What with the recent attention the House of Representatives has brought upon itself, it is a good time to bring back House prediction models. Sam Wang has been busy over at Princeton Election Consortium, and he points to various sorts of prediction pontification discussed by Andrew Sullivan. Here's my two cents as far as models go: if the election were held today, Democrats would have a 50/50 chance of taking back the House based on recent polls and the graph below. As generic ballot polling changes over the next year, you can use this chart to see how predicted seat gains would change.

**Where did this graph come from?**The graph comes from the idea that the two biggest predictors of the change in the number of House seats for Democrats are this cycle's popular House vote and the current distribution of House seats. Of course, the current seat distribution will be related to last election's popular House vote. So, this election's *change* in the number of Democratic seats should be a function of the *change* in the margin of the popular House vote (see graph here). See here for a full explanation.

**My Interpretation.**

If the House election were held today, the model shows Democrats would have a 50/50 chance of taking back the House with a popular vote margin mirroring current polling of about D+6, without taking into account additional factors. *In my opinion*, additional factors known at this time suggest Democrats would need a popular vote margin of about 3 points to have even a slim chance of taking back the House, and Democrats would be virtually assured of taking back the House at a popular vote margin of about 9 points. (Please note I was too optimistic in assessing additional factors in 2012!)

**Quick instructions for using this graph:**

1. Find generic ballot polls here. As of this writing, Democrats lead by 4 to 10 points in October's polls. This will quite likely change significantly over the next year, although I would also note that in the last three election cycles, generic ballot polls have underestimated Democratic performance by a few points.

2. Use the generic ballot number you found to find how many seats Democrats would gain in a typical year, using the black line above.

3. Adjust for the **change** in this year's circumstances compared to 2012. (See below the fold for more discussion.) Your adjustment should keep you within the bounds of the green lines, which are essentially historical limits.

**Did this model work last cycle?**

Yes. Last year's final polls showed a slight Republican advantage of 0.3 points on the generic ballot. The model predicted a Democratic gain of 23 seats for a normal year. However, we were confident that conditions had changed for the worse for Democrats with the 2010 redistricting. This meant we should have expected that the results would come in somewhere below the black line, or somewhere between 0 and 23 seats gained for the Democrats - not enough to take back the House. Actual result: 8 seats gained. (Using poll numbers makes the model seem a bit better than using the actual popular vote, but as far as assessing predictions, we of course didn't have the actual popular vote available *before* the vote.)

** Other factors to consider.** This is where you get to use your judgement. Whether we will be above or below the black line this year depends on the

*in conditions since 2012. In other words, the 2010 gerrymandering is already built into this graph and does not need to be accounted for. So what do we need to think about? Here's some examples:*

**change**a. *De-gerrymandering.* The gerrymanders were designed based on the 2010 census. People move. This probably helps us a little, but I don't know how much. Is it worth an extra 2 seats? Half a seat?

b. *Money.* Are the Koch brothers pouring more or less money into individual House races than 2012, for example?

c. *Enthusiasm.* How has the grassroots participation (door knocking, phone banking) changed for both parties since 2012? This is very difficult to evaluate, although small donor numbers are likely a good proxy.

d. *Realignment.* What do Independent voters think? Are they swinging to one party or another? Sam Wang suggests winning gerrymandered districts depends on voters in the middle of the ideological spectrum, something I agree with and hope to analyze in further detail soon. Currently, the Pew poll shows the generic ballot tied among Independent voters, much better than Pew polls showing R+13 in 2010 but worse than D+7 in 2006.

e. *Recruitment.* This may be the biggest factor. A strong challenger with great fundraising prowess is usually necessary to win any Republican-held seat. The shutdown seems to have helped Democratic recruitment efforts, but is recruitment better or worse than 2012?

f. *History.* Using this prediction method, 2012 was the third worst year since 1932 for Democrats, as they won 20 fewer seats than predicted by the black line. On two other occasions the result have been more than 15 seats away from the regression line. In 1996, Democrats won 23 fewer seats than the black regression line would calculate; this was followed by winning 10 *more* seats in 1998. In 1952, Democrats won 22 fewer seats; this was followed by winning 1 more in 1954. Very bad or very good years (more than 15 seats away from the line) *may* be more likely to be followed by 'normal' years (within a few seats of predicted), and so far very good or bad years have always been followed by years within about 10 points of predicted. In other words, the residuals *may* have a non-random pattern of swinging from one extreme to the other. (The residuals certainly do not have a normal distribution, making the 95% prediction lines in the graph above a little inaccurate, but I don't *think* the effect is worth worrying much about.)

**Polling data.**

Just for fun, here's all the generic ballot polls for the year, and all the generic ballot polls for the 2010 midterm cycle, too, for comparison. A year can be a long time.