Epidemics and elections. If the correlation coefficient of these two words could be measured, it would most likely be 0. What could an epidemic possibly have to do with an election? Well, this report by Alexandria Volkening et al from Northwestern University claims that disease modeling--yes, the famous SIR--may provide some insight into the unpredictable phenomenon of voting.
(Link: https://epubs.siam.org/doi/abs/10.1137/19M1306658)
In the report, the primary inspiration is establishing the relationship between the SIR model (read more here) and how voting works. In the figure (a) below, an analogy is drawn between the SI model and how individuals shift political opinions from Democrats, Republicans, or to undecided. Then, the researchers, as in figure (b), modelled how people would shift opinions due to interactions with others. A clear value of this research is that it would allow us to predict election outcomes from pre-election polls that have been taken a considerably long time before. In many occasions, we use only poll data that ended at a time very close to the actual poll, because people's opinions are volatile. If we can model that volatility using this model, it would allow us to extend that time frame of useful data.
However, whether this is an apt analogy is, in my opinion, open to challenge.
Granted, the argument of the paper is very logical and convincing. Consider the following scenarios.
Person A is a young man who exercises regularly and is less likely to get the disease.
Person A is an obstinate republican who hasn't changed his opinion for a considerably long time.
Person A is a "shy Trumper."
Person A is asymptotically infected, so people that come into contact with him are very unlikely to be infected even though he is a patient.
Person A spends a lot of his time talking to others, so he is more likely to be swayed by others' opinions.
Person A has low immunity / meets lots of people, and hence has a greater chance of meeting someone infected.
The second scenarios are no different from the first scenarios in a mathematical point of view, and hence because the second scenarios can be successfully modelled with a SI model, the latter would be so. Clearly a compelling argument.
However, there are three main flaws that I identified in the assumption:
1. A systematic shortcoming of the SI model itself is amplified when applying it to elections.
The SI model cannot account for external factors. In the case of disease modeling, this is not really a great problem, since external factors are mostly possible to model, such as seasons or the distribution of a new vaccine. Even if there are unexpected events, such as mass infection, its effects are geographically limited. On the other hand, politics are entirely comprised of human actions, which are simply near-impossible to model. It is impossible to determine when a sex scandal of a high-ranking democrat politician would be exposed, let alone estimating its effects. In the case of elections, these events affect the entire population due to the news, and people react in unexpected ways.
2. Spreading disease is an involuntary choice; talking about politics is a voluntary choice, and often a rare one.
If one has a disease, unless it is the rare case of an asymptotic infection, one has no choice in spreading the disease. On the other hand, talking about politics and persuading someone else to their political ideology is an action that is quite rare considering the sensitivity of the issue. Also, one could easily feign others claiming that they are democrats such as the example of a Shy Trumper, whereas revealing their real views around people that they know, hence unexpectedly reinforcing beliefs in both the anti-Trump region and the pro-Trump region.
3. Social Media.
SNS feeds us with information that the algorithm would evaluate to be interesting to us, and hence creates a political echo chamber. There is no equivalent to this online reinforcement in disease modeling. Even more, the formation of online herd mentality also easily shapes our beliefs; in the modern era, we would have to model interactions in SNS as well.
Overall, I believe that the report is a very, very creative and novel approach to the variable outcome of elections. However, the phenomenon seems to be too random for disease modeling to be an sufficiently accurate outcome, at least in my opinion.