Another nice episode! You explain super well why we should keep in mind the narrow research question. For example, “Changing the Facebook algorithm to reverse chronological for 20,000 people for three months in 2020 has little effects”.
Ongoing work by two colleagues at PSE (and others) suggests that we may need to be even more specific. They ran an experiment in 2023 with two treatments on X, paying users to change their algorithm settings. The key thing is that some X users are still on the chronological feed (maybe not a fully representative population, but still interesting).
The results are more nuanced: Again, switching OFF the engagement algorithm (i.e. switching to reverse chronological) has little effects on outcomes, but switching ON the algorithm makes users more conservative. The explanation for the asymmetric effect is that the X algorithm makes you follow more conservative accounts and the effects of this persist even when you switch off the algorithm for a short period.
Oh that's interesting -- that there are delayed effects from timeline recommendations. I look forward to reading more!
I guess there's two hypotheses here: about FB and 2020, and about SM and elections generally. This research may definitely influence my view on the latter.
Fascinating, thanks for the reference! I'd have to read the paper in detail, but the results are surprising to me, since I think these sorts of attitudes are hard to shift.
Another nice episode! You explain super well why we should keep in mind the narrow research question. For example, “Changing the Facebook algorithm to reverse chronological for 20,000 people for three months in 2020 has little effects”.
Ongoing work by two colleagues at PSE (and others) suggests that we may need to be even more specific. They ran an experiment in 2023 with two treatments on X, paying users to change their algorithm settings. The key thing is that some X users are still on the chronological feed (maybe not a fully representative population, but still interesting).
The results are more nuanced: Again, switching OFF the engagement algorithm (i.e. switching to reverse chronological) has little effects on outcomes, but switching ON the algorithm makes users more conservative. The explanation for the asymmetric effect is that the X algorithm makes you follow more conservative accounts and the effects of this persist even when you switch off the algorithm for a short period.
The abstract is here (paper on request I think): https://www.cognition.ens.fr/en/agenda/political-effects-xs-recommender-algorithm-17955
Curious if this affects your posteriors at all.
Oh that's interesting -- that there are delayed effects from timeline recommendations. I look forward to reading more!
I guess there's two hypotheses here: about FB and 2020, and about SM and elections generally. This research may definitely influence my view on the latter.
Fascinating, thanks for the reference! I'd have to read the paper in detail, but the results are surprising to me, since I think these sorts of attitudes are hard to shift.