Which test to use - correlation or differencies in biology?

[Joachim]

Hello!

First question here, hope someone of you can give me a helping hand.


Me and a friend have performed a basic study on peregrine falcons.
We have studied 35 different locals suitable for the species, and looked at 2 parameters:
"vertical drop" and "vicinity to water". We've given these parameters points; "0" for "insufficient", "1" for "sufficient", which leaves us with two different points for each local.

Our purpose with the study is to see which types of locals (terrain, vertical drop, vegetation, vicinity to water) is prefered by the bird.

How can we use this data in a statistical test, and what test should we be using?

We have 35 observations, does this equal 35 samples?
Should we investigate "differencies" between locals, or "correlation" among "likelihood for the birds to occupy" and the two parameters stated above?

If I've got this right, this would leave us with two suggestions:
Kruskal-Wallis test
or
X2-test (chi-square)

Or am I mistaking?

With regards
/Joachim

[victorxstc]

Hi dear Joachim

So you are telling that you checked 35 locations with different characteristics in terms of vicinity to water and vertical drop (what is this one?), and recorded whether or not they were occupied by falcons?

In that case you will have a dependent variable (if a location is occupied or empty), and three independent variables (vertical drop, vicinity to water, and types of locals [terrain, etc.]).

Right?

[GretaGarbo]

I am glad to se your post since I care especially much about eagles and hawks here. (I get the impression that some participants don’t. So I hope that especially (the user) “Trinker” join in with suggestions.)

Maybe, maybe your threads could be moved to Biostatistics to maybe get more readers?

I would not run Kruskal Wallis. I can’t se that it is relevant here.

Of course just run through the cross tables and chi-squarred test. You need to have that as descriptives anyway.

If the response variable is 0/1 (“there is no falcon”/”there is a falcon”) I would run logit regression (the same as logistic regression).

Normally in a regression model, you need all relevant explanatory variables (other wise the estimates will be biased and inconsistent (ie “bad”).

Think of other missing variables. I would suggest “closeness to humans” (ie disturbances).



@victorxstc, I hate waving little figures that distracts attention when reading a difficult text. Please, please edit and delete that one!!! (/Edit: thanks for having deleted that one.)

Maybe you both ecologist can interact with each other and make suggestions:


http://www.talkstats.com/showthread....-linear-models

Maybe, the above threads could also be moved to Biostatistics to maybe get more readers?

I especially hope that you will se many raptors!

[hlsmith]

It seems like one sample of 35 observations out of countless locales (total population). I do not see much information that correlations could provide. Ideally, you may want to run three Fisher's exact tests, 2 X 2 tables. Then if significance is found in all three tests then run multivariable logistic regression (bird present = dependent, then three variables are independent variables) and see if all three remain significant. However you may be a little underpowered to have 3 variables (not enough observations) in your model. If two variables are significant in Fisher's Exact tests then perhaps follow-up with a Cochran Mantel Haenzel (CMH) and if only one variable is significant then use that variables Fisher's exact test only.

Just suggestions.

[victorxstc]

@victorxstc, I hate waving little figures that distracts attention when reading a difficult text. Please, please edit and delete that one!!!

lol, sure I would delete it

[Joachim]

Hi dear Joachim

So you are telling that you checked 35 locations with different characteristics in terms of vicinity to water and vertical drop (what is this one?), and recorded whether or not they were occupied by falcons?

In that case you will have a dependent variable (if a location is occupied or empty), and three independent variables (vertical drop, vicinity to water, and types of locals [terrain, etc.]).

Right?

Swift response!

We've checked 35 different locations, and given these points for 2 characteristics: (1) vicinity to water and (2) vertical drop (steep, how steep the cliff is). (Characteristics 2 also includes how much vegetation the cliff has, I was unclear on that one, sorry).

This gives us 35 X 2 = 70 values.
Out of these 35 locations, 4 locations have mating birds.

[victorxstc]

Swift response!

We've checked 35 different locations, and given these points for 2 characteristics: (1) vicinity to water and (2) vertical drop (steep, how steep the cliff is). (Characteristics 2 also includes how much vegetation the cliff has, I was unclear on that one, sorry).

This gives us 35 X 2 = 70 values.
Out of these 35 locations, 4 locations have mating birds.

So as guys told, you can run a binary logistic regression to analyze the associations between the independent variables and the presence/absence of mating birds. Other methods such as chi-squared are good too, but I personally prefer a multivariate test.

[Joachim]

So as guys told, you can run a binary logistic regression to analyze the associations between the independent variables and the presence/absence of mating birds. Other methods such as chi-squared are good too, but I personally prefer a multivariate test.

Great. Thanks for all the help, I'll look into that.

Sorry if this was posted in the wrong section.

[vijay]

Hi Guys

this is my first post on this website and i am not sure if i have reached a correct thread but i need your help to complete my dissertation analysis. i am in serious trouble so please help me out of this situation

I have to analyse relationship between:-

1. Monetary Reward (Independent Variable) and Employee Retention (Dependent Variable)
2. Non Monetary Reward (IV) and Employee Retention (DV)
3. Monetary & Non Monetary rewards and their relationship with retention

I have to do this in SPSS, i have very little knowledge on SPSS. all the three sections Monetary, Non Monetary and employee retention are on likert scale with 10 questions in each section. i am having problem in entring data in spss.


I really apologies if i have posted in a wrong section


Thanks

[victorxstc]

dear vijay, based on my learnings from experts here, may I suggest that you can run an ordinal logistic regression.

For doing so in SPSS, you can consult Google, and find resources such as this one.

[vijay]

Thanks for replying

i have gone through the link as given by you but still i am not clear as to how enter data in SPSS. can you please message me your mail id so that i can forward you my questionnaire and data set that i have created and then can you please suggest where am i making mistake.

Thanks

[GretaGarbo]

@vijay, It is nice that you are posting. But in this thread we discuss raptors and Joachims problem. I suggest that you create a thread of your own. Just copy what you had and maybe add something about how many observations. If you want advice about spss, then maybe add a few data values and variable names and code if you have. Insert it in code-marks (look around to find that.) (And skip the “ in serious trouble so please help me out”, I mean many of us are in trouble. Frankly I read if I think it is interesting. Sure I would like to help you but…)

@Joachim. (I don’t think you posted wrong but maybe you could gotten more reader in Biostatistics. I don’t know)

But, you only have 35 observations. That’s small. And even worse only 4 falcons (or observations of nests). I don’t think you can get any conclusions here. I am really sorry to be a bringer of such bad news.

By the way, the tests and formulas in logit are based on larges sample asymptotics (that is not valid in this case) so logit will not work here.

The formula below could be used on larger samples (say more than 20 falcons each) so really it is irrelevant



P +/- 1.96*sqrt(p*(1-p)/n) = 4/35 +/- 1.96*sqrt((4/35*(1-4/35)/35)) = 0.11 +/- 0.10

Proportion nest is about 11% and margin of error is about 10%.

But you really want to compare say 3 nests in high slopes with 1 nest in small slope. That is a very small sample.

Maybe something could be done with cross tables with Fishers exact test, but frankly I would not hope too much (really sorry!).

I am really sorry that there are so few raptors!