We’ve been using data, now published publicly by Facebook via their Ad Library, to try and better understand how campaigns are targeting their ads for the midterms.

We’ve done this in two ways, using data from the last three months.

First, we’ve built a map of the US, state-by-state, showing how and where Democratic (i.e. candidates, PACs and other progressive organizations) and Republican (i.e. the same, but for Republicans) have used zipcode-level targeting to direct their Facebook ads over the last 90 days (they sometimes also use zipcodes to avoid showing ads in some locations).

Second, we’ve built a chart showing all the interests that Democrats and Republicans have targeted AND excluded from their Facebook ad targeting. Why don’t Democrats want people with an interest in Elon Musk to see their ads? How are campaigns using proxy interests to get around Facebook’s ban on targeting political interests?

You can explore the data here:

https://whotargetsme.shinyapps.io/midterms2022/

(The service we’re using to display the data has quite limited bandwidth and connectivity. You may need to be patient to view the data and/or refresh a few times.)

To build it, we:

• Categorised, by partisan affiliation, several thousand US political Facebook advertisers.
• Grabbed ad targeting data and relative spending from the ‘audience’ tab in Facebook’s political ad library and added it all up.
• For the map, overlaid all of this on top of the entire US, sub-divided into zipcodes and states.
• For interest targeting, we added up all the spending on each interest, across all of the pages.

Taken together, the data gives a sense of where people are spending their money to try and reach and persuade users ahead of the midterms.

Unfortunately, there are some limitations, caused mostly by the imperfect data provided by Facebook. Let’s be up front about those:

1. Only 10% of the total ad spend is on zipcode-level targeting. Obviously that means over 90% doesn’t use it.
2. About 11.5% of ad spending uses any interest-based targeting.
3. Custom (70% of spend) and Lookalike audience targeting (19%) are the most commonly used. Sadly there’s no data we can use to help visualise who is and isn’t seeing ads targeted in these ways.
4. Different types of targeting can overlap with each other (i.e. none of these numbers will add up to 100%).

Our plan is to release another version of this once we have data up to and including election day later in the week.

Thanks to Fabio Votta for his help building this.