Wrong d.f. in statistiXL l.r. for slope

I entered the following data:
X Y
10 3.5
18 2.8
12 3.3
15 3.7
16 2.9
16 3
22 2.4
9 3.1

The regression coefficients returned:
Y -0.067 0.028 -0.703 -0.135 0.001 -2.423 0.046

The probability seems to have been computed using 7 d.f. instead of 6. I haven't tested the multiple linear regression, but I would guess it it using the total d.f. instead of the residual d.f.. The correct probability should be 0.051671855. This is an error that would lead me to accept the slope when it shouldn't.

I am using the demo version of 1.2

David Gosselin

Comments

  • Hello David

    First I must apologise for the delay in replying to your mail. I am usually informed by email when a new post is made and for some reason this didn’t arrive in your instance.

    You are quite right, an error is apparent in version 1.2 of statistiXL that results in the incorrect probability being reported for individual regression coefficients. The overall ANOVA for the regression as a whole and its probability remain correct however, as do the t values for the regression coefficients. It is only the coefficient probabilities that are being displayed incorrectly. The error was down to a single letter typo that caused the probability calculation to assume it was for an analysis without a constant term (i.e. intercept=0). I have corrected the error in Version 1.3 which will be released within the next few days. I apologise for any issues this may have caused people. We have a stringent testing procedure for all code but unfortunately this slipped by us and our beta testers. FYI Version 1.3 will also include a number of new features such as Installation support for Excel 2003 and a Fisher's Exact Test for 2 x 2 Contingency Tables.

    Regards

    Alan
Sign In or Register to comment.