Johannesburg, 31 Oct 2016
The recent local government elections in South Africa brought about change on a scale that has not been seen before, with the ANC losing several metros to opposition coalitions, and their overall percentage of the vote dropping from 63.6 to 55.6%. The question is: could political parties have improved their percentage if they had been better equipped to identify 'swing' voters and to target, at ward level, specific issues that were of concern to voters?
Aneshan Ramaloo, Senior Business Solutions Manager for Advanced Analytics at SAS Institute, suggests that 2016 may have been slightly too early for the various parties to make effective use of analytics for such a purpose, but he adds that Predictive Analytics will certainly play an important part in elections in the years to come, including the 2019 national elections and various local by-elections that are currently under way.
As an example, Ramaloo points to how the Obama 2012 campaign had already begun using a modelling technique called "net lift" or "incremental modelling" to determine swing voters. Swing voters, explains Ramaloo, can be described as citizens who could be positively persuaded to vote for a party by some kind of political campaign, such as TV advertisements, phone calls, personal visits, and direct marketing flyers.
Because Obama's campaign was able to adopt a more targeted approach, volunteers were able to focus closely on voters that had a higher probability of being persuaded to support their candidate. This was an improvement on wasting time and effort on a larger group of people, the majority of whom would already have decided on their choice of candidate.
"The concept of 'voter churn' is not really any different from the customer churn that many private sector organisations face in highly competitive markets. SAS has proven itself capable of helping organisations decrease attrition, by accurately predicting the customers most likely to leave and developing the right proactive campaigns to retain them. It seems logical that a similar approach to potential 'swing' voters may well have an impact on elections in the not too distant future," he says.
"In the case of South Africa, using data analytics to target swing voters in elections remains theoretical at this stage. Nonetheless, the technology certainly exists to help political parties increase their voter base. In addition to determining swing voters, analytics can be also used to predict the winner of the elections using sentiment analysis."
Sentiment analysis, explains Ramaloo, enables organisations to quantify insights into the perceptions that customers have of their products, service offerings and brands. Using statistical techniques and linguistic rules, it can reveal valuable information on consumer opinions, trends and potential problems. This can help organisations identify and correct problematic issues before escalating further.
With SAS sentiment analysis, organisations are better equipped to understand their customers' opinions and take action on this knowledge, and this concept can be applied to political parties as well. They could quite easily utilise the same principles to improve service delivery."
Sentiment analysis works by collecting various textual data inputs from Web sites, social media outlets and file systems, and then putting them in a unified format to assess relevance to predefined topics.
"In the context of South African elections, parties could look to use sentiment analysis to address specific and relevant problems with customised messaging to the voting base, instead of generic ones. A customised message is more likely to strike an emotional chord with citizens than a non-specific one," says Ramaloo.
Furthermore, he adds, if analytics can demonstrate that a significant population in a certain ward where Political Party Y currently holds the seat are unhappy about specific issues - such as electricity provision, service delivery or housing - a rival political party would then be able to use this information to target its campaigns more aggressively. It could do this by explicitly focusing marketing materials such as posters and flyers distributed in a ward on the issues that are of most concern to the citizens living in that ward. Such an approach would inevitably garner them additional votes.
He does add the caveat that in the South African context, the population that uses social media is not currently representative of the entire voting population, and this might diminish the usability of the sentiment analysis.
"There is most likely bias in analysing social media data at present, which is something that can be very difficult to correct for. However, as broadband continues to become more accessible and cheaper, together with an increasing adoption of smartphones and the adoption of social media, it is likely that this will be less of an issue in future elections," concludes Ramaloo.
Share