Custom Search

Saturday, March 06, 2010

Handicapping Control of the House

From The Marston Chronicles

Can you really accurately predict this far in advance which party will control the House of Representatives after the November elections? The answer is only in midterm election years can you do it this far in advance. As we wrote in Will GOP Regain Control of the House, there is a known correlation between the generic ballot results and the percentage of votes that a given Party will get if the election were held today. You can find a table of how good of a predictor of the result the generic ballot is at Election Polls -- Accuracy Record Midterm Congressional Elections on the Gallup website. You will notice that only a generic ballot poll that uses likely voters works well.

Unfortunately Gallup does not switch to a likely voter model until just before the election. Indeed, there is only one pollster that always uses a likely voter model when polling and that is Rasmussen Reports. If you look at where Rasmussen lists the results of its generic ballot polling, you will see that the results of the polls have changed drastically since the last election. That means that who would have controlled the House and by how much if an election had been held at the same time the poll was taken has varied considerably over that same time period. This is not surprising because voter sentiment has varied greatly over that time as we are sure you have noticed.

That brings up the question of how to predict the makeup of the House. The Gallup table gives the relationship between the generic ballot result and the vote total a Party will get nationwide in all the House races combined. We know what the makeup of the House was after each midterm election so we ought to be able to correlate the two. It turns out that directly using the generic ballot to predict the House makeup does not work that well. It requires a two step process to first predict the vote percentage and then use that to predict the House makeup. For you mathematical wonks, the best fit in the first step is a quintic equation and then use a straight linear regression to get to the House makeup from there.

Since most of you never liked algebra in high school anyway, we will skip to the chase and give the results in the form of a handy-dandy table and not strain your brains with polynomial equations and least squares regression analysis.

GENERIC BALLOT DEM PCT SEATS MARGIN M.O.E.
Democrats +20 58.73% D 292 R 143 D 149 +1
Democrats +19 57.75% D 284 R 151 D 133
Democrats +18 57.09% D 279 R 156 D 123
Democrats +17 56.65% D 276 R 159 D 117
Democrats +16 56.35% D 274 R 161 D 113
Democrats +15 56.13% D 272 R 163 D 109
Democrats +14 55.93% D 271 R 164 D 107 - 12
Democrats +13 55.72% D 269 R 166 D 103
Democrats +12 55.46% D 267 R 168 D 99 - 9
Democrats +11 55.15% D 265 R 170 D 95
Democrats +10 54.78% D 262 R 173 D 89 + 3 / -7 / -15
Democrats +9 54.34% D 259 R 176 D 83
Democrats +8 53.85% D 255 R 180 D 75 -12
Democrats +7 53.32% D 251 R 184 D 67 +18
Democrats +6 52.76% D 247 R 188 D 59 -8
Democrats +5 52.19% D 243 R 192 D 51 -4
Democrats +4 51.62% D 238 R 197 D 41 +27
Democrats +3 51.08% D 234 R 201 D 33 +2
Democrats +2 50.57% D 230 R 205 D 25 -5
Democrats +1 50.11% D 227 R 208 D 19
Tied 49.69% D 224 R 211 D 13
Republicans +1 49.33% D 221 R 214 D 7
Republicans +2 48.99% D 219 R 216 D 3
Republicans +3 48.67% D 216 R 219 R 3
Republicans +4 48.33% D 214 R 221 R 7
Republicans +5 47.92% D 211 R 224 R 13
Republicans +6 47.39% D 207 R 228 R 21 +3
Republicans +7 46.65% D 201 R 234 R 33 -3
Republicans +8 45.63% D 193 R 242 R 49
Republicans +9 44.20% D 183 R 252 R 59

A few things jump out at you from this table. First, the Republicans have to be 2.5 points ahead to gain control of the house. The margin of error column tells you that this prediction can be anywhere from 1 to 27 seats off which can make a rather large error in some cases. In three midterm elections, 1962, 1978 and 1982 the Democrats were 10 points ahead in the generic ballot and they achieved 259, 277 and 269 seats respectively. The mathematical equation cannot be in three different points at the same time so the best fit predicts 262 seats.
Even so, this table is better than nothing since none of the errors changed which party controlled the House. The last point is that the Republicans only had control twice out of 15 midterm elections since 1950.

The Rasmussen Reports have had the Republicans ahead by 3 points or more ever since July 5th and our table says they get control of the House if that result held up at the time of the next election. The Republicans have been ahead by as much as 9 points since then. That would translate to a 59 seat margin in our table and even the maximum error of 27 seats would still give them control. Any generic ballot result of 7 or more points in their favor gives the Republicans control even with the maximum error applied.

Now you know why we have had the Republicans regaining control of the House as a result of the November elections ever since last October unlike the other analysts. That was when the Republicans got ahead 4 points in the Rasmussen generic ballot and never fell below it. That was when we projected a gain of 46 seats which would give the Republicans control by 8 seats. After the election of Scott Brown, we revisited our projections because by that time the Republicans had been ahead by at least 8 points for 4 weeks in a row. That caused us to adjust our formula for projecting which way a seat will go (a cubic equation this time) to give the Republicans another 17 seats for a margin of 47 seats just as the table would predict.

Our current prediction is from 63 to 72 seats for margins of 47 to 56 seats depending upon a 8 or 9 point generic ballot lead for the Republicans. Can we be off in our projection? Of course, we could be off by as much as 27 seats but that still gives the Republicans in a worst case scenario a margin of 20 seats. We do not make these projections up nor do we overestimate for the Republicans just because we are Republican analysts. There is a solid mathematical basis for our projections because unlike other analysts, we do not make "seat of the pants" calls.