R squared adjusted in DOE
Six Sigma – iSixSigma › Forums › Old Forums › General › R squared adjusted in DOE
 This topic has 4 replies, 3 voices, and was last updated 16 years, 5 months ago by Robert Butler.

AuthorPosts

March 3, 2006 at 4:31 am #42590
TierradentroParticipant@john Include @john in your post and this person will
be notified via email.Hi
I have 2 significant X’s that impact the Y. I am trying to manipulate the input data when running a one way anova to achieve a >80% R Square adjusted figure. When ‘playing’ around with the individual inputs and then running the anova I am getting a p value <0.05 though a low R squared adjusted figure OR a higher (though not high enough) R squared adjusted figure and a p value >0.05.
I am trying to understand how to manipulate the input data to influence the output and achieve >80% R square adjusted
Really, really need to understand from a learning perspective
HELP !!
John0March 3, 2006 at 4:50 am #134572Hi John,
you need to be careful with the Rsq & Rsq adjusted values. You probably know that Rsq describes the amount of variation in Y that is explained by your X’s. It is possible to over inflate this by having an excess amount of X’s in your model, hence Rsq adj is used. Rsq adjusted is adjusted by the amount of X’s in your model. I’d be more concerned about getting your Rsq adjusted to within 5% of Rsq. We were told this is the “Monkey” test, because any monkey can get a high Rsq, but not a high Rsq adjusted. I’d much prefer to have a Rsq =0.6, and an Rsq adjusted = 0.55 then an Rsq=0.9 and an Rsq adjusted = 0.6.0March 5, 2006 at 11:32 pm #134687
TierradentroParticipant@john Include @john in your post and this person will
be notified via email.Hi Pat
In experimenting with my figures I am finding that once I start to increase the adjusted Rsq % result then the pvalue becomes >0.05
I am told that I need an adjusted rsq fiqure of >80%, I am getting nowhere near that. I have only 2 significant X’s to work with
???
John0March 6, 2006 at 12:17 am #134688Hi John,
an Rsq value less then 80% means that less then 80% of the variation in your Y is explained by your chosen X’s. It doesn’t mea your model is inadequate, remember the other checks you need to do such as residual analysis.What Rsq values are you getting?. Possible reasons why Rsq is lowExternal noise in DOE caused variation in Y
Your model did not include all significant X’s
You cannot mess around with your model to artificially inflate Rsq values.0March 6, 2006 at 3:38 pm #134701
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.John,
Pat has given you good advice. As he said, you need to look at your residual plots – they will tell you what you need to know with respect to missing X’s, influential data points and noise. As was noted, you can make the R2 go as high as you want by just throwing terms in the model. Assuming you have no replicates you can drive it all the way to 1 if you want – of course the model won’t mean a thing.
As for the assertion that you need “adjusted rsq fiqure of >80%” Whoever offered that advice doesn’t know what they are talking about. There are many processes with a level of ordinary variation so high that you will be lucky to get an R2 of 30%. Does this make a model invalid or useless – no – it just means the uncertainty associated with its predictions will be large. It is nice to have a good predictive model with a high R2 but a high R2, by itself, guarantees nothing.0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.