>> Well, I am a bureaucrat, is that not my position to be a pain in the...?
>> :~)
>
>Eehr, Bruce, I am sorry. This is not what I meant, and certainly not to
>insult you. I thought I had written : Your anonymous ...., as a wink
>at your own signature. So if anybody is a pain ..., it's me. Sorry for
>the mix up. That'll teach me to try to be funny! :-)
>
No problem, I saw your other message...thought I'd let you sweat a little bit :~)

>>
>> You don't have to know your distribution before selecting a sample, you
>> just have to have clear guidelines as to what to sample. You test for
>> sampling distribution after sampling if you don't have prior knowledge.
>>
>> The median is not the greatest measure statistically, you would always
>> prefer the mean, but if distribution is non-normal, the median is a better
>> simple measure. Non-parametric methods may be better, but they aren't so
>> great for the statistical novice...as it happens, housing prices are
>> generally non-normal and median is the generally accepted measure...
>
>I agree, but I guess the usefulness of the median must decrease these
>days, because calculation constraints are less relevant. If you are
>going to build some software I my idea would be to use the mean, or any
>other standard statistical entity, be it only because it is easier to
>validate theoretically.
>
Yes, I agree, we would always prefer to use the mean and variance from a statistical standpoint. However, if you have a highly skewed distribution, the mean is not a great measure of central tendency, the median, though statistically rather crude, is really a more accurate measure. I actually have available tables to create confidence intervals based on median, but the math is messy (I recall once having to prove this :~()

Anyway, that's why the median is used sometimes -- the mean is just not a good statistic for certain distributions.