CHI SQUARE - Ignore cases if grouping is not clear?

[Mase]

Hi everyone!
This is my post, so let me say hello to everyone: Hello world!
Ok, now to the business .
Context. I'm studying if there are differences by sex in my subjects' anxiety, which is measured by a scale (HADS-A) with quantitative direct score and then cathegorized into Normal (no anxiety), clinical (anxiety) and doubtful (not clear if there's clinical anxiety). I applied Chi-Square and found that 2 cells (women in the 'doubtful' and 'clinical' groups) had an expected frequency under 5.
My attempts and reflections. The Manual I work with suggests that if the assumption is not satisfied the categories should be grouped. However, I don't think it's correct grouping the 'doubtful' group with either the 'normal' or the 'clinical' group.
Question. Can I perform the Chi-Square only with the cases of the 'normal' and 'clinical' groups ignoring the 'doubtful' ones? Or should I just skip to Fisher's exact test?
Thank you very much in advance for your time!

[Karabiner]

then categorized into Normal (no anxiety), clinical (anxiety) and doubtful (not clear if there's clinical anxiety).

Why that? Categorization does not seem necessary, and it wastes information.

I don't think it's correct grouping the 'doubtful' group with either the 'normal' or the 'clinical' group.

It depends on the exact research question you want to answer.
If you really just want to investigate sex differences regarding
anxiety, then you don't need to categorize at all (or just for
descriptive purpose).


With kind regards

K.

[Mase]

Why that? Categorization does not seem necessary, and it wastes information.

Hi Karabiner! Thank's for the quick response.
Typically, anxiety is higher in women than in man, but it doesn't have to be clinically important. Patients in my sample are facing surgery and I want to know if there are sex differences in the number of participants who fall in the different categories (specially between normal and clinical anxiety). Another reason for categorizarion is because we want to define groups based on their anxiety level.
I hope this additional info helps
Thanks again for your time!

[hlsmith]

So you have a reason and want to categorized them, this is part of your purpose.

If it seems inappropriate to consolidate the groups then don't. Both options you described are feasible (i.e., Fisher's Exact or the single pairwise comparison). I would recommend just using the Fisher's Exact for all of the statistical procedures and forget about the chi-square.

[Mase]

Thank's!
Fisher said: no differences!
(By the way, I did the Mann-Whitney U test on the uncategorized data, and got a statistically significant result)

[hlsmith]

Just curious, who had a higher anxiety score in your sample per the nonparametric test?

[Mase]

Women (median = 7; vs median = 5 in men)
I'm not sure if it's the right data to give. As far as I underatand the U test Orders the cases by its value, so maybe I should answer you with the mean rank of each group? (112,20 for women; 86,63 for men)