Analysing count data? PLEASE HELP!

I'm an MSc student and not particularly statically minded and I would really appreciate some help on going about analysing count data! Please try to keep is simple!

I have two species, the common frog and marsh frog and I have count data on both species since 1980. The marsh frog is non-native and I am trying to ascertain whether since the introduction of the marsh frog they are negatively impacting/outcompeting the common frog. If they were outcompeting you would expect to see the population counts of the marsh frog increase whilst the population of common frogs decrease. I want to statistically test this "model/hypothesis"!

How is it best to go about this? I have already ascertained that counts for both species is non-normally distributed so I'm assuming non-parametric tests. I don't really know where to start, maybe a Mann Whitney U test or Kruskal-Wallis test? Is it best to group the data counts i.e 1980-1984, 1985-1989, 1990-1994 etc!

(Sorry I cant upload the actual data to a public forum, I have not got permission for this but please find attached fabricated data!)

Many thanks in advance!


TS Contributor
as a first shot, just to see what happens, you could try a simple regression model with the count data as the independent variable and year and species as dependent variables. If you had a significant interaction term between year and species that would be an indication that the two species evolve differently in time.

If the model is unsatisfactory, e.g. the residuals have a significant autocorrelation you could try more sophisticated models.

Your simulated data yields a linear model that is almost ok, except that the common frog values are unusually large in a few years.

I wouldn't use a Mann Whitney test when I have your data. Even though they are discrete, you have more than 30 observations, just do a time-series analysis.
A MW test is usually indicating some potential difference in distribution, median between two samples but your hypothesis seems like you're not only comparing but you want to prove a causal relationship. It seems like the introduction of the marsh frogs actually increases the number of common frogs.
Thank you once again! rogoje

kaisermuhle what you posted is pretty much what I'm looking for! From the very simple count against year graph that I constructed I could see that actually the population of Marsh frogs was quite stable, whilst the common frog population actually increased regardless of the introduction of the non-native frogs! Thank you for posting a screen shot of the time analysis, could you please give details of how to set this up in SPSS/Excel? and then how to interpret it? I very much appreciate the help! My email address is
rogojel To do a regression with one independent variable and two dependent variables am I right in thinking that a multivariate regression is required, which in SPSS is achieved through the GLM-multivariate option??

Please see attached pictures of the multivariate regression of my data!

As all my P-values are less than 0.05 what does this say about the significance between species? Is there a relationship? What further tests (if any) would I need to do? My email is ,please contact me there if you prefer!

Many thanks!


TS Contributor
unfortunately i could not read the thumbnails, but the residuals do not look bad. If you have a reasonable model, you should look at the coefficient of the interaction term. That will tell you by how much the slope of the regression line changes if you go from one species to the other. BTW if you can send some more realistic data to look at, my email address is