Fisher's Exact Test

#1
Hello,:wave:

I'm currently a doctoral candidate in wildlife ecology. I'm analyzing habitat selection data for bullsnakes collected over three years.

Initially, I ran a chi-squared test to determine if observed habitat use differed from expected values. However, I have several expected values below 5, and several observed cells with zeros. I was told that perhaps a Fishers exact test would be the best way to analyze my data.

However, I can't conform to a 2X2 contigency table. I have to analyze habitat use data for males and females in 14 different habitat types...and repeat these analyses for 2003, 2004, 2005.

At any length, I'm having trouble determining: (a) if this will work for me and (b) if there is a statistical software package available that can analyze it.

Thank you for your time,
Josh
 

JohnM

TS Contributor
#2
Even though your data violates some of the criteria for using chi-square, and since it sounds like you cannot combine categories or whittle(sp?) it down to a 2x2 table, then I would just run a chi-square anyway....and see how it goes.

I don't think you have many other options....
 
#3
Thanks, John.

In the chi-squared analysis, what should I do in habitat types where I have no observations (ie, zero in the observed cells)? Eliminate them for that analysis?

I'm assuming I can run a G-stat test (logliklihood analysis) to "correct" the fact that several of my expected values are below 5.

Thanks,
Josh
 
#4
Hi John,

I know you must be ridiculously busy, but I was just curious if you had gotten a chance to think about my previous question?

Thanks alot for your time,
Josh
 

JohnM

TS Contributor
#5
I would keep them in the analysis. Just because your sample is a 0, doesn't necessarily mean that it will always be 0.
 
#6
JohnM said:
I would keep them in the analysis. Just because your sample is a 0, doesn't necessarily mean that it will always be 0.
Thanks for the reply, John.

Just curious: isn't it a prerequisite for chi-squared tests that observed values can't be zero...in other words, no "observed" cells can have a value of zero? Thought I read that somewhere, but could be completely wrong.

Thanks...I should be just about done pestering you.

Josh
 
#9
Hello John,
You seem very knowledgeable and I was really hoping you could help me.
I have a 2x4 table with nominal data (the columns are simply YES/NO, the rows are four categories)

Category A: 7, 13
Category B: 15, 5
Category C: 15, 5
Category D: 19, 1

I am hoping to test the significance of a couple of the categories to each other (2x2), but also to assess the significance of the whole table (2x4, although I am not really sure how to interpret the meaning of the significance here).

I understand that as my sample sizes are small (as you can see, Category D features "1" under NO, and each category has only 20 people) so I should be using Fisher's Exact Test. Is this correct? Can Fishers also be used in a 2x4? And what does it mean if this result is significant?

Many thanks!