By Angela Loyola ’21 and Sam Feuer ’23
A prior Delta Lab research project on the amount of spending by presidential candidates Joseph Biden and Donald Trump on Spanish language Facebook ads found that while Trump was leading in spending until a few weeks before the election, Biden’s spending skyrocketed as the election neared, eventually surpassing Trump’s lead. As Hispanic voters were recognized to be a crucial voting bloc in the 2020 presidential election, especially in swing states, we wanted to continue researching the sorts of appeals that advertisers were using to target this community. How did Spanish language spending by the two candidates compare to other sponsors? What type of ads were run, and where were Spanish ads targeted?
Data
Data used in this analysis came directly from the Wesleyan Media Project, which tracks advertising through Facebook’s Ad Library API tool and Ad Library Report. Using SQL and Google’s Detect the Language feature, we filtered the data set for ads in Spanish that ran during the traditional fall general election period (September 1 to Election Day, November 3). As we were particularly interested in group activity, we further filtered the dataset to non-candidate advertisers that were also active on television and/or were easily identifiable as election spenders by WMP and/or OpenSecrets (this is not an exhaustive list of sponsors but is likely to contain the most prominently known outside groups active during the election). Groups include organizations like Service Employees International Union and AARP as well as PACs like Priorities USA Action and The Lincoln Project.
Sponsor by Sponsor
The following table shows spending by group:
Table 1: Total spend on Spanish ads by sponsor among non-candidate groups.
Click here to view an interactive version of this table.
It is worth noting that a majority of the groups shown above to be spending on Spanish language ads are Democratic or anti-Trump groups. The biggest spenders include Everytown For Gun Safety Victory Fund and CASA in Action PAC, which are blue-leaning PACs. The Lincoln Project and Republican Voters Against Trump also notably make the list as Republican groups supporting Biden’s candidacy.
However, Priorities USA, a Democrat-leaning super PAC, dwarfs most other organizations in terms of Spanish language spending. It is also closely connected to another one of the top spenders, Latino Victory Fund, and it is the only group to spend more than Joe Biden’s campaign. It is important to note that another organization, Senate Majority PAC, co-sponsored a significant portion of Priorities USA’s ads, and it is unclear what portion of the spend was from that PAC. The majority of the co-sponsored ads are likely related to this partnership.
Trump’s campaign also falls near the top of the list in terms of spend, albeit behind a few more groups.
Figure 1: Top Non-Candidate Sponsors in Spanish Ad Spend (10/01-11/03) Compared to Presidential Candidates
What are these groups talking about?
As we dived deeper into researching the groups that were sponsoring Spanish language ads, we also wanted to know what topics they were discussing in Spanish. In order to uncover this, we ran a Latent Dirichlet Allocation (LDA) topic model on the texts of the ads.
Topic modeling is a machine learning technique that takes a set of documents and creates a set of “topics.” Each topic contains a list of words with scores based on how commonly that word appears together with other words from that topic. The words with the highest scores can give an idea of the ideas that each topic is focused on. By running this statistical model, we hoped to determine what the most common “topics” were in the data set.
Figure 2: Most common words in Spanish ads among all topics from a 10 topic LDA model.
Note: Many of the odd looking words in the topics are unicode representations of letters with accents (é, for example).
This model demonstrates that some of the Spanish language ads are written with a combination of English and Spanish words, as evidenced by the existence of both languages within each topic.
More common, though, are “dynamic” ads, which advertisers use to create multiple versions of the same ad with slight differences. Facebook uses machine learning to determine which of these ads to show the user. In our data, the texts of the different ad versions are concatenated, which led us to believe at first that these ads had both English and Spanish. In the future, we hope to look further into this subset of ads.
Figure 3: Example of an ad run in two different languages using Facebook’s dynamic advertising feature.
A topic that appears repeatedly throughout the model is the concept of voting by mail. In the topics, this is evidenced through the words “correo” (“mail” in Spanish) and “mailbox.” Additionally, pages with the highest spending had names such as “Vote Early 2020” and “Vote By Mail 2020,” highlighting how much organizations valued voter mobilization.
Table 2: Total spend on Spanish ads by page name among non-candidate groups.
Click here to view an interactive version of this table.
Where are they targeting?
The last question we wanted to answer was where these groups were targeting with their ads. In order to do this, we used the region distribution variable in Facebook’s dataset. This variable maps each state to the percentage of the ad’s impressions that occurred in that region, which gives a general idea as to where the organization was targeting the ad. We hypothesized that most ad campaigns would be targeted at swing states and those with high populations of Hispanic or Latinx voters.
Figure 4: Targeted spend on Spanish ads in U.S. States, Puerto Rico, and Washington, D.C.
Figure 5: Top 10 states in targeted Spanish ad spend by non-candidate groups, colored by general party leaning in presidential elections.
Florida has much more spend than any other state, and its status as both an electoral battleground and a state with a large Hispanic population speaks to the compounding nature of these two factors. However, Pennsylvania and North Carolina are not far behind, despite their relatively small population of Latino voters [1]. The six states with the most targeted ad spend are all swing states, even though all of these states, besides Florida and Arizona, have relatively small Spanish speaking populations. This emphasizes just how close organizations believed the election was in these states, and the amount of work they put in to get every vote possible.
There is also a relatively significant amount of spend in Puerto Rico, which is all from one group: Alianza for Progress, a progressive organization that claims “to unite the Puerto Rican and Hispanic population in the state of Florida” [2]. Even though residents of Puerto Rico cannot vote in the United States presidential election, some of this organization’s ads tell them to ask their relatives on the mainland U.S. to vote. This effort further highlights just how important the Hispanic and Latinx communities in Florida were in the election.
Conclusion & Next Steps
Most of the Spanish language Facebook advertising in the 2020 presidential election by non-candidate groups was on ads about voting by mail. These groups were mostly blue-leaning PACs, with the Priorities USA Action PAC outspending all other groups by a large margin. The ads were targeted mostly at swing states, generally, regardless of the Spanish-speaking population in those states. Florida was targeted significantly more than any other state, likely due to its position as a swing state and its large population of Latino voters.
One limitation of this study was the choice to study the “prominent groups” subset of advertisers (i.e., advertisers that were also active on television and/or were easily identifiable as election spenders by WMP and/or OpenSecrets). In future projects, we might simply exclude candidate and party sponsors from the complete set of advertisers to create a more complete and less subjective set.
This research confirmed a lot of our suspicions about Spanish language political advertising on Facebook—it’s mostly about mobilization and it occurs more frequently in swing states or states with large Spanish-speaking populations. In the future, we hope to look further into the “dynamic” ads that publish both Spanish and English versions. Which advertisers utilize this feature? What other purposes do they use it for?
We are also interested in looking at how spending on Spanish language Facebook ads has changed over time. Hispanic people accounted for nearly half of American population growth this decade [3], so we expect that Spanish language advertising will become even more important than it is now. Perhaps tracking the spending over time can give us a better understanding of the future of political campaigns.