R2 vs lack of fit

[need]

Hello:
I am working with a multiple regression model where I see a 91% R2 but significant lack of fit with pure error. I also have a slighly S-shaped normal probablity plot of the residuals. Can anyone suggest how to interpret R2 in light of the lack of fit and additionally suggest what the S-shap in the residuals plot is telling me?

[Dason]

Possible cross-post: http://stats.stackexchange.com/quest...ple-regression

Can you tell us what you think the interpretations of R^2 are and what your lack of fit tests are telling you?

[need]

Message deleted

[Dason]

I didn' t realize the two sites were related. They are related in more ways than one too. The responses were equally helpful..

Anyone else have any comments that aren't an utter waste of time?

A good day to you too! The two sites aren't related but it's good etiquette to at least inform people that you're cross posting - otherwise somebody could spend a great deal of time formulating a response to a question that has already been answered well.

Cardinal's response at the stack exchange might have seemed rude to you but their response is quite appropriate for the stack exchange network (and I really think they were quite nice about it!). You're expected to put a little bit of work into searching to see if an identical/similar question has been asked at SE (although ideally you'll do the same anywhere).

My response here was actually attempting to be helpful - I sincerely want to know what your interpretation of those two things is before attempting to provide an answer that may or may not be misinterpreted (due to misunderstanding of what the quantities represent).

However, your response suggesting that both Cardinal's and my posts were wastes of time even though we were both trying to help you is highly suggestive of an unappreciative nature. I would suggest putting a little more thought into your posts asking for (free) help from people who aren't paid to do this and realize that if you don't get a response that you like or answers your question then you might want to consider the possibility that you didn't do a good enough job asking the question! This is a good read to help with that in the future: http://www.talkstats.com/announcement.php?f=2

[need]

My apologies Dason, as I misinterpreted your response. I thank you in advance for your help. Let's start over.

I have done some research on both sites and can't seem to get the right answer. On the surface the high R^2 would lead me to think the model is explaining the relationship quite well. However the fact that the lack of fit test is significant confuses me as I don't know how the model could explain such a high degree if the variation but not be fit.

To add the residuals plot suggest curvature in the relationship which lead to more confusion. If the equation from the model predicts such a high level of variation how could there be curvature. I also tried conversion with Box-Cox and modeling iwth th natural log of both sides of the equation to no avail.

So there is a better description of the problem: I see a high level of prediction of the variation but in contrast are the apparent curvature of the residuals and the lack of fit.

I hope this helps.