Categorical Count Data and ANOVA

wy38

New Member
#1
Hi, I hope this is in the right section.

I heard that ANOVA shouldn't be used for analysing categorical count data. If this is the case, can someone explain to me why and suggest ways I can analyse these count data? Any help will be greatly appreciated, thanks.
 

Dason

Ambassador to the humans
#2
Because ANOVA is used for a continuous outcome. It's possible to use it for count data but there are much much better ways. You'd have to explain more about what you're doing exactly for us to recommend anything though because there are A LOT of ways to do an analysis on count data.
 

wy38

New Member
#3
As an example, say I have an independent sample of 40, within each sample, a certain event occurs at a certain rate. The samples are split into 4 treatment groups, and I count the number of events that occur on each sample within a test period. The aim to is see if there is significant difference between the mean events occurred under each treatment group, and to give a conclusion on each treatment's effect.

Would carrying out separate t-tests be sensible? What happens if assumption for equal variance is not valid?

As you can see, I am a total amateur. I hope I've explained what I wanted to do well enough.
 

Dason

Ambassador to the humans
#4
I still say t-tests and anovas aren't necessarily the best route because those are assuming a continuous response. It's possible to do that stuff but there are better ways. Modeling counts using the Poisson distribution is better than assuming a normal distribution.
 

wy38

New Member
#5
I'm sorry, I'm afraid my knowledge of methods for testing significant difference in means is very limited. I don't know any way other than the two methods I mentioned. Could you give me a description of some ways you can use to tackle this? I really appreciate that you are taking your time to help me :)
 

Dragan

Super Moderator
#6
As an example, say I have an independent sample of 40, within each sample, a certain event occurs at a certain rate. The samples are split into 4 treatment groups, and I count the number of events that occur on each sample within a test period. The aim to is see if there is significant difference between the mean events occurred under each treatment group, and to give a conclusion on each treatment's effect.

Would carrying out separate t-tests be sensible? What happens if assumption for equal variance is not valid?

As you can see, I am a total amateur. I hope I've explained what I wanted to do well enough.
If you are dealing with frequencies (count data), and in view of your statement that your knowledge of statistics is limited, then why don't you attempt to do a basic chi-square test?
 

wy38

New Member
#7
Ah, yes I can try the chi-squared test, thanks a bunch. :)

You seem to be suggesting this method because I'm a beginner. What would a veteran do for this kind of tests?