Which statistical test? Please help...

#1
Hi, I'm just wondering if anyone can help. I haven't done any statistics in such a long time and I am really struggling to find the appropriate test, I have been looking online, through books, help guides and I just can't figure out which test to do.... I am trying to statistically show whether or not 23 different animal species show a difference across / preference towards four different management programs.... The table is the total number of sightings of each species, for each management... Hope that makes sense.
Any help would be absolutely amazing. Thanks in advance.
 

hlsmith

Not a robit
#5
What is your sample size and expected cell counts were < 5, but you can still use it given they aren't zero. I believe the 5 rule has been shown to be trivial. Also, you can try to use Fisher's Exact test, but it may take awhile to run if you have large sample.

Do you have to look at the 4 outcomes at one time? If not this can be broken into pieces and perhaps Poisson of similar model used.
 
#6
(21 species I'm comparing sorry not 23)...
Total sample size is 3316. And 22 cells are <5, as you mentioned none are 0, but 10 if them are - 0.3/0.1/0.2/0.4/0.8 etc. But with a Chi-square value of 211.516, surely that is too high to use..?

N = 3316, is too large for Fisher's?

My main aim is to statistically support that X species show preference towards X management program, as from the table/graphs some species show such an obvious preference towards certain managements - but I just don't know how to statistically show that.
 

Karabiner

TS Contributor
#7
My main aim is to statistically support that X species show preference towards X management program, as from the table/graphs some species show such an obvious preference towards certain managements - but I just don't know how to statistically show that.
Did you have precisely this idea in mind before you inspected the table/graphs,
or do you want to attach a signifcant test result to an interestingly-looking
pattern in the data?

With kind regards

Karabiner
 
#8
Did you have precisely this idea in mind before you inspected the table/graphs,
or do you want to attach a signifcant test result to an interestingly-looking
pattern in the data?

With kind regards

Karabiner
Thanks for responding. The latter of the two...
I have managed to do a test on the change in total abundance (all species observations accumulated) to highlight the overall success of a particular management strategy, whilst also showing the apparent failure of another.
But I want to go into a little more detail, and look at species-specific preferences. And as I mentioned, from the table and a simple split bar chart some species show such obvious favour towards certain managements, but I am struggling to find a way to statistically show this...

Thanks.
 

Karabiner

TS Contributor
#9
If you eyeball a complex pattern of results, and then tailor a test to proof a statistically significant
relationship, then the p value is distorted or even worthless. Eyballing means that implicitly a large
number of "tests" are performed by the researcher, until something interesting is found. But these
large number of preliminary "tests" are not accounted for by the statistical test of significance
(the data dredging problem). Since the hypothesis you investigate is customized to fit the sample
data, common solutions to your problem with the analysis (e.g., collapsing of categories) would
only exacerbate the underlying problem, I'm afraid.

With kind regards

Karabiner
 

Miner

TS Contributor
#10
Karabiner makes a valid point. What you have done is called exploratory data analysis, which is a common graphical approach in industrial statistics and quality control. However, the next step is to use that to formulate a theory, design and run a new, more focused experiment to test that theory, then perform the statistical analysis. That may be difficult in this situation as I assume this data were collected over a long period of time? Industrial experiments can usually be completed in a short period of time.
 
#11
Sorry maybe I have confused the matter slightly. After re-reading your question, I totally understand where your responses have come from.
The question of whether I had this question in mind prior to seeing the results, I did.

I have been collecting the data over the last year, and one of my proposed hypotheses is looking at species diversity. I have generated Shannon Wiener Diversity Index's for each management type, providing evidence that there is a bias in diversity across the managements.
I then wanted to go into further detail, as from a biological stand point there are a huge amount of species-specific mechanics for such habitat preference. But for me to do so, I would like to generate statistical backing. As I have generated a number of visual graphics and descriptive statistics which support species-specific preference but I am just not sure which test/s I could use to support the evident difference in preference.

Hope that has made sense, and I fully understand and appreciate that a non-significant result is as valid, if not more valid than a significant one. If this was the case I would happily report one, but I just get frustrated with myself as I cannot remember which test to use to show the statistical significance I know is present.