States that Move Together

The website from the Economist has provided a map showing how the states' votes are correlated. This dependency is incorporated into their model, but it is not a part of the Sim 538 model. They state:

"Our model also simulates what would happen if the race moves, or the polls are biased, in similar amounts in like states. We calculate similarity between states by comparing their demographic and political profiles, such as the share of white voters who live there, how religious they are and how urban or rural the state is."

My interest is in the swing states, where a small shift in the polls could really make a difference. The correlation between five swing states is the highest: Pennsylvania, Ohio, Michigan, Wisconsin and Iowa, with correlation coefficients, ranging from 0.79 to 0.88. Please note on the map below, I arbitrarily colored in these 5 states with red, just locate them on the map, and not to show 70 electoral votes to any candidate.

Click the map to create your own at 270toWin.com

The correlation coefficients from Economist are presented below:

I believe the Economist model uses the full correlation matrix, which would be much larger than this one. I will provide the result of some experiments using this small set of values.

It is also strong motivation for the candidates to campaign in these states, as improvements in one state likely translate into better polling numbers in the others.

I do not have more background information on the calculation of the values, other than the single sentence provided by the Economist.

David Lord

October 17, 2020

David Lord

Links:

Economist Webpage (scroll down until you reach "States that move together" heading.

Note since only the upper triangular part of the matrix contains unique values, 50 states require (50 x 50 - 50)/2 = 1225 unique values. For the selected five states, unique values = (5 x 5 -5)/2 = 10.