Before the campaign the Berlin Left barely registered. Election night is the high point; the coalition talks, the Spiegel interview and the Emde row each drew less attention than the one before.
Every headline naming the Left or Eralp, coded by its main frame. Each column is a period, each band one frame’s share carried from period to period. The bands show how the mix changed, not articles moving between frames. Choose a frame to follow it.
The summer is about policy and government. The campaign is about polls. After the election, governing and extremism run neck and neck. The interview week adds threats against the party. In the Emde phase extremism is nearly the only topic, but none of those headlines is negative towards the party: they report it acting against an MP.
Share of each party’s headlines whose main frame is extremism, antisemitism or criminal ties.
Did articles on the Left change after the election relative to articles on CDU/CSU, SPD, Greens and AfD in the same outlets and weeks? Outcome: whether an article contains accusation vocabulary (antisemitism, extremism, scandal, investigation and similar stems). Party, period and outlet fixed effects; standard errors clustered by day; 3,546 article × party rows.
Before the interview the controversy was pinned on the party: among articles naming Eralp, only 26% put her in the headline, against 65% in the Krach (SPD) and Spahn (CDU) scandals (OR 0.18, 95% CI 0.09–0.39). The interview made it personal for one week. Since 4 November the headlines belong to Emde.
Every headline naming the Left or Eralp from 28 October, one square each, by main frame.
The interview in Der Spiegel (28/29 October) is the clearest single trigger in the data: personal headlines jumped the next day, and six of seven Left headlines on 29 October were negative. The taz pushed back twice, criticising the Spiegel’s coverage on 30 October and again on 6 November. Since the interview the taz covered the Berlin Left most (8 articles), ahead of the FAZ and SZ (6 each) and the Spiegel (4).
| Test | p (Holm) | Holds |
|---|---|---|
| Accusation vocabulary rises after the election | 1.0e-5 | yes |
| Party, not person, in headlines (to 29 Oct) | 1.5e-5 | yes |
| Story outlasts the Krach scandal | 3e-15 | yes |
| Eralp coverage shifts to extremism | 0.062 | no |
| Eralp’s role in events shifts | 0.25 | no |
| Measure | Agreement |
|---|---|
| Headline frames | α 0.83 |
| Extremism frame alone | α 0.75 |
| Accusation dictionary | P 0.73 · R 0.71 |
| LLM blame score | ICC 0.71 |
| Second coder | blind AI |
Data. Politikradar collects the top articles from the RSS feeds of 11 German online outlets every day (Welt, Spiegel, FAZ, tagesschau, SZ, Zeit, taz, Handelsblatt, heute and others), about 80 a day. Each article is stored as headline plus a short AI-written summary; the full text is not kept. From October, Zeit and part of SZ provide only teasers.
Berlin Left. Articles naming Eralp, Emde, Koçak or the red-green-red coalition, or the Left together with Berlin.
Coding. Frames, tone and critics were coded by an AI (Claude) with a fixed codebook. The newest period covers three days and about ten articles, so read it as a direction.
Limits. This measures what outlets reported, not whether allegations are true, and not what newsrooms intended. The AI summaries sometimes get details wrong; the counts here do not depend on those details.