Merkel is one of the most prominent figures in international politics. Her successor, whoever they may be, will have the power to shape global politics just as much. So let’s look at the candidates’ track records. We analyzed:
- which words each candidate uses more often than others in parliamentary speeches
- how each candidate voted on various issues in parliament, and how these votes compare to Merkel’s
In this repository, you will find the methodology, data and code behind the story that came out of this analysis.
Read the full article on DW: English | German
Story by: Kira Schacht
| Name | Content |
|---|---|
scripts/btw_candidates.Rmd |
The main R markdown script. Run in RStudio to reproduce this analysis. |
scripts/data.RData |
The R Data file containing the imported datasets. Use if csv import doesn’t work. |
data/... |
Data files |
Data on voting behaviour is taken from the Abgeordnetenwatch.de API.
Parliamentary speeches of the candidates are downloaded from:
- https://www.bundestag.de/services/opendata
- https://www.landtag.nrw.de/home/dokumente_und_recherche/parlamentsdatenbank/
- https://www.buergerschaft-hh.de/parldok/formalkriterien
Here is a step-by-step-explanation of the code we used in this analysis.
You can explore it yourself by opening btw_candidates.Rmd in RStudio.
Data from Abgeordnetenwatch.de API extracted in early June 2021. The latest votes included are from 19th May 2021.
Number of polls by candidate:
| candidate | n |
|---|---|
| baerbock | 297 |
| laschet | 38 |
| merkel | 443 |
| scholz | 88 |
Share of polls by candidate and response:
| abstain | no | no_show | yes | |
|---|---|---|---|---|
| baerbock | 0.12 | 0.31 | 0.14 | 0.43 |
| laschet | 0.03 | 0.66 | 0.11 | 0.21 |
| merkel | 0.00 | 0.06 | 0.65 | 0.29 |
| scholz | 0.05 | 0.14 | 0.22 | 0.60 |
See btw_candidates.Rmd for code.
From the list of voting topics, those relevant to DW audiences are identified manually:
2 coders label each topic independently as either a “DW topic” or a “non-DW topic”. They then discuss and resolve any discrepancies.
The list of topics identified as DW topics in this way is:
- 17 Entwicklungspolitik (development policy)
- 20 Energie (energy)
- 43 Menschenrechte (human rights)
- 48 Naturschutz (nature conservation)
- 35 Frauen (women)
- 4 Europapolitik und Europäische Union (European policy and European Union)
- 9 Umwelt (environment)
- 11 Außenwirtschaft (foreign trade)
- 13 Verteidigung (defense)
- 21 Außenpolitik und internationale Beziehungen (foreign policy and international relations)
- 25 Ausländerpolitik, Zuwanderung (immigration policy)
- 33 Humanitäre Hilfe (humanitarian aid)
| dw_topic | n |
|---|---|
| FALSE | 289 |
| TRUE | 281 |
Around half of all polls fall into DW topics.
All votes that concern DW topics are then labeled manually to identify which specific issues they concern and what stance on these issues a “yes” vote on the poll conveys.
The manual content analysis was conducted as follows:
3 coders read the polls labeled dw_topic == TRUE and created a list of
10-20 key issues the polls relate to. The 3 lists are then compared,
discussed and consolidated into one list.
The 15 key issues identified in this way are:
| group | code | label |
|---|---|---|
| Militär | 1 | Bundeswehreinsatz |
| Militär | 2 | Mittelmeereinsatz |
| Umwelt | 3 | Atomenergie |
| Umwelt | 4 | Erneuerbare Energien |
| Umwelt | 5 | Emissionsausgleich |
| Umwelt | 6 | Kohleenergie |
| Umwelt | 7 | Klimaschutz |
| Umwelt | 8 | Naturschutz |
| Umwelt | 9 | Gentechnik |
| Außenpolitik | 10 | EU-Kooperation |
| Außenpolitik | 11 | EU-Finanzhilfen |
| Migration | 12 | Flüchtlingspolitik |
| Migration | 13 | Zuwanderung |
| Außenpolitik | 14 | Handelsabkommen |
| Militär | 15 | Rüstungsexporte |
| Andere | -99 | Nicht zutreffend |
| Anderes | -90 | Zu komplex |
Polls could also be labeled as concerning neither of the 15 key issues
(-99 = Nicht zutreffend) or as too complex to discern whether a
“yes”-vote can be counted as a stance pro or contra any of the
issues (-90 = Zu komplex).
2 coders independently assigned each poll to up to 3 of the key issued identified in the previous step. For each poll and issue, they also note whether a “yes”-vote in the poll indicates a stance for or against the issue. The labels are then compared and discrepancies discussed and consolidated.
##
## -99 -90 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
## 25 9 145 30 11 4 3 3 12 14 5 16 15 20 23 2 1
9 polls were labeled as too complex to include (-90), 25 didn’t have
anything to do with the selected key topics (-99). Those are excluded
from the analysis.
Column vote = {"yes", "no", "abstain", "no_show"} records how the
candidate voted (Source: Abgeordetenwatch API). Column pro.contra = {0,1} records whether a “yes” vote means the candidate voted for (1)
or against (0) the labeled key issue (Source: manual labeling).
stance then records the stance indicated by the vote (TRUE if the
candidate supports the issue with this vote, FALSE if they don’t) as
follows:
vote = yesANDpro.contra = 1–>TRUEvote = yesANDpro.contra = 0–>FALSEvote = noANDpro.contra = 1–>FALSEvote = noANDpro.contra = 0–>TRUEvote = abstainORvote = no_show–> N/A
To allow a solid basis of comparison between candidates, stances are included in the following analysis if:
- at least two different candidates (incl. Merkel) have participated in at least one vote on the issue and
- the individual candidate has participated in more than one vote on it
For each issue, the candidate’s stance is calculated as the share of votes that supported the issue.
The candidates have given speeches in 3 different parliaments. To get data on all speeched, we proceeded as follows:
Bundestag (Annalena Baerbock ab 2013, Olaf Scholz 2005-2011):
- Download XML data from https://www.bundestag.de/services/opendata
- Filter for candidate’s speeches
Landtag NRW (Armin Laschet):
- Find a list of speeches in the parliamentary databases and download it as a HTML document.
- URL: “https://www.landtag.nrw.de/home/dokumente_und_recherche/parlamentsdatenbank/Suchergebnisse_Ladok.html?dokart=PLENARPROTOKOLL&redner=LASCHET%2C+ARMIN*&view=detail&allOnPage=true&wp=17”
- Go through the pages for wp=14 through wp=17
- Save each as a HTML file
- Convert the PDF documents to text and save
- Filter for candidate’s speeches
Hamburgische Bürgerschaft (Olaf Scholz):
- Find a list of speeches in the parliamentary databases and download it as a HTML document.
- URL: “Formalkriterien”-search under “https://www.buergerschaft-hh.de/parldok/formalkriterien”
- For Wahlperiode 19 and 20, choose “Urheber (Personen): Scholz, Olaf” and tick “Reden”.
- Save each as a HTML file
- Convert the PDF documents to text and save
- Filter for candidate’s speeches
See btw_candidates.Rmd for code.
See btw_candidates.Rmd for code.
The parliamentary protocols from the federal parliament are available as XML files here.
See btw_candidates.Rmd for code.
See btw_candidates.Rmd for code.
See btw_candidates.Rmd for code.
See btw_candidates.Rmd for code.
See btw_candidates.Rmd for code.
See btw_candidates.Rmd for code.
To analyze which words are most characteristic for each candidate, we use a smoothed odds ratio:
We separated their speeches into individual words, removing stopwords (filler words that carry no meaning) and grammar (using the Snowball stemming algorithm for German). We then calculated the relative frequency of each word in the candidates’ speeches and compared it to its relative frequency in their competitor’s speeches. This ratio, with a small constant added to avoid division by zero, is called the “odds ratio” and describes how much more likely a candidate is to use a given word than their competitors are.
If the odds ratio is bigger than 1, the candidate is more likely to use the word than their competitors are, if it is between 0 and 1, they are less likely to use it. If the value for “change” is 2, for example, the candidate is twice as likely to use the word “change” than their competitors. If the value is 0.5, they are hald as likely to use it.
For more info on odds ratios, as well as other measures of “characteristic” words, see here:
Monroe, B., Colaresi, M., & Quinn, K. (2017). Fightin’ Words: Lexical Feature Selection and Evaluation for Identifying the Content of Political Conflict. Political Analysis, 16(4), 372-403. doi:10.1093/pan/mpn018 Link
See btw_candidates.Rmd for the
code.
