Extrapolating Changes in Distributions - how?

Factual Questions
1

Odd question, but I can’t think of a better place to ask it.

I’m reviewing a set of objects weekly. Each object has a number of attributes (arrival date, complexity, status etc.) but the important one right now is age.

When I review the data weekly it is painfully obvious that the age profile is not normal, we’re bound at 0 days old and there is no theoretical upper limit but practically speaking nothing over 300. Arrivals are independent and fairly constant week over week.

Weekly review of P95, Median, Mean, Stdev Inter-quartile range --> no problem

Here’s my problem. How can I extrapolate what the population will look like in X weeks? Taking a linear fit and extrapolating forward into time a non-normal distribution’s Mean or Median seems dumb. Anyone have any ideas?

2

Why can’t you just forecast the new mean based on the series of old ones? If the linear model fits what you have reasonably well, it might not be a bad thing to do.

3

Well the fitted plot only has about 10 weeks of data and the extrapolation out to the end of the year stretches it past were I’m comfortable. I want error bars damn it! :slight_smile:

I guess I have a gut feel that more and more (relatively speaking) of the oldest cases will move out. That in turn will heavily move my StDev and Mean values but the relative impact is going to change as the age profile bunches up towards the left hand side of my histograms.