Multiple Regression for Compensation Planning

I have very little statistics experience, so what I am about to ask may actually be a very simple question, but it is driving me crazy.

I am working on creating a job evaluation system (for compensation purposes) using Job Component, which is a statistical method. This method requires for several factors to be evaluated for each job (which eventually become the x values), along with the estimated market pay rate for each job (which become the y values).

I have chosen several factors to use (in order to determine which will be the best ones to stick with when I am finished), and gathered the data for each job in my "test" group (which consists of about 30 jobs). It seems that no matter what combination of the factors I use, or even if I use all of them that I cannot seem to get an r square value of above about .76 or so.

I have been doing the multiple regression in Excel, but I also have Minitab (which I have only used a couple of times, so I am not very familiar with the processes).

Has anyone else tried something similar, or possibly have any advice on what I can do to get this to work? Do I need a larger sample size (more than 30)?

Any pointers or advice would be greatly appreciated.
So you have some variables X1,X2,...,Xk that contain scores on several aspects of evalution (I would guess for example: being on time, not exceeding deadlines, personal sales, etc...)

And you have an (estimated) pay rate Y

So we would all expect the higher the X scores (better performance), the higher the Y score (at least in the US or UK: countries that are very performance orientied. In Europe or Japan I would expect to relation to be less strong). But anways...

First of all, an R² value of 0.76 is already VERY NICE! In finance (what I do) we are jumping in the air if we find R² values over 0.10 :D

So I guess you are estimating the folowing regresion:

Y = a + bX

So you are looking to explain the pay rate with the skills. If you can explain 76% of the variation in pay rates by performance, that would be very nice! The other 24% could be due to length of service, gender, or simply luck.

But I would be pretty happy with your results :)
Last edited:
Thanks so much for your response!

Let me clarify a little... the job evaluation is to find a pay value for the job itself (not necessarily the person in the job), so the factors are things like educational requirements, work experience, level of decision making, responsibility, etc.

The problem with only being at a .76 is that I need to use the formula that I ultimately come up with to price the jobs for which I do not have market data available. In almost every example I have seen the r square value is in the 90's.

Also, when I run the analysis, I always get a couple of factors (and sometimes the intercept, depending on which factors I choose to run the analysis on) with coefficients that are negative.

The "degrees" within each factor are something like this: Education Level = 0-5 (no education required through PhD); # of Departments Responsible For = actual number of departments responsible for; Years of Experience Required = actual years of experience required. Those are just a few examples.

I guess what I am looking for is a way to test, statistically, which of the factors are going to be the most useful to me in this model.
Last edited:
In SPSS you can select methods for this I remember... Like the Enter method, or Remove *something something* method... Then SPSS removes bad explanatory variables automatically and stuff like that...