A cornerstone of our ability to accurately forecast the outcome of the Presidential elections is a simple concept we call “Regional Partisan Lean”.

What it means:

  • Regional Partisan Lean is an estimation of the margin by which a party should win or lose in any given region in a generic election.

  • Think of the partisan lean as a set of internally calibrated expectations that our Polls model uses to assess whether a given candidate is over-performing or under-performing in a given region. 

  • That assessment is crucial to deciding whether or not to award a bounce to the final forecast.

  • Our partisan leans are computed by analyzing all past election results in all constituencies within a given region, and then computing a score that measures how much that region leans towards a party. 

  • For example, the NDC has a lean of +15 in the Northern Region. This means that in any given election the NDC should win the region by about 15 points (or win about 57% of the total votes in the region).

  • As such if we run a poll in the Northern Region and the result shows the NDC winning 60% of the support, then our Polls Model interprets it as the NDC over-performing relative to expectations and therefore assigns a bounce to the forecast. 

  • Conversely, if our poll showed the NDC winning 55% of support in the region, then the Model concludes that the party is under-performing and as such deducts some points from the final forecast.

  • Without the computed leans, our model will naively interpret a 55% support for NDC in the Northern region as a good thing, even though it’s actually not very good news for the party.

%d bloggers like this: