Congressional Prediction for 2018: October 29, 2017
The equation used to predict the Democratic share of the two-party House vote in 2018 is equation 3a in Table 2 in Presidential and Congressional Vote-Share Equations: November 2014 Update. This discussion assumes that the paper, Presidential and Congressional Vote-Share Equations, American Journal of Political Science, January 2009, 55-72, has been read.

All the right hand side variables in this equation are known except the inflation variable and the good news variable. The Democratic share of the two-party presidential vote in 2016 was 51.11, and the Democratic share of the two-party House vote in 2016 was 48.44. Given these values, given the constant term in equation 3a, and given the value of variable I of -1 (the party in the White House is Republican), equation 3a can be written:

VOTECC = 51.09 + .429*INFLATIONCC - .602*(15/7)*GOODNEWSCC

where VOTECC is the Democratic share of the two-party House vote in 2018, INFLATIONCC is the growth rate of the GDP deflator in the first 7 quarters of the Trump administration, 2017:1-2018:3, at an annual rate, and GOODNEWSCC is the number of quarters in the first 7 quarters of the Trump administration in which the growth rate of real per capita GDP is greater than 3.2 percent at an annual rate.

Using the forecasts of October 29, 2017, from the US model for 2017:4--2018:3 and the actual data for 2017:1--2017:3, the value of GOODNEWSCC is 0 and the value of INFLATIONCC is 1.8. Given these values, the prediction of VOTECC for 2018 is:

                     INFLATIONCC    GOODNEWSCC    VOTECC
October 29, 2017         1.8            0          51.86

The Democrats are thus predicted to get 51.86 percent of the two-party House vote in 2018. This is higher than the 49.44 percent that they got in 2016 and higher than the 46.50 percent that they got in 2014. This is probably close to what the Democrats need to gain control of the House, although I do not have an equation that translates the vote share into House seats.

If you want to compute your own prediction of the vote share using different assumptions about the economy, go to:

Compute your own prediction for 2018.

Post-Mortem: December 14, 2016
The final ex ante prediction below was for the Democrats to receive 44.0 percent of the two-party vote. It looks like Clinton will receive 51.1 percent of the two-party vote, so the error is 7.1 percentage points. The mean absolute error (MAE) of the ex ante predictions of the 9 elections beginning with 1980 is 3.53 percentage points, as shown in Table 5 in Presidential and Congressional Vote-Share Equations: November 2014 Update. The current error is thus about 2 MAEs away from zero. The ex post error will, of course, be somewhat different as the equation is reestimated through the current election and the lastest revised economic data are used.

The 7.1 percentage point error is probably not large enough to warrant any specification changes when the equation is updated through the current election, but it is not great. Why such a large error? While this is not possible to test, most people would probably say that it is due to Trump's personality. Had the Republicans nominated a more main stream candidate, they may have done much better---much closer to what the equation was predicting. The prediction from the equation from the beginning in November 2014 was that the Republicans were heavily favored. The election was theirs to lose because of the economy and the duration effect, and they almost lost it!

The house vote equation did better. The last ex ante prediction was for the house Democrats to get 45.0 percent of the two-party vote, and it looks like they will get 49.0 percent, for an error of 4.0 percentage points. The MAE of the ex ante predictions of the last two on-term house elections in Table 5 is 2.80, so the error is 1.4 MAEs away from zero. Again, probably not large enough to warrant any specification changes when the equation is updated.

Vote-Share Equations: November 2014 Update
Background: The paper, Presidential and Congressional Vote-Share Equations: November 2014 Update, discusses the November 2014 update of the vote-share equations. Forecasts for 2016 are also made. This discussion assumes that the paper, Presidential and Congressional Vote-Share Equations, American Journal of Political Science, January 2009, 55-72, has been read.

Compute your own prediction: The following link allows you to compute your own predictions of the 2016 presidential and House elections. Compute your own predictions for 2016

Predictions: The current and past predictions of VP and VC using the economic forecasts from the US model are:

                        G      P      Z      VP     VC 
November 11, 2014     2.97   2.14     6     48.7   47.6 
January 31, 2015      3.04   1.86     3     46.0   46.1 
April 29, 2015        3.22   1.14     5     48.6   47.6 
July 31, 2015         3.03   1.33     3     46.4   46.3 
October 31, 2015      2.16   1.37     3     45.8   46.0 
January 30, 2016      1.97   1.37     3     45.7   45.9 
April 28, 2016        0.87   1.28     3     45.0   45.5 
July 29, 2016         0.94   1.40     2     44.0   45.0 
October 28, 2016      0.97   1.42     2     44.0   45.0 
VP is the Democratic share of the two-party presidential vote, and VC is the Democratic share of the two-party vote in the House. Click the above "Compute your own predictions for 2016" for the definitions of G, P, and Z.

October 28, 2016, comment: This is the final vote prediction before the election. It is based on the NIPA data that were released on October 28, 2016. No economic forecasts from the US model are needed for this vote prediction, since all actual data are available. The vote predictions are essentially unchanged from the July 29, 2016, predictions. The US model forecasts for the third quarter of 2016 were fairly accurate, and so the changes in the values of G and P were quite small. See the discussion in the following comment (July 29, 2016), especially the second and third paragraphs, for my view of these predictions.

July 29, 2016, comment: The NIPA data have been revised back to 2013:1, and they now show that there have been only 2 good news quarters during the second Obama administration: 2014:2 and 2014:3. The quarter 2015:2 was a good news quarter in the non revised data, but not now. The quarter 2013:4 is close to a good news quarter, with the per capita growth rate equal to 3.08 percent, but not quite. The current US model forecast for 2016:3 does not show a good news quarter, so Z is now 2 rather than 3. The new values for G and P (based on the revised data and the US model forecast dated July 29, 2016) are very close to the April 29, 2016, values, so the only main change is for Z. The change for Z from 3 to 2 lowers the Democratic predicted vote share by 1 percentage point, from 45.0 to 44.0. The prediction for the two-party House vote falls from 45.5 to 45.0. If you want to count 2013:4 as a good news quarter because it is close, the two predictons are unchanged at 45.0 and 45.5 respectively.

I have been asked whether I believe any of this? The coefficient estimates are based on data going back 100 years---25 elections. They show that the economy has significant effects on voting behavior, as does the duration variable. In this election the duration variable is working against the Democrats, and G and Z are quite low by historical standards, which also works against the Democrats. P is low, which is positive for the Democrats, but this is the only bright light for them. If one assumes that the empirical regularities gleaned from the past 25 elections, as reflected in the coefficient estimates, hold for this election, one would conclude that the Democrats' chances are quite poor.

It is no secret, of course, that Donald Trump is an unusual choice for a candidate. It may be that people who would otherwise vote for the Republicans because of the sluggish economy and a desire for change will vote for the Democrats because of Donald Trump's characteristics that they don't like. If so, then one might say that personalities overwhelmed the economy in affecting voting behavior for this election, which means that the equation's predictions could be way off. The econometric analysis behind this work assumes business as usual, which may not be the case this time. There is no way I can test this before the fact, and even after the fact it is only one observation.

April 28, 2016, comment: The forecast for G is now lower (0.87 versus 1.97 for the previous forecast), and so the predicted vote share for the Democrats is lower (45.0 for VP versus 45.7 for the previous forecast). The main message has not changed, however, and so there is nothing new to add to the previous comments below. The economy in terms of the growth rate of GDP is clearly not a plus for the Democrats in 2016. This could, of course, be trumped by other factors.

January 30, 2016, comment: The US model forecast dated January 30, 2016, is almost unchanged from the forecast dated October 31, 2015, and so the vote-share predictions are essentially unchanged. There is nothing new to add to the previous comments below.

October 31, 2015, comment: In the current NIPA data there are three good new quarters: 2014:2, 2014:3, and 2015:2. The US model forecast dated October 31, 2015, is predicting no more good news quarters, so Z is 3. G is down to 2.16 compared to 3.03 for the July 31, 2015, forecast. P is little changed. The lower G results in the prediction for VP falling from 46.4 to 45.8 and the prediction for VC falling from 46.3 to 46.0.

The October 31, 2015, forecast from the US model is close to the current consensus forecast (remember that G is per capita growth; the growth rate non per capita in the first three quarters of 2016 is 2.81 percent, which is close to consensus). If this turns out to be roughly the case, the Democrats are predicted to lose by a fairly large amount. See the earlier comments for more discussion.

July 31, 2015, comment: In the revised NIPA data that were released on July 30, 2015, 2013:3 is no longer a good news quarter, and so there have so far since 2013:1 been only two good news quarters: 2014:2 and 2014:3. The US model forecast dated July 31, 2015, is predicting one more good news quarters, 2016:3, compared to two in April 29, 2015, 2016:2 and 2016:3. So Z is down from 5 to 3. There is little change in G and P, but the fall in Z is important. The prediction for VP is down from 48.6 to 46.4, and the prediction for VC is down from 47.6 to 46.3.

A further comment about Z. As noted in the April 29, 2015, comment below, Z is sensitive to small changes in growth-rate predictions. I am using 0.65 percent as the rate of growth of population going forward, and if I used 0.66 instead, 2016:3 would no longer be a good news quarter. This would subtract a point from VP, making it 45.4 percent. Also as noted in the April 29, 2012, comment, the consensus forecast does not have any more good news quarters, so Z is two for it.

April 29, 2015, comment: The US model forecast dated April 29, 2015, is predicting two more good news quarters: 2016:2 and 2016:3. (There have so far been three: 2013:3, 2014:2, and 2014:3.) So Z is 5. G is little changed from the previous two forecasts, although P has been falling. The prediction for VP is now 48.6 percent, and the prediction for VC is 47.6 percent.

The US model forecast is more optimistic than the current consensus view. For example, the Survey of Professional Forecasters of the Philadelphia Fed has a predicted growth rate of 2.9 percent in 2016 (and about the same for the rest of 2015). Using a population growth rate of 0.75 percent, this is a per capita growth rate of 2.15 percent. So G is 2.15. Given this growth rate for the year, there are unlikely to be any good news quarters, so Z remains at 3. Assuming that P is, as for the US model, 1.14, this implies a prediction for VP of 45.9. (Use the "Compute your own predictions for 2016" link above and plug in 2.15, 1.14, and 3.) The current consensus economic view thus implies a fairly large loss for the Democrats.

A comment about Z. The US model forecast has predicted growth rates in the first three quarters of 2016 of 3.91, 4.02, and 3.97. Using 0.75 as the population growth rate, which is in line with recent rates, these are per capita growth rates of 3.16, 3.27, and 3.22. The cutoff for a good news quarter is 3.2, so two of these three are good news quarters. In fact, of course, these three rates are essentially the same. Each good news quarter is worth about one percentage point of vote share, so the vote prediction is somewhat sensitive to which side of the cutoff a growth rate is on. Historically the best fit is obtained using a cutoff of 3.2, but the fits for values near this are similar. This sensivity is not a problem using the current consensus economic forecast because the per capita growth rates are not near 3.2 percent, but for the US model forecasts one should be aware that the vote prediction is somewhat sensitive to the use of 3.2 as the cutoff.

January 31, 2015, comment: The US model forecast dated January 31, 2015, is not predicting any more good news quarters. There have been 3 so far (2013:3, 2014:2, and 2014:3). So Z is 3. In the previous forecast Z was 6. For the current economic forecast G is slightly larger and P is slighter smaller, both good for the Democrats, but the lower Z is bad. The net effect for VP, the two-party presidential vote share for the Democrats, is that it has fallen from 48.7 to 46.0. VC, the two-party vote share for the Democrats in the House, has fallen from 47.6 to 46.1. Conditional on the present forecasts of the economy, the two vote share equations are thus pessimistic about the Democrats chances in 2016.

November 11, 2014, comment: The economic forecasts from the US model dated October 30, 2014 are used for this prediction. The US model economic forecasts are fairly optimistic about the U.S. economy between now (November 2014) and the third quarter of 2016. Conditional on these economic forecasts, the prediction for VP is 48.7 and the prediction for VC is 47.6. So even with a fairly good economy, the Democrats' predicted shares are less than 50 percent. See the discussion of Table 6 on page 12 in the first paper linked above for alternative predictions.

Data for downloading:
Data back to 1876. (November 2014 Update used data back to 1916 only.) Data back to 1876
Quarterly data back to 1877:1 on nominal GDP, real GDP, population. Quarterly data back to 1877:1

Original paper:
1978#2---data through 1976.

Previous update papers:
1982#1---data through 1980.
1988#5---data through 1984.
1990#3---data through 1988.
1996#1---data through 1992.
1998#1---data through 1996.
2002#4---data through 2000.
2006#3---data through 2004.
2010#3---data through 2008.

Non technical discussions:
"Econometrics and Presidential Elections"
Predicting Presidential Elections and Other Things (Chapters 1, 3, and 4)

Web site material for previous elections:
Old site material for the 1996 election
Old site material for the 2000 election
Old site material for the 2004 election
Old site material for the 2008 and 2010 elections
Old site material for the 2012 and 2014 elections

Related work:
Interpreting the Predictive Uncertainty of Elections
The Ranking Assumption