it is not totally wrong, but mean is not well defined in the right tail if the largest observation is censored. (in technical word :since estimate of S(t) is inconsistent in the right tail if the largest obs (say yn) is censored). due to this many (including me) peoples prefer median. if you really want to report the mean their are several proposal for this issue and you can use one of them

thanks for your answer Belay, but i didn't get your mean by saying their are several proposal ???????? would you explain more

as a variable, time is always non-normally distributed and hence calculated and expressed as median. As far as i have seen, time variable is analysed on median values to the point of interest.

You will notice that the distribution of survival time is skewed, survival time is usually described using median

Note however that If less than 50% of the observations are uncensored and the largest observation is censored, the median survival time cannot be estimated from a

Kaplan Meier curve.

SPSS does lots of wrong things: more precisely, it calculates many things which are usually irrelevant, and sometimes misleading. It was initially designed for the social sciences, not for biostatistics.

Why would you want to know the mean, anyway? Who is interested in the average survival time of thousands of people? The median means something to an individual person: 50 percent chance of surviving that long. The mean means something to an insurance company, perhaps.

@Richard: Who would want to know the mean? Well, it's sometimes useful. Not when modelling actual death, but when using time-to-event analysis to model other less lethal endpoints. People tend to forget that recovery, employment, marriage etc are all suited to analysis by time-to-event methods.

The mean time to event requires that all times to events are known. This happens often – when analysing such endpoints as time from onset of symptoms to admission, or delay between admission and discharge, we will generally have complete data.

In such cases, the mean can be useful. The mean admission time will allow us to estimate how many hospital beds are needed to cope with given admissions incidence.

The median, and other quantiles, do not depend on all the times to event being known, just that q% of the observations have known times, where q is the desired percentile.

Good day Eman. Yes, you can use mean in any situation of survival analysis. If you put here a dataset of less than 10 periods of time, without explaining details of treatment I will work it in excel. It is valid to express any Cummulative distribution function in "medias" as units of itself, keeping the real value of media U# to reintegrate it at the end of analysis, so CDF real = U# (real value per media)*CDF ( in medias) for transversal analysis and for longitudinal one.

I read your question several months after you published it. Sorry, emilio

Thanks for giving valuable insights on survival data

Thanks alot for your fruitful answers and comments I agree that most of survival times are non normal and the censored data mostly even if my endpoint was not death such as remission or response to therapy exceed 50% however, i find CI missing for median survival times in spss output and i wonder why???