Applying Cox's Porportional Hazard model I have a dataset on two groups. Group 1 (18 subjects) and group 2 (5 subjects).
Each subject in the groups had to perform a number of tasks which was timed. The time restriction was 60 seconds. If they didn't complete the task within 60 seconds they were asked to stop. Otherwise the time variable is the time of completion.
This means I have a right censoring at 60 which constitutes of about 10 % of the data points.
Group 1 has around 380 data points (380 tasks with a time between 0 and 60, or 60 as the censoring) and group 2 has 52 data points.
What I want is to do an analysis on this to see if group 1 and group 2 are different in their mean time of completion.
I'm pretty sure I can do this with Cox's proportional hazard model.
I have the time variable, the censoring variable (0 or 1) and the group variable (group 1 or group 2). Can I do an analysis solely on these variables?
Could the censoring variable be used as the Event Indicator and the group variable as the variable. Obviously I would use the time variable as time.
I'm not that strong in statistics, so any help would be greatly appreciated. This is for my Master's Thesis which is due end May, so this darn statistics is an annoying hurdle for me right now.
 A: In principle that sounds right. Obviously, the censoring indicator is 1 when the event indicator (i.e. actual time was observed) is 0 and vice versa (most software will let you indicate what the observed and what the censored observations are, depending on the software there may be different ways of doing this). You will then get a hazard ratio (likely with 95% confidence interval and p-value) with a hazard ratio of 1.0 indicating that the two groups complete the task about equally fast (>1 would be one group doing this faster than the group used as the reference group, while <1 would be slower). Which way around the comparison is depends on what group is used as the reference group, which you can probably choose in your software. The software may also alternatively report log-hazard ratios (i.e. taking the log to the base e of the hazard ratio).
If these were randomized groups that are being compared, this may be very straightforward. If that is not the case and these are e.g. people doing with certain characteristics, you may have to look into whether other things ("confounders") than this characteristic could explain what you see. E.g. it could be that something else (which you have hopefully recorded data on) explains differences that you see. You may have trouble doing much to adjust for this given the small sample size, but you may want to try and depending on whether it makes sense in the context mention it in the discussion of your results.
