Interpreting an interaction effect

#1
Hi All,

Last year I ran an experiment where I measured the behaviour of a songbird (how long it took them to approach a speaker playing their song) in urban and rural sites along a north south transect covering 20 cities in the UK. I was interested to see if the behaviour of birds differs according to how far north they are (latitude) and if they live in an urban or rural habitat.

My results show a significant effect of latitude and also a significant interaction between latitude and habitat (latitude X urban/rural).

The results were:

Latitude: P = 0.001
Habitat: P = 0.78
Latitude*Habitat: P = 0.007


Where habitat is a factor with two levels (urban/rural) and latitude is continuous.

I used the package glmmADMB in R with the ID of the birds as a random effect to control for repeated measures. As my data are overdispersed I used a negative binomial distrubtion and a log link.

My question now is how should I interpret the interaction term? Is it acceptable to run separate tests on urban an rural birds to determine if the slopes for each group are significant?

If I do this I find a significant result for urban birds and a non-significant result for rural birds. When I plot this on a graph you can see that the slope for urban birds is steeper.

Alternatively should I just state that there was an interaction which suggests that latitude affects the behaviour of urban and rural birds differently?


 
Last edited:

kiton

New Member
#4
Note, to add an interaction effect to your model (study), you should have strong theory behind it. I.e., it shouldn't come just from the empirical analysis, but be based on some plausible explanation. Also, what is the correlation between habitat and latitude?
 
#5
Hi Kiton,

I agree there should be a strong theory behind it, and I am quite unsure about using this result for that reason. I didn't set out to test this, I just noticed it in my results. That said, other studies have found latitudinal effects on behaviour in this species of bird so I thought it might be worth cautiously looking into.

I am not sure what you mean about the correlation between habitat and latitude. I ran experiments across the country in paired urban/rural sites. So for every urban site there was a rural site no more than three kilometres away. If you mean what could explain the result, then I am not sure. The behaviour of urban and rural birds does differ in a lot of different ways though.
 

rogojel

TS Contributor
#6
Hi,
just my 5 cents: if your analysis is done correctly and your daza does show an interaction, I would see no reason not to report it. Of course it is a lot more elegant to have a theory to explain it but you might not be able to find it, unless you delay publication until you find the explanation.


regards
 
#7
Hi Rogojel,

Thanks for your comment. I think what I will do is report that I found this but without much interpretation. I think that as long as I am cautious about how I present it it should be ok. This was not the aim of the experiment and I have other results too, it's just something potentially interesting that turned up in the data...