THE POLITICAL-ECONOMY MODEL: THEORY AND MEASUREMENT

The underlying theory of our model contends that the incumbent government gains voters from good performance and loses voters from bad performance. The performance involves economic and political issues, for which the government reasonably can be held accountable. The modeling of the chancellor party and the opponent party most clearly fits this retrospective reward–punishment idea, as first articulated by Key (Reference Key1966). Conceptually, the specification of our equations aligns with standard vote functions containing macroeconomic data and executive-popularity survey measures (Stegmaier and Lewis-Beck Reference Stegmaier and Lewis-Beck2013). For the German case, the main incumbent coalition party (i.e., the INC-MAIN) dependent variable measures the party-seats share for the CDU/CSU or the SPD chancellor (or the designated candidate of his or her camp). Likewise, the main opposition (OPP-MAIN) dependent variable measures the party-seats share for the main opponent of the outgoing chancellor. Although the CDU/CSU and the SPD are nominal opponents, the grand-coalition strategy sometimes forces them to work together; however, the chancellor would be from only one of those parties. It is noteworthy for this election that the chancellor is not running for reelection.

With respect to the independent economic variable, we use prior unemployment level (i.e., U_Q-2, measured two quarters before the election); as it increases, we hypothesize that it harms the main incumbent party and helps the main opponent (following the Clarity of Responsibility theory; see Powell and Whitten Reference Powell and Whitten1993). Similarly, as in Norpoth and Gschwend (Reference Norpoth and Gschwend2010), the popularity independent variable is voter preference (i.e., one quarter before the election) for the chancellor (KANZ_Q-1^INC) or the opponent (KANZ_Q-1^OPP).

Our SUR model, estimated on the elections from 1961 to 2017, is constructed of a set of equationsFootnote ¹ beginning with these two dependent variables. They constitute the two “standard” equations: to explain the seats won by the lead incumbent party and to explain the lead opposition party. The remaining equations explain seat shares for the FDP (Liberal Democrats), the Green Party (Grune), the Left Party (Linke, formerly the Party of Democratic Socialism), and other parties (Others, including the AfD). The smaller parties should be held less responsible for government performance (Powell and Whitten Reference Powell and Whitten1993). Therefore, the vote for the FDP (for example) should be less retrospective and more prospective, expressing the hope that it will be included in a new coalition. (Perhaps surprisingly, this variable has been measured by German pollsters for 50 years.) Thus, for this variable, we distinguish between when the FDP is incumbent with the CDU/CSU (CO_INC^FDP/CDU) or with the SPD (CO_INC^FDP/SPD) as opposed to its being an opponent of the CDU/CSU (CO_OPP^FDP/CDU).

What about the equations for the other smaller parties? Green Party seat share, we assume, is driven mostly by the strength of its opposition (PGRUNE_OPP), as measured in the polls (IFD Allensbach and ZDF–Politbarometer). To clarify, we noted “preferred coalitions” in which the Green Party is incumbent with the SPD (CO_INC^SPD/GRUNE) or an opponent (CO_OPP^SPD/GRUNE). (Evidently, when the Green Party is included in a preferred coalition with the SPD while it is in opposition yields it no electoral benefit.) For the Left Party (PLINKE), we recorded the strength of its vote intention in the polls (IFD Allensbach and ZDF-Politbarometer). Finally, other seat-party share (Others) also appears accounted for by strength in the vote-intention polls for the other parties (POTHERS)—specifically, the AfD (IFD Allensbach and ZDF-Politbarometer), which comprises two thirds of the vote intentions in this category.

In addition to these theoretical specifications for the equations, we consider certain relevant electoral-rule changes over the period under study. One consideration is the 5% threshold. At different times in German history, small splinter partiesFootnote ² have achieved marks of between 2% and 5%; an obvious consequence is the reduction this causes in the marks of the large parties. Furthermore, since 2013 and 2017, these splinter parties have been increasingly heard—for example, the anti-Euro German party AfD’s threat to CDU/CSU voting. (Regarding our current coding of the AfD in the Others category, if none of the other small parties exceed 5% of the vote, all of the seats predicted in the Others equation might attach to the AfD.)

In general, German parties that do not exceed the 5% vote threshold cannot win seats. In addition to the AfD, this has affected the FDP and the Left Party. To incorporate this information into our model, we used different dummy variables (i.e., GRUINF5, FDPINF5, AFDINF5, and LINKINF5). Additionally, other dummies were used if the Green Party or the Left Party did not contest the election (i.e., NOGRUNE from 1961 to 1976 and NOLINKE from 1961 to 1987). Relatedly, dummy variable DUM83 codes the SPD’s split from the FDP; DUM6180 marks the period 1961–1980 when the SPD and the FDP were allied in “social-liberal” coalitions.

Along with consequences from electoral rules, there have been major political–institutional shifts that altered outcomes at the ballot box. First, the grand coalitions CDU/CSU/SPD (GCOAL^09/17) adversely affect the vote shares of the CDU/CSU and the SPD even more; however, they have had a positive “opportunism effect” on the other parties (Others) such as the AfD. Moreover, when the SPD was incumbent (from 1961 to 2017), it had greater losses compared to the CDU/CSU when it was incumbent. The theoretical rationale for this is party identification, which—in the aggregate for the SPD—is decreasing over time; with fewer individual identifiers, it has fewer voters. Accounting for this differential effect, we included the partisan dummy SPDINC. Finally, a major political change occurred with reunification (REUNIF) in 1990, which ushered in the Green Party—an ongoing presence since then.

THE POLITICAL-ECONOMY MODEL: SUR ESTIMATES

Using these equation specifications and measures, we estimated the SUR model for each seats-share equation. These estimates serve as the basis for making our forecasts for the party seats in the 2021 election. The SUR model reads as follows in table 1: Goodness-of-fit statistics indicate the performance potential of the equations. For the two initial equations predicting support or opposition for the main parties (i.e., INC^main and OPP^main), the Adjusted R², respectively, is 0.73 and 0.92. Also, the standard errors of the regression (SER) are encouragingly low, at 2.37 and 2.17 points, respectively. For the smaller parties, these statistics also are encouraging, with the Adjusted R² ranging from 0.97 to 0.69 and the SER from 0.71 to 1.86. As shown, it is the other parties (Others) that appear least predictable.

Table 1

SUR Estimates Seats Share (%)

T-stats between brackets (two-tailed).

Could the model have correctly predicted past elections? The out-of-sample forecasts conducted from 1990 to 2017 for the parties seem to show in table 2 that it could. In each case, the winning chancellor was well predicted, including for the year 2002. (At that time, our single-equation model could not anticipate the victory of Gerhard Schröder over Edmund Stoiber.)

Table 2

Out-of-sample predictions Seats share (%)

FORECASTS FOR 2021

To carry out seats-share forecasts for the parties,Footnote ³ we inserted into the equations the following current values:

$$ {\displaystyle \begin{array}{l}{\mathrm{U}}_{\mathrm{Q}-2}=6.2\;\left(\mathrm{June}\hskip0.5em 2021\right);\hskip0.5em {{\mathrm{KANZ}}_{\mathrm{Q}-1}}^{\mathrm{INC}}=16\%(\mathrm{May}\hskip0.5em 28,\hskip0.5em 2021)\\ {}\;\left(\mathrm{A}.\mathrm{Laschet}\right);\hskip0.5em {{\mathrm{KANZ}}_{\mathrm{Q}-1}}^{\mathrm{OPP}}=18\%(\mathrm{May}\hskip0.5em 28,\;2021)\;\left(\mathrm{O}.\mathrm{Scholz}\right);\\ {}\mathrm{SPDINC}=1;\hskip0.5em {\mathrm{GCOAL}}^{09/17}=1;\hskip0.5em {{\mathrm{CO}}_{\mathrm{OPP}}}^{\mathrm{FDP}/\mathrm{CDU}}=11\%;\\ {}{\mathrm{PGRUNE}}^{\mathrm{OPP}}=24\%;\hskip0.5em {{\mathrm{CO}}_{\mathrm{OPP}}}^{\mathrm{SPD}/\mathrm{GRUNE}}=13;\hskip0.5em \mathrm{PLINKE}=6.5\%;\\ {}\mathrm{OTHERS}=16\%\left(\mathrm{wherein}\ \mathrm{non}-\mathrm{AfD}=6\%\right)\end{array}} $$

If the election had taken place as of this writing (June 2021), (see figure 1) the forecast would be 228 seats (i.e., 32.1% of the seat share) for Armin Laschet and he would more likely than not be chosen as the next chancellor, according to our calculations in figure 2.Footnote ⁴ Nevertheless, that would mean attaining fewer seats for the party than in 2017 (i.e., 246 seats). This result supports the notion of a continued decline for the CDU/CSU; concurrently, the AfD is expected to take a firmer hold, with 93 seats. Laschet’s popularity as “favorite chancellor,” initially low at about 16%, suggests a weakness at the ballot box. However, by coincidence, he stands to benefit from a similarly deficient popularity for his SPD competitor, Olaf Scholz (18%). Furthermore, the Green Party appears poised for a breakthrough, which would put it in an approximate tie with the SPD, which would come in second—although still at a considerable distance from the CDU/CSU (i.e., a gap of 82 seats).

Figure 1

Seats share forecast (%) June 2021

Figure 2

Seats forecast June 2021