ability each one demands. When they ran the clustering, the three approaches did not converge. They produced noticeably different groupings of the same jobs, which echoes what earlier studies had found.

Then came the practical question. Did those differing clusters actually change anything for the uses people care about? They checked three: validating selection tests, evaluating jobs for compensation, and helping people explore career paths. For those specific purposes, the gap between the cluster solutions did not translate into a clear-cut difference.

That is not the same as saying the data type is harmless. The particular ability measure you pick changes the clusters you end up with, and the clusters are what personnel decisions rest on. Change the input, change the output, change the call that gets made from it. The lack of a sharp practical divergence in these three cases does not erase the fact that the classification itself shifted.