On 10/2/05, Ryan Delaney ryan.delaney@gmail.com wrote:
DF wrote:
For those around here that like statistics, I have updated my automated statistical analysis of AFD,
This is fantastic, thank you. I would like to call >
everyone's attention to something, though. On your
table of the most common AFD closers, there is User:Jni, who closed 117 AFDs. Of each AFD he closed, 100% were deletes, with a 30% estimated threshold for delete votes.
If I'm reading this correctly, that means he only needed 30% of the votes to be delete to decide that there was consensus to delete.
<snip other stuff>
Jni is something of a special case. In 117 closes, he never closed as Keep. Hence at some empirical level, his threshold might equally well be anything below his weakest consensus. The algorithm being used picks a number in the middle of the many equivalent values, so it may be skewing a lot lower than is sensible. A more useful column to look at in cases like this is the "Est. Deviation", this shows how many AFDs he might have been expected to close differently had he been using a 63.5% value which is the average. For Jni, that value is 2, meaning that he closed as delete roughly two AFDs with less than 63.5% votes for deletion.
Sorry for the confusion.
-DF
On 10/2/05, DF dragons_flight@yahoo.com wrote:
Jni is something of a special case. In 117 closes, he never closed as Keep. Hence at some empirical level, his threshold might equally well be anything below his weakest consensus. The algorithm being used picks a number in the middle of the many equivalent values, so it may be skewing a lot lower than is sensible. A more useful column to look at in cases like this is the "Est. Deviation", this shows how many AFDs he might have been expected to close differently had he been using a 63.5% value which is the average. For Jni, that value is 2, meaning that he closed as delete roughly two AFDs with less than 63.5% votes for deletion.
Thing is, some admins specialise in early closes, and some specialise in late closes--you choose which day log to process and depending on its position in the list of day logs you'll find yourself processing old or new listings. The early closers will have a very high number of straight keep or delete closes (mostly deletes), the late closers will have a fairly high number of no consensus keep closes.
I'm shocked at the extremely low figure at which the sample seemed to close with delete. 63% is absolutely not to be confused with consensus.
On the other hand, 75% of closes beingdelete is a much lower hit rate than I expected before I went and examined a few completed day logs myself. I naively expected the delete rate to be around 90% or more.
DF wrote:
Jni is something of a special case.
Thanks for the clarification. Still, I could just as easily make the same point by comparing, say, Tony_Sidaway (80%) with JeremyA (43.9%), or ABCD (47.3%) with Sjakkalle (70%).
Incidentally, I'm on there twice, since I was renamed from User:Malathion to User:Ryan_Delaney somewhere in the middle of when you started gathering these stats, it seems.
- Ryan