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.
International Journal of Forecasting 26 (2010) 42–53
www.elsevier.com/locate/ijforecast
The chancellor model: Forecasting German elections
Helmut Norpotha,∗, Thomas Gschwendb
aDepartment of Political Science, Stony Brook University, United States
bCenter for Doctoral Studies in Social and Behavioral Sciences, University of Mannheim, Germany
Abstract
Our forecast model for German Bundestag elections relies on three predictors: (1) the popularity of the incumbent chancellor
(hence the christening of it as the “Chancellor Model”); (2) the long-term partisan balance in the German electorate; and (3)
the cost of ruling, as captured by the tenure of the government in office. The model forecasts the vote share of the governing
parties (typically two, such as Social Democrats and Greens, or Christian Democrats and Free Democrats), except for instances
of a grand coalition. The coefficients of the predictors are estimated based on elections since 1949, the beginning of the Federal
Republic. The out-of-sample forecasts of the model deviate from the actual results by just over one percentage point, on average.
The first real-time test of the model came in 2002. The forecast issued three months before Election Day picked the incumbent
vote share to the decimal (47.1% for the SPD-Greens coalition); for the 2005 election, called a year early, our forecast three
weeks before Election Day was just three-tenths of a percentage off the mark. For the upcoming election, we offer separate
forecasts, conditional at this moment, for each of the two parties in the grand coalition.
c
2009 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved.
Keywords: Election forecasting; Government popularity; German elections; Multivariate models
1. Introduction
Forecasting elections in Germany is no cottage
industry. Pollsters and academic students of elections
in that country view the endeavor with bewilderment,
disbelief or undisguised disdain. When we introduced
∗Corresponding author. Tel.: +1 631 632 7640; fax: +1 631 632
4116.
E-mail addresses: helmut.norpoth@sunysb.edu (H. Norpoth),
gschwend@uni-mannheim.de (T. Gschwend).
URLs: http://www.sunysb.edu/polsci/hnorpoth/ (H. Norpoth),
http://www.sowi.uni-mannheim.de/gschwend/ (T. Gschwend).
our forecast model with a prediction for the 2002
German Bundestag election, one critic gleefully
predicted that our “model will not survive the next
[2002] election” (Quoted by Neumeyer, 2002). In the
end, the model did more than survive. The forecast
posted three months before Election Day in 2002
picked the result, the vote share of the governing
parties, to the decimal (47.1%). No poll, not even
the Election-Day exit polls, beat or even equaled
the performance of our forecast model. In 2005, the
forecast of our model came within three-tenths of a
percentage point of the outcome, again beating pre-
0169-2070/$ - see front matter c
2009 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.ijforecast.2009.02.008
H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53 43
election polls. Granted, a broken clock may be right
twice a day, but our model’s performance in those
elections was not just a stroke of luck.
Our forecast formula combines predictors of
the vote that are familiar to students of elections
anywhere. We owe a special debt, of course, to
forecast models developed for American elections (as
featured by Campbell & Garand, 2000, and Lewis-
Beck & Rice, 1992; and, most recently, in PS: Political
Science & Politics, 2008. However, our formula
should not be mistaken for a German replica of any
particular American model. The predictors of our
forecast model for Bundestag elections, in a nutshell,
are long-term partisanship (a normal-vote baseline),
short-term chancellor approval, and a medium-term
dynamic of declining incumbent support over time
— call it the “cost of ruling” or simply government
fatigue (Gschwend & Norpoth, 2001). Using election
returns and measures from opinion surveys, we
estimated the statistical influence of these variables on
the vote in Bundestag elections. Forecasts after the fact
confirmed that the model would have been capable
of predicting German election outcomes from 1953
to 1998 quite closely (Norpoth & Gschwend, 2003).
The out-of-sample forecasts of our model picked the
results of those elections with an average deviation of
just over a single percentage point. What broken clock
can do that?
2. The forecast target
Anyone who tries to forecast an election in a multi-
party system must first settle on what exactly it is that
is being forecast. With no single party likely to control
the government in Germany, our efforts focus on the
politically most telling party combination: the parties
forming the government before the election. For each
Bundestag election we obtained the combined vote
of the governing parties. The vote of any party that
belonged to the government at some time during
the term preceding the election but departed before
Election Day was not included. A few elections,
however, posed some problems.
One is the 1969 election. CDU/CSU and SPD
entered the election campaign as partners in a grand
coalition, but they parted company as soon as the
votes were tallied. What is more, the combined share
of those two parties (88.8%) is an extreme case
that would require special consideration or else the
exclusion of that case. The most practical solution
was to use only the vote share of the dominant party
(CDU/CSU) of the coalition for that election. The
upcoming election (2009) presents us with the same
problem. Again, CDU/CSU and SPD are governing
in a grand coalition. This time, however, we will be
offering a forecast each for CDU/CSU and SPD.
Another sort of tough case is the election of 1983.
Technically, the coalition government going into that
election consisted of CDU/CSU and FDP, and yet
this was an election that was specially called to let
the public decide on the replacement of the previous
government of SPD and FDP, headed by Chancellor
Schmidt, in the Bundestag a few months earlier.
With the FDP newly aligned with the CDU/CSU, we
decided to consider only the vote share of the SPD, the
dominant party of the Schmidt government, for that
election.
Yet another problem arises from unification, in
1990. Should we use the vote in the newly unified
Germany for elections since then or stick to the vote
in the old Federal Republic? Again, for practical
reasons, we decided to keep the continuity of the
vote series intact until 1998. Beginning with the 2002
election, we have ended this exclusion and begun
relying on the vote in the unified Germany. What then
determines voter support for governing parties in any
given Bundestag election? What makes the difference
between winning and losing?
3. Theoretical foundation
Our main point of departure from other approaches
is the premise that the voting decision is determined by
the confluence of long-term, short-term, and medium-
term components. The influence of long-term partisan
loyalties on electoral decisions is an undisputed fact
(Campbell, Converse, Miller, & Stokes, 1960;Lewis-
Beck, Jacoby, Norpoth, & Weisberg, 2008), and the
German electorate is no exception. What determines
the vote choices in Bundestagswahlen to a large
extent is long-term stable attachments to political
parties (Falter & Rattinger, 1982;Norpoth,1978). The
aggregate distribution of party identification in the
German electorate provides a “normal vote” baseline
for the outcome of a given election under normal
circumstances.
44 H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53
While the notion of long-term forces, in principle
at least, is not hard to grasp, the notion of short-
term forces is more elusive. What tips the electoral
balance in the short run, as we see it, is a matter of
retrospective judgment. The electoral calculus boils
down the following: “If the performance of the
incumbent party is ‘satisfactory’. . . , the voter votes
to retain the incumbent governing party. . . ; while if
the government’s performance is not ‘satisfactory’, the
voter votes against the incumbent.. . ” (Kramer, 1971,
p. 134; see also Fiorina, 1981). Or, to put it more
apocalyptically, the electorate acts as a “rational god
of vengeance and reward” Key (1964, p. 568).
Aside from long-term partisanship and short-term
performance judgment, our vote model also includes a
cyclical dynamic. In US presidential elections, a party
rarely occupies the White House longer than two or
three terms. The two-term limit for presidents, either
by tradition or law, makes for competitive contests
that open the door to the opposition party getting back
into the White House. Some forecast models have
taken account of this dynamic in U.S. presidential
elections (Abramowitz,2008;Norpoth,2002,2008).
In parliamentary systems like Germany without such
term limits, the “cost of ruling” (Paldam, 1991) is set
to ensure that parties in office lose rather than gain
votes from election to election.1
4. Measures for the predictors
Measuring electoral variables over a span of
50 years or so, like making sausages, is not something
for purists. Opinion surveys track few if any questions
over such long periods. The business of survey
research in Germany is barely that old. For measuring
long-term partisanship, we have altogether forsaken
survey data and instead derived an estimate from
election returns, as was done for U.S. elections
under similar conditions by Tufte (1978, ch. 5). It
is undeniable that the long-term support of a given
party registers in its mean vote percentage over
several elections. For the still quite young Federal
Republic it seemed prudent to rely only on a few
1While the Federal Republic came up as an exception to that rule
in Paldam’s analysis, our replication, including elections since 1983,
shows the German case to be a perfect fit, with an estimate for the
key parameter of −1.7, which is close to the average of −1.6.
past elections. Hence, the long-term partisan support
for the governing parties in the German electorate is
measured as follows:
Long-term partisanship =
Average vote in the last three Bundestag elections.
For the early Bundestag elections, it is necessary
to modify this measure of long-term partisanship. The
first three simply have not enough previous elections
to fall back on, and there is just no way we can
afford to drop them all from the analysis. Those early
elections also witnessed a massive transformation of
the German party system that realigned the partisan
loyalties in the German electorate. This makes it
defensible to use only the immediately preceding
election for a measure of long-term partisanship for
the elections of 1953 and 1957. With nothing to guide
us for the first election, we have no choice but to let
1949 go.
The relationship between long-term partisanship
and incumbent vote is quite strong (r=0.55), as can
be seen in Fig. 1. To be sure, some elections do not
fit especially well. One would not expect long-term
orientations to offer much guidance for voting in 1953,
only the second election of a new political system.
This is an incumbent victory that one would predict to
derive primarily from short-term forces. Similarly, an
incumbent defeat, as in 1998, registers in an electoral
showing far below the normal-vote prediction. All in
all, long-term partisanship is not correlated with the
vote in any given election strongly enough for it to be
used as the sole predictor.
What matters for the vote in the short-run, we
contend, is public satisfaction with the performance
of the incumbent government. Forecast models for
U.S. elections have typically captured this factor by
relying on measures of economic performance and
presidential approval (Abramowitz,2008;Holbrook,
2008;Lewis-Beck & Tien, 2008), or combining
economic measures with the trial-heat standing of
presidential candidates (Campbell,2008;Erickson &
Wlezien, 2008). Regarding Germany, the only other
forecast model besides our own that we are aware of
relied on economic predictors such as unemployment
and the budget deficit, not direct measures of
popularity (Jerˆ
ome, Jerˆ
ome-Speziari, & Lewis-Beck,
2002), and was based on only 11 elections. Our model,
in contrast, relies solely on chancellor support as a
H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53 45
Fig. 1. Partisanship and the vote.
short-run predictor of the vote, omitting the economy.
That is not to say that we consider the economy
of no importance for electoral outcomes. Our view
is that the electoral effect of the economy is fully
mediated by chancellor support. Preliminary tests
have shown that the inclusion of economic measures,
be it economic growth, unemployment, or inflation,
adds no predictive leverage to our forecast model
with chancellor support as a predictor (Gschwend &
Norpoth, 2000, pp. 404–406). The federal chancellor
is the most visible figure of any federal government,
and his reputation mirrors that of the government he
leads. While purely personal considerations may play
a part in evaluations of the chancellor, matters of
the economy, as well as foreign policy, leave their
mark on such evaluations (Gschwend & Norpoth,
2001, pp. 492–493).
Lacking a single measure of chancellor approval
for the last 50 years, we have fashioned a series
of comparable items from several survey sources
(Norpoth, 1977). For the early elections (1953 and
1957) and the special case of 1983, we had to turn to
the Allensbach question, “Are you satisfied with the
policy of chancellor [name]?” For the other years, we
have relied on the question in the German Election
Studies about chancellor preference (incumbent vs.
challenger), which is also used in public opinion polls
of the Forschungsgruppe Wahlen. Wherever possible,
our measure of chancellor support for a given election
averages the values one and two months before
Election Day.
The correlation between chancellor approval and
the vote for the governing parties is quite strong
(r=0.75), as can be seen from Fig. 2. Nonetheless,
it would be foolish to make forecasts of German
elections solely from chancellor support. What is
more, the less than perfect fit shows that chancellor
support and vote choice are not two sides of the
same coin. It is not a tautology to say that someone
votes for one of the governing parties because she
approves of the incumbent chancellor. There is enough
room for additional factors, quite aside from long-term
partisanship.
In the medium-run, our vote model specifies a
fatigue effect that prophesies electoral decline to
incumbent parties. The longer a government is in
office, the more its electoral support diminishes, to
the point where defeat in an election brings on a new
46 H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53
Fig. 2. Chancellor support and the vote.
government. The federal governments in Germany
have lasted three terms, on average. We measure
voter fatigue simply by the number of terms a
government, or the leading governing party if the
coalition composition changed, has been in office.
Fig. 3 shows that the vote of governing parties
declines as the number of terms of that government
increases. The correlation is fairly strong and has
the proper negative sign (r= −0.44). To be sure,
some of that electoral decline is prompted by the
withdrawal of parties from the government. The
departure of several parties from government in the
early years (1953–1957) had to diminish the vote of
the remaining governing parties at the next election.
However, since then, voter attrition has been the
primary source of the electoral decline of governing
parties. That is true for the social-liberal coalition
(1969–1982) under Chancellors Brandt and Schmidt,
the Christian-liberal one under Kohl (1982–1998), and
the red–green coalition under Schr¨
oder (1998–2005).
The tendency of the governing parties in Germany to
lose electoral support confirms the general “cost of
ruling” proposition put forward by Paldam (1991).
The reader may wonder, however, whether this
medium-term fatigue is really something independent
of the short-term evaluations of incumbent perfor-
mance, or just an artifact of the latter. If such fatigue
exists, should it not register in declining satisfaction
with the incumbent chancellor, which in turn drives
down the vote of governing parties? Chancellor sup-
port and term of office are indeed correlated with each
other (r= −0.34), but not to such a degree that the
relationship between them would seem in imminent
danger of proving to be spurious. It remains to be seen
how much of the fatigue factor remains with chancel-
lor support held constant, and vice versa. That takes
us to the next step, where we put all three predictors in
one electoral basket.
5. The forecast model
Using all three predictors in a single vote equation
lets us determine whether long-term partisan strength,
a short-term evaluation of incumbent performance,
and the medium-term decline of government support
each exert an independent effect on the incumbent
vote in Bundestag elections. As the results in Table 1
H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53 47
Fig. 3. Terms of office and the vote.
Table 1
Estimates for the forecast model.
Forecast target:
Vote of governing parties
Predictors Parameter (SE)
Long-term partisanship 0.75*** (0.08)
Chancellor support 0.38*** (0.04)
Term −1.5*** (0.28)
Constant −5.6 (4.5)
R20.95
Standard error 1.3
(N) (15)
Durbin–Watson d1.80
Lijung–Box Q(5 lags) 2.78 ( p>0.70)
Note: Model estimation based on elections 1953–2005.
*p<0.05.
** p<0.01.
*** p<0.001.
show, the coefficients of all three vote predictors are
statistically significant beyond a doubt, and their signs
are all in the expected directions. Each of them brings
something distinct to the table. We can rule out the
possibility that partisanship has such a tight grip on
chancellor approval that the latter has no vote leverage
left. It is also not true that the decline of support with
an increasing number of terms in office manifests itself
in chancellor support. However unwelcome it may be
to the coalition governments in the Federal Republic,
the cost-of-ruling effect proves extremely helpful to us
in coming to grips with election outcomes, and hence
for forecasting.
What do these results tell us about election
outcomes in the Federal Republic? For one thing,
long-term partisan strength does help a governing
coalition get re-elected, but it does not guarantee
it. The governing parties can expect to retain about
three fourths of their “normal” vote in a given
election. Contrary to claims about dealignment, long-
term partisanship is a powerful electoral factor in
Germany. Still, it is not strong enough to secure
reelection. Second, chancellor approval has a strong
effect, above and beyond long-term partisanship.
For every point gained in chancellor approval, the
incumbent parties can expect to gain close to one-half
of a percent in votes. Granted, the incumbent parties
cannot count on the vote of everyone who favors the
chancellor. Let us not forget that partisanship counts
for a lot, too, in German elections. But chancellor
48 H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53
Fig. 4. Stability of estimates in synthetic out-of-sample predictions.
approval adds a critical margin to the base support of
the governing parties. A chancellor whose approval
rises from, say, 40% approval to 60%, adds eight
percentage points to the vote total of the parties in
his government. That can spell the difference between
winning and losing. And finally, the government
fatigue associated with increasing terms in office
proves costly for the governing parties at election
time, and does so independently of the incumbent
chancellor’s approval and partisanship. Taken all
together, the three predictors capture the actual vote of
incumbent parties in Bundestag elections from 1953
to 2005, with an average error no larger than 1.3%. By
another measure of fit, the explanatory power of the
vote equation with the three predictors reaches 95%.
Very little of the variance of the actual vote in elections
between 1953 and 2005 is left unexplained. Such a
performance inspires strong confidence that the model
can make reasonably accurate forecasts about future
elections.
But how robust are the predictor coefficients,
given only 15 cases (elections) from which to extract
information about electoral outcomes? One sign of
reassurance is the pattern of the relationship between
each of the three predictors and the vote that is
visible in Figs. 1–3. There are no signs of any outliers
that could disturb the estimates in any particular
election. This is suggestive of robustness, but is not
conclusive. A harder test is provided by examining
the stability of coefficient estimates for synthetic out-
of-sample predictions. As can be seen in Fig. 4, the
estimates for a given predictor vary very little across
elections, with none even close to straying beyond
the 95% confidence intervals. In other words, none
of the estimated coefficients used for out-of-sample
predictions differ significantly from the benchmarks
(dashed lines) of the full model presented in Table 1.
The out-of-sample forecasts themselves, which are
displayed in Fig. 5, track the actual results extremely
closely. Apart from the notoriously difficult case of
1953 – the first instance of a federal government
seeking reelection – there is only one other election
(1965) where our model prediction is off by more
than two percentage points. Luckily, in close elections
(such as 1976) the margin of error is far smaller than
the average, to allow us to make the right call. What is
H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53 49
Fig. 5. The vote of governing parties with point forecasts.
more, the forecasts shown in Fig. 5 pick the winner of
every single Bundestag election since 1953, at least in
hindsight.
6. Real-time tests of the model
The ultimate test of a forecasting model is its
ability to offer precise predictions ahead of time. How
did our model perform in one-step-ahead forecasts?
The first time the model was tested this way was in
2002. Three months before Election Day we posted
the forecast, derived from our model, that the SPD-
Greens coalition, led by Chancellor Schr¨
oder, would
get 47.1% of the vote (Neumeyer, 2002). Such a
vote would be enough to ensure victory over the
combination of CDU/CSU and FDP, given the realistic
assumption that all other parties combined would
muster at least 6%; the share of all other parties had
averaged just above 8% during the last three elections.
On Election Day 2002, the red–green coalition won
47.1% of the vote—exactly the share predicted by our
model three months earlier. Red–Green defeated the
combination of CDU/CSU and FDP while securing a
majority in the Bundestag.
To be sure, scoring a bull’s-eye hit with a forecast
model is a stroke of luck, but coming within about
one percentage point would have been no surprise for
our model, given its overall fit to German elections.
What was surprising in 2002, however, was that the
model did as well when all the polls predicted the
opposite. Relying on opinion polls, the headlines in the
German media practically called the election for the
opposition CDU/CSU and FDP. The governing parties
were trailing so far behind in the horse race that defeat
seemed inevitable. Like Schr¨
oder, our model was fated
not to survive the 2002 election (Neumeyer, 2002).
The second real-time test of our model came
in 2005. The SPD-Greens government was running
for reelection again, and facing a combination of
CDU/CSU and Free Democrats (FDP) as a likely
successor government once again. This time, however,
the arrival of a new competitor, the Left Party (DIE
LINKE.PDS), complicated the electoral competition.
This party – shown in public opinion polls securely
above the critical 5% threshold – was cutting into
50 H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53
the support for the Social Democrats. How could
our model cope with this new wrinkle? Creating a
variable just for this novelty would be very costly,
given the small number of cases (elections in the
Federal Republic). Could any of the model predictors
cope with the impact of a new party? Which one would
be most suitable?
The government fatigue variable provides no
leverage for any such adjustment, and it seemed
too early to tweak long-term partisanship since the
party had just been formed in the election year. So
we decided to use chancellor support to make the
adjustment. Our reasoning was as follows: Supporters
of the new Left Party, as was evident from polls,
preferred the SPD chancellor candidate (Gerhard
Schr¨
oder) but would not vote for his party. In so doing
they would distort the normal relationship between
chancellor approval and voting decision in 2005.
Thus, we adjusted the popularity rating of the SPD
chancellor candidate (vs. his CDU/CSU-challenger,
Angela Merkel) by the support for the Left Party
recorded in polls.2
Our forecast, based on the left-adjusted popularity
rating, along with the other variables of the model,
predicted that the SPD and Greens together would
receive 42.0% of the vote, with a standard error of
the forecast equal to 1.7.3The actual SPD-Greens vote
share (42.3%) was just three-tenths of a percentage off
the mark — and well within the expected margin of
error of our forecast. In a historical perspective, the
2005 forecast ranks among the best showings of our
model, using out-of-sample predictions for previous
elections (Fig. 5).
The polls, in contrast, fared less well in 2005.
Three months before the 2005 election, support for the
red–green coalition registered in polls at about 37.2%,
on average, and even a month before the election at
about 38%. At the same time, those polls pointed to a
decisive victory for a CDU/CSU-FDP coalition. Our
forecast disagreed with that prospect (Gschwend &
Norpoth, 2005a). As it turned out, our model predicted
2This step resulted in a left-adjusted chancellor approval rating
of 43% (52.4% overall minus 9.4%, the level of support for the Left
Party).
3The 2005 forecast was issued three weeks before Election Day.
The 2005 election was called a year early and did not give us as
much time to prepare a forecast.
the outcome not only earlier, but also more precisely
than the pre-election polls, and our forecast was on
par with exit-poll results on Election Day (Gschwend
& Norpoth, 2005b).
7. Forecasting the 2009 election
To use this model for the upcoming election
(expected in September 2009), we first have to
overcome two obstacles. One has to do with the rare
occurrence of a grand coalition (Christian Democrats
and Social Democrats) in the federal government;
the other concerns the continued presence of the
new Left Party. Regarding the first obstacle, the
federal government since 2005 has been made up of
Germany’s two major parties. This is only the second
time that such a grand coalition has been formed in
the Federal Republic, the previous occasion occurring
in the late 1960s. Though the Christian Democrats
and Social Democrats are governing together, each
of them has nominated its own chancellor candidate
(Merkel, the incumbent chancellor, for the CDU/CSU,
and Frank-Walter Steinmeier, the vice chancellor, for
the SPD). And each would surely part company as
soon as alternatives become available. So a forecast
for the combined share of the governing parties, which
is what our model is designed to produce, would be of
little interest for the upcoming election. We must adapt
the model to generate separate forecasts for CDU/CSU
and SPD in the 2009 election. We do so by entering
separate values for the CDU/CSU and SPD for each
of the predictors of the forecast equation.
The second obstacle deals with the Left Party.
As noted above, this party suddenly rose during
the 2005 election season, with dissident leftists
in the SPD joining the post-Communist Party of
Democratic Socialism, which had strong support in
the eastern states. Rather than create a new variable
to account for this contingency, we adjusted one of the
predictors in making our forecast for the 2005 election.
This involved deflating the chancellor approval in
proportion to the strength shown by the new Left Party
in polls before the election. We assumed that such
voters would favor the SPD-chancellor (Schr¨
oder), but
not vote for his party. While such an adjustment is an
option for the upcoming election as well, by now the
impact of the new party should have registered in long-
term partisanship. As a rule, long-term partisanship
H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53 51
Table 2
Conditional forecasts for the 2009 election.
Approval rating (%) Vote of CDU/CSU (%) Vote of SPD (%)
Chancellor SPD candidate
30 70 31.4 (2.1) 42.0 (1.9)
40 60 35.2 (1.9) 38.2 (1.9)
50 50 39.0 (1.8) 34.4 (1.9)
60 40 42.8 (1.8) 30.6 (2.0)
70 30 46.6 (1.8) 26.8 (2.1)
Note: Forecast standard errors are in parentheses. The predictor values for long-term partisanship and term of office are 36.3 and 1 for all
forecasts of the CDU/CSU vote, and 34.2 and 3 for the SPD.
is measured by the average vote in the last three
Bundestag elections. This may not be a wise rule
for the SPD these days. Its vote declined from 40.9
to 34.2% between 1998 and 2005, as the Left Party
took root. While we cannot determine for sure what
proportion of former SPD voters have defected to the
new party, it is certain that the average of a declining
phenomenon is not the best estimate of its long-term
prospect. A better move is to take its last showing as
an indicator of future performance. Hence, we have
decided to use the SPD’s showing in the last election
(2005) as the best measure of its long-term strength.
Even that level may be inflated, given the boost the
SPD received in 2005 from the popularity of then
Chancellor Schr¨
oder.
Table 2 presents a range of model forecasts for
each of the two major parties in the upcoming
(2009) election. These are conditional forecasts,
subject to approval ratings for Chancellor Angela
Merkel (CDU/CSU) vs. her SPD-opponent, Frank-
Walter Steinmeier. What is known for certain at this
moment is the long-term partisanship and term of
office for each of the parties. Assuming an approval
rating of 70% for the Chancellor (Merkel), the model
would predict a landslide victory for her party, the
CDU/CSU, over the SPD. Even such a favorable
result, however, might not guarantee a majority of
seats for the CDU/CSU in the Bundestag. Seats are
allocated in proportion to the national vote, provided
that a party secures at least five percent of the national
vote. Though it will take a little less than 50%
of the national vote to capture a majority of seats
in parliament, recent experience suggests that in all
likelihood at least 47% of the national vote would be
necessary. The CDU/CSU will be able to top that mark
with the help of a willing coalition partner, the Free
Democrats (FDP), provided that that party clears the
five-percent hurdle. In the end, a moderately popular
Chancellor Merkel, along with an FDP getting its
normal share of about eight percent of the national
vote, will be able to control the next Bundestag.
In contrast, the prospect for an SPD-led govern-
ment is less rosy. Even with an exceptionally popular
chancellor candidate (70% approval), the SPD would
fall short of controlling the next Bundestag outright.
The help of the Greens, its coalition partner during
the Schr¨
oder years (1998–2005), would no longer be
enough for an SPD chancellor candidate with a less
impressive approval, assuming a normal level of the
Green vote (about eight percent). Embracing the new
Left Party would be a move that is fraught with too
many risks for the SPD to contemplate. In any case, all
this will remain academic unless the SPD-candidate
Steinmeier manages to rise above the anemic levels of
approval recorded in polls so far (37% in the January
2009 poll of the Forschungsgruppe Wahlen).
8. Conclusion
Our forecast model for German elections has
admittedly taken a page or two from models developed
for U.S. presidential elections. That is quite obvious
from the reliance on chancellor approval, echoing
the presidential-approval measure. However, few U.S.
models incorporate a partisan predictor, which seems
to make good sense in the case of German elections.
After all, German voters are not given a choice of
chancellor candidates on the ballot. The chancellor
approval nonetheless supplies the most forecasting
leverage.
The “Chancellor Model” has performed admirably
in one-step-ahead forecasts, issued for two German
elections thus far. In 2002, we picked the incumbent
52 H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53
vote share to the decimal, and for the 2005 election we
were just three-tenths of a percent off. What is more,
in both years our forecasts stood in stark contrast to
what the trial heats in polls portended. While a broken
clock may be right twice a day, the synthetic out-
of-sample forecasts of our model for previous cases
do just about as well. German elections prove to be
highly predictable amidst contrary horse-race polls
and media coverage. The upcoming election presents a
special test for the model. The parties of the governing
coalition will be competing against each other to see if
they can govern without the other one, so we have had
to tweak our model to produce a separate forecast for
each of them.
Parliamentary elections in countries like Germany
do pose special problems for forecasters. Presidential
elections, by nature, produce a single winner. Parlia-
mentary elections using a proportional representation
rarely do so. The party winning the most votes need
not be “the winner” of such an election. A coalition
of parties is typically required to govern. But which
coalition emerges after an election as the new gov-
ernment depends on many factors. Ideally, a forecast
model for elections in a multiparty system should be
able to predict the vote shares for each of the parties,
but neither electoral theory nor data makes such an en-
terprise feasible yet.
Given the small number of Bundestag elections and
five parties with a parliamentary representation these
days, there is no way to use our model to predict the
vote shares of all of the parties. What we have done
is focus on the governing parties, which makes for the
most compelling forecasting target. If the governing
parties win enough votes in an election to command
a majority of seats in the Bundestag, they win the
election and stay in office. If they do not win enough
votes for a majority of seats, some other combination
of parties will try to govern.
The problem is predicting what proportion of the
vote is “enough”. It need not be a majority, since,
as a rule, seats in German elections are apportioned
on a proportional basis only to parties winning more
than five percent of the national vote. The larger the
combined pool of “wasted” votes (Gschwend, 2007)
going to parties falling short of the five-percent hurdle,
the lower the requirement for a vote share below 50%
to ensure a majority of seats. To take an example,
with six percent of the vote going to such parties, any
coalition of parties with more than 47% of the vote
would capture a majority of seats, and hence win the
election.
There is no question that the governing coalition
composed of CDU/CSU and SPD will get more than
that vote share in the next election, so they will have
“won” the vote battle. But that is no guarantee that
they will keep governing. Grand coalitions are the
exceptions to the rule in the Federal Republic. Each
of the major parties would rather be the bigger partner
in a coalition with a smaller one. Whether one of the
major parties will be in a position to do so after the
next election depends not only on its own vote share,
but also on the vote of the likely partners. We do offer
a forecast of the vote of each major party, but do not
have a model to forecast the vote of the minor parties,
just educated guesses based on past experience. For
the upcoming election the following outcomes appear
most likely. Whichever major party receives more than
40% of the national vote will get a majority of seats
with the help of its traditional junior partner (the FDP
for the CDU/CSU, and the Greens for the SPD).4If
neither the CDU/CSU nor the SPD clear the 40%
mark, the best bet would be a continuation of the grand
coalition. If both clear it, the outcome depends on
which of the junior partners does better in the election.
References
Abramowitz, A. I. (2008). Forecasting the 2008 presidential election
with the time-for-change model. PS: Political Science and
Politics,41, 691–696.
Campbell, A., Converse, P. E., Miller, W. E., & Stokes, D. E.
(1960). The American voter. New York: Wiley.
Campbell, J. E. (2008). The trial-heat forecast of the 2008
presidential vote: Performance and value considerations in
an open-seat election. PS: Political Science and Politics,41,
697–702.
Campbell, J., & Garand, J. C. (2000). Before the vote. Forecasting
American national elections. Thousand Oaks, CA: Sage
Publications.
Erickson, R. S., & Wlezien, C. (2008). Leading economic
indicators, the polls, and the presidential vote. PS: Political
Science and Politics,41, 703–708.
Falter, J. W., & Rattinger, H. (1982). Parties, candidates and issues
in the German federal election of 1980: An appraisal of normal
vote analysis. Electoral Studies,1, 65–94.
4Recent polling gives FDP and the Greens at least 8% each in the
upcoming election (http://www.wahlrecht.de/umfragen/index.htm),
so the threshold for the major parties need not be as high as 40%.
H. Norpoth, T. Gschwend / International Journal of Forecasting 26 (2010) 42–53 53
Fiorina, M. (1981). Retrospective voting in American national
elections. New Haven: Yale University Press.
Gschwend, T. (2007). Ticket-splitting and strategic voting under
mixed electoral rules: Evidence from Germany. European
Journal of Political Research,46(1), 1–23.
Gschwend, T., & Norpoth, H. (2000). Soll und Haben: Die
deutsche W¨
ahlerschaft rechnet mit den Parteien ab. In
M. Klein, W. Jagodzinski, E. Mochmann, & D. Ohr (Eds.), 50
Jahre Empirische Wahlforschung in Deutschland. Entwicklung,
Befunde, Perspektiven, Daten (pp. 389–409). Wiesbaden:
Westdeutscher Verlag.
Gschwend, T., & Norpoth, H. (2001). Wenn am n¨
achsten Sonntag
. . . Ein Prognosemodell f ¨
ur Bundestagswahlen. In M. Kaase, &
H.-D. Klingemann (Eds.), Wahlen und W¨
ahler: Analysen aus
Anlass der Bundestagswahl 1998 (pp. 473–499). Wiesbaden:
Westdeutscher Verlag.
Gschwend, T., & Norpoth, H. (2005a). Schr¨
oder lebt. Financial
Times Deutschland, 30.
Gschwend, T., & Norpoth, H. (2005b). Prognosemodell auf dem
Pr¨
ufstand: Die Bundestagswahl 2005. Politische Vierteljahress-
chrift,46(4), 682–688.
Holbrook, T. M. (2008). Incumbency, national conditions, and the
2008 presidential election. PS: Political Science and Politics,41,
709–712.
Jerˆ
ome, B., Jerˆ
ome-Speziari, V., & Lewis-Beck, M. S. (2002).
´
Elections allemandes: La situation ´
economique fera la
diff´
erence. Le Figaro, p.17.
Key, V. O. (1964). Politics, parties and pressure groups (5th ed.).
New York: Thomas Y. Crowell.
Kramer, G. (1971). Short-term fluctuations in U.S. voting
behavior, 1896–1964. American Political Science Review,65,
131–143.
Lewis-Beck, M. S., & Rice, T. W. (1992). Forecasting elections.
Washington DC: CQ Press.
Lewis-Beck, M. S., Jacoby, W. G., Norpoth, H., & Weisberg, H. F.
(2008). The American voter revisited. Ann Arbor: University of
Michigan Press.
Lewis-Beck, M. S., & Tien, C. (2008). The job of the president and
the jobs model forecast: Obama for ’08? PS: Political Science
and Politics,41, 687–690.
Neumeyer, J. (2002). Zauberformel f¨
ur Wahl: Forscher entwickeln
Prognosemodell. dpa. June 23, 2002.
Norpoth, H. (1977). Kanzlerkandidaten. Wie sie vom W¨
ahler
bewertet werden und seine Wahlentscheidung beeinflussen.
Politische Vierteljahresschrift,18, 551–572.
Norpoth, H. (1978). Party identification in West Germany:
Tracing an elusive concept. Comparative Political Studies,11,
36–61.
Norpoth, H. (2002). On a short leash: Term limits and the economic
voter. In H. Dorussen, & M. Taylor (Eds.), Economic voting
(pp. 121–136). London and New York: Routledge.
Norpoth, H. (2008). On the razor’s edge: The forecast of the primary
model. PS: Political Science and Politics,41, 683–686.
Norpoth, H., & Gschwend, T. (2003). Against all odds? The
red–green victory. German Political Studies,21, 15–34.
Paldam, M. (1991). How robust is the vote function? A study
of seventeen nations over four decades. In H. Norpoth, M. S.
Lewis-Beck, & J. D. Lafay (Eds.), Economics and politics. The
calculus of support (pp. 9–31). Ann Arbor: The University of
Michigan Press.
PS: Political Science & Politics, (2008). Symposium—Forecasting
the 2008 national elections. Vol. 41 (pp. 679–732).
Tufte, E. R. (1978). Political control of the economy. Princeton:
Princeton University Press.
