0
$\begingroup$

I could really use some help to determine what statistical test to use in my particular case here.

I work with insects and I have two sets of individuals, a test set and a control set, on which I applied the same treatments.

The data collected is a proportion type one: for each repetition in each treatment I get a yes/no answer to the question "has it emerged?". I then compile everything in a global proportion per treatment per data set).

My treatments are elapsed time before evaluation (development time), each hour from 28h to 35h. I have n=40 for each treatments, but they are not the same individuals as we dissect them everytime.

I get this kind of results (sorry for the ugly table...) :

Treatment(h)___28___29___30___31___32___33___34___35
Control________0____0____0____0___59___70___100__100
Test__________50___70___90__100__100__100___100__100

I'm trying to find a test to test the hypothesis that the control and the test are/arent emerging at the same time / speed.

Could anybody give me a hint about the test to use? I'm really stuck here, I praise to the stats gods to send me their avatar quick!

Cheers!

Improve this question
$\endgroup$
1
  • $\begingroup$ Values in the table are % emerged individuals $\endgroup$
    – lerussophile
    Commented Jun 27, 2017 at 19:20

1 Answer 1

Reset to default
0
$\begingroup$

You could do this using a logistic regression. Use one predictor as the time/treatment (possibly with a polynomial or spline term for non-linearity) and another predictor as a 0/1 variable indicating control or test, also include the interaction between them. If there is no difference in the groups, then the coefficients on the 0/1 variable and the interaction will not be significantly different from 0. I would suggest a full/reduced model test comparing the model with the 0/1 term and interaction to a model that only includes the terms for time/treatment.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Hello again! I've been working on my logistic regression to finally realize that 0 cells are problematic. I happen to have structural zeros, because looking at my data from the first message, you can see that the control never emerges earlier than 32 h. Is there a way i could work around this? $\endgroup$
    – lerussophile
    Commented Jul 5, 2017 at 13:18
  • $\begingroup$ @lerussophile, structural zeros are where the science says that something is impossible (count of pregnant males must be 0, etc.), your data looks like observed zeros (nothing emerged, but there is no scientific reason that they couldn't have). If they are just observed then logistic regression will still work fine. $\endgroup$
    – Greg Snow
    Commented Jul 5, 2017 at 15:08
  • $\begingroup$ You're right, I'm the one saying there can't be emergence before 32h, and they are probably observed-type zeros. Should I take any measures to prevent overestimating my coeffs or odds ratios? Like Firth's bias-reduction in the brglm function, in R? $\endgroup$
    – lerussophile
    Commented Jul 5, 2017 at 16:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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