1
$\begingroup$

I have an experiment to look at habitat preferences of beetles and the effect of competition. In the first test, a beetle is on his own in an arena and I measured the % of time he spends in each of THREE types of habitat (open, bush and underground). I repeat this experiment with several beetles. In the second test, beetles are placed together with a beetle of a competing species in the arena and they are again observed for their % time in each habitat.

I want to test if the amount of time spent in each habitat is affected by the presence of the competitor.

But how can I compare the % time in each of the three habitats when the competitor is absent (test 1) and present (test 2)? Which test can answer this question? Do I need to test the means?

My expectation is that in the presence of the competitor the use of Habitat 2 would increase.
Apologies, I dont know how to import data.

data: Habitat 1=

$\endgroup$
  • $\begingroup$ Even in test 1, I'm not sure a chi-square test of association would be helpful. This is because time is not decidedly countable. For example, if you observe a beetle for 12 hours, if you count where it is each hour you will get a different result than if you count each minute than if you count each second. $\endgroup$ – Sal Mangiafico Jan 15 at 9:28
  • $\begingroup$ time was standardised - so every test ran for 2 hours and location in every second was recorded .. does that make it appropriate? $\endgroup$ – Birdonawire Jan 16 at 10:06
  • $\begingroup$ for me, I wouldn't treat time as something countable... In your case, I think you might find the hypothesis test with a chi-square analysis isn't very useful simply because you have 7200 (?) observations per subject. With that many observations, even small differences many be statistically significant. So, I wouldn't put much weight in a small p-value without thinking critically about the difference in percents or time. $\endgroup$ – Sal Mangiafico Jan 17 at 15:02
0
$\begingroup$

Using a $\chi^2$ test you would always check 33 %. A linear model would be able to capture a more complex situation. With it you could set up hypotheses to test the % time in each habitat and whether is changes with number of competitors present.

$\endgroup$
  • $\begingroup$ Wouldn't it be problematic for the linear model that the percents in a scenario always add up to 100%? That it, as the percent in one habitat increases, the percent in another must decrease. So the observations of the dependent variable are not independent? $\endgroup$ – Sal Mangiafico Jan 15 at 9:30
  • $\begingroup$ @SalMangiafico you are right, I overlooked that. Maybe if the variable was not transformed and kept in raw values (time) it would work. Afterwards the raw results could be transformed into percentages. $\endgroup$ – user2974951 Jan 15 at 9:50
  • $\begingroup$ I think there is a type of model that is used for data that sum to a whole. Unfortunately I don't remember the name and can't find anything online about it. ... Whether as a percent or as a time, I would think that focusing on just one habitat would be approachable with a linear model, beta regression, or something similar. $\endgroup$ – Sal Mangiafico Jan 17 at 15:14
  • $\begingroup$ It might be "compositional data" $\endgroup$ – Sal Mangiafico Jan 17 at 22:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.