Poisson distributed data

cjm

I have two data sets for a behavioural expt. with two sets of birds 1 set equal sexes the other unequal. They have been offered the same 3 separate options of a total variable. This was offered 7 times a day over 16 days with 2 diet options before & after microwaving (unequal days 4,5,4 & 3 respectively mistake). While I can do most stats individually Jasp doesn't seem to like zero's when an option is not taken by a subject? This is a very low frequency behaviour, hence the hours spent collecting data on freeze frame footage! Can someone please help with:

Correct analysis for (I assume GLM) & inputting data correctly as Jasp keeps returning unable to compute & remove factors? Many Thanks if you can HELP!!

EJ

In order to help I think it would be good to see some more background -- maybe a screenshot of the data, and a more precise description of what it means for JASP "not to like zero's". It is also not clear to me why you would model this as Poisson -- if they get three options repeatedly then the data appear to be multinomial (trinomial, in fact)

E.J.

cjm

Thanks for your reply E.J.

Just been working through exploratory factor analyses & covariance structures, I need the Anova ( Kruskal Wallis) for two substrate options by birds the third & total work fine. however I assume it is birds that it is individuals that don't use a substrate option that cause this error message.

Is there a way to get around this? Do you want the CSV file attached?

BudgiesOriginalData NewCSV.csv

cjm

Sorry should also have explained all substrates aren't normally distributed appear to have poisson distributions.

EJ

I've downloaded the csv but I am unclear about what analysis fails -- what is the DV and what are the predictors? Can you email me the .jasp file? Judging from the error message, you probably have too few observations in one of your design cells to carry out the desired analysis...

E.J.

cjm

Hi EJ, I realised in the data 2 male birds did not take 1 or other of the options (invariably zero). So I've have split by sex for the analysis of the interested female group. They have a diurnal usage pattern for all substrate options for all birds but the mud usage pattern changes on day 10. I have 2 jasp files attached which keep giving failed response codes? I have also attached an image of what I'm trying to demonstrate in jasp?

EJ

Hi cjm,

The forum does not seem to take the .jasp files. Feel free to email me the files. I would need more handholding though -- in your figure:

(1) what is the DV, exactly?

(2) the points are averages?

(3) the pattern across time is measured for each bird? (a "within-bird" design?)

(4) what exactly are you trying to demonstrate? That there is a U-shaped pattern?

(5) What do the different colors refer to?

Cheers,

E.J.

cjm

1. The DV has three options: mud, fine, & large all forming the total substrate. I was, initially, interested how the three options compared with 2 different seed feeds with respect to microbial loadings. However I am fairly confident that the variances between all 4 options are a function of the daily variance between days not diet. Then I have investigated the other factors used in the design. Time of day (hours from dawn), sex, between days & between birds.
2. Yes, they are the means generated from the one way Anova (Kruskal Wallis) I had used to unpick the data set to find which are the significant factors.
3. Yes, the pattern across each time frame within each day over 15 or 16 days for 4 female birds. ie 4 birds x 7 time frames/day x 15 or 16 days = 60 or 64. Day 10 having been controlled for 15 days, as being only significant for mud by day 10. Hopefully keeping all other variance factors equal. As the single plot for 1 day is only 4 birds for each time frame, a bit meaningless statistically, I guess.
4. I'm fairly confident of this usage pattern, however there is an anomaly within the data. The number of 0 frequencies. A mean plot of frequencies ie 1/Zero's = motivation (do the behaviour or don't) gives a constrained distribution curve. Obviously zero's are relatively meaningless but makes me wonder whether this is inverse gaussian data requiring some other complex analysis? I remember reading about Poisson data & their frequency distributions tending towards normality with increased counts. Is this the reason for the note on the Linear Modelling analysis about a "competing model" which gives a lower R squared fit for the quadratic fit?
5. The colours are the plots for the means for both frequency & quantity for 15 day set & 16 day set, I hope this helps you see where I am going with it? Will attach the excel too! The other thing to be aware of is I am currently working on an old system so stuck with the trial version of Jasp & excel too, sorry! Thanks again CJM

cjm

Screen shot pic of linear modelling for point 4.

EJ

Hi CJM,

This is a more involved modeling project. When I look at the data it does seem as if a zero-inflated Poisson model is appropriate for your data. So personally I would check out a program like brms (a wrapper for Stan) and go from there.

Cheers,

E.J.

cjm

cjm

Thank you for your help I'll give it a go, cheers.