2 ms·

> ...microtargeting and segmented audiences... Asking out loud: Is voter profiling basically k-means clustering? Given the USA's FPTP elections, k=2. (Per Du

> ...microtargeting and segmented audiences... Asking out loud: Is voter profiling basically k-means clustering? Given the USA's FPTP elections, k=2. (Per Duverger's Law.) Quick peek at wiki wrt Spain's govt, featuring a parliament and using plurality block voting, apparently yielding k=7. -- I've also participated in politics (campaigns). But I have zero idea how our consultants do their profiling and targeting. Closely guarded secrets. I recently read Jill Lepore's book If/Then about the history of Simulatics Corporation and their initial voter behavior modeling efforts and opinion polling. One of my takeaways is that Cambridge Analytica is just the 2016's iteration of the evergreen quixotic effort to model, predict, shape voting behavior. As you said, just like marketing. I couldn't find the initial research papers online. (I haven't yet made the pilgrimage to my uni's library.) Nor do I know how outfits like dataforprogress.org work. Would love to find a primer on this stuff. Because I have questions. What is a wedge issue, really? The stuff Karl Rove is famous for. Did he identify wedge issues? Or did he just recycle tropes and use advanced targeting? Or maybe it's case-by-case? Like racial animus is evergreen whereas angst over abortion became a fault-line over time. How malleable are the clusters? Ezra Klein in Why We Are Polarized argues there's some kind of feedback loop happening. People sorted, adopted the positions of their new cluster mates to align more closely, then became polarized. Lather, rinse, repeat. Can we do the opposite analysis? Identify unifying issues to counterbalance the wedge issues? Sorry, I could go on, but this is enough for now.

spaniard89277 5y ago
Well, I did voter profiling, and in my case I tried some clustering statistical methods, but IMO the best clustering involves a lot of human in the loop, keywords etc. It's a lot of cualitative work, I couldn't find any way to streamline clusters that made sense both in an operational level and understandable for non-tech/stats people. I was supposedly an assistant with the data pipeline but in the end was pretty much in charge of 90% of it's logic. Devs made the scrapers and some automated data adquisition, that got dumped into a data lake, I took the data from the data lake, I made a huge DB of voters with many columns (+30 don't remember exactly) from which I made a ruleset for initial clustering, based on conversations with creatives and operations and some of my own ideas, and that data got used for Ads people and operations people (basically, the ones that handle campaigns and volunteers). I know that devs eventually made a nice front-end for it, but it was basically ETL work, and instead of "buyer persona" it was kinda "voter persona". It was fairly sofisticated and honestly, it worked pretty well.

spaniard89277 5y ago

Well, I did voter profiling, and in my case I tried some clustering statistical methods, but IMO the best clustering involves a lot of human in the loop, keywords etc. It's a lot of cualitative work, I couldn't find any way to streamline clusters that made sense both in an operational level and understandable for non-tech/stats people. I was supposedly an assistant with the data pipeline but in the end was pretty much in charge of 90% of it's logic. Devs made the scrapers and some automated data adquisition, that got dumped into a data lake, I took the data from the data lake, I made a huge DB of voters with many columns (+30 don't remember exactly) from which I made a ruleset for initial clustering, based on conversations with creatives and operations and some of my own ideas, and that data got used for Ads people and operations people (basically, the ones that handle campaigns and volunteers). I know that devs eventually made a nice front-end for it, but it was basically ETL work, and instead of "buyer persona" it was kinda "voter persona". It was fairly sofisticated and honestly, it worked pretty well.