[R-sig-ME] Reporting den of df for f like AOV table using Sattertwaite's method

[Phillip Alday]

I didn't see an answer go past for this question yet so let me make a
few comments:


- If your ddf are that large, then the F distribution is very close to a
chi-squared distribution and the approximation used by e.g.
car::Anova(model, KR=FALSE) will be faster and nearly as accurate.

- (This is why one of the suggested ways for dealing with the missing
p-values in lme4 is to simply treat the t values as z values -- at some
point with a few tens of groups and a few hundreds of observations, the
effective degrees of freedom are so large that you can treat them as
infinite and thus t -> z and F -> chi-square)

- If you decide to stick with the F distribution, then you need to
report all digits of your ddf before the decimal point and maybe 1 or 2
after the decimal point (if they aren't integers).

Best,
Phillip


On 19/10/20 7:00 am, Salahadin Lotfi wrote:
> Hi everyone,
> I have run several MLM models with over 200,000 observations. I intend to
> report f values calculated using lmerTest package for each model
> (Sattertwaite's method). I am trying to learn best practices to report
> denominators of df estimated by Sattertwaite's method as I am reporting
> usual f(NumDF, DenDF)=xxx. The obtained den of df is pretty large and I am
> not sure it does make sense to report 6 digits values. I have run several
> of these models and it takes a big chunk of the result section if I keep
> reporting 6 digits.
> I am also aware that many researchers report beta/ES/t/z estimates,
> however, it makes sense to report f values in the context of my study,
> hence I am here with my question. :-)
> 
> What is the best practice when it comes to report pretty large DenDF of f
> models?
> 
> Any input will be greatly appreciated.
> 
> Thanks,
> Sala
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>