Help needed with counter intuitive result in OLS

#1
Hi All,

I have a simple linear regression run between say, statements about the impression with a brand on its favorability

What i am getting however is, one of the independent variables is coming out as a negative driver which is not expected.

The IV is a agreement question, Basically siomething like " how much do you agree that the product is fast"

What i am geting is , its coming out as a negative driver. In the sense, if people agree more that its a product with good speed, they would have lesser favorability with the product (Dep,variable)

This is something i am having trouble understanding. Can you please help me out here. All other variables have intuitive impact scores (beta coefficients)

That is another variable, for example is, "Brand is reliable", which means if people agree more that the brand is reliable, their brand favorability will increase by the amount of the beta coefficient.

Please solve this doubt of mine. how do i tackle this. I understand this could be because of multi collinearity in Independent variables, but how do i eliminate certain variables from the analysis (based on the correlaion matrix?)
Or, even the VIF score would do the same. None of the metrics have a VIF of more than 5..

Please help me out here

Thanks a ton

Satish
 

Dragan

Super Moderator
#2
Hi All,

I have a simple linear regression run between say, statements about the impression with a brand on its favorability

What i am getting however is, one of the independent variables is coming out as a negative driver which is not expected.

The IV is a agreement question, Basically siomething like " how much do you agree that the product is fast"

What i am geting is , its coming out as a negative driver. In the sense, if people agree more that its a product with good speed, they would have lesser favorability with the product (Dep,variable)

This is something i am having trouble understanding. Can you please help me out here. All other variables have intuitive impact scores (beta coefficients)

That is another variable, for example is, "Brand is reliable", which means if people agree more that the brand is reliable, their brand favorability will increase by the amount of the beta coefficient.

Please solve this doubt of mine. how do i tackle this. I understand this could be because of multi collinearity in Independent variables, but how do i eliminate certain variables from the analysis (based on the correlaion matrix?)
Or, even the VIF score would do the same. None of the metrics have a VIF of more than 5..

Please help me out here

Thanks a ton

Satish
What is the correlation between the dependent variable and the IV in question?...If this correlation is close to zero, then this IV may possibly be acting as a suppressor variable.
 
#4
Do you think you could have accidentally coded your "fast" variable backwards? That is, assigned high numbers to low values of "how much do you agree?" If this is a possibility, it might be worth checking.
 

spunky

Doesn't actually exist
#6
What is the correlation between the dependent variable and the IV in question?...If this correlation is close to zero, then this IV may possibly be acting as a suppressor variable.
i 100% agree on this one... it feels like a suppressor variable's doing something funny... if we could have a look at the correlation matrix it would be very helpful..