Multiple linear regression

GustavoPires

New Member
Hello friends,
I have a situation that I don't understand the results:
I'm doing a multiple linear regression with 1 dépendent variable and 4 independent variables 1 VDep e 4 VInd (A, B, C, D)

- The results are clear with them alls:
VindA β = 0.734 p<0.001
VindB β = 0.489 p<0.001
VindC β = 0.285 p<0.05

VindD β = 0.734 p = 0.062 ( so, variable VindD exclued of the model)

But when I perform a linear regression 1 to 1 with VDep and VindD the result is ViD β = 0.453 p < 0.05

Even if I put the variables VindB et VindC within the multiple linear regression with VindD, it is still significative. BUT when I put the VindA, VinD turns non-significative.

Do you have any idea?

CamilleJosion

New Member
Hello, this happens because when A,B,C are in the model what's brought by D is not significant.

GustavoPires

New Member
Hello, this happens because when A,B,C are in the model what's brought by D is not significant.
Thanks Camille, but do you know why when I put B,C and D they all have significant effects and when I introduce A D becames non significant? Is it a negative moderation effect? ( I am guessing here…)

obh

Member
Hi Gustavo,

Did you check the correlation between the independent variables?

GustavoPires

New Member
Hi Gustavo,

Did you check the correlation between the independent variables?
obh, just did it, thanks. D is significantly correlated to A (Pearson Correlation 0.548 p = 0.012)

So no luck for me?

GustavoPires

New Member
I don't know a lot about it. So it's a good thing that they are correlated?

obh

Member
Hi Gustavo,

First I wouldn't necessarily count p = 0.062 as insignificant ...I don't really see a big difference between 0.06 or 0.05.
I know that people usually use 0.05 as alpha.
Also if you have a good theoretical reason to assume that D should be in the model you probably should keep it even if p=0.2.

But back to your main question, let's assume that p=0.2 and you don't have a good theoretical reason to assume that D should be in the model.

0.548 is a large correlation. but it doesn't necessarily say it is multicollinearity. I recommend you read the following:
http://blog.minitab.com/blog/unders...ling-multicollinearity-in-regression-analysis

A significantly predict Y (VDep ).
If the part that predicts Y in D has a high correlation to the part that predicts Y in A adding D to the model will not improve the prediction.
That why D is not significant in the model with A, B, C, D.

Or it may be more complex the part that predicts Y in D already included in the combination of A,B,C.

Is all clear now?