Friedmans test

[EL KIMON]

Dear all,
i need some clarification about Friedmans test for cross sectional dependence.
Between two results that both suggesting that there is no cross section dependence, a higher p value suggesting a better model? Can i use that as interpetation too?
Or just that there is no cross sectional dependence?
Thank you for your help.
Best,
El.

[Nick Cox]

This is in essence https://www.statalist.org/forums/for...nal-dependence reposted

Please read and act on https://www.statalist.org/forums/help#adviceextras #1 which explains that just repeating a post is not as helpful as expanding on a question. There are now two threads, which is likely to confuse, except that I will post a diversion.

My guess is that no-one has replied to this because it is almost totally unclear what you are doing. No example data, no code, no exact results, just questions with no context. Again FAQ Advice #12 spells out what is expected in a good question.

[EL KIMON]

Dear Nick,
You have absolutely right my question wasn't clear and not proper.
I am running a model using panel data analysis and i am using Friedmans' test to check for cross sectional dependence.
The results of the Friedmans test are:

Friedman's test of cross sectional independence =    24.032, Pr = 0.2915

meaning that there is no cross sectional dependence.
Although, when i add more explanatory variables in the model the friedmans' test result are:

Friedman's test of cross sectional independence =    18.677, Pr = 0.6058

which still means that there is no cross sectional dependence.
My question is: Can i characterize the second model as proper choice to use since the p-value of the Friedmans test is higher? Therefore the cross sectional independence is further supported under that model?
I hope my question to be more clear now.
Thank you,
Best,
El.

[Carlo Lazzaro]

El:
the choice between the two models shoud not be based on Friedman's test higher/lower P-value, as both the outcomes are telling you that, according to your data, there's no evidence for rejecting the null hypothesis of no cross-sectional dependence. Obviously, as you do not provide data on your sample size and so on, the usual cautionary tale applies: the absence of evidence is not evidence of absence (https://www.ncbi.nlm.nih.gov/pubmed/7647644).
Be as it may, you should prefer the regression model that gives a truer and fairer view of the data generating process: the literature in your research field can help you out in this respect.

[EL KIMON]

Dear Carlo,
Thank you for your help.
The choice of my model of course will not depend on the Friedmans test. I just wondering if i can also say this, among other reasons, to support my choice.
Thank you,
Best,
El.

[Carlo Lazzaro]

El:
in my opinion the main issue is: given that both Friedman's test outcomes fail to reject the null, which is the model that gives a truer and fairer view of the data generating process?
Personally, I would not mention Pr = 0.6058 among the criteria for preferring model 2 vs model 1.