Articles and Analysis


Converting Gallup's Generic Ballot to Seats

Topics: Gallup , Generic House Vote

On Saturday, political scientists at the American Political Science Association conference in Washington, D.C. will debut their models for predicting results in the House of Representatives. In this article, I put forth my own model that translates Gallup's final generic ballot result into seats. While the generic ballot has generally not been fantastic at forecasting House outcomes, Gallup's final likely voter generic ballot poll has proven itself to be a great predictor in midterm elections.

In 2002, Alan Abramowitz created a model based off of Gallup's final likely voter poll in every midterm election since 1950. What I have done here is recreate that model and included 2002 and 2006 data. Abramowitz's model estimates the amount of seats the Republican party will gain by how many seats they won in the prior Congressional election, the party in the White House, and the Republican lead (or deficit) on the final generic ballot. I have also added my own variable: whether the party in the White House has been in power for more than one term. It is important to keep in mind that this model is based off only 14 elections and the final Gallup likely voter poll (before then Gallup's polls can be bouncy). If past trends hold, the model will do very well at predicting the 2010 final House seat count.

This simple model is quite robust and explains a little over 97% of the variation in the amount of seats won by Republicans in midterm elections from 1950-2006. In addition to gaining more seats when they do better on the generic ballot, Republicans are also more likely to perform well when Democrats control the White House, and the party in the White House has been there for more than a term. In 12 of the 14 elections*, the regression's error is 5 seats or less. The model's error is never greater than 9 seats for any of the 14 elections.

So what type of lead do Republicans need on Gallup's final likely generic ballot to take back the House? Amazingly, they only need to be leading by 3% to be slated to garner 218 seats and win a majority by the slimmest of margins. If Republicans have a 6% lead, an error in estimate larger than this model has ever seen would be needed for Republicans not to gain back the House. A likely voter lead of 10% like Republicans had on Monday with registered voters, not likely voters who Republicans will do better among, translates into a cosmic 240 seats.

Other possible generic ballot margins to seat translations are

Screen shot 2010-09-02 at 4.13.25 AM.png

When Gallup makes the transition to a likely voter model, we will have a very good idea which party is destined to control the House chamber. Considering the current Republican position on the registered voter ballot, and how that has historically translated to the likely voter model, the Republicans look to be in mighty good position. Of course, November is still two months away.

*Gallup did not have a likely voter model in the fall of 1986.



That seems about right. The 6-8 percent might be a little high. I could see the Republicans winning by 6 and ending up with a half a dozen less seats.



Finally this week there have been a few articles showing the fallacy of the democrat denial of a bloodbath coming in November. I was beginning to think that huffington post had banned all articles that showed democrats in a negative light.

Mr. Enten has it right in ignoring the "pickup count" and going straight to what matters - the total count. The pickup count is the emotional component of the election and the democrats can't seem to handle things when people start talking about 50, 60, 80, or even a 100 seat pickup.

Forget the pickup numbers. According to this model, the minimum numbers necessary in the final gallup poll of LV's is 48.5% democrats, 51.5% republicans.

That's it. The republicans only have to be at 51.5% and Nancy Pelosi is finished.

What about the wild end of the spectrum? Gallup's RV poll this week has republicans up by 10%. That equates to at least 14% using a LV model.

If this was the final poll before the election, this model would show a more than 100 seat pickup by republicans and within spitting distance of a veto override super-majority of two thirds.

I'm not predicting anywhere near that, but those who now scoff at an 80 seat pickup were scoffing at a 60 seat pickup 2 months ago, a 30 seat pickup 6 months ago, and had declared the republican party dead 2 years ago. Now there are multiple analysts showing a pickup in the 60 seat range and the chances of the democrats holding the house nearly impossible.



Why in the world would such things as prez job approval, current seat count, and party in WH be used in a model that modifies generic ballot poll?

Generic ballot already has all those factors baked into it. "In your district, will you vote for the Repub or the Dem?" That is the question. How does one rationalize adjusting the result of that poll? Those other factors would be great to use in a model IF ONE DID NOT HAVE GENERIC BALLOT POLLS. Or a model created with those factors can be used in parallel with generic ballot polls, just as an alternate way to get to the same place. Although I doubt such a model will be as accurate a predictor as an aggregate of surveys from reputable pollsters.

A few other details in my next post..


Chris G:

Is this based on the generic ballot poll just before Election Day? Some kind of average? How predictive are other variables + generic ballot 2 months out? Another intermediate question is how much "drift" we in the generic ballot week-to-week.



Repubs would need a THREE point popular vote advantage just to break even in seats won? Based on what? 1950 election? Why wouldn't winning 50% of the popular vote get you 50% of the seats? So assuming 100 Mil turnout, and ignoring third party votes, Dems only need 48.5 Mil votes to GOP's 51.5 Mil to break even in the seats won? What causes that? Uneven party distribution in districts. Meaning - higher concentration of GOP voters in fewer districts. Which translates to GOP winning fewer districts by higher margins than Dems. That's the only explanation I can come up with.

Is that really the case? The famous model above suggests it is, apparently. But the model looks at elections going all the way back to 1950. More recent data suggests that the exact opposite is true. It is the Dems that have higher concentrations in fewer districts.


Since this is not something I do for a living, I have not bothered to check Barone's data for accuracy. Whatever common sense I may or may not possess, it tells me that yes, that sounds right. There are likely more congressional districts with 70%+ Dem concentrations than there are such GOP districts. The opposite may have been the case in 1950's, 60's, 70's, but not today.

I did take the time to glance at another relevant data point. 2006 House election versus 2004. Those two cycles offer a terrific contrast for this discussion, since they were near perfect mirror images of each other.

In 04 GOP won 232 seats with 49.2% of the total popular congressional vote. Dems won 202 seats with 46.6%. There was one Indie seat.

In 06 Dems won 233 seats with 52.0% of the total popular congressional vote. GOP won 202 seats with 44.1%. No indies this time.

So it took somewhere around 2.5 to 3 percentage points more for Dems to reach the same seat totals that it took GOP.

So I would argue that Barone's data and the 04/06 comparison breaks down the claim that GOP needs a 3 point popular vote advantage just to get to 218. It looks like something of a reverse is true. It is the Dems who need a 2-3 point generic ballot advantage just to hold the House by the slimmest of margins.



People are greatly overvaluing the Gallup generic ballot number this week. A few weeks ago, we had the Democrats leading by 6 in consecutive weeks. Anyone remember this? Even Charlie Cook (who has been writing the Dems obit since last August) was talking about a possible comeback. Gallup is experiencing a lot of noise in it's generic ballot polling. The Dems have had a bad month but seriously, it hasn't lead to a 16 point flip. That's just nuts.



What would a discussion like this be without someone making the argument that the Republican lead can't possibly be ten points? The Republicans have had a lead in the Gallup generic ballot for five straight weeks now, something that is as unprecedented as their 10% advantage. While it may not be 10%, there is a half-way decent possibility that it is. The newly released USATODAY/GALLUP poll shows the lead at 6% while the newly released Fox News poll has it at 9%. Rasmussen has it at 6% as well. While 10% may, or may not be overstating it, it is obvious the 10% lead is a distinct possibility. The less plausible explanation for the 16% turnaround is that Democrats ever had a 6% advantage, not that the Republicans have a 10% one. Of course my argument may well be rendered moot when the Generic ballot polls come out on Monday (or Tuesday given the holiday).



@StatyPolly: The incumbent party is favored. If the overall vote is 50% - 50%, then the incumbent party will usually win. In 2004 and 2006 the Republicans were the incumbents, so they were favored. Now the Democrats are the incumbents, so the Republicans need more than 50% to win.



masanf - I never said that the Republicans don't have a ten point lead. I said that people are overvaluing the gallup generic. It simply has too many violent swings to be very reliable. People were reading too much into the Dems 6 point lead a few months ago and they are reading too much in to the GOP 10 point lead today.


Harry Enten:

Hey folks,

I just want to reiterate that this conversion is for the *final* likely voter poll only. Gallup does have many swings, which makes this week's number no more reliable than last week's. Still, I think most of us can agree that the Democrats are in worse shape than they were 2 or 4 years ago. The question is how bad? Many indicators outside of this conversion scale say it is really bad. In the next two months, we'll continue to gain an understanding of the environment the Democrats are facing. And just before election day, we can use this scale to give us a very good idea of whether the Democrats can hold onto the majority or not.



The reason that Republicans can pick up a lot of seats with a three point lead in the generic ballot is because, in part, many congressional districts in the country thanks to the gerrymandering by the republicans have a greater number of Democrats concentrated in them than needed to hold a seat instead of haveing them a little more spread out so that Democrats would have a better chance in more races.



So I played with some numbers from this site, that Barone used for his analysis.


I did not verify those numbers, but the site is endorsed by none other than Nate Silver himself, so the numbers must be good.

So I took the entire three election table and dumped it into a spreadsheet. Just sorted it by margin of victory for each election. Barone has already done this for 08 and 04, but in a way, 00 Bush v. Gore was even more interesting since Gore's margin of victory was only half a percent. Turns out, even though Gore won the popular vote, he lost congressional districts 193 to 238. Four districts were tied at either 49 or 50.

I also took the 08 election and forced the popular vote in each district to the nearest tie. Meaning, since BOBO won by 7.2, in first case I deducted 4 points from his district totals and added 4 points to McCain's. In second instance I added and deducted 3 points. Since I wanted to be as close as possible to that 7.2 margin but without making the exercise messy with fractions. The results were McCain winning 232 to BOBO's 198 using the 8 point adjustment, and 222 to 207 with the 6 point one. The unaccounted for seats were ties. Taking the exercise to a perfect 7.2 adjustment and splitting tied seats, gets you somewhere around 230 for McCain and 205 for BOBO.

To reiterate, had BOBO and McCain split the popular vote perfectly at 50/50, McCain would have won 230 congressional districts to BOBO's 205.

Obviously, the way a congressional district votes in a presidential contest does not match the way it votes in a congressional contest. But it appears to be a great measure of voters' concentration by party. And of course the example of 04 vs. 06 congressional election that I posted earlier, supports this narrative perfectly. It took GOP 2.8% fewer popular votes that it did Dems in subsequent elections to get to the same numbers of seats.

So my argument, as it relates to the table Harry worked up, is that should GOP end up with the actual vote advantage of 0, they will end up winning 225+ seats and not the 209 figure that modified Abramowitz' model suggests.

Interestingly, Abramowits model may end up predicting the election fairly close, but it will be more of a case of a broken clock being right once a day (in case of military clock) rather than a measure of model's overall effectiveness. Reason being, it starts out forecasting too few seats at the low ranges of GOP popular vote advantage, but ends up gaining seats at an accelerated rate. Why +3 for every point? One percent of the House is 4.35 seats. So if both parties' concentration by district were even, each additional point advantage in generic would yield 2.175 seats. Abramowitz model has it 3. Since Dems concentration is higher, that causes GOP to have a lower starting break even point. Meaning - GOP needs less than 50% of popular vote to get to 50% of seats. The downside of this is GOP's gains for every additional point advantage in generic will be something less than 2.175. So since Abramowitz model starts out too low at 0 GOP generic lead, but than adds 3 seats, instead of something that is less than 2.175, it will eventually cross or at least come close to my model. Somewhere at 10+ level. I predicted earlier a 7-8 popular GOP win translating to about 240 seats.

Abramowitz's model would, in fact, work pretty accurately for the Dem side of things. Start out slow (less than half seats at 50/50 generic) and gain more than 1% of seats for each marginal percentage point lead in generic. But, the way things are looking now, should GOP continue to lead by 8-10%, the model will end up being only about 10 seats short of the actual outcome.

Sorry about the rambling post, but I hope someone can find something to challenge in my rational.




Nate Silver's numbers don't take into account several factors such as seats NOT up for re-election that should be taken out of the "calculus" and the fact that the resulting numbers should only be whole numbers, aka you can't have half an independent senator. There are actually two independent senators, who will remain two independent senators because they aren't up for re-election.

His numbers actually are very poor in my estimation given the glaring error.

Statty Polly it's the same reason why the Republicans have to do better than 50/50 to take the majority because not all the seats are up for re-election.


How many seats did the Republicans win in the mid-term elections following the great depression after Franklin Delano Roosevelts first two year's as President?

This voter model is based on normal years when their was not an economic crisis prior to the election. Why would people vote for Republicans again when their policies lead to the worst economic crisis since the great depression?

Our country gained 67,000 private sector jobs in August, and the job losses have stopped. That beats losing 500,000 jobs per month when Bush was our President.




Nobody here is talking about the Senate. Only the House. Generic ballot polling is only useful for forecasting House elections, not Senate.



I still say the republican gain will be in the mid 40s at best.

If they DO win more, then they'd better hope the economy gets a whole lot better really fast, which is unlikely, since I just read articles today predicting that it will take until 2014 to recover all the job losses from 2008-09. Mainly because real estate, construction, and financial services may never come back they way they were in the mid 2000s. It's going to take a lot of time for those people to transition. About half of the 8 million job losses were in those 3 sectors.

It's not because of what the democrats are doing... they're actually helping the health care sector long term by creating a lot of new customers. I expect health care to be the next bubble that will pop when Boomers start their demographic decline in about 15-20 years. What will all those health workers do without all those elderly to care for?

The slow recovery is because of globalization that requires people to become more and more specialized. A hs diploma is now worthless beyond retail/food service or military, a master's is the new bachelors... it's getting ridiculous. Republican OR democrat policies are not going to affect that process except on the margins.



The lead will likely not be R+10. There's a good chance republicans in the house will be about where they were after 2004. 230-205 or something like that. According to this model that's a lead of 5-7%. I think that's a little over-optimistic but we'll see.

I'm looking forward to it because then it will be the republicans who have struggles between their moderates and more hardcore factions. They need to have that discussion.



"How many seats did the Republicans win in the mid-term elections following the great depression after Franklin Delano Roosevelts first two year's as President?"

The problem is that Obama is no FDR by a long shot. He defers to congress too much on domestic policy - which is why Pelosi had to drag him kicking and screaming on health care (NYT reported he was ready to capitulate and take ANY republican HCR plan after the Scott Brown win).

On foreign policy he's done a pretty good job. But then so did G.H.W. Bush. Public doesn't care much about FP, even though that's what the president can directly affect, and do so quickly.

The stimulus is perceived as a failure because people expected more out of it. It succeeded in its limited goals. It was designed to shore up state budgets that would have been overly stressed by medicaid/medicare resulting in mass state layoffs, gave the middle class a tax credit hoping they would spend it (Bush policy holdover), and increased or maintained some grants, programs (the pork everyone's so upset about).

If it were me I would have kept the state aid part (necessary), but instead of the latter two I would have directly put a couple million people to work building a national high-speed rail network, national wi-fi infrastructure, and various other large-scale projects (green tech infrastructure? medical infrastructure?). We would have gone into the same amount of debt but would have had something tangible as a benefit instead of just "not-worse" conditions that most people find unacceptable. That would have been the FDR strategy. But Obama is no FDR, and never really indicated he would be throughout the campaign.

Now we've got nothing, and will have republican control which will bring us more nothing.



Staty Polly:

From Real clear politics, only 129 out of 435 seats for the House are "in play", which is essentially the same as 70% not being up for re-election, aka they just aren't a factor.

Of those 129 seats, a whopping 113 are held by Democrats. Only 13 are currently Republican. Hence the needed increase in Republican support to shift the 129 seats that went Democrat last time.

I'm pretty sure if you divide the margin of win in the closest 40 seats held by Democrats by three you will come very close to the 3% needed difference between Republican and Democratic votes to re-take the house.

That's some very crude math, but that's how you get the discrepency I believe.



Sorry my crude math is wrong. Eventhough only about a third of the races are in play, this actually has no effect on the increase in votes needed to win.

It's the 88% of seats in play that are held by Dems that requires the generic increase for Republicans.

So 9 out of 10 (crude math) seats in play needed to win are Democrat, and roughly 40 seats needed to win are out of the roughly 120 in play (1/3) or 3/9 Democratically held seats.

So yes, as a straight probability, the Republicans should only need 50% plus 1 of the vote to win. But these votes need to increase more in the 9/10 Democratically held seats to shift the balance of seats.

And the model takes into account previous voting patterns.

I'm not exactly sure what you divide by three, because the more I think about it the more addled I get, but you divide something by three, (in crude math) which is the increased percentage needed for Republicans to take more than 50% of the seats. It's either the percentage of seats in play, the amount of those seats needed to win, or the increase in voting % needed to shift 1/3 of the seats in play.

Sorry not to be of more mathematical help, but I'm pretty sure that's the reason more Republican votes are needed.



Statty Polly:

By your reckoning a Republican generic ballot support of 50% plus one vote should win 100% of the seats in the House, because each house district is individual and all the votes are distributed equally.

But that's not the way it works.




That RealClear House count has 129 seats in play, 143 safe Dem, and 164 safe GOP. I am not sure how they figure what's safe, but they don't mean that "safe" seats are necessarily incumbent seats. It just means that one of the candidates has a lead large enough to deem the seat "not in play". The candidate with that large lead could be either the incumbent or the challenger. Obviously there are some open seats and retiring incumbents, and obviously incumbents or the incumbent party have those large leads in majority of "safe" seats.

Second point on RealClear is that those 143, 164 and 129 seats are not cast in stone by any means. Just because a seat is considered safe today, does not mean it will be safe tomorrow. A couple of weeks ago, RealClear had 148 safe Dems, 164 safe GOP and 124 in play. So it can move back and forth.

"So yes, as a straight probability, the Republicans should only need 50% plus 1 of the vote to win. But these votes need to increase more in the 9/10 Democratically held seats to shift the balance of seats."

I don't see why GOP would need more than 50% plus 1 in Democratically held seats.

"And the model takes into account previous voting patterns."

I know it does, and I think that's what makes it inaccurate. It goes all the way back to 1950 election, but voting patterns in individual districts have changed dramatically since then. Usually the bigger the sample, the closer the outcome, but in this case, less would be more. Instead of using 14 mid-term elections to come up with the model, I am sure that something like 4 most recent mid-term elections would yield more accurate results.

"By your reckoning a Republican generic ballot support of 50% plus one vote should win 100% of the seats in the House, because each house district is individual and all the votes are distributed equally."

Not at all. All I am doing is looking at existing data. Gore v. Bush is one good example. While Gore won total popular vote by half a percent, Bush won popular vote in 238 congressional districts while Gore won in only 193. (4 seats were ties when rounded to whole percentages). The same theme runs through 04 and 08 Presidential elections. GOP candidate win more than than 50% of congressional districts with less than 50% of the popular vote. GOP needed only 49.2% of the popular vote to win 232 seats (over 50%) in 04, while Dems needed 52.0% to win a similar number of seats (233) in 06.

So, no, I am not saying that "50% plus one vote should win 100% of the seats in the House". Only that 50% wins something more than 218 seats needed for majority. And even 48.5% should get GOP 218 seats. The model above states that GOP needs 51.5% to get to 218. I disagree. Of course, we are not likely to find out who was right on this point, since it's not looking like it's going to be a very close election.

Just as a "fun with math" bonus point, one party can win the controlling 218 seats of the House with mere 25% plus 218 individual votes. Which is something like 25.05% of total votes. That's assuming equal number of votes cast in each of the 435 districts. A party can even win with less than 25% of total votes, if votes cast in districts where it wins are fewer than the average district.



LBJ type landslide coming for Obama in 2012! Can't wait for Obama to go toe to toe with Johnny Bonehead. The GOP taking the house is going to be the best thing that ever happened to Obama mark my words.



"LBJ type landslide coming for Obama in 2012"

Sure, he'll be more popular in 2012 than he was in 2008. Did you get a chance to clean out that crack pipe this weekend? It's hitting really smoothly.

"The GOP taking the house is going to be the best thing that ever happened to Obama mark my words."

Sure, genius, your words just got marked right bellow a million of other identical words, coming from both, the left and the right.



Mark my words StatePolly you can crow now, but you won't be come 2012. I have consistently said the dems were going to lose in 2010 and never thought otherwise. With that said John Bohener is going to be an amazing target for Obama. I don't want the dems to lose 30-35 seats and keep marginal control. I want them to lose 45-50 seats to give Johnny Bonehead and his henchmen marginal control. These next 2 years are going to be a blast to watch!


Post a comment

Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.