Yes, Internet communities can augment and expand thinking. Case in point: I want to thank Jean-Paul Bretons who replied to my blog “2016 Presidential Election: Did Big Data Just Get Lazy?” with the following observation:

“Changing the way the pollsters asked the question might have changed the insights: just rating from 0 to 100 the odds people being polled were to vote for a particular candidate would yield more actionable data.”

I think Jean-Paul is spot on about the importance of using a rating system versus a simple “Yes/No” polling system. A “Yes/No” polling system – while maybe easier to implement – glosses over the insights buried in the granular data such as nuances into the strength of a voter’s preference for a particular candidate and campaign issues. A “Yes/No” poll yields an “absolute” answer as to one’s support for a particular candidate, but cannot distinguish between the nuances in voter sentiment towards candidates and key issues (see Figure 1).

Figure 1: Blah

Instead of a “Yes/No” polling question, campaign managers would have gained more actionable insights if the polls had used a rating scale of 0 to 100 to measure a voter’s “strength of preference” for each candidate. For example, a voter rating a candidate in the 85 to 100 range would have a substantially strong preference for that candidate. However, a voter rating a candidate in the 51 to 60 range would have a weak preference for that candidate, which potentially makes that voter a target for switching.

The same 0 to 100 scale could then be used around campaign issues to measure the “strength of preference” that particular voters have for particular campaign issues (e.g., immigration, jobs, education, crime, gay rights).

A rating yields a “relative” answer; the strength of one opinion or preference relative to the strength of another opinion or preference. “Relative” is a comparative measure that ascertains the strength of an opinion or a preference comparative in proportion to something else.

For example, we might have learned that voters who were saying “Yes” to Candidate A were only in 51-55 rating scale, which would indicate very weak support for that candidate. Or we might have learned that voters saying “No” to Candidate B were in 40-49 rating scale which could indicate a slight dislike for that candidate. Using ratings would have yielded additional insights into the strength of the voter’s candidate preferences and the fragility or vulnerability of their support (see Figure 2).

Figure 2: Rating Yield Comparative Values Showing Risks and Opportunities

These insights are actionable in a number of ways, including:

• Flagging voters (e.g., 40 – 49 range) that one candidate might be able to swing into their camp
• Flagging at-risk voters (e.g., 50 – 59 range) who are at-risk of being persuaded to vote for the other candidate

And if the candidates used ratings to judge the strength of voters’ convictions across a range of topics or issues (e.g., immigration, jobs, education, crime, gay rights), then the candidates would have the information that they need to not only know what voters to target (swing voters and at-risk voters), but also which issues are the most effective ones to emphasize to each audience.

But wait, there’s more!

Keeping a detailed history of voters’ preferences and strength of their convictions, candidates could spot changes in “voter sentiment” around candidates and key issues (see Figure 3).

Figure 3: Tracking Changes in Voter Sentiments

For example in Figure 3, one might be able to track when the candidate’s relative position with a particular issue is getting stronger (moving to point #3 in Figure 3) or getting weaker (moving to point #2 in Figure 3). With these detailed insights, candidates could respond more quickly targeting the right message to the right voters at the right time, and importantly, this would enable pollsters to account for the power of momentum, by issue, rather than just compiling point-in-time snapshots of likely Clinton, or Trump, voters.

Election Learning Ramifications to The Real World

The techniques described in this blog could be leveraged in any industry to better understand the strength of your customers’ preferences, convictions and sentiments towards your product, services and company especially relative to your competitors. Using ratings (instead of polls) provides not only a means to track changes in customer sentiment, but also yields the actionable insights necessary to take more timely and targeted corrective actions.

These are concepts and approaches that we use to help our business users as part of our Big Data Vision Workshop to “Think Like A Data Scientist.” When the business users (or campaign managers in this example) understand the decisions that they are trying to make (What customers/voters to target? What message/issues to apply? What channels to use? What time of day to reach out to the customer/voter? What associations or affiliations to leverage?), then the business users can work with the data scientists to identify those variables and metrics that might be better predictors of customer or voter behaviors.

The post More Election Lessons: Value of Ratings Versus Polls appeared first on InFocus Blog | Dell EMC Services.