Content uploaded by Thomas Gschwend
Author content
All content in this area was uploaded by Thomas Gschwend on Aug 27, 2018
Content may be subject to copyright.
686 PS • July 2017 © American Political Science Association, 2017 doi:10.1017/S1049096517000415
........................................................................................................................................................................................................................................................................................................
........................................................................................................................................................................................................................................................................................................
POLITICS SYMPOSIUM
Chancellor Model Predicts a Change
of the Guards
Helmut Norpoth, Stony Brook University
Thomas Gschwend, University of Mannheim
........................................................................................................................................................................................................................................................................................................
When the votes for the Bundestag have been
tallied on September 24, 2017, Germany
will most likely get a new chancellor.
This is the forecast of the Chancellor
Model (Norpoth and Gschwend 2013),
as of March 2017, to be updated throughout the election
year. Machtwechsel, a change of the guards, in the Federal
Republic is in the air. Martin Schulz, the Social Democratic
candidate for chancellor, is poised to take over from Christian
Democrat Angela Merkel, ending her 12-year tenure as German
chancellor, spanning three full terms with a varied cast of coa-
lition partners.
Schulz will have an embarrassment of riches to choose from
in assembling the coalition pieces of the new government
(figure 1). The chances of a red-red-green coalition (Social
Democrats, Linke, and Greens) commanding a majority of
seats in the next Bundestag are 83 out of 100, according to our
model. If that is not the government option Schulz wants to
pursue, or if this cannot be worked out among the prospective
partners, he can entertain a “Traffic Light” coalition (Social
Democrats, Free Democrats, and Greens). This one also has
83 out of 100 chances, according to our model, of securing a
majority of seats. And if that option proves elusive, Schulz can
fall back on staying with the CDU/CSU, the SPD’s partner in
the Grand Coalition, though with him as the chancellor now.
The reason: the chances of the SPD beating the CDU/CSU in
the 2017 election are 66 out of 100, according to our model.
One way or the other, it looks very promising right now
for Schulz to be elected federal chancellor this September.
The best chance for Merkel to retain the chancellorship is
through a coalition of the CDU/CSU with the Free Democrats
(FDP) and the Greens. We rate the chances of that combina-
tion to command a majority of seats as 69 out of 100. We are
very doubtful, however, that the Greens would prefer a coali-
tion with the CDU/CSU to one with the SPD, especially if the
latter comes out ahead in the election.
The main reason our model rates the prospect of a Chancellor
Schulz so high is that the German electorate heavily favors him in
a one-on-one chancellor duel over Merkel. Schulz leads Merkel
49–38 in the February poll of the Forschungsgruppe Wahlen;
other polls have shown similar leads. Granted, Germany is not
a presidential system, where voters elect the chief executive.
They vote for members of the legislative branch who then
elect a chancellor. But, as the Chancellor Model has convinc-
ingly shown, the marks on the ballots in Bundestag elections
bear the stamp of chancellor preferences.
The other reason to be bullish on a challenger like
Schulz is that after three terms in office, the German elec-
torate is in a mood for change. This is not unusual. Donald
Trump greatly benefitted from that sentiment in the recent
US presidential election after two terms of Democratic
control of the White House. Demand for change along with
Trump’s doing better in presidential primaries than did
Hillary Clinton predicted his victory early on in 2016 (Norpoth
2016; http://primarymodel.com/). Incumbent fatigue and
an appealing challenger augur well for a changing of the
guard this year in Germany, too.
THE FORECAST RECORD
This is not the first time we are making a forecast of a
Bundestag election. We introduced the Chancellor Model,
as we called it, in time for the 2002 Bundestag election,
and have used it for every Bundestag election since. For
the record, the model forecasts have proved uncannily
accurate and/or correctly picked the chancellor of the next
government. In 2002, with polls and pundits writing off
Chancellor Schroeder’s red-green coalition, we predicted its
reelection. Our vote forecast, issued three months before
Election Day, got the combined vote of SPD and Greens right to
the decimal—47.1%—a feat unmatched by any poll or election-
night projection (Norpoth and Gschwend 2003). In the elec-
tion of 2005, called one year ahead of schedule, our model
correctly predicted that the red-green coalition would fail to
get reelected but prove strong enough to prevent the forma-
tion of a black-yellow coalition (CDU/CSU and FDP). Our
vote forecast came within three-tenths of a single percent-
age point for the governing parties, closer than other poll or
projection (Gschwend and Norpoth 2005). In 2009, with a
Grand Coalition (CDU/CSU and SPD) in office under Chan-
cellor Merkel, we predicted that a new coalition (CDU/CSU
and FDP) would win enough votes to secure a majority of
seats in the Bundestag (ZEIT Blog 2009). This came to pass
and Chancellor Merkel formed a new government with the
FDP as her coalition partner. In 2013, our model predicted
that Merkel’s coalition would capture enough votes to stay
in office (Norpoth and Gschwend 2013). Two-tenths of a per-
centage point separated this forecast from its target. That
was the gap by which the FDP, notching 4.8% of the vote,
missed the 5% threshold of the vote required for getting
seats in the Bundestag. Nonetheless, as predicted, Merkel
remained in office as chancellor, though with another junior
partner to replace the FDP.
Downloaded from https://www.cambridge.org/core. Universitaet Mannheim, on 24 Jul 2017 at 09:11:36, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1049096517000415
PS • July 2017 687
........................................................................................................................................................................................................................................................................................................
........................................................................................................................................................................................................................................................................................................
THE PREDICTORS
By now, election forecasting is a worldwide phenomenon,
encompassing an ever expanding variety of approaches and
formulae (Stegmaier and Norpoth 2017). Building on core
determinants of the voting decision familiar from classics like
The American Voter (Campbell et al. 1960, updated by Lewis-
Beck et al. 2008), the Chancellor Model puts a premium on the
popularity of the top candidates for the job of chief executive.
It is a more inclusive indicator than economic performance,
a widely used predictor in forecast models. Our operational
measure of chancellor popularity, for most elections, takes
the preferences for the two chancellor candidates recorded in
polls one to two months before Election Day. This measure
correlates quite strongly (0.70) with the incumbent vote in the
17 Bundestag elections since 1953. Still, this leaves room for
other predictors to make their weight felt.
Needless to say, in German elections as elsewhere par-
tisan attachments, whatever one may call them, hold a strong
grip on voters (Campbell et al. 1960, Lewis-Beck et al. 2008).
Partisanship affects how voters feel about the candidates for
the highest office but the effect is not so powerful that one
is the simply the other side of the same coin. For a model of
the aggregate vote, as used here, we constructed a measure of
long-term partisanship by averaging a party’s vote share in
the last three Bundestag elections, except for the first two.
This measure correlates fairly strongly (0.59), though less so
than chancellor popularity, with the incumbent vote in the
17 Bundestag elections since 1953.
And finally, incumbent governments face longer and
longer odds for reelection, the more terms they accumu-
late in office. In American presidential elections time
is up for the White House party after two terms. During
the last 70 years, the spell was broken only once—in 1988,
when George H.W. Bush extended the Republican hold on
the White House for a third term. While German politics
during the nearly 70 years of the Federal Republic’s exist-
ence has not been afflicted with an aversion to third terms,
Table 1
Statistical Estimates of Vote Predictors
Vote Predictors Parameter (SE)
Chancellor Support .39*** (.05)
Long-term Partisanship .79*** (.08)
Term -1.18** (.32)
Constant -8.95 (4.8)
2
R
.93
Root Mean Squared Error 1.49
(N) (17)
Lijung-Box Q (5 lags) 2.08
Note: Model estimation based on elections 1953-2013.
*p<.05 **p<.01 ***p<.001
Figure 1
Chances of Each Coalition Winning a
Majority of Seats, and Chances of SPD
beating CDU/ CSU
the governing parties nonetheless lose electoral support as
their tenure lengthens (r= -0.38).
As is clear by now, any vote forecast of interest in
German elections will not be about a single party or candi-
date. With no party strong enough to command a majority,
one must forecast the combined vote of the coalition parties
forming the government. This is what our model typically
does except for the unusual situation of a Grand Coalition.
In that case, each of the partners will be seeking an alter-
native after the next election, as happened after the 1969
and 2009 elections, and most likely after the 2017 election
as well.
We have estimated the size of the three predictors with
data from all but one of the 17 Bundestag elections from
1953 to 2013; there was no poll about chancellor popularity
in 1949, the initial election. As can be seen in table 1, each of
the predictors proves highly significant and robust.
The parameter estimates for 1953–2013 are very consist-
ent with those obtained for forecasts of previous elections,
beginning in 2002. We used the same operational definitions
measures for chancellor support and long-term partisanship
as before. For the term-effect, however, we took the unlogged
number of terms this time so we would be able to predict the
vote for parties (or party) in a future government headed by a
brand new chancellor (Martin Schulz). A Merkel-led govern-
ment gets a term-value of ‘3’ while a Schulz-led government
gets a term-value of ‘0’. The in-sample predictions of this
model deviate by no more than 1.5 percentage points from
the actual vote and the overall fit stays above 0.9. Past perfor-
mance, of course, as the saying goes, is no guarantee of future
success, but it cannot help build confidence.
THE 2017 FORECAST
There is no suspense that the two parties in the federal gov-
ernment right now, the CDU/CSU and the SPD, will win a
majority in the 2017 election. The chances of that happening
are practically 100 out of 100. The only suspense about a
Grand Coalition is which of the two partners will come out
on top in the election and thus make the stronger bid for the
Downloaded from https://www.cambridge.org/core. Universitaet Mannheim, on 24 Jul 2017 at 09:11:36, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1049096517000415
688 PS • July 2017
Politics Symposium: Forecasting the 2017 German Elections
........................................................................................................................................................................................................................................................................................................
........................................................................................................................................................................................................................................................................................................
Right now, we predict that a red-red-green coalition (SPD, Linke, Greens) will get 49.3%
of the vote in the 2017 election.
chancellorship. At the same time, it is no secret that each part-
ner will be tempted to leave the marriage for a liaison with a
younger one. This is where election forecasting this year turns
into an intriguing exercise. So we have tweaked our model to
come up with vote forecasts for quite a few scenarios along
with their chances of success.
Right now, we predict that a red-red-green coalition (SPD,
Linke, Greens) will get 49.3% of the vote in the 2017 election.
Is that enough to win a majority of seats in the Bundestag
and thus be able to form a government? Assuming that
the votes for all the other parties failing to get seats in the
Bundestag add up to at least 5%, one can safely stipulate half
of 95% of the vote as the threshold for a majority of seats.
By that standard, we estimate that the chances of Red-Red-
Green winning a majority are 83 of 100, as shown in figure 1.
In order to assess the chances of winning we employ a para-
metric bootstrap approach (King et al. 2001). When we say a
coalition has a 83% chance of winning a majority we simulate
10,000 predictions for the 2017 election and in about 8,300 of
those the combined vote share of this coalition is predicted to
be greater than 47.5% (= 95/2).1
Our model makes the same forecast with the same prob-
ability of success for a “Traffic Light” coalition (SPD, FDP,
Greens). It’s up to the Social Democratic chancellor candi-
date Martin Schulz to choose which one, if any, he prefers
and what he can work out with those parties. He is also in a
privileged position vis-à-vis the CDU/CSU. Our model pre-
dicts, as of early March, that his party will edge the CDU/
CSU 34. 5 to 33.6% of the vote—enough to rate the chances
of the SPD coming out ahead as 66 of 100. Short of topping
the SPD and continuing the Grand Coalition, Merkel’s
best hope is for a coalition with the FDP and the Greens.
It would get 48.3% of the vote and its chances of winning
a majority of seats would be 69 of 100. German politics
after this election is headed for suspense and intrigue.
Our forecasts right now, of course, are based on the values
of the predictors, one of which is subject to change—the
popularity of the chancellor candidates. Martin Schulz is
a new face in German politics. His lead over Merkel in the
chancellor duel may be fragile. Any shrinkage of that lead
would diminish his prospect of becoming Germany’s next
chancellor. n
NOTE
1. Why do we get so many different predictions for the same election? Our
simulations of the predicted election outcomes differ ever so slightly because
they reflect two types of uncertainty inherent in every prediction. First,
there is estimation uncertainty of our model parameters that accounts for
the historical uncertainty since 1953 (see table 1) we face when predicting a
typical election. Every concrete election prediction such as the 2017 election
depends additionally on the influence of innumerable chance events that
arise during election campaigns. These chance events potentially influence
the outcome in September but are not systematically part of our model.
Even if we knew the exact regression parameters (i.e., without error) the
second type of uncertainty, fundamental uncertainty, would prevent us
from predicting the outcome perfectly. Thus, the variability of our simulated
election predictions reflects both types of uncertainty.
REFERENCES
Campbell, Angus, Philip E. Converse, Warren E. Miller, and Donald E. Stokes,
1960. The American Voter. New York: Wiley.
Gschwend, Thomas and Helmut Norpoth. 2005 “Prognosemodell auf dem Prüf-
stand: Die Bundestagswahl 2005. “ Politische Vierteljahresschrift 46 (4): 682–88.
King, Gary, Michael Tomz, and Jason Wittenberg. 2000. “Making the Most
of Statistical Analyses: Improving Interpretation and Presentation.”
American Journal of Political Science 44 (2): 347–61.
Lewis-Beck, Michael S., William G. Jacoby, Helmut Norpoth, and Herbert F.
Weisberg. 2008. The American Voter Revisited. Ann Arbor: University of
Michigan Press.
Norpoth, Helmut. 2016. “Primary Model Predicts Trump Victory.” PS: Political
Science & Politics 49 (4): 655–58.
Norpoth, Helmut and Thomas Gschwend. 2003. “Against All Odds? The Red-
Green Victory.” German Politics and Society 21 (1): 15–34.
———. 2013. “Chancellor Model Picks Merkel in 2013 German Election.”
PS: Political Science & Politics 46 (3): 481–82
Stegmaier, Mary and Helmut Norpoth. 2017. “Election Forecasting.” In Oxford
Bibliographies in Political Science. New York: Oxford University Press.
ZEIT Blog. 2009. http://blog.zeit.de/zweitstimme/2009/10/14/das-
kanzlermodell-bei-der-wahl-2009-diesmal-kein-volltreffer/
Downloaded from https://www.cambridge.org/core. Universitaet Mannheim, on 24 Jul 2017 at 09:11:36, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S1049096517000415
