Radio broadcasts and the 1960s race riots in the U.S. South

The role of social media in recent periods of unrest comes as no surprise to us. From the Arab Spring uprisings in 2011, to the U.S. Capitol riot in January this year, to anti-lockdown protests in Melbourne, social media have helped turn a small spark of resistance into a large-scale anti-establishment event. It is tempting to think that the role of media (as exemplified by social media in modern times) in these sorts of incidents is a recent phenomenon. However, that seems unlikely, and this 2020 working paper by Andrea Bernini (University of Oxford) provides one example. Bernini studies the effect of black-appeal radio programming on the 1960s race riots in the U.S. South. For those of you who aren't intimately familiar with those riots (which included me):

In the eight-year period between 1964 and 1971, 752 riots occurred. These totaled to 1,802 days of civil unrest, leading to 228 deaths, 12,741 injuries, 69,099 arrests, as well as 15,835 episodes of arson and other destructive events.

Bernini collected data at the county level on the occurrence and severity of riots (where severity was and index calculated as the average of the number of days of rioting, number of deaths, injuries, arrests, and episodes of arson occurring in each county, as a proportion of the total across all counties and all years). He also collected data on the coverage of black-appeal radio stations (defined as those "airing at least 12 hours of black-appeal programming a week"). The radio coverage was calculated based on a model of electromagnetic signal propagation that incorporates the terrain, using satellite data from NASA - each county was then classified based on the share of the county that was coverage by black-appeal radio broadcasts, or a dummy variable that classified each county into whether or not more than 50 percent of the county was covered.

Looking at the relationship between radio coverage and riots, Bernini found that:

In the preferred specification, a marginal increase in the share of the county receiving the signal from a black-appeal radio station is estimated to lead to a 7% and 15% rise in the mean levels of the likelihood and intensity of riots, respectively.

Here's where the paper goes a little off the rails though, as the interpretations are not quite intuitive. That result isn't the effect of all black-appeal radio stations, but is the effect of black-appeal radio stations that have 100 percent black-appeal programming. The effect for radio stations with less black-appeal programming is smaller (which is, in itself, an important finding). Also, I believe that in saying "a marginal increase", Bernini is actually referring to a one-unit increase in the radio coverage variable. Since that variable ranges from zero (no coverage) to one (complete coverage), then the correct interpretation of these results is that the difference between a county with no black-appear radio coverage and one with complete coverage, holding other variables constant, is a 7 percent increase in the likelihood of a riot, and a 15 percent increase in riot intensity. Those effects are large, but not necessarily as large as the phrase "a marginal increase" in coverage would imply.

Taking things a bit further, and comparing counties covered by radio stations in the top tertile (the top one-third) of the distribution in terms of the proportion of programming devoted to "news, interviews, religion, and public service" with counties in the middle and bottom tertiles, Bernini finds that:

...the content of the programming does have an important effect on the likelihood of a city experiencing a riot. In fact, it is only in the subset of counties receiving the signal from a 100% black-appeal radio station with a high level of political content programming, that a riot does emerge. The coefficient on the other 100% black-appeal radio stations, albeit positive, does not reach statistical significance at conventional levels. On the other hand... there is not a meaningful difference on the severity of riots between the two categories [low and high political content] of radio stations.

It is interesting that 'more political' radio stations increased the likelihood of a riot occurring, but not the severity of the riot. The radio stations provided the spark that ignited the riot, but after that, it was up to the rioters (and the authorities) what happened.

Bernini then went on to look at variations over time. All of the above analyses were based on the distribution of radio stations in 1964 (and not a year-by-year change). He also has data on the 1968 distribution of radio stations. However, from the paper it is difficult to understand exactly how he treated the panel data, since he had only two annual observations of radio coverage, but eight years of data.

Bernini presents an array of different econometric specifications in the paper. The instrumental variables (IV) analysis relies on a number of instruments. However, one of the instruments is the enrolment in Historically Black Colleges and Universities (HBCUs), measured in 1976. This is likely a good example of what econometricians refer to as a 'bad control', because HBCUs enrolment in the 1970s might have been affected the extent of riots in the 1960s. So, I don't find the IV analyses to be particularly convincing (although perhaps dropping the HBCU variable as an instrument would have little effect - maybe journal reviewers will test this point as this paper moves towards peer-reviewed publication).

Despite those gripes, this is a really interesting paper. It surprised me that the research literature hadn't previously established the importance of this relationship quantitatively. Bernini concludes that:

Black-appeal radio stations significantly changed the fabric of southern counties, through their sizable and robust impact on the likelihood and on the severity of the 1960s race riots.

The role of media (and now, social media) in influencing social movements is important, then as now.

[HT: Marginal Revolution, last year]