I’m stuck on a CFA problem. I’ve got a 25-item scale and I’m trying to test a 2-dimension model. But when I checked the data, the skewness and kurtosis values are way off the charts. I’m not sure if my code is messed up or if there’s a way to fix this so it doesn’t mess with my CFA results.
Here’s a snippet of what I did:
data_subset <- data[, c('q1', 'q2', 'q3', ..., 'q25')]
data_clean <- na.omit(data_subset)
data_clean[data_clean == 999] <- NA
skew_result <- calc_skewness(data_clean)
kurt_result <- calc_kurtosis(data_clean)
print(skew_result)
print(kurt_result)
The output looks crazy:
Skewness: 1710 (chi: 133356, df: 2925, p: 0)
Kurtosis: 3668 (z: 881, p: 0)
Are these numbers normal? What should I do next? Any help would be awesome!
Whoa, those numbers are totally wild! 
Have you checked if there might be some sneaky outliers hiding in your data? Sometimes a few extreme values can really throw things off.
I’m kinda curious - what’s the nature of your scale? Is it measuring something that tends to be super skewed naturally? Like, I dunno, how often people win the lottery or something? 
If the data’s legit and not just a coding hiccup, you might wanna look into some robust CFA methods. I’ve heard good things about the Satorra-Bentler correction for non-normal data. Ever tried that?
Also, have you thought about visualizing your data? Sometimes a good ol’ histogram can give you insights that numbers alone can’t. Might help spot what’s causing all that skew and kurtosis craziness.
Keep us posted on what you find out! This sounds like a really interesting challenge. 
wow, those numbers are insane! have u tried data transformation? maybe log or square root could help tame that craziness. also, check for outliers - they might be messin things up. if nothing works, u might need to consider non-parametric methods or robust CFA. good luck mate!
Those skewness and kurtosis values are indeed extreme. Before jumping to transformations, I’d recommend double-checking your data and code for any errors. Ensure there are no miscodings or data entry issues. If the data is truly that skewed, you might consider robust maximum likelihood estimation in your CFA, which can handle non-normal data better than traditional methods. Alternatively, you could explore parceling techniques to create more normally distributed item clusters. Just be cautious with interpretation if you go this route. Lastly, if all else fails, non-parametric approaches like PLS-SEM might be worth investigating as they’re less sensitive to distributional assumptions.