The political relevance of the Cambridge Analytica/Facebook data

Do you use Facebook?  You probably do because you aren't me.  That means the Trump campaign got data about you.  Hah!  Sorrynotsorry.

Anyway, today's topic is the use of Facebook's data for micro-targeted ads.  What can these accomplish?  Buried within this question are two separate questions.  First, what do we learn about you beyond your basic demographics, and second, can that knowledge be exploited?

Let's start with the first question.  If I know that you are African-American, how much do I know about your partisan preferences?  A lot.  Why, because around 90% of African-Americans are Democrats.  Tell me that you are African-American, and in particular, an African-American woman, and I can make a really really really good prediction about your vote choice.  In the 2016 American National Election Studies survey, 95% of African-American women went Clinton over Trump in the two-party vote.  Now, you are all individuals.



Fine, fine.  We are all individuals.  However, this is about baseline statistical probabilities, and the baseline statistical probability of an African-American woman voting for a Democrat is so high that the additional information on Facebook doesn't give you that much more.

I'm just talking about voting behavior.  As a person, what does Nora K. Jemisin have in common with Bettye Lavette?  No clue.  I've never met either.  I really really really want to meet both.  I would expect them to be very different people.  The way Jemisin writes, and the way Lavette sings, I doubt they have much in common.  But, I bet neither voted for Donny-boy.

White people?  White males?  Statistically, we are more likely to identify as Republicans, and vote Republican, but statistically, we are more divided than African-Americans.  According to the 2016 NES, white males favored Trump by about 60-40 in the two-party vote.

Um... what was I saying about statistical inferences and groups?  Uh...

I'm going to quote the great philosopher, Frank Zappa here:  "I'm not black, but there's a whole lotsa times I wish I could say I'm not white."

Regardless, there is an important statistical point here.  There is less politically relevant information conveyed by the fact that I am a white man than by the fact that Nora Jemisin is an African-American woman.  How much other basic information could be added to simple demographics to get my preferences correct, though?

I'm weird, for a lot of reasons, but this is the basic issue for Cambridge Analytica.  How much can you learn from personality-based information, above demographic baselines?  Basic political science will tend to get you a lot further than "big data" bullshit.

Second, what about micro-targeted ads?  Here's the basic fact that I beat into my students' heads, metaphorically, every time I teach about voting behavior.  90% or so of the electorate are partisans.  Forget that bullshit about how 30-40% are independents.  Those people are liars.  The basic political science method of asking about partisanship in a survey is to ask two questions rather than one.  If someone first claims to be an independent, you must then ask if that person leans towards one party or the other.  Most "independents" admit to a leaning, and most "leaners" act just like partisans, hence the line that so many of my students can recite:  "leaners are liars."  Can I trademark that?

Somewhere around 85-90% of votes in any given election are just partisans voting their party ID.  The rest have messages coming at them from all directions and incoherent preferences about politics.  The idea that they are susceptible to narrowly tailored messages relies on the idea that they have consistent and coherent preferences, and think in consistent ways about politics, but if they did that... they'd be partisans, and have their votes determined by partisanship, in which case, none of this Cambridge Analytica stuff would matter anyway.

Does the micro-targeting of ads influence outcomes?  I strongly doubt it.  James Comey?  Yeah, that motherfucker gave the White House to Trump.  Cambridge Analytica?  Nope.

So, here's the storyline.  Trump's campaign used Cambridge Analytica data from Facebook, and Trump won.  Therefore, he won because of micro-targeted ads.  Post hoc, ergo propter hoc.  Nope.  That's a logical fallacy.

Trump bragged about committing sexual assault and getting away with it.  Trump won.  Therefore, Trump won because he bragged about getting away with sexual assault.  Post hoc, ergo propter hoc.

Put shit on Facebook, and someone's going to get it.  Did it swing the 2016 election?  Not likely.  Privacy, though?  Yeah, that's an issue.

Tweet

Subscribe to receive free email updates: