HORSERACE BLOG - Prediction #1: Bush Wins NM
No more "News of the Day" posts, ladies and gents. At this point, there is very little news left to report. We know where the candidates are going. We know what they are going to say. The only thing left is digesting polls -- which I shall continue to do -- and making predictions. Of course, there is the "inside scoop"/rumor mill stuff that other sites do. While I am a voracious consumer of this information, I do not publish it around. (Thus, those of you with insider dirt, feel free to email me :-D)
So here is the first one.
George W. Bush wins New Mexico.
The reasons for this are the following:
There has not been a single poll in a month in NM that shows Kerry with a lead.
The Washington Times notes today that Kerry has pulled his advertising from the state (and he plans no more visits). They also note that Bush is over-performing among Hispanics relative to 2000.
Bush only lost NM in 2000 because of depressed turnout in the heavily Republican southeast, i.e. Roosevelt, Chaves and Lea counties. The state average turnout was 63.46%. Roosevelt turned out at 56.23% (68.75% of these votes went for Bush). Chaves at 58.52% (68.75% for Bush). Lea turned out at 55.70% (60.20% for Bush). If turnout was on par with the rest of the state, Bush would have netted about 895 votes, enough to give him NM's five EVs in 2000. I expect BC04 has fixed this problem.
These three counties have also increased their registration rolls by some 3,700 votes -- but an overwhelming number of these new voters are Republicans. The Democrats suffered heavy registration loses in each county.
While the Democrats have gained more voters than Republicans in New Mexico, the Republicans managed to reduce the percentage difference between each party. Thus, if Bush carries the same percentage of Democrats in NM as he did in 2000, he will net several hundred votes.
Bush's strategy here will be to boost support and turnout in the southeast and just not get killed in Bernalilo County. I believe this will be quite do-able. That Kerry has pulled out of NM completely (no more visits, no more ads), indicates, to me, that Kerry has ceded the state to Bush.
More states to come as the day wears on.
posted by Jay @ 1:59 PM 4 comments
Don't Sweat the Fox Poll...
...it is suffering from a common problem with weekend tracking polls. It has to do with the fact that their first poll was conducted Tuesday through Thursday and today's poll was conducted Thursday through Saturday. It makes a big difference. Now, since Fox does not give us their per-day results, it is impossible to figure out each day's actual samples. But we can figure out the differences between certain days and certain other days.
Specifically, we can determine the difference between the Tuesday sample and the Friday sample; and the Wednesday sample and the Saturday sample. There are many possible scenarios to explain Fox's drop with Bush. Here is one such possibility.
10/26 Sample: Bush 50%
10/27 Sample: Bush 50%
10/28 Sample: Bush 50%
10/29 Sample: Bush 41%
10/30 Sample: Bush 47%
Note the wild swing on their Friday sample. While this is just one possible situation, we can be certain that the Friday sample saw Bush's margin 9% lower than the Tuesday sample. Thus, if Bush was polling at 52% on Tuesday, it must mean that he was polling at 43% on Friday. This is still within the margin of error, but it is at the tail-end of it. More likely is either Bush is losing ground or the poll is suffering from methodological problems.
How do we explain this? Well, first let's consider some other internals of the polls.
(1) Republicans
In Friday's sample, Bush improved his margins with Republicans, Independents, and Democrats relative to Kerry. Why, then, did he lose so much ground? It must have been that there were fewer Republicans in Friday's sample relative to the Tuesday sample. Why would this have occurred? There are two possible explanations: (A) is that there was a statistically significant decrease in the number of people who call themselves Republicans; (B) there was a sample error in relation to Republicans.
(2) Men
There was also a drastic drop-off in his support among men when you compare the Tuesday sample with the Friday sample. Specifically, the Friday sample showed Bush receiving 15% less support among men than the Tuesday sample. The Saturday sample saw another drop-off in support among men. It was a 12% decline among men relative to the Wednesday sample. Again, either we can hypothesize (A) that Bush's support among men is statistical significant decline; (B) methodological issues are driving this change.
(3) Independents
There also seems to have been a wild swing in terms of Independents in the Saturday sample. The Saturday sample must have been 18% lower for Bush among Independents than the Wednesday sample. Again, it is likely not a result of normal statistical variation. It is more likely Bush either (A) lost statistically significant ground among Independents on Saturday or (B) Bush-leaning Independents were undersampled on Saturday.
So, we must find a way to explain three phenomena: A. The smaller number of Republicans in the Friday sample; B. The drastically shrinking support of men for Bush on both Friday and Saturday; C. The shrinking support of Independents for Bush on Saturday.
Democrats would love to think it is the UBL tape indicates explanation (A) for all three, but this cannot possibly cut toward Kerry (not matter what Democracy Corps might want to think...the fact is that the "terrah" issue helps Bush). It presumes that people have bought Kerry's, "Bush has failed to make us safer argument, and all the polls suggest that this is not the case."
Beyond that, there are no reasonable causal factors that possibly explain this three-pronged drop (remember that this is not a matter of undecideds migrating to Kerry, it is a matter of people leaving Bush and the Republican party).
My guess is the following: football. The high school football season is wrapping up. Thus, southern and midwestern men were out on Friday night, rooting their teams to the playoffs. Where I come from, Western Pennsylvania, Friday night football is a big deal throughout the region, especially in the northern suburbs. North Allegheny, North Hills, Seneca Valley, Butler -- all of these school districts love their football. All of these school districts are, to varying degrees, pro-Bush. There is an even greater love of high school football in rural Ohio (heavily GOP) and Texas. The same is probably true Saturday, as men were at their local college game (e.g. Penn State's stadium holds more than 50,000 people, basically sapping Centre County and the surrounding counties of men for the whole afternoon). In other words, Fox ate some bad sampling pie.
This is why I do not like tracking polls. They use the same method every day, and this is quite unwise. You must adjust your method to correspond with the day of the week as well as the time of day. This is why national, non-tracking polls that are regularly released are regularly released on the same day of the week. Gallup always comes out on Monday or late Sunday. Ditto for Time. Newsweek always comes out on Saturday. Battleground comes out on Friday. There is a reason for this -- they conduct their sampling on the same days of the week every week. The method that the pollster uses fits the day of the week that the poll is conducted. These tracking polls, however, apply the same methodology every day of the week. This can produce the sort of wild swings we have seen this weekend in the Fox poll. This is just bad method.
Here is an important point. The methodology which Fox uses is a methodology for Sunday and Monday sampling. Fox always releases their polls on Tuesday, and that poll does 500 people on Sunday and 500 people on Monday. Their method is designed for those two days with that number of people per day. And yet, so far they have sampled for this tracking poll the five days during which they did not sample this year, and they have sampled 100 fewer people every day. This can make a big difference.
Honestly, I am disappointed Fox has started a tracking poll. This seems to me to be a cheap trick to make news this weekend. The bottom line is that it is methodologically unsound.
In general, what I would like all my readers to do is to begin thinking like social scientists, not like journalists. Journalists look at a result and just report it. They do not take the time to consider how the result was produced. They are generally unskeptical. If you get a source with "credibility" who says X, Y and Z -- well that is good enough for the journalist. Social scientists do not do that (at least, the good ones do not). They are skeptical of every result until the method of that result has been understood and evaluated. Journalists on Fox will yim-yammer all day about this poll, about how the "race is getting (GASP!) even (GASP! GASP! GASP!) closer" without even looking at the method. The Fox journalist figures, "Opinion Dynamics knows what it is doing. Let's run with this!" But, if you are interested in reporting the truth, this is exceedingly unwise. St. Paul in 1 Thessalonians exhorts us to "test everything. Hold on to the good." Now, he was speaking of more weighty matters than sampling Republicans, Independents and men over the weekend, but the logic applies here as well!
It is more than this, though. Journalists are, to varying degrees, dishonest brokers in the truth market. A good social scientist will make a hypothesis that does not pan out, and will still publish his result, saying, "My hypothesis did not pan out. Honestly, I have no idea why this effect is occurring." He has made no "news" (beyond eliminating one hypothesis) among social science circles because he has left the question unanswered. In journalistic circles, the hypothesis must always pan out. "News" must always be made. Questions must always be "answered." Truth must always be "found." This is, I believe, one reason why almost everybody who is involved in some story that gets covered by the news considers themselves to have received a "bad rap." Journalists over-simplify. They slur over the details. They have little concern for nuance. After all, they must draw an impressive conclusion in under 1,000 words.
In my estimation, this was a good weekend for Bush. In a nutshell: Mason-Dixon redeems all of this crap. They had the best track record in calling races in 2002, upwards of 90%. The kosher-seeming Newsweek poll and the Battleground poll indicate national results along the same lines (a Bush +5% win yields 306 EVs). They also indicate stability in this week's results. Zogby, meanwhile, has been totally discredited thanks to the wild swings. I am guessing that Gallup and Time come out tomorrow with polls in line with Battleground and Newsweek. Thus, on Tuesday we will see the national, non-tracking surveys with Bush around 50% and the tracking surveys with a closer race thanks to methodological problems. |
