October 01, 2008

States and Probability: A Simple Method

(Note: I don't know why the table starts so far down, feel free to fix it if you can)

OK, so I've been foiled a few times. An earlier version of this post wasn't saved into the comments on the EV Master Thread, and the HTML-intensive nature of this post means I can't make it a comment on that thread.

So as the picture gets complicated (take NC, where everyone in the country thinks it's a toss up but my math still shows it's solid McCain) I like to have a common sense test, so I added a feature that simply counts the number of times each candidate has been ahead or tied in a poll in the given state.

Clearly this method has problems. If a state has two polls, a bad poll from May putting Obama up by 5 and a good poll from October putting McCain up by 5, they will each have 1 "point". So I then added another line to weight each poll by pollster quality and date. Then, I took the percent of times Obama was in the lead under each methodology and I also added the current probability from 538 into the following table. The table is in order by my actual mathematical average.






















































StateSignsSigns2538
DC100100100
HI100100100
VT100100100
MD100100100
DE10010099
NY9610099
MA8910099
RI10010099
IL10010099
CT10010098
CA10010098
OR9510096
IA969394
ME10010093
NJ8810095
WA9610096
NM768489
MI798888
WI899990
PA759284
MN899490
CO717284
VA465178
NH747071
NV412867
OH383167
FL293358
IN252450
MO17343
NC132544
WV0022
MT201117
ND251316
AR009
GA0011
LA008
TX008
MS006
SC005
AZ007
SD006
KY003
TN002
KS003
AK005
NE002
AL001
WY001
OK000
ID001
UT001

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.