Statistical Analysis Advice Required

#1
Hello,

I posted a topic yesterday, and I assumed I had the issue "solved" so to speak. However, after much deliberation last night I think I may be back to square one, plus I have a few other queries I would would be ever so grateful for help with.

I want to establish whether covering affects the abundance of insects onto a corpse. I was using a Chi-Squared test, but as the majority of my data values were below five, it was decided that I would use Fisher's exact test. When I performed this test I came up with a p-value of 1.000, which would suggest that there was no affect by covering; but to look at the data it would be obvious that there is a difference in abundance.

An example of my data set is attached as an excel spreadsheet. I believe the issue to be the number of data points with zero. Essentially, I would like to know where to go with this data; i.e. what tests can I do?

I also have to investigate the correlation between temperature and rainfall on overall abundance; can I do both factors at the same time?

I also have access to Excel with XLStat and Minitab.

Sorry I had to post a separate thread, and thank you for any help; i'm a complete statistics novice.
 

gianmarco

TS Contributor
#4
A significant result is returned by:
a) XLStat (chi-square test; chi-square with Monte Carlo simulation)
b) PAST (Monte Carlo p)

Regards
Gm
 
#5
Within your XLStat sheet the fisher's exact is also 1.000; however you have not highlighted this as where you obtained your p-value.

I'm not sure what the Monte Carlo simulation is? Does it allow for expected frequencies less than five?

Thanks.
 

gianmarco

TS Contributor
#6
I was just stressing the results similar to that obtained by Dason.
As for Montecarlo simulation applied to chi-square and its context of use, this is an issue that I am trying to investigate...
I would like to have feedbacks on this by more experienced people here.

Gm
 
#7
Right, ok. I'm afraid I'm completely out of my depth with this level of statistics. I really only want to know whether the covering type is significant on the abundance of arthropods.

FM
 

Dason

Ambassador to the humans
#8
I was just stressing the results similar to that obtained by Dason.
As for Montecarlo simulation applied to chi-square and its context of use, this is an issue that I am trying to investigate...
I would like to have feedbacks on this by more experienced people here.

Gm
I'm guessing the Monte Carlo simulation you're talking about is a just a randomization approach to get an estimate of the sampling distribution under the null hypothesis? I like that approach especially for low cell counts and/or low expected counts.
 
#9
I'm guessing the Monte Carlo simulation you're talking about is a just a randomization approach to get an estimate of the sampling distribution under the null hypothesis? I like that approach especially for low cell counts and/or low expected counts.
This is all getting a bit double Dutch to me...
 

gianmarco

TS Contributor
#12
Thanks Dason,
I was guessing something similar, but I lack the appropriate stat terminology.
I would like to learn more about its context of use and, in particular, if it can be used when chi-square "assumptions" are not met (as in the case object of this thread).
Any bibliographical reference is welcome.

Thanks
Regards
Gm
 

Dason

Ambassador to the humans
#14
The results I returned and partially confirmed by gianmarco are pval < .05. I get something like .00001 actually but I don't know what GM ended up getting (he only said significant). Is there a reason you want a p-value of 1?
 
#15
Hello,

There is no reason why I want a p-value of 1.000, it's just what I keep obtaining using XLStat for Fisher's exact. gianmarco used a monte carlo simulation, which I assume isn't the same thing, although I have no idea as i've never heard of it before.

I investigated using R, but alas, have no idea how to use it i'm afraid.
 

Dason

Ambassador to the humans
#16
I've never used XLStat but I think something must be going wrong. Regardless of significance - with the data that you have you shouldn't be getting a p-value of 1.
 

gianmarco

TS Contributor
#18
hmmmm....this discussion is getting more and more intricaded by the minute....
I do not know if something is wrong with the way XLStat is running the test. So, I would leave is aside for the time being.
Secondly, the chi-square test and the same test performed using the Montecarlo simulation is pointing to a significant result, or (to keep with the chi square terminology) to a significant dependence between rows and columns, that is between the two categorical variables under analysis.

This significant result should be (if I am not mistaken) compatible with the one Dason got using (again, if I am not mistaken) Fisher test with R.

So, if chi-square with Montecarlo simulation can be' used in this context (table with sparse values), I think that your issue should be solved: you are in the position to test a significante association between rows and columns.
If this is true, you could report the Xlstat output for the chi-square with Montecarlo simul, or the Past output.

I hope this helps
Regards
Gm

I look forward to hear about the use (and validity) of montecarlo simulation in case for chi square in case of tables with sparse values.