Informational Cascades in U.S. Politics and Elections

Information cascades, or herding, is the phenomenon of people sequentially making decisions based on previous people’s decisions. We see this commonly in everyday life: restaurants that are busier garner more customers because people observe that previous people decide to eat at that restaurant, social media users will follow a popular influencer because they observe previous social media users enjoy the content of this influencer. It is important to note that these informational cascades form from what people observe of previous people’s behavior, not previous people’s decisions that compelled them to take such action. For example, a person choosing the busier Restaurant A over Restaurant B makes their decision because they observe more previous customers chose Restaurant A (behavior), not because they know previous customers personally enjoy the food (their private signals) at Restaurant A more than Restaurant B. 

The result is informational cascades and herding materialize from people’s inferences of what previous people know (behaviors observed) that led them to make such a decision. Specifically, it is based on the first few people who start the cascade. An informational cascade starts (and never ends) when enough consecutive people choose the same option. Generally, assuming rational behavior, this number is small (a few or several people, depending on the precise scenario). Informational cascades more often than not lead to folks choosing the universally popular decision because from a Bayesian perspective, if enough earlier people consecutively choose the same signal to give a significant prior probability to a specific decision then everyone after will also choose this same decision, regardless of the signals they are given. And the correct (or actual universally popular) decision is not always made–there is a possibility that an information cascade is triggered by the first bunch of people receiving the unlikelier, universally incorrect signal and acting upon it, thus triggering the proceeding people to choose the incorrect behavior in an informational cascade regardless of those later people’s private signals. This is a case where informational cascades cause people to choose the incorrect, or skewed, behavior or option.

We see information cascading in elections and government, namely sequential voting in political elections and congressional votes on legislation. In this research paper, S. Nageeb M. Ali and Navin Kartik show through mathematical analysis that sequential voting systems cause information cascades that impact how later people vote in sequential voting system. Sequential voting is the process where an electoral contest is taken over a period of time where preliminary results are observed by voters before everyone votes. We see this often in American politics: presidential primaries hold contests state-by-state over a period of several weeks, Congress holds roll-call votes on pieces of legislation, early voting results are released to the public before Election Day. This practice of showing results of a population of earlier voters, according to his paper, actually alters the final election results by impacting how later voters vote who observe early results.

Specifically, they show that voters are more likely to support a candidate if they observe previous decisions that that candidate is performing well (getting more votes) because it shows that that candidate is perceived by others as better. Thus proceeding voters will consider the fact that such a candidate is perceived as better by others, and choose to vote for (or increase their likelihood of voting for) the leading candidate from earlier votes despite their private opinions (signals) on which candidate is the best. This is an informational cascade within voting in political contests. Ali and Kartik define this phenomenon as momentum—the idea that a candidate that performs well earlier will have “momentum” behind their campaign and will perform increasingly well as later voters vote because of these information cascades.

We can see real life examples—beyond mathematical proofs—of informational cascades affecting political results in sequential voting in the 2020 Democratic Presidential Primaries. According to this analysis in Fortune, 2020 Democratic presidential candidate Joe Biden won the Democratic Primary after having low expectations because of “voters making their choice in the few days before the elections, in what turned out to be a reflection of the momentum coming out of South Carolina. Across eight of the states with presidential primaries that day, 37% of voters said they made up their minds in the last few days. About half of them went to Biden.”

The momentum Biden received after his win in the South Carolina primary election, according to this analysis, is what led him to his subsequent victories in proceeding contests and his eventual win in the primary and general elections. The article specifically cites “late deciders” (voters who were still considering candidates within a few days of their state’s election) as voters impacted by the South Carolina results and crucial to Biden’s success. This is a clear example of how the sequential voting system of the 2020 Democratic Primaries and informational cascades that result from them impacted the 2020 presidential election in a very real way. Joe Biden’s consequential victory in a pivotal early contest (and its resulting observed decision to later voters that people behaved to choose Joe Biden as the best candidate) was enough to influence millions of later voters to switch their vote for Biden enough for Biden to win proceeding contests despite Biden’s poor polling going into those contests before his victorious South Carolina primary (or, the original private signals of those later voters pointing against Biden). In technical terms, the private signals from previous voters started an informational cascade that caused later voters to switch their vote for the candidate with more observed positive behavior decisions despite those later voter’s (contradictory) initial private signals on who’s the best candidate.

Whether or not sequential voting is a good system is subjective. Some may argue that letting later, more undecided voters observe behaviors of other earlier voters may allow them to make a more informed assessment of the candidates, thus is a net positive. Some may argue on the contrary that sequential voting creates these informational cascades and social pressure that push later voters to vote for a candidate that they don’t actually prefer, but instead the candidate that won the most early voters. This could create a weaker democracy where unequal power or influence lies with the earlier voters who impact how later voters vote in favor of their choice instead of a democracy where each person has equal power and votes for their personal preferred candidate unaffected by other voters’ behaviors. Nonetheless, there is clear evidence of informational cascades and their very real effects on our elections that utilize sequential voting—it may have indeed been the deciding factor of the 2020 presidential election.