Next I will work on my "Top 50 Posters at the End of Time" extrapolation, which is always a challenge because I have to guesstimate the post rate each member will have from now on. Once a rate is established for each member, the rates alone determine the results, regardless of current post count, because eventually somebody with a faster rate will overtake somebody with a slower rate (ignoring human life expectancy).
In the past I've tried various ways to compute the "future post rate" for each member. I've tried using the most recent post rate, but that gives too much weight to members having a rush of posts they can't sustain and too little credit to members who are temporarily posting less. I've tried using the overall post rate since joining, but that sometimes reflects past habits more than future ones.
So this year I'm going to get fancier and borrow my method from
Achilles and the Tortoise. I am going to compute a weighted average for each member's post rates, based on their rates over as many preceding 6-month intervals as possible, using a geometric progression.
Details:
Posters who joined less than 6 months ago will not be included.
For posters who joined between 6 and 12 months ago, I will use their latest 6-month rate.
For posters who joined between 12 and 18 months ago, I will use the average of their rates over their last two 6-month periods, i.e., 1/2 x rate1 + 1/2 x rate2.
For posters who joined between 18 and 24 months ago, I will use a weighted average, based 1/2 on their most recent 6-month rate and 1/2 on the average of the preceding two 6-month periods, i.e., 1/4 x rate1 + 1/4 x rate2 + 1/2 x rate3.
For posters who joined between 24 and 30 months ago, I will use a weighted average, based 1/2 on their most recent 6-month rate, 1/4 on the preceding 6-month rate, and 1/4 on the average of the further-preceding two 6-month periods, i.e., 1/8 x rate1 + 1/8 x rate2 + 1/4 x rate3 + 1/2 x rate4.
And so on.
That way, past history will be accounted for, but the further a 6-month range is into the past, the less impact it will be assumed to have on future post rates.
But if the results look completely screwy (less realistically believable than my usual guesstimates), I'll abandon that method and go back to one of the simpler methods.
The method I described here doesn't take into account the post rate from the day of joining to the start of one of my 6-month measurement intervals. I'm still pondering whether that should be factored in as well. Omitting it keeps the complexity down, and I could excuse it based on the idea that new members aren't "up to speed" at first, but it's also true that some members could have 1 day of posting ignored (because they joined 6 months + 1 day day before my measurements) while others could have almost 6 months of posting ignored (because they joined 1 year - 1 day before my measurements). Hmmmm....
