43292 Multicollinerity in the moderation model -

43292 Multicollinerity in the moderation model

Last post 04-06-2009 9:46 AM by pwestfal. 1 replies.

Page 1 of 1 (2 items)
	Sort Posts: Previous Next

04-04-2009 10:17 PM

43292 Multicollinerity in the moderation model

Reply Contact

ID: 43292On 3/24, at the approximately 00:08:30 mark, we were discussing “Multicollinearity”. In here and the materials you provided, “Multicollinearity exists when the X’s are correlated. The multicollinearity problem does not involve the Y’s.”I can understand that we cannot measure the unique effect of X1 regarded as the main predictor on Y when X1 and X2 are closely related. And I also know that this is related to problems (correlations) between independent variables (Xs).However, in the moderation model, the moderator (M1) can be considered as a kind of independent variable influencing Y. Hence, when X1 and M1 are perfectly correlated in the model, can we say “there is a multicollinearity problem.”?For example of the tire type in Homework 5, the speed moderates the relationship between the type of tire and operating cost. In this example, both the speed and tire type affect the operating cost. If we assume that there is perfect correlation between speed and tire type, can we say this modes has multicollinearity? In addition, if multicollinearity exists in this model, should we remove one of two variables even though one is independent variable and another is moderating variable?Otherwise, we can say: “This is not multicollinearity problem. That is because this model has only one X (predictor), and because the main effect we want to know is only one, and the moderator is only related to the interaction effect, not main effect.”

04-06-2009 9:46 AM In reply to

pwestfal
Joined on 08-12-2008
Posts 199

Re: 43292 Multicollinerity in the moderation model

Reply Contact

Anonymous:
ID: 43292On 3/24, at the approximately 00:08:30 mark, we were discussing “Multicollinearity”. In here and the materials you provided, “Multicollinearity exists when the X’s are correlated. The multicollinearity problem does not involve the Y’s.”I can understand that we cannot measure the unique effect of X1 regarded as the main predictor on Y when X1 and X2 are closely related. And I also know that this is related to problems (correlations) between independent variables (Xs).However, in the moderation model, the moderator (M1) can be considered as a kind of independent variable influencing Y. Hence, when X1 and M1 are perfectly correlated in the model, can we say “there is a multicollinearity problem.”?For example of the tire type in Homework 5, the speed moderates the relationship between the type of tire and operating cost. In this example, both the speed and tire type affect the operating cost. If we assume that there is perfect correlation between speed and tire type, can we say this modes has multicollinearity? In addition, if multicollinearity exists in this model, should we remove one of two variables even though one is independent variable and another is moderating variable?Otherwise, we can say: “This is not multicollinearity problem. That is because this model has only one X (predictor), and because the main effect we want to know is only one, and the moderator is only related to the interaction effect, not main effect.”

The question is very rambling and confused! Consider proof reading.

So, how can "speed" and "type" be "perfectly correlated"? Give it a thought. The only possibility is that all the type=A are at one speed (eg 30MPH) and all the type=B are are another speed (eg 50MPH). In that case you cannot isolate the effect ot type from speed.

Now, moderation is a separate issue. But if you cannot isolate the effect of type from speed, you certainly cannot go the next level and isolate moderating effects.

As far as recommendations about MC, please review the class notes an the materials I placed on the class web, notably "Multicollinearity Notes."

70 80 100 40

Professor

Page 1 of 1 (2 items)