Suppose age is normally distributed with mean 20 and standard deviation 5. How do you ensure that you get only positive values when you sample age from this distribution?
I am trying to impute missing data. I am assuming that it is MAR (missing at random) since the missingness depends on gender. The missing ages come from both males and females. Thus I am sampling from two normal distributions (one for males and one for females).
I cannot suppose age to be normal, since age can only be positive, while normal distributions are on the range $-\infty$ to $\infty$.
Since 41.25 is only 3.43 standard deviations (12.02) from 0, you will get some negatives by supposing it to be actually normal.
If you simply want some set of random numbers with mean 41.25 and standard deviation 12.02 that looks vaguely normalish, you might consider generating a gamma random variable.
Here's a histogram of a sample of 100000 ages from a gamma distribution with mean 41.25 and standard deviation 12.02:
The smallest age generated was about 7, while the largest was 110.
If you want a strict upper limit as well as lower limit, you might consider using a beta distribution.
It looks like in SAS you could do it something like this:
x = beta*RAND('GAMMA',alpha)
Where the shape parameter alpha can be set to $(\frac{\text{mean}}{\text{sd}})^2$ and the scale parameter beta can be set to $\frac{\text{sd}^2}{\text{mean}}$.
