Hello people,
I want to know how to use GLM to compare mean number of granivore birds for "high water level" years and "low water level" years as shown in the picture provided below. This is an arbitrary data set I made up but the data I have is similar and they are not normally distributed. What step should I follow? Where should I start? Should I use GLM or something else? Should I first determine whether the data fits neg.binomial or Poisson distribution? If so, how can I do it with R?
I tried using Mann-Whitney U-test but I think I should use something stronger. I would be glad if somebody can explain to me what to do in plain language. Thanks in advance.
Kaan,
You might start with an informative data exploration. Zuur et al. (2010) is a great place to start (there nice supplementary material for R user at the Wiley page too).
Just to clarify some terminology–so your problem might be better communicated to others–general linear models and generalized linear models are similar, but use different error distributions. You're probably familiar with members of the general linear model family: ANOVA, ANCOVA, linear regression, t-test, and more. Generalized linear models relate a linear model to the data via a link function—this allows for linear relationships to be made from data that have error distributions other than a normal distribution.
It's hard to make specific recommendations without having the data or a reproducible example, so I hesitate to say use one method or the other. I would advise, however, that diagnostics for the assumptions of a general linear test should be made on the residuals. It's common to check for normality (homoscedasticity and distribution) before the model is applied, but we really should be checking the residuals for such assumptions. (Don't forget independence of observations, e.g., spatial and temporal dependence).
An R package that I find useful for exploring distributions and fits of data to distributions is the fitdistrplus package (link to vignette below).
Best of luck with your analysis.
Caleb
Zuur, A. F., Ieno, E. N., & Elphick, C. S. (2010). A protocol for data exploration to avoid common statistical problems. Methods in ecology and evolution, 1(1), 3-14. Article Zuur AF, Ieno EN, Elphick CS.. A protocol for data explorati...
Delignette-Muller, M., & Dutang, C. (2015). fitdistrplus: An R Package for Fitting Distributions. Journal of Statistical Software, 64(4), 1 - 34.
Article Fitdistrplus: An R Package for Fitting Distributions
Hello Kaan,
Modelling count data can be done with Generalized Linear Models. Typically, the counts are modelled with a Poisson or Negative Binomial distribution. Tests like the Kolmogorov-Smirnov can be done to test the appropriatness of a given probability density function.
A site which does a good job of showing how to model count data In R using Generalized Linear Models is:
http://datavoreconsulting.com/programming-tips/count-data-glms-choosing-poisson-negative-binomial-zero-inflated-poisson/
Best,
Aran
Dear Aran,
I followed your advice and checked the link. I used it to perform the model fitting. What comes next? What should I do now to see if the means of the two groups are significantly different from each other? Thanks in advance.
Best wishes,
Kaan
The objective (i.e. what is wanted to know) is not clear. It is required to describe the objective clearly and specifically.
The data, shown in the table, is a kind of time series data. It may be possible to extract information from the data. However, the size of data is small. It will be convenient if more data are collected.
- Badges
- Science topic