Timeline for Why is it possible to get significant F statistic (p<.001) but non-significant regressor t-tests?
Current License: CC BY-SA 3.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 24, 2021 at 0:38 | comment | added | NoName | THANK YOU, THIS IMPROVED MY P-VALUE BY 72% !! | |
| Jan 23, 2021 at 0:23 | comment | added | whuber♦ | @UsDAnDreS I can't tell. In my example there's a lot of multicollinearity--my nine variables are confined almost to a space of just five dimensions. Perhaps in your case your ten variables have only one or two degrees of redundancy among them, but we would need better diagnostics than VIF to determine that. | |
| Jan 22, 2021 at 22:50 | comment | added | UsDAnDreS | Nice work! I just wonder about one specific example I ran into (on real data): logistic regression with p=10 predictors, n=500 observations (balanced response classes), largest VIF < 1.5, where one ends up with significant full model Likelihood Ratio test (at 0.0003) and all insignificant predictors (two smallest p-values at ~0.11, rest are > 0.25). I understand the main point of "It takes very little correlation among the independent variables", but in your example VIFs average 4.8, which is whopping 3.5 times what I have (avg VIF ~ 1.3), and it's still an issue... due to same reasons? | |
| Jan 25, 2017 at 19:21 | history | post merged (destination) | |||
| Jul 3, 2013 at 17:58 | history | edited | Scortchi♦ | CC BY-SA 3.0 |
fixed typo, improved formatting
|
| May 4, 2012 at 22:11 | comment | added | Michael R. Chernick | Great answer. A plus 1 from me. I would have liked to give it more. | |
| Aug 19, 2011 at 22:49 | history | answered | whuber♦ | CC BY-SA 3.0 |