# How to account for the subjects who are no more available for recording observations over time?

#### artiga

##### New Member
Hello,

I have taken a simple random sample in January, which is 10% of the population. Every month there is 10% people leaving the population, which will make my sample size also shrink. Over the time, there will a time where I will not have any subject left in my sample to record the observation on.
How should I deal with this kind of problem in sampling?
How to deal with this kind of problem?

#### GretaGarbo

##### Human
What is it that you want to estimate?

In survival analysis it is called censoring when people leave the study area.

Or is the event that you want to study statistically independent of the event of leaving the area?

#### artiga

##### New Member
I want to study their spend features. Yes, the the event I am studying is statistically independent of the leaving area.

#### GretaGarbo

##### Human
Yes, the the event I am studying is statistically independent of the leaving area.
Then it doesn't matter if they are leaving. You can just estimate the parameter from the remaining. And then you could also recruit a partly new sample, as compensation for those who are leaving. That is based on the assumption that leaving does not matter.

On the other hand, if leaving matter, for example that those who will leave spend much less on furniture (since that would just cause trouble), then there is a difficulty.

#### artiga

##### New Member
@GretaGarbo Thank you for the response.

situation 1(Indpendent):
As I mentioned my population is changing one. So the sample I have taken at time t1 won't be representative of population at time t2. So even if I take the partly new sample (which will be representative of current population) and mix with my remaining sample it will won't be representative sample of current population.

situation 2 (not independent):
I would still like to know, how can I address this problem?

#### GretaGarbo

##### Human
situation 1(Indpendent):
Then sample 1 describes the situation as it was then, and sample 2 describes the situation as it was when they were sampled. So that means that you will have a time series. Maybe a short time series. Possibly an exponential smoothing could be of interest.

situation 2 (not independent):
It is much more difficult. Let's think about that one. But the situation must be very carefully specified.