2

I used scipy.stats.linregress function for my dataset. It returns 5 output variables.

Slope = 0.0453

Intercept = 14.905

r-value = 0.91

p-value = 0

standard error of estimate = 0.00015

This is the documentation link if anybody would like to see

Linear Regression Scipy

Can somebody help in understanding the output? Questions I would like to ask are the two variables highly correlated. The r-value is something that I think points to the correlation, but since it is not 1.0, what would be ways to find out which ones do not satisfy this constraint, can I use any of the parameters here?

In general when you see such values how to grasp them?

Community
  • 1
gizgok
  • 569
  • 5
  • 9
  • 4
    You should probably take a course in regression, there is much more to it than just looking at the output, at least when the output is so limited. You have assumptions to check, graphs to make, etc. – Peter Flom Dec 20 '13 at 23:43
  • As Peter correctly suggested a short course in regression will probably benefit you greatly. For self-study, very approachable and hands-on expositions of linear regression with R coding examples are presented in J.Faraway's books: `Linear Models with R`. Additionally, the following thread : http://stats.stackexchange.com/questions/170/free-statistical-textbooks offers a variety of free e-books that can take you a very long way. – usεr11852 Dec 21 '13 at 00:16
  • 1
    Your question amounts to "teach me to understand regression", which would take a book to answer. "*...are the two variables highly correlated?*" - that depends on what counts as high, and there's no single standard; a chemist might call that $r$ very low, while a psychologist might say it's unbelievably high. It depends on what we're dealing with (but then why would a high $r$ matter in any case?). "*The r-value is something that I think points to the correlation*" -- no, it *is* the correlation. "*what would be ways to find out which ones do not satisfy this constraint*?" -- what constraint? – Glen_b Dec 21 '13 at 00:33
  • @Glen_b Since the correlation coefficient is not 1, there are values which are not showing direct relationship. Is there a way to figure out such values. – gizgok Dec 21 '13 at 01:23
  • If there was a perfect linear relationship, you wouldn't need regression at all. Because there's going to be variability about any underlying linear relationship, you don't expect it to be 1.0 even when the population relationship is linear. – Glen_b Dec 21 '13 at 01:51

0 Answers0