They like it! Now prove they wouldn't have liked this other thing

I have done extensive research (1:1 interviews, concept testing, and validation) that shows employees like to be paid based on X, Y and Z variables. Now leadership wants to know for certain that employees wouldn't be just as satisfied if we only paid based on X and Y (same net, though). Ideas for how to do this?

One catch - I can't do a survey.

My plan is to,

A) show the enthusiastic response to Z we saw in the research, and

B) do a few lightweight sessions using the same method I used to prove they liked X, Y and Z - but this time with just X and Y. Subjectively assess their acceptance / enthusiasm for XY, which I expect to be lower than it was for XYZ.

I would like to have more concrete, less subjective proof, but without the ability to send a survey and get satisfaction ratings for each variant independently, I'm not sure how to do this...I welcome any suggestions!

• Without revealing what X, Y and Z are, how do you know that employees actually like to be paid based on just X or Y if you are comparing condition A (X + Y) to condition B (X + Y + Z)? Jun 20, 2018 at 5:41
• In arriving at your conclusion that XYZ is the ideal set of variables, what other variable combinations did you research? I would have imagined in getting to XYZ that you already investigated XY, YZ, XZ, etc. You may want to review the work you've already done to see if you already have the answer. Dec 17, 2018 at 16:23

It sounds like your initial research may have been a little too prescriptive, unless I'm not understanding your premise correctly.

If you started with a defined set of options, I believe the best solution is a card sort exercise (kind of). Give the subject all available cards and allow them to choose up to 3 factors (as few as one) and sort them according to how the factors should be weighted.

You would then have a clearer picture of which factors are most desirable and which are completely irrelevant.

This is an opportunity to leverage secondary research. You can quite reasonably extrapolate from a proxy model: "In this similar case, this approach worked well..." I've never met a research question that didn't already have some data against it out there in the world. Someone, somewhere has done a relevant study. Just pick one that isn't a completely false analogy, a study that is similar to your research problem in a meaningful way. Argue from theory and from existing data before you have to reinvent a wheel.

Design of Experiments is an effective approach for answering questions like the one you’ve described.

Use DOE when more than one input factor is suspected of influencing an output. For example, it may be desirable to understand the effect of temperature and pressure on the strength of a glue bond.

DOE can also be used to confirm suspected input/output relationships and to develop a predictive equation suitable for performing what-if analysis.

(from ASQ.org)

In your case, you have an output variable (willingness to do job) and three input variables (x, y, z). You’ll want to test the relative impact of x, y, or z and whether combinations of x, y, and z matter. For example, x may not matter on its own, but xy might matter a lot. In Design of Experiments (DOE) this is called a three factor design.

It is also important to understand what levels of x, y, and z need to be tested. When determining levels, it will matter if x, y, and z are continuous or discrete variables. Salary is an example of a continuous variable since it could be any salary on the number line, bounded by a low value (\$20,000) and a high value (\$90,000). Project might be a discrete variable, since your company has a limited number of projects and an employee must be on one of them (project A, B, or C).

Once you know how many factors you’re testing (3) and how many levels there are for each factor you’ll be able to use a template to plan an experiment. In the experiment, you’d offer people jobs with different values for x, y, and z. Based on which jobs people pick, you can infer how much impact x, y, z, and their interactions have on job offer acceptance.

DOE can be a little complicated when there are lots of factors. But in your situation with 3 factors it is pretty easy to use. In college we used the text book Response Surface Methodology.

Note: You may already have a good enough answer. It sounds like you’ve done a lot of research already. Before doing any more research, it’s worth thinking about the expected value of perfect information. Basically ask yourself two questions and stop doing more research if 2 is greater than 1.