Could Frrole’s algorithms have predicted Indian elections results?

Back in May, we worked with a couple of the top English news channels in India to provide insights and intelligence about Indian elections as mined from social media data. One of the challenges that we took up was to show if elections results could be predicted by applying AI to 100s of millions of social media conversations.

For the first time, we are publicly reproducing the analysis done by Frrole BA team. We will let you judge how much we got it right (now that we all know the results of the elections)

Our work shows that if accurate algorithms and AI can be leveraged with expertise, they can become powerful methods for predicting consumer trends, behavior and possible results of major campaigns - for public organizations or consumer brands alike.

Below is the analysis done on May 20, 2019, reproduced almost verbatim :

Observations & Recommendations

National Level

• When analyzed at a national level on both the party and leader fronts, PM Narendra Modi and BJP respectively seem to hold the edge.
• At a leader level, Rahul Gandhi seems to have gotten the nations’s attention and has put up a strong show against Narendra Modi in terms of SOV garnered. However, he still does not seem to have the nation’s confidence, with a large difference in the final scores for both the leaders.
• At the party level, Congress packs a surprise, outgunning BJP in terms of SOV, leading 55:45 in the month and half preceding the elections. During elections, it actually manages to increase its lead slightly to a more comfortable 59:41. However, overall, it still scores lower than BJP on account of more negative sentiment and weak trends.
• This is in stark contrast to what was observed during Rajasthan assembly elections analyzed earlier where both the Congress and its leader outscored BJP and its leader on account of stronger positive sentiment even when being in a weaker position in terms of SOV.
• While Frrole is not predicting the number of seats that each party is likely to win, it can be said that Congress is likely to put up a stronger show at the national level than what might be anticipated. However, it is still unlikely to be enough to carry the day.

State Level (UP only)

• When analyzed at a UP state level on both the party and the leader fronts, PM Narendra Modi and BJP respectively seem to hold the edge. While the edge clearly seems to be stronger (relatively speaking) compared to the edge it holds at a national level, it is hard to predict the overall seat-wise performance of either party as data for the MahaGathBandan (MGB) has not been taken up for analysis.
• PM Modi scores even higher when analysis is limited to UP state, compared to the scores at a national level (15.88% instead of 10.64%). Rahul Gandhi scores slightly better too, at -5.63% instead of -7.43%
• At a party level, BJP scores much higher as well, scoring 2.81% instead of -6.73% that it scored at the national level. Congress does only marginally better, scoring -7.85% instead of -10.58% that it scores at the national level.
• BJP clearly is in a much better position vs. Congress in UP, compared to its position vs. Congress at a national level. However, its final seat tally would be defined by how well the MGB performs

Overall

• PM Modi is clearly the BJP’s winning card, both at the state and national levels. The sentiment for him in UP stays net positive all through the period, a feat that is quite hard to achieve during a strongly polarized elections season. Even at the national level, he consistently scores very close to 0%, which is really good given the context . Comparatively, sentiment for Rahul Gandhi is around -25 to -32% all through, sentiment for Congress party is -28 to -53% all through and sentiment for BJP is -18 to -33% all through.

Methodology

• We have analyzed data at two levels party, and leader. The leader data should also be seen as a proxy for the party.
• We have used a composite predictive score that takes a set of factors into account, mainly including: 1. SOV preceding elections, 2. Sentiment preceding elections 3. Delta change in SOV from the preceding election period to the elections period 4. Delta change in Sentiment from the preceding election period to the elections period. The other factors can be shared if needed.
• The score for each component that makes up the prediction score can have a maximum value of +100% and a minimum value of -100%.
• We believe that ‘delta change’ is a crucial factor as it would be hard to game that in a manner that it is selective. It also accounts quite well for historical activity levels of each group.

- Amarpreet Kalkat, CEO

If you would like to learn more how we can help you predict the results for your campaigns, brand activations or other organizational initiatives, you can reach out to us at marketing@frrole.com. If you like to go through a similar analysis on the market side, read how we have predicted the sales volumes for passenger cars in India by leveraging only social data.