Hacker News new | comments | ask | show | jobs | submit login
Politics in the Facebook Era: Evidence from the 2016 US Presidential Elections [pdf] (warwick.ac.uk)
18 points by longdefeat 61 days ago | hide | past | web | favorite | 5 comments



I think this paper's methods are suspect. They are 'measuring' politic spend/targeting in 2016 by assuming that price increase for a specific targeting segment = more targeted political spend to that measured audience. But political spending is a tiny % of FB spending. I strongly doubt this method gives any actual insight, first into what/who these campaigns spent/targeted in 16, and second on the persuasion effect of that spend. Also how would you control to only FB ad exposure effect using the survey data mentioned in the paper...

I actually buy FB ads for political candidates as a big part of my business. We've done some 'lift/persuasion' type experiments where we do 1:1 targeting with a control group and do an IVR to measure changes pre and post exposure. This feels like the best way to do this experiment - but that would require academics to work with actual spenders (or to spend themselves).

I actually really wanted to use FB's new attribution reporting with voting results this year but our accounts weren't given access until like a couple days before election and it wont let me look back in 'time' (e.g. we uploaded 'votes' as a purchase with the voted date timestamp, but FB will only show attribution window from date we setup the reporting tool, which is after the votes...).


> [O]nline political campaigns targeted on users’ gender, geographic location, and political ideology had a significant effect in persuading undecided voters to support Mr Trump, and in persuading Republican supporters to turn out on polling day.

No mention of any targeting beyond what has been on user’s profile for a decade, and it transparently used by most ads.

> [W]e use variations in advertising prices across political audiences as a measure of the intensity of Facebook political campaigning.

Political advertising represents less than 5% of ads on Facebook at the peak of a campaign, in swing states: I’m puzzled you can measure anything. Actually:

> As illustrated in Figure 4, average worldwide prices sharply rose during the last two months before the elections (an average increase of around 50 USD cents, which corresponds to about a 25 percent increase in prices), and steadily dropped over the three months after the election date.

That’s unlikely to be the election unless people spend money for three months after. That’s more likely to be US Holiday Season.


> No mention of any targeting beyond what has been on user’s profile for a decade,

The story about Cambridge Anal. was specifically that they had build a dataset that could infer a lot of information, such as sexual orientation and political affiliation from a small set of apparently harmless information.

> Political advertising represents less than 5%

A market's reaction to such variation can be widely different. For bread, it's probably nothing. If it's the last batch of the only medication that is effective against some just-discovered deadly pandemic, the price could rise from %1 to a billion.

> That’s unlikely to be the election unless people spend money for three months after. That’s more likely to be US Holiday Season.

The chart & data here are average over 40 different worldwide elections. Unless elections cluster around certain dates, the data should be pretty well spread out over the year.


> The story about Cambridge Anal. was specifically that they had build a dataset that could infer a lot of information, such as sexual orientation and political affiliation from a small set of apparently harmless information.

Yeah. That story is false. They used magazine subscriptions and car model. Source: talked to the employees.

> The chart & data here are average over 40 different worldwide elections. Unless elections cluster around certain dates, the data should be pretty well spread out over the year.

I’m fairly confident no political advertising is significant after the election in most of them. The big secret here is that US ads still represent the largest share of Facebook’s market and that commercial markets have a strong yearly pattern. I don’t know how they averaged to get a ratio number, but I would compare years with and without elections.

And, yes, elections tend to be around the same date in most countries (except, notoriously, the UK): it doesn’t have to be the same in each country for an aggregated commercial pattern to be prevalent.


100% agree. Even if they could measure political spend, there is no way to tie that to persuasion. Maybe those who use FB are more likely to support trump. WAY too may variables without a control group 1:1 voter experiment (i've done survey before, match 1:1, withhold 20% random, run ads to 80%, run follow up survey, compare difference between two groups)




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: