This Paper Will Die in Darkness
I think I have a great idea. I'm the only one...
Hi,
I think maybe an introduction is in order before the angry rants begin. My name is Arjun Chaudhuri. I’m a student of political science at the LSE. My work is largely comparative, South Asia focussed and I specialise in examining system-fragilities. I believe it’s also fair to ask why you, dear reader, ought to care about what I intend to write. Well firstly, because system-fragilities means nothing. Not yet anyways. I’m deluded enough to try and will this subfield into existence, and for my sanity’s (and your entertainment’s) sake, I’d like a written record of my attempts. Secondly, I’m often writing proposals that pass into academic purgatory. Perhaps better minds (or deeper pockets?) find these puzzles to be of some interest. Ever so often I’m able to conduct the work and produce a result that I believe is worth noting. I think that some of these might invite debate, discussion, or are worth a raised eyebrow. Lastly, life as a young academic is frustrating and can often feel like yelling into a void. For those like me, I hope that my struggles and many failures can be a source of comfort. Therefore, welcome to the void and today’s post: my grand attempt at divining Indian elections.
Indian elections are famously impossible to predict. I think it’s because the pollsters are trusting the wrong source of data: people. Collecting samples that are representative of the country is incredibly hard, trusting they’ll give you a truthful reply is even harder and factoring in whether or not they’ll change their mind is the hardest. I propose an alternate idea. Instead of trusting their words, look at their actions.
Actually, look at the words they’re choosing to trust. Media consumption is at an all time high, helped along the way by increased accessibility to the internet through cheap data rates. Media production, is similarly high, with a myriad of newspapers, “news” channels, and independent commentators working on YouTube, Facebook and Instagram. This creates a competitive market for information, where people demand news on the topics that are most on their mind and the myriad of agents respond to this signal by producing content on these exact subjects. By effectively reading this market in a pre-election period, we can identify the issues playing on the minds of voters. Two assumptions are necessary here:
Media producers are responding to the market and not acting independently to create and set an agenda. That is to say, the media has abdicated its role as a thought leader and has given up trying to influence what people are thinking about. I think this is largely true, especially because of the level of stratification in the current ecosystem. Sustainable economic success, beyond the handful of household names, is only possible by responding to what the people want to hear. And I think this can be measured and proven.
Issues voters care about influence their voting intentions. This is a slightly trickier assumption. Each party, in any democracy, has a base vote share. That is, the percentage of votes it is guaranteed to generate from the party faithful. These voters already have their news producers of choice tied down, and prop them up as well. What is of interest is the floating voter, those who vote depending on the issues they deem important. My model is based off of a political science theory called the Saliency Theory. The idea is that floating voters don’t choose parties based on their positions on certain issues. Rather, these voters associate certain issues as a whole with parties. Depending on the issue on their mind, they vote with the associated party. A clear example of this is immigration in “the West.” The (far) right has been banging on about immigration since the first periods of mass movement in the post-war period. Consequently the right is associated with immigration, and as immigration has become salient, they have reaped strong electoral benefits, regardless of what the left does to capitalise on the same issues (looking at you Keir).
My model seeks to use the media, and their relative success as a result of covering certain issues, as an indicator of saliency. I would track the issues as being associated with certain parties, in order to determine how the election might swing. Right about now, you’re probably asking, “what drives the saliency of certain issues if not the media?” Great question, and if I ever manage to conduct the research, addressing this potential reverse causality will be my first priority. If you’re interested in reading the methodology in full, I’ve linked it as a PDF at the end of this post.
I reckon that’s the end of this post. The next one will be an account on the many issues I ran into while trying to get some attention for such a project, both from within the LSE, as well as from external agents. I hope you’ll be back for that. I’m going to get back to the essay I’ve been procrastinating on!
Aside from a few minor technical oversights, the methodology is quite compelling. It reads like a strong candidate for a cascading machine learning prediction approach, first modeling the frequency of specific topics to estimate their likelihood of media pickup, and then using that output to predict the probability of favorable voter behavior!