6 factors and a response

Hello All,

I have a spreadsheet with 6 factors and 1 response... i'm not sure if all the factors effect the response but I'm hoping there is a way to figure out which ones are the biggest contributors... How do I go about doing this ? I have access to minitab matlab and excel.

Any help would be appreciated.


TS Contributor
I played a bit with your data and by using backward elimination I got a model where factor 5 and factor 2 are significant, R-sq(adj) is about 50%.

factor 3 and factor 4 are very strongly correlated, but in the end this is not interesting because neither is part of the model. You also have two suspicious observations - lines 21 and 24 - you might want to look at those, maybe they are special in some way?

I hope this helps a bit.

Hi rogojel,

How did you go about doing this? Did you just plot it and see? I'm not really understanding the backward elimination you're talking about :)

I think I know after a little googling... Stepwise regression am I right?

But how can something be strongly correlated but not part of the model... kind of a correlation not causation thing?


TS Contributor
I first took out the f4 because the VIF was an atronomical 340 and it was highly correlated to f3.
then i went stepwise, always eliminating the term with the highest p value and recalculating the regression.

Thanks that was really helpful.

Now that I have the variables that affect my response...

How do I go about generating main effect curves... basically a curve that shows the effect of say factor 4 vs response while all the other paramters are fixed?
Sorry another followup question... if I want to know the percentage or importance of these factors with relations to each other... what I use the value that is spit out in minitab above the T value

Factor 20 -2260 -1737 -1706 -1801
T-Value -3.41 -4.04 -3.95 -4.19
P-Value 0.001 0.000 0.000 0.000

The bolded bit and relate them to each other?


TS Contributor
I think in case of a regression the effect is actually shown by the coefficient of the factor - the interpretation of the coefficient would the average change in the Y if all other factors are held constant and the factor in question is increased by one unit. This will be affected by the units of measurement of the factor, so just by looking at the magnitude of the coefficient might be misleading.

If you have comparable units of measurement for all factors then the coefficients will give you an idea of the relative importance as well, but you would need to double check this against reality, this interpretation can be tricky.

Hi all,

the stepwise and general regression are giving me two different results... in that the factors it selects as having the largest influence are different. Why would this happen?