Thread: Student in dire need of help - Multinomial logistic regression

Hello everybody!


I am a student in biology working on my master thesis. I am facing a huge problem, the statistical analysis.
The due date is in 1 month (a bit less) and I am not sure of what I got and this is why I turn to you today.


I used SPSS 20 to run my multinomial logistic regression on my data. So here are the screenshot of the data :


My data


Where Temp is the temperature, Behavior is a 4 choices variable, Valley is a 2 choices variable, Month and Year are when the particular behavior was recorded. Lastly for the individual, the number correspond to the tag (so to a particual individual, I have 12 different individual).


My variables


for the non french speaker :


in type : Chaine is a string
numérique : numeric
Aucun = none
If you have other doubt just ask!


So now the important part : here is what I ran :


Multinomial logistic regression


And here is what I got :


Results


Here is the output of SPSS :


Number1
Number 2




Now my questions/concerns :


I have, for all my effects, a 0.000 p-value and it seems odd (particulary, I wasn't expecting a p-value < 0.05 for the valley) so here is my main concern, is my data formated in a good way ? do I need to enter some specific value somewhere in the options?


and my second question, how can I put those results in a graph? I have no idea and couldn't figure how to make a graph with a string as main axis in SPSS.


I really hope some of you here will be able to help me as I run out of time.


Thanks also for the one that will take some time for me,


Happy holidays


Lapive

Hey lapive.

That data looks like an absolutely horrible fit with your AIC values being ridiculously high and the p-values being ridiculously low.

If variables do not have a good dependency (and are closer to being independent than dependent), then this can happen.

Did you produce a correlation (i.e. a scatter-plot) of your dependent variable against the independent ones you wish to test?

Hey Chiro!

No I haven't produced a scatter plot.
Actually I came across an internet course about multinomial regression, that said that we don't need to look at that kind of things, maybe they are wrong then...
If they are indeed independant, what options do I have ?

I will try to make the scatter plot today and post it here!

Thanks a lot

Lapive

Hi, Lapive.

Could you please provide some context on your experiment? WHAT question are you trying to answer? Also, how large are your data? Do you have prior experience with binary or multinomial logistic regression? How married are you to using multinomial logistic regression? Is it too late to consider changing the design?

-Andy

Hey Andy!

So what I am trying to do is to check what factors have an influence on the behavior of my model (Alpine chamois, Rupicapra rupicapra rupicapra).
I have 11 indivudual and a total of 271k (a bit more if you need the exact number I can give it to you) observations (all individual mixed).

I am not "married" to using logistic regression at all, I have been told (on reddit) that it is the test I was looking for given my data.
My behavior column has 4 different cases (feeding, runing, moving and resting) therefor binary regression is out of the question. But unfortunatly I have no prior experience with that kind of statistics (therefore I have a lot of problems).

I can change my design if it doesn't take more than 3-4 days to test, otherwise I won't be able to write my thesis and this would be a huge problem.

Do you have some ideas concerning my problem ?

I want to thanks all of you that help me, especially at that time of the year

Olivier