Julius Kölzer

Delighted to announce that today is the first day in weeks that no one told me I should use this new AI tool and this new causal inference estimator

Das Problem der SPD ist nicht, dass sie zu sehr Arbeiter oder Mittelschicht anspricht. Sie erreicht programmatisch keine dieser Gruppe. Ihre Gruppe sind Renter:innen. Auch in den Ländern liegt sie überall unter 20% außer sie stellt den Ministerpräsidenten oder die Ministerpräsidentin.

On Sunday, one of Germany’s largest states went to the polls in a closely contested election. What can we learn from the results & how do they relate to looming deindustrialization in one of Germany’s industrial heartlands? A thread with some decriptive patterns from data on 1,101 municipalities. 🧵

😭

Thread.🧵Fascinating read in the light of UK politics/polls, w many strong echoes. German Greens c 15-20 yrs ahead of UK ones, w the SDP's decline prefiguring UK's Labours dire polling, Also note how the AfD's vote correlates w lower educ'n levels, un-/semi-skilled work & fear of deindustrialization.

I was asked to run some horse-race regressions with all covariates included (which is not always the best idea and very sensible), but here they are. The estimates for the Greens look fairly unremarkable, apart from the share of employees with tertiary education.

And here are the same models for the Linke and FDP.

Besides the patterns discussed earlier, the AfD’s gains in places with larger shares of voters with a migration background are noteworthy. This association appears to be driven by small & mid-sized towns with sizable Russian German & resettler communities (e.g., Bubsheim, Pforzheim, Lahr, Rastatt).

The SPD’s geographical coalition appears to be largely reduced to areas with older voters, who make up a large share of the party’s remaining voter base.

The same applies to the CDU. There also seem to be some gains in more peripheral areas farther away from major cities.

I was asked to run some horse-race regressions with all covariates included (which is not always the best idea and very sensible), but here they are. The estimates for the Greens look fairly unremarkable, apart from the share of employees with tertiary education.

We also observe some AfD gains in municipalities with a higher share of voters with a migration background—though not in large urban centers, but rather in small and mid-sized towns with sizable Russian German and resettler communities (e.g., Bubsheim, Pforzheim, Lahr, Rastatt).

thanks!

Thanks! I had to smirk a bit reading the comment about the center-left being ‘squeezed,’ as the SPD’s vote share in the scatterplots sort of literally looks like it’s bleeding out

Auf Gemeindeebene gibt es bei Industriebeschäftigung aus Gründen der Geheimhaltung seitens des statistischen Landesamts sehr viele missings. Alle Modelle/Plots nutzen immer den jeweiligen Share auf Wahlkreisebene (die kleinräumigste Ebene für diese Daten).

Danke!

A must-read thread
👇

Fascinating thread - one thing that leapt out for me was a formerly dominant centre left party being squeezed by Greens in diverse graduate heavy cities and by the radical right in economically insecure manufacturing towns. Certain patterns keep repeating

Dieser Zusammenhang zwischen dem Anteil des produzierenden Gewerbes und AfD-Stimmen und Stimmgewinnen ist wirklich bemerkenswert stark.

Thanks for reading!

6) Similar to the Greens, the Linke makes gains in larger cities where it already performed above average in 2021. However, these gains are not enough to offset its consistently low vote shares in rural constituencies, leaving it short of the threshold for entering the state parliament.

5) Finally, the FDP: Although liberal Baden-Württemberg is the party’s historic stronghold, it is being almost completely wiped out across much of the state and fails to clear the 5% threshold. The FDP still performs somewhat better in affluent municipalities and Protestant upper-middle-class areas.

Despite announced manufacturing layoffs, SPD vote shares are negatively correlated with industrial employment. However, the state’s industrial areas are dominated by SMEs and have historically leaned CDU (the SPD’s decline in the state’s manufacturing base also appears to have occurred long ago).

However, the relationship between the share of industrial employment and AfD gains is not significantly stronger in areas with SME-dominated labor markets, declining business tax revenues, or higher insolvency rates. Data for 2025 is however not yet available.

Obviously, manufacturing in Baden-Württemberg is largely concentrated in more rural areas. However, also in a simple municipality-level regression model controlling for different spatial characteristics, a higher share of manufacturing employment remains a strong predictor of AfD gains.

While areas with larger manufacturing sectors were already AfD strongholds, the party appears to be making further gains here (often by double-digit percentage points). The CDU also appears to makes modest gains in these places, while the incumbent Greens seem to suffer substantial losses.

The AfD performs particularly strongly in parts of rural Baden-Württemberg affected by prospects of manufacturing layoffs (especially in areas reliant on automotive & mechanical engineering industries). Constituencies with higher manufacturing employment show a very strong shift toward the AfDs.

AfD gains are also notably lower in areas with higher shares of residents with tertiary education. Conversely, this suggests stronger shifts toward the AfD in municipalities with larger shares of low- and medium-skilled workers.

A further increase in support compared to 2021 is found in municipalities with higher homeownership rates. As is typical, the party performs worse in areas where the Catholic milieu is more strongly rooted. However, this negative association is rather moderate.

4) The far-right AfD gains 9 percentage points, expanding its support primarily where it already performed strongly in the last state election 2021. As the scatterplot and the map show, these areas are, unsurprisingly, rather rural and peripheral.