Finally, the classic Left-Right vs. pro-anti Europe plot of party positions, to confirm that at the party level the pattern still holds in 2024, with some small but significant variations across regions.

You can explore the country-level relationships in the app here:
dimiter.shinyapps.io/ches/

Next, we look at the measure of internal party dissent with regard to EU positions. Dissent is of course related to the clarity of positions, but the two are not the same.
Western European parties with moderate positions on Europe have the highest levels of internal dissent, on average.

Next, blur of positions on European integration:
Again, Eurosceptic and pro-European parties have the clearest positions, whille the parties in the middle are the most likely to have blurred or unclear positions on Europe. This is especially true for parties in Eastern and Southern Europe.

First, let's look at salience:
A rather pronounced inverted U-curve, with the most Eurosceptic and the most pro-European parties having the highest levels of saliance. Not much difference across Europen regions, but in Western Europe, the dip in saleince for centrist parties is the deepest

That's a great way to put it!

In your example, the experiment primes identity, so it manipulates its recency and availability in the minds of respondents. This might be important for real world interventions, such as political speeches.

I don't think that's correct. First, most survey experiments control 'which' vignette is presented, not the order. Second, the differences in answers to a question following exposure to alternatives vignettes is not necessarily the same as a stated preference.

I think that's what practitioners of conjoints will say, only in their case the 'behavior' is actually responding to a different survey rather than an actual purchase, vote, etc.

Not a fan of conjoints, but for the sake of argument: asking directly would *not* reveal your preference for women due to social desirability bias or imperfect introspection. So the conjoint identifes a version of a causal effect (with limited transportability to the real-world IMO, but still).

In your example, you have estimated the causal effect of exposure to the word 'coffee' or 'tea' on my tendency to say 'yes' or 'no'. Whether this causal effect has any real-world or theoretical relevance is a different question. Meanwhile, you have *also* measured my preference for tea or coffee.

The blog post puts 'priming interventions' as measurement-oriented. But they identify the causal effect of the prime, hence causal inference. And, in principle, this effect can be useful for policy purposes, for example in the design of information or communication strategies.

If conducted properly, a survey experiment with randomized exposure to the treatment identifies the average causal effect of the treatment relative to the alternatives, no? This is 'measuring quantities' in the same sense that any other design is 'measuring quanitities', incl. lab exp-s, d-in-d, etc

I don't follow: survey experiments estimate the causal effect of an experimental intervention, which is often in the form of a vignette, audio stimullus, video message or whatever.
So I agreee that they aren't central to the credibility 'revolution', but they aren't purely descriptive/measuerment.

The effect was ... 2 points on a 100-point scale 🙄. It was due to a LLM-based extension to X designed by the researchers to max polarization, not to changes in the native X alghorithm.

Funny that the link in the Guardian article is to a commentary that is not very enthusiastic about the results.

Yeps, 2024

yeps, it's telling that the Renew vote on the Omnibus Simplification Directive was split, with 17 MPs siding with the EPP.