« WaPo Op-Ed Questions CW About Battleground States | Main | Kerry & Bush in Arkansas »
Monday, May 10, 2004
General Election Cattle Call, May 10: New and Improved
Posted by Chris BowersNational Two-Party Vote Projection
Kerry: 50.2
Bush: 49.8
Electoral Vote Projection
Bush: 275
Kerry: 263
States changing hands from 2000: Ohio, Pennsylvania, New Hampshire
After completing a round up of the past five weeks of polling on Saturday, I realized a huge flaw in my methodology. While I still believe I am correct in factoring trial heats, approval ratings and favorable / unfavorable numbers equally at this point in the election, placing theses three types of polls into three separate, equally weighted categories does not allow me to properly measure the central mean of the entire data set. In other words, instead of measuring the central mean of the three separate categories and then taking the central mean of the categories themselves, I should include all of the data in a single central mean calculation. Thus, I have eliminated the separate categories, thereby allowing the three types of polls to be weighted equally.
Also, I will no longer project any states as ���too close to call.��� Instead, I���ll be gutsy, look as deep into the numbers as I can, and make my best guess. (Man, I do not like predicting Kerry slightly ahead in the popular vote and slightly behind in the electoral vote).
I know that changes to my formula and a common occurrence these days, but please bear with me. I am trying to develop the best snapshot / predictor possible, and it will take a while to work out all of the kinks.
(Also posted on MyDD)
Posted at 07:43 PM in General Election Cattle Call | Technorati
Comments
It really is disconcerting to have Kerry ahead in the popular vote but behind in the EC.
I think your new methodology is probably better, but I don't think it makes sense to try to predict each state at this point. Its way too far out, and I think its more valuable to see which states are too close to call.
The only other suggestion I would make would be to factor in the right track/wrong track question in polls, because that is a pretty indicator of presidential success (or failure). I would probably be hard to find as much right track/wrong track data at this point, but it is one of the most valuable pieces of data in predicting the election.
Posted by: jjbman1121 at May 10, 2004 08:58 PM | Permalink | Edit Comment | Delete Comment
A complex but accurate way to calculate summative poll numbers is to convert the percentages back into raw data by multiplying the decimal equivalents of the percentages by the number of people in the survey. I.e- if 1000 people were polled and Kerry is at 46% you multiply 1000 times .46. giving you a raw total of 460 votes. You can then combine any two (or more) surveys for a raw total and then put them into a simple spreadsheet total for a good estimate of where national polls are at. I used this process during the primaries...go to the link below and you can see that using the method, I darn near nailed it.
http://iliketowrite.blogspot.com/2004_01_01_iliketowrite_archive.html#107513184761199867
Posted by: steve at May 10, 2004 09:34 PM | Permalink | Edit Comment | Delete Comment
The Democrat ahead in the popular vote but the Republican ahead in the electoral vote? That's crazy-talk!
On a more serious note... I don't like the idea of trying to call each state this early on. If the data on a large swing state (say, Florida) oscillates even a little from week to week, you could have wild swings in your cattle call, making it meaningless.
If you really want to have your numbers adding up to 538 at each cattle call, I have another suggestion: Use the raw data to find confidence percentages. If, for example, a poll has 527 Kerry voters and 503 Bush voters (for a total of 1030), then the standard deviation is the square root of (527*503)/1030^3, or 1.558%, and Kerry has a mean of 51.165%. Since Kerry is then 0.748 standard deviations above 50%, a normal distribution table says that he has a 77.3% chance of winning that state. If the state has, say, 27 electoral votes, you can assign 27*.773 ��� 21 to Kerry and the remaining 6 to Bush. It'd be easy to mesh together different polls for the same state just by adding together the Bush/Kerry numbers. I don't know immediately how approvable ratings and favorable/unfavorables could fit into this scheme, but something could probably be done.
Posted by: Benjamin Schak at May 11, 2004 09:12 PM | Permalink | Edit Comment | Delete Comment