Storm on the Horizon or a Pink Mirage
Publicus Asia’s latest poll suggests BBM Is still ahead by 34 percentage points over VP Leni. Their survey covered 1,500 respondents in March 9-14 period. Meanwhile, our Social Intelligence suggests otherwise. Based on our analysis of Facebook data, VP Leni has taken the lead in Net Positive sentiments, and the difference is in fact widening over BBM.
My heretical post starts by first noting that Social Intelligence and the surveys do have some common grounds. Both methods agree that Isko Moreno, Manny Pacquiao and Senator Panfilo Lacson will not make it. You can find my explanation of what creates a viral effect and who failed at a later blog.
This congruence alone is enough to show both methods have some common grounds, but there are people who persist in championing one method over another. Where both methods differ is in precision and recency. I will explore these issues within four contexts.
First is the nature of the beast. Social intelligence is designed to be hyper-sensitive to sudden changes because social intelligence finds opportunities and threats in real-time. Remember canaries used in the coal mines to detect noxious gasses? This is because the canaries are much more sensitive than humans to carbon dioxide. The same concentration of gas that can kill a canary will only be bad for a human, but we err on the side of caution. By the same reasoning campaigns use Social Intelligence to be the canary in the social media. Social intelligence alerts campaigns to opportunities and threats before it becomes ‘toxic’.
By the time there is a precise and measured understanding, the opportunity usually has passed and the threats may become raging wildfires -hard to put out. Successful campaigns leverage social intelligence to be proactive and respond faster to unfolding events. Social intelligence, with its speed, is akin to google maps for navigation, while surveys are like paper maps in this age of instant media and digital communication.
Second context is timing. Quality opinion research like OCTA’s face-to-face method is very time-intensive; it takes time to plan non leading questions, determine sampling methods, locations, and train the survey takers. And then more time is needed to compile, aggregate and correlate the data and build a rigorous statistical model to balance the results. Its scientific process produces a high-quality representation of human opinion, albeit slowly.
At AutoPolitic, we use QSearch, the leading social data provider in Asia, to conduct our analysis, dashboard, and in-depth research. We can evaluate an issue within a two-hour window after people have been exposed to it. In contrast, typical opinion research takes 2 to 3 weeks. For stable opinion and enduring attitude, OCTA’s survey is the gold standard, its method rigorous and its conclusion impeccable. For campaign managers who plan for next week, they use Social Intelligence.
Third context is the sample size to faithful representation. The two methods differ in the amount of data and the assumption how the quantity represents reality. There are 67.5M registered voters in 2022. The opinion research industry believes a sample of 1500 to 2500 respondents sufficiently represents these 67.5M voters. Social intelligence doesn’t use statistical adjustments. Our six-month presidential tracking has over 659 million data points to approximate the “Honest Signals” of 67.5 million voters. We rely on the law of large numbers to remove biases and make adequate representation. Survey relies on statistical methods to measure its vulnerability to misrepresentation.
Finally, geographic coverage. Surveys are confined to the country residents/respondents. Sentiment analysis of FB users can be geographically dispersed. It CAN ALSO cover the 10% of Philippine population that are currently working abroad. In a close contest, this 10% of the population is a significant figure that is missed by traditional surveys.
In conclusion, I want to talk about interpretation and projection. Most firms in the Philippines will release a final survey a week before the voting date, tracking attitudes from a week earlier. A lot of survey results ‘wrapped up too early’ as can be seen in US polling. While the practice measures the general attitude well, they will miss any ‘11th hour’ events that may tip voter sentients. Because typical surveys have a time-lag between its sampling and its reporting, those firms that proclaim the winner before voting day have an unexamined assumption that the voter’s attitude is stable and will not change between the proclamation and voting day.
Social Intelligence will keep monitoring sentiments right before voting day. This is simply the limitation between model and reality. In this aspect, Social Intelligence assumes voters’ mind are malleable and their hearts are subject to change. While both surveys and social intelligence have many flaws that could be discussed and improved upon, it is these unexamined off-model issues that will be more critical.
This election is an ongoing experimentation between Survey and Social Intelligence. Which one is more representative? BBM has decided not to attend the debate because he believes he is in the lead based on SWS’s survey. Leni supporters are excited because they believe our Social Intelligence is showing a surge of support. This is the wretched heart of human endeavor; sometimes, people need to see to believe. Most of the time, people realize the truth when it’s too late to do anything about it.
"Storm clouds over Highway 97 near Valentine, Nebraska" by diana_robinson is marked with CC BY-NC-ND 2.0.