A survival analysis using interval censored data. If you have actual times for the 0/1 data (i.e., right censored data) rather than intervals, you will get a higher quality analysis.
I have a data set of two groups of individuals that I want to compare. The data contains who survived and who didn't after 40 days and after 1 year. So this is not a numerical data but a kind of 0/1 data (survive/did not survive).
What is the test I should use?
Thank you very much,
Roy.
A survival analysis using interval censored data. If you have actual times for the 0/1 data (i.e., right censored data) rather than intervals, you will get a higher quality analysis.
Royan (11-09-2017)
Thank you very much. Do you happen to know which function does this test in R?
Sorry. I am not an R user. Hopefully. one will respond.
Added: I did find this article, but cannot advise whether these are the best R packages.
Royan (11-10-2017)
Royan (11-10-2017)
So for clarification, your description seemed like you may not have a time variable - just group and dead: yes/no.
For survival analysis you need the time piece as well, so the day they died, if they died. If you have this, than survival analysis can be used for say 40-day survival where you censor all non-dead at 40 days and the same thing can be done for the 1-year mark.
However, if you just have group and death status at 40 days and then again at 1-year without the time piece, this information is less informative and you would most likely use logistic regression.
Stop cowardice, ban guns!
Royan (11-10-2017)
Yes that is my case. Due to the limitations of the experiment I had limited excess and was only able to examine the subjects twice: once after 40 days and another after a year. Each time I wrote who died and who survived.
The thing is that I feel like regression might be problem since I have only two sampling times (three if you count time 0) and that is too low right? is there an ANOVA equivalent test that can maybe help me here?
I think you are fine just running two logistic regression models:
y (death at 40 days) = Group.
y (death at 1 year) = Group.
What is your sample size? Also, you will want to correct your alpha level, perhaps use 0.01, since you are conducting two comparable tests on your data sample.
Side question, how was death status collected? If you do not have a direct source, how do you know if you missed anyone to loss to follow-up?
Stop cowardice, ban guns!
I started with 12 individuals (each) of two species of corals. the death status was collected by observation.
I am not sure I understand. If I will run them in two separate models I am losing some data about the connection between them...? or maybe I there is something that I don't understand?
Can I use a non-parametric rank test like Wilcoxon?
If your dependent variable is binary Wilcoxon tests wouldn't work.
Stop cowardice, ban guns!
Hi hlsmith,
how about the following idea, that could work with one continuous IV: build two groups, one for the status Dead and one for the status Survived and do a two sample test for the IV (t-test or Wilcoxon, depending). If there is a significant difference between the two groups we have a possible effect, if not then not.
I have been wondering if this could actually work instead of a univariate logistic regression?
regards
Tweet |