This week I attended the Swiss Political Science Association Annual Conference at the University of Basel.
It's one of my favorite conferences to attend. It's small, filled with familiar faces and I enjoy seeing what my colleagues are working on. This year, I presented 2 ongoing projects: One on the automatic detection of populist elements in political speeches and one on issue engagement in the Swiss parliament. The latter paper is quite exploratory in nature. We're trying to show which topics the Swiss parliament has dealt with over the past decades.
We are using all bills that ever moved through the parliament as our base. For every bill, we use the bill text (the submitted text, or for federal drafts, the media summary provided by the Federal Council) to group the bills by topic. Additionally, we have labeled bills using data from the Année Politique Suisse + bills we hand-labeled ourselves.
Then, we train a BERT-model to classify all bills into one of 21 policy areas. We're still tweeking the model, but for now we can classify bills with an accuracy of 85% (which is quite high, given that we have 21 categories).
The Figure below shows the distribution of bills by topic over the past 20 years. Since before 2000, the labeleld bills are still few, I don't quite trust the BERT-model fully. That's why I only show the past 20 years here. The policy waves are quite beautiful - you can clearly see the financial crisis of 2008, or the COVID-19 pandemic of 2020-21.
You can also see that some topics are quite stable (e.g., taxes), and some have show bumps along the way where issue attention was stronger (e.g., migration).
If we only look at bills that can be submitted by members of parliament (MPs) or parliamentary groups (Fraktionen), we can group them by sponsor + topic:
See how the left parties are increasingly sponsoring bills on agriculture? While liberal parties and middle parties have a relatively stable amount of bills they sponsor on agriculture, left parties have started to increase their share.
Now there's no need to overinterpret these figures: they are descriptive, do not contain any statistical tests for basis, nor are they final. But they do give some indication of how we can describe issue engagement at the parliament level with the DemocraSci data in future.