Comments on: When

By: arw

arw — Tue, 16 Dec 2008 05:15:58 +0000

Is there a distribution function that comes to mind for a peaked curve like this?

By: teddy

teddy — Tue, 16 Dec 2008 04:07:02 +0000

arw,
I don’t think a normal distribution captures the skew and kurtosis that appears in the data.

By: arw

arw — Mon, 15 Dec 2008 20:24:35 +0000

RQ, because of the random nature of the referral dates, I’m not sure if just next month’s result we’ll be a very good indication of whose method is more accurate, but if you test Teddy’s method over the past 12 months, you’ll find his mean (the rolling twelve-month average day-of-the-month) tends to be a more accurate predictor than the mean based on the rolling twelve-month average duration between referrals. I hope that makes sense.

By: arw

arw — Mon, 15 Dec 2008 15:16:48 +0000

So Teddy am I doing this right?

Looking at the prior 24 months, and setting the last day of each month as the integer 0, I’m getting a mean of 3.92 and a standard deviation of 3.17.

Then I plot it on a normal distribution curve with this calculator:.

http://davidmlane.com/hyperstat/z_table.html

And, I’m getting that there is a 56% chance it falls between the 2 and 7th next month and only a 2.7% chance we would see it go to the 10th or beyond.

By: teddy

teddy — Mon, 15 Dec 2008 13:55:40 +0000

actually, the table says it is most likely to fall between the second and the seventh day of the month. (one to six days after the first).
for the probability density, divide the number of occurrences by 24.
Days after—
the first of—probability
the Month—
-1 to 0——0.125
1 to 2 ——0.292
3 to 4 ——0.208
5 to 6 ——0.208
7 to 8 ——0.083
9 to 10 ——0.083

By: arw

arw — Mon, 15 Dec 2008 02:37:28 +0000

teddy

I’m curious, could you develop a probability density function from your histogram and assign probabilities to each day off from the first of the month?

By: RumorQueen

RumorQueen — Mon, 15 Dec 2008 01:33:26 +0000

We’ll find out soon enough. His way points to between the first and sixth, right? Mine says the soonest is likely to be the 6th, with the most likely being closer to the 12th.

By: windthrow

windthrow — Mon, 15 Dec 2008 01:27:49 +0000

I think Teddy’s method is a better predictor of what is going to happen in the next month, but thats just me.

By: RumorQueen

RumorQueen — Mon, 15 Dec 2008 00:59:41 +0000

That’s a really shortsighted way of looking at it. The CCAA sends them out about once a month, and over the long run the time of the months shifts around. Back in 2006 it would have been accurate to say batches came between the 25th and the end of the month. But then there was a shift and then they started coming slightly later (or earlier, depending on how you look at it, since it shifted to the beginning of the month instead of the end). A year or two from now it might be accurate to say they come between the 8 and 15th. Over the long run, the way I do it works the best, by looking at the number of days between batches and stating the min, max, and average with however the weekends fall taken into consideration.

By: teddy

teddy — Sun, 14 Dec 2008 23:37:14 +0000

the ascii formatted table looked bad. here it is again below. This kind of analysis says simply that it is most likely to happen one to six days after the first of the month, independent of when the last match was.

Days after— Number of
the first of—occurrences
the Month—
-1 to 0——-3
1 to 2 ——-7
3 to 4 ——-5
5 to 6 ——-5
7 to 8 ——-2
9 to 10 ——2