This thread saying that Kamala is one of the "most progressive" senators based on "metrics" is going around.
As someone who works with stats, I'm triggered when political actors abuse popular trust in stats to spread misinfo, so let's go thru this 1 by 1: https://twitter.com/cmclymer/status/1291379835332448257?s=20
As someone who works with stats, I'm triggered when political actors abuse popular trust in stats to spread misinfo, so let's go thru this 1 by 1: https://twitter.com/cmclymer/status/1291379835332448257?s=20
ProgressivePunch's score is based on how closely a legislator votes with an algorithmically selected (long story) "progressive" cohort. Who's in the cohort? Kamala Harris, Bernie, and others. So Kamala scores high because she votes like Kamala. Cool!
https://progressivepunch.org/whatIsProgScore.htm
https://progressivepunch.org/whatIsProgScore.htm
http://govtrack.us starts with a similarity matrix of bill cosponsorship, so two legislators are similar if they cosponsor bills together, then uses a technique called PCA to transform this matrix...
https://govtrack.us/about/analysis#ideology
https://govtrack.us/about/analysis#ideology
...into a single number such that legislators who cosponsor bills together are closer together. In our two-party system, you'll generally see one cluster of negative numbers and one of positive numbers which correspond to D and R.
The authors call this number an "ideology score", meaning one extreme is conservative and one is progressive, but a better interpretation is as a partisanship score; a low score really means you're less likely to cosponsor with R's or with D's that cosponsor with R's.
Voting with Trump by a few percentage points less doesn't make you more progressive. Kamala voted no on Bernie's amendment to slash Pentagon funding by 10% to fund social services. Is that progressive?
http://voteview.com uses DW-NOMINATE, another clustering technique. The math is different than the technique used by http://govtrack.us above, and this one is based on votes instead of bill sponsorship, but result is conceptually similar: ...
https://voteview.com/about
https://voteview.com/about
... You're assigned a score close to the people you vote similarly to. This leads to strange situations like AOC and Ilhan Omar being seen as moderates because of a few rank-breaking votes.
https://voteview.com/articles/Ocasio-Cortez_Omar_Pressley_Tlaib
https://voteview.com/articles/Ocasio-Cortez_Omar_Pressley_Tlaib
In conclusion:
1) Cluster analysis doesn't measure ideology, it measures affinities and allegiances. All this data shows is that Kamala works with D's the most and R's the least (which is weird bc I was told that you had to reach across the aisle to get anything done).
1) Cluster analysis doesn't measure ideology, it measures affinities and allegiances. All this data shows is that Kamala works with D's the most and R's the least (which is weird bc I was told that you had to reach across the aisle to get anything done).
2) Ideology cannot be measured through votes & bills co-sponsorships without analyzing the substance of those votes & bills.
3) Being more Democrat doesn't make you more progressive.
4) Distrust anyone who touts these metrics without describing how they work in detail.
3) Being more Democrat doesn't make you more progressive.
4) Distrust anyone who touts these metrics without describing how they work in detail.
5) Lol unrelated to the rest of the thread, but remember the time Kamala fought against a Supreme Court order to release nonviolent offenders early from severely overcrowded prisons, arguing that the state needed cheap firefighters for all the wildfires? https://www.thedailybeast.com/kamala-harris-ag-office-tried-to-keep-inmates-locked-up-for-cheap-labor
One more general point about political data science shit is that the more complex the methodology, the more the authors need to just that the metric really measures what they claim. The math may not be wrong, but there's plenty of room for ideology and bias in the interpretation
It's funny that the authors of these studies use these clustering methods because they're more "objective." But the choice to measure ideology with clustering, to cluster on votes and cosponsorship - belie assumptions about how ideology is expressed...
...and the relationships between party affiliation and ideology. These assumptions must be questioned. Eliding these questions while pretending these are all objective, scientific numbers IS a key component of liberal/Democrat dominance
Anyway, this blew up (relative to my tiny account). Thanks for the love and follow me if you want - I mostly tweet about leftism, race/identity, hating the cops, snacks I'm eating, and occasionally this math stuff when it really pisses me off