Cortico’s mission is to help foster a healthier public sphere. We’ve been developing technologies that give journalists new access to citizen voices in the cities and towns they cover. The goal: local journalism that better reflects the concerns, issues and stories of those communities.
One type of data we’re capturing is talk radio shows from across the country. We pull the audio from radio stations throughout the day and night, run automatic speech recognition and diarization, and save the shows as easily searchable transcripts.
Recently, with the midterm elections approaching, we began tuning our speech recognition software to ensure it picked up the names of all candidates running in every congressional election. As of Election Day, we had collected a month’s worth of talk radio conversations mentioning midterm candidates, from across the set of 162 stations that we currently track in 36 states plus the District of Columbia. This includes talk radio shows from across the ideological spectrum.
We counted the number of mentions of each candidate in the month-long corpus. We then identified the election-issue phrases (such as “gun safety” and “abortion”) most correlated with discussion of each candidate.
The list below shows the 25 most mentioned candidates. The far right-hand column shows the issue-related phrases most correlated with the mention of each candidate. For example, for the candidate mentioned more than any other during this period, Sen. Elizabeth Warren of Massachusetts, the top issue phrases in order of frequency were: 1) “immigration,” 2) “affirmative action”, 3) “open borders”.
“Immigration” was the most frequently mentioned issue phrase. It was among the top three issues for 76% of the most-mentioned candidates. If you include other phrases related to immigration such as “sanctuary cities” and “amnesty,” this rises to 92%.
In short, in American talk-radio discussions of the midterms over the last month, immigration loomed larger than any other topic.
Work at Cortico
If you like working on exciting, creative projects, enjoy things like machine learning, natural language processing and data visualization, and want to help make the public sphere a bit healthier, come work with us at Cortico! We’re hiring several positions and would love to hear from you.
Thanks for reading! If you have any questions or comments, feel free to reach out to us here or on twitter @cortico.
Other notable findings
- Hottest races by total number of mentions of candidates:
TX Senate (7,115 mentions)
MA Senate (7,004)
FL Senate (4,564)
VT Senate (3,917)
CA House District 43 (3,869)
- Incumbents got 63% of the mentions while challengers got 37%
- Democrats got 57% of the mentions, Republicans got 41% of the mentions, and third party candidates got under 3%
Here are some important caveats about the data:
- These findings are strongly influenced by the nature of the programming on the stations we cover. Right now we cover 36 states plus the District of Columbia. While we have tried to cover a demographically and geographically balanced set of stations, the candidates from the 13 states we don’t cover are less likely to make the cut for the table above.
- Complicating our analysis, certain candidates have the same names as well-known celebrities — for example there was an Elvis Presley running for congress in Arkansas — and there were a few candidates with the same names as radio hosts. Well-known candidates such as Bernie Sanders may be mentioned for reasons other than their candidacy in the midterm elections.
- Syndicated talk radio content (including commercials) that aired on many different stations was counted just once, to the best of our ability to recognize it as such.
- The set of social issue phrases that we considered were not chosen manually, but were discovered algorithmically using word embedding models trained on the radio transcripts.
- We derived our insights by tracking talk radio stations from AK, AL, AZ, CA, CO, CT, DC, FL, GA, IA, ID, MA, MN, MO, MS, MT, ND, NE, NH, NM, NV, NY, OH, OK, PA, RI, SC, SD, TN, TX, UT, VA, WA, WI, and WV.