Regression Analysis
Six Sigma – iSixSigma › Forums › Old Forums › General › Regression Analysis
 This topic has 10 replies, 7 voices, and was last updated 15 years, 5 months ago by Dr. Mikel Harry.

AuthorPosts

July 27, 2006 at 2:25 pm #44142
Hi,
Can anyone explain me how can the end result of a regression analysis be interpreted?
ie., what should be value of RSq and adj RSq normally? should it be above certain %?
Thanks,
Mark0July 27, 2006 at 2:32 pm #141053If Rsq is above 90% then you have a good model.
0July 27, 2006 at 2:35 pm #141054Sorry I should also mention that if it’s below 90% then there’s something else that’s not accounted for in your analyse, some special cause variation or a missing factor. Or maybe an interaction that’s been taken out.
0July 27, 2006 at 2:36 pm #141055Interpreting regression analysis includes a lot more than RSq and adj Rsq. Search this forum for Robert butler’s posts. He posted some good detailed explanation regarding this. Deep
0July 27, 2006 at 3:45 pm #141068
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.You should never forget that regression is a geometric fit of a line/surface to a cloud of data. Because it is geometric and not based on first principles you will always get a regression equation.
Consequently, the issues surrounding the interpretation of the regression effort should focus on the quality and correctness of the fit. To that end you will need to do regression analysis. The heart of regression analysis is the examination of the residuals.
The following post may be of some help with respect to understanding the folly of trying to use R2, by itself, to assess your regression equation.
https://www.isixsigma.com/forum/showmessage.asp?messageID=43683
0July 27, 2006 at 4:03 pm #141072Robert:How do you find your old posts? I tried the search option at the top right hand corner of this website and i could not get that work properly. Can i know how did you find this old post? Are there any other way to search this forum?
This is what i do, Type the key word in the search option at the top right hand corner, Select discussion forum from the drop down and hit search.
But it does not give me a lot of results. It only give very few results. Please let me know if you use any other technique.Deep0July 27, 2006 at 4:12 pm #141073
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.As far as I know Deep you can’t. When I first started posting you could look up an author either by using an author search key or by typing in their name on the keyword. That function disappeared a long time ago. As recently as a few months ago you could still type in a keyword like “regression” and get pages and pages of listings but this too appears to have been eliminated. As a result, the only way I know to find my posts is by saving them to a separate file. I only save those posts of mine (and others) that I think might be worth citing at some future date.
0July 27, 2006 at 4:19 pm #141074Deep,
This is the link that Robert Butler provided in a previous discussion about Regression. It succinctly summarizes the major points and is extremely well researched.
https://www.isixsigma.com/forum/showmessage.asp?messageID=516250July 27, 2006 at 4:34 pm #141078Thanks a lot Hans, I saved the discussion on that thread. I think you did not forget about the simulation :)Thanks once againDeep
0August 3, 2006 at 3:01 pm #141369Mark,
What are you trying to use the regression analysis for? Regression can be used as a tool in different stages of a project. Ultimately you should remeber that Rsquared is just a statistic to explain the goodness of fit of your data to a linear equation of “best” fit. You should never look blindly at Rsquared. ALWAYS look at the graphical representation of your data. An Rsquared value will be given when you ask for it, no matter what the graphical representation looks like.
If you are using regression for correlation evaluation, there is no certain number to shoot for. Some references say 0.7 or higher, but I say it depends. If you see a correlation/regression of 0.5, that means that the input explains 50% of the output key measure. What a find!
If you are near the end of your project and are using regression as a statistical model of your key measure, then the higher Rsquared the better. Typically, 0.9 or better is requested for a “good” linear model of the data. Hope this helps
Paige0August 3, 2006 at 6:16 pm #141380
Dr. Mikel HarryMember@MikelHarry Include @MikelHarry in your post and this person will
be notified via email.ideally, the Rsq should be close to 1.0 to ensure that a big % of your data is being accounted for in the experiment.
Rsq and Rsqadj should not have a difference of 0.2 because this is alarming meaning that the no. of your insignificant data are close to the ones accounted for. Also, Rsqadj should not be greater than Rsq0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.