This is an illustration of narrative analytics applied to 3,902 statements made by the US president in COVID-19 press briefings. After breaking them into clusters based on our deep learning technology, we correlated each cluster with changes in daily presidential approval ratings - both positive and negative: The results revealed 7 clusters with significant correlations to daily polling results, at 99% confidence. We reveal two of them here. Each dot on the below map represents a statement from the President. Hover your cursor over the dots to read the underlying statement for each. For easy reading, some statements are highlighted with stars.
Correlated with higher approval: The 'City/state leaders' narrative
The City/State leaders narrative features statements from the President in which he references productive engagement and cooperation with State and city leaders - particularly governors. Several references are made to constructive work with Democratic as well as Republican governors, and the ability to engage across party lines. Hover the dots within the cluster to read them for yourself.
For each statement within this narrative, the number of people approving of the President increased as below:
​​
Each statement the president made in the "Deserve more credit" cluster correlated with the above changes in daily presidential approval ratings, as noted by the red arrows. As an example, every statement the president makes in this cluster is correlated with an average of 840,000 Independents ceasing to somewhat or strongly approve of the President. Hover over the "deserve more credit cluster above to learn more about the type of statement correlated with these changes.
Correlated with lower approval: The 'deserve more credit' cluster
For each statement within this narrative, the number of people approving of the President decreased as below:
Obvious or Aha? (the value of quantifying both)
People with specific knowledge of pandemic era politics may find the above aligns with their observations. In these cases, narrative analytics adds value in 3 ways:
​
-
Empirical backing in an often subjective space. No matter how well-informed expert judgement may be, it is likely to be viewed as subjective.
-
​Proactive actionability: By identifying narratives drivers known to impact polling, teams can monitor shifts in known narrative drivers to act earlier. (As opposed to the status quo of waiting for polling number to change, then debating why.)
-
Speed of alignment: The approach is not just empirical - it is transparently empirical. The numbers are linked back to every single underlying statement - with a clear line of sight via the maps. That transparency builds confidence and aligns teams faster.
Notes on methodology
This analysis is based on 3,902 statements by the President in COVID-19 press briefings, from Jan 5 2020 until July 30 2020.
​
Sources:
https://www.rev.com/blog/transcript-tag/coronavirus-update-transcripts
https://today.yougov.com/topics/overview/survey-results
​
These statements were parsed into themes through deep learning-based narrative analytics. The categorization is not based on human input - it is purely based on our proprietary algorithm, and is unique to the dataset. Selected narrative themes were correlated with decreased or increased Presidential approval by specific voter groups. The lag time between the statements and the polling results varied, with most in the range of 1-5 days.
​
Note: This analysis is based on limited publically available data from limited sources. Deeper, stronger insights should be expected when the tool is run on larger datasets
Average across all days: 1 statement by the president = 2.3% of daily statements
Total number of US adults: 209.1mm
*CCF = Cross Correlation Function (interpolated)
** Approval defined as somewhat approve or strongly approve
Using 2,000 simulations for each correlation we compute 99% confidence intervals on the CCF values. The correlations above only include those where the 99% confidence intervals are both either below zero or above zero (i.e. those for which we have 99%+ confidence that there is a significant correlation).
Example calculation is below for the “Deserve more credit” cluster:
