01-28-2015 08:21 AM
I'm new user to Modelling and now I've struck whilst interpreting the GLM output.
What does F value and Pr > F denotes? I've F value as 49.63 Pr > F as <.0001? When I should reject my null hypothesis for 95% CI and what is the significance behind F value?
Thanks for any help you offer.
01-28-2015 08:47 AM
The F-statistic compares two variances by dividing them, in the simple case of a one-way PROC GLM model, it would be the variance* of the model divided by the variance* of the residuals. A large number for the F-statistic indicates that the variability of the model term being tested is larger than the error variability, and if this number is large enough, then you say the F-statistic is statistically significant (in layman's terms, the F-statistic wouldn't get this large by random chance, and so we conclude that the result is due to a statistically significant effect). The Pr>F idnicates the probability of getting this large of an F-statistic if the null hypothesis was true. A Pr>F of less than 0.05 would indicate the effect is statistically significant with 95% confidence.
All of this is basic hypothesis testing, and if my paragraph above is not clear to you, then you probably ought to read a text on basic statistics and hypothesis testing.
* — well actually these are not variances but rather mean squares from the model, but they are very similar to variances
01-29-2015 03:09 AM
What if the F value is more than 1000 ,e.g.2000. Still would be significant?
How the critical value being calculated in F-statistic? Whether the critical value keeps changes when degrees of freedom changes.?
01-29-2015 04:00 AM
"What if the F value is more than 1000 ,e.g.2000. Still would be significant?"
Yes. I think so. since all the statistical estimator is estimating a deviation. 1000 is the F value when H0 is true. F=1000 is already significant , so F=2000 would be more significant (i.e. the deviation is bigger).
How the critical value being calculated in F-statistic?"
F estimator is a combo of two Normal distributions , check its formula at wikipedia.com
"Whether the critical value keeps changes when degrees of freedom changes.?"
No. I don't think so . It is depend on your significant degree ALPHA . i.e. is base on the ALPHA percentile .
It have nothing to do with DF .
01-29-2015 04:40 AM
Thanks for your inputs.
'1000 is the F value when H0 is true' - if F value is significant then we should rejecting the null hypothesis. isn't? So it means H0 is NOT true. Apologize if I'm wrong.
'P' value will be 0.05 for 95% CI. Do we've any standard (like P value) F critical value for 95% CI?