Regression interpretation

RaMa09

New Member
Hi there,

I have done a linear regression with 5 IVs and 1 DV.

In the output under coefficients, it shows one of the variables is not signficant (.825).

However if I do a regression with just this IV against the DV, under coefficients it is significant (.000).

Does this mean that the variable does not contribute within the group to the DV, however when taken alone it is? I am a little confused about how to interpret this?

Many thanks,

Rama

hlsmith

Not a robit
It is likely a collinearity issue. Examine the Variance Inflation Factors. Also, did the coefficient change much or was it just the SE estimates?

RaMa09

New Member
What do you mean by a collinearity issue?

Does this mean that the IV significantly predicts the DV, but when with the group of IVs it doesn't? How to interpret it?

The coefficients did change quite a bit yes.

hlsmith

Not a robit
Multiple coefficients can be explaining the same variability. Also, there could be confounders. It is all about knowing the relationships between variables. We don't know the context of your study!

RaMa09

New Member
The context is we are measuring the Overall Satisfaction of Customers.

IVs are such as 'feeling like a valued customer', 'given confidence staff knew what they were doing' and 'staff take ownership of request'.

If the three above explain 70% of the variance, are significant as a model in the ANOVA, but the 'staff take ownership of request' IV is not significant (.875) it means that all variables except for this variable add statistically significantly to the prediction, p < .05.

However, if I do a regression on just 'staff take ownership of request', this then makes the variable add statistically significantly to the prediction. It also has an r square of .545 which isn't bad.

I am just not sure whether to say that 'staff take ownership of request' is a significant predictor of overall satisfaction or not?

hlsmith

Not a robit
It is not a significant IV when controlling for the other variables, which apparently steal its thunder. Why are you stuck on that variable, just imagine I am trying to predict your weight, I know your gender, height, age, education level, and salary. Knowing one of those variables could be significant. But throwing them all in, some are going to lose steam, since age and education and salary may all be partially accounting for the construct of age.

How are your variables formatted (e.g., continuous [any number] or likert (integer scale)?

RaMa09

New Member
Hi hlsmith,

Thanks for that, starting to make more sense now.

The variables are all based on a likert 1-10 scale of strongly disagree to strongly agree.

Basically we are looking for questions that we can remove from the questionnaire if they do not add anything to the current model. It would appear that the 'staff take ownership of request' question could be removed and not affect how the other questions contribute 70% to the prediction of overall satisfaction.

hlsmith

Not a robit
There is a whole field related to examining instruments, look into principal component analysis, factor analysis, cronbach alpha, and item response theory!