Minimum data points required for Regression
Six Sigma – iSixSigma › Forums › General Forums › Methodology › Minimum data points required for Regression
 This topic has 4 replies, 5 voices, and was last updated 11 years, 4 months ago by Robert Butler.

AuthorPosts

June 1, 2010 at 4:40 pm #53469
Hello everyone,
What is the minimum number of data points required to get a valid Rsquared value on a regression line for a scatter plot?Thank you for your assistance!
0June 3, 2010 at 3:07 am #190263
Venugopal. GMember@Venugopal.G Include @Venugopal.G in your post and this person will
be notified via email.Hi,
Please find the below.
” Unless the relationship between X and Y is very strong, a small sample (<15) may not be large enough to detect it. Also, small samples do not provide a very precise estimate of the strength of the relationship, which is measured by adjusted R2. If a precise estimate is needed, larger samples (typically 40 or more) should be used."
Extracted from Minitab 16 – Assistant Regression.
Venugopal. G
0June 4, 2010 at 1:14 pm #190286
Marty Y.Participant@MartyY. Include @MartyY. in your post and this person will
be notified via email.I am not a statistics guru, but I am pretty sure that the pvalue takes sample size into account. I.e. the smaller the sample size the less confidence in your results and the higher the pvalue. So, if you have a pvalue that is less then 0.05, then I think this is all the information you need. Your sample size is adequate. If you have a pvalue greater then 0.05, then you either need more samples or your data is not correlated.
0June 4, 2010 at 9:06 pm #190287If you are talking about multiple linear regression, it will also have to do with how many independent variables you have.
I’m no statistician, but I thought that you needed at least 2 more data points than there were Xs. You need 1 for each X, 1 for the intercept and 1 more for the error term. More points beyond that gets you a more reliable error term. (again, I’m no stats person, but that is something I recall from my training).
0June 7, 2010 at 2:08 pm #190293
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.As asked your question is meaningless. An Rsquared or an adjusted Rsquared is a computed statistic and they are whatever they are – there is no such thing as a “valid” Rsquare.
If you could elabortate on what it is you are trying to do perhaps I or someone else could offer some additional thoughts.
0 
AuthorPosts
You must be logged in to reply to this topic.