F-statistics so high.

[Azlin Ahmad]

Hi. I'm PhD student. I have problem to analyze my result. Can i know, why F-statistics is so big. It's ok? or have some problem?

Can someone help me to explain why this value so big? F-statistics = 8231.39

Thank you.

[Carlo Lazzaro]

Azlin:
A sky-rocketing R-sq (and roughly one half of your predictors lacking statistical significance notwithstanding) makes me suspect that your model is overfitted.

[Azlin Ahmad]

Hi Carlo.

Thank you for this comment. So, can know how to solve this problem? I need to change another variable or something wrong with my data?

[Carlo Lazzaro]

Azlin:
a trivial recomendation would imply a more parsimoniuos model.

[Nick Cox]

Focus first on oss and npm. What do they mean? Are they related? How well does a regression of those alone perform?

[Jesse Wursten]

Do you have a cross section or a panel? Huge F-stats sometimes arise from ignoring the panel structure of your data.

[Azlin Ahmad]

Hi Nick Cox and Jesse Wursten. thank you for your comments.

1. # Nick, I have remove npm in my regression because I have check back formula for npm is related to oss. However after I remove npm, F-statistic show lower amount.

2. #Jesse, I have use panel data.

After remove npm, F-statistic lower but R-square also lower..Value R-square can be acceptable or not? My data use secondary data.

[Carlo Lazzaro]

Azlin:
why considering -regress- as a first st choice if you actually have panel data?

[Azlin Ahmad]

Carlo:

Based on my understanding, OLS, Random effect and Fixed effect is most appropriate to use for panel data. To check either random effect or fixed effect , I need to conduct OLS test first. So that why I'm considering OLS result first (regress).

I'm still new in using STATA..If mu understanding is wrong, kindly inform me. Thank you.

[Carlo Lazzaro]

Azlin:
if somebody taught you that OLS (like the one you shared with the list at #7 that actually ignores the panel structure of your dataset and considers each observation as independent) is good for panel data analysis, she/he was simply wrong.
OLS with -i-panelid- can give you the same results of -xtreg,fe- (but this is not what you did in your example).
Pooled OLS with standard errors clustered at -panelid- level can be a choice if no evidence of panel-wise effect comes alive after -xtreg-.
See the difference between -regress- and -xtreg- (I still asume that you have a continuous regressand).
I would recommend you to take a look at and decent handbook on panel data econometrics.