ANOVA table - Practice Question (Par 4 difficulty)

idalton

Director
Can you explain to me the reasoning why you wouldn't just use the basic correlation coefficient (I see that you used the Correl function to backup your answer) way to come up with the answer instead of the ANOVA approach which comes up with the same result?

I'm trying to understand where I could use the ANOVA approach in a real world situation instead of Correl.

Thanks
Ian
 

David Harper CFA FRM

David Harper CFA FRM
Subscriber
Ian,

I agree, I don't see a need for ANOVA in a simple regression. (Nor, btw, is there a need for the F test in a two-var regression - there isn't really a "joint hypo" around the single explanatory variable).

I am just trying to hit on 2008 AIMs. ANOVA & R^2 (ESS/TSS) & F test are in the AIMs/reading. So, i was just trying to "connect the dots: between ANOVA, R^2, and r (i.e., that R^2 = SQRT(R^2)). I think the relevance is less practical and more for the concepts. I think ANOVA is more relevant in a multivariate regression. In practice, I prefer =LINEST() for everything--because like CORREL(), it's dynamic. Whereas ANOVA and the add-in regressions have to be re-run each time. But, again, i'm just looking at the AIM document.

(It's not to your point, but i do find ANOVA helpful as a reminder that the whole regression business is about breaking squared sums into two sources of variation. Just for myself, once i understand ESS and RSS, some of the other pieces come more naturally. Like, the standard error of regression/estimate, I personally have a hard time with it "directly" but it's easer from the "two pieces" approach)

David
 

idalton

Director
thank you for the response. I wasn't questioning why you were going through ANOVA I know it's because of the AIM's. I was just trying to understabd when/where it would be useful within a everyday risk management tool.
 
Top