As promised, we are continuing the look at the 2020 election.  This newsletter digs into the election polling and a quirk I noticed when looking at state polling and national polling. Thank you for subscribing, and if you enjoy reading this, please forward the newsletter to your friends. ~ Kevin Something I have noticed, but have not seen discussed elsewhere, is a trend that I was curious about during the primary season. This trend may be an indication that national polling and state polling have some differing results which are interesting. First, this is a post that is not a diatribe about polls.  Some firms do a better or a worse job per generally accepted methods. However, there is a nuance of the art of polling that many have debates over, but they still have difficulty in substantiating.  Let’s put all that drama in a box and revisit it later. I must make a brief tangent into statistics to put us all in the same headspace.  Let us focus on two concepts in statistics: average and variance.  The calculation of average is commonly used in our conversation in a multitude of areas.  We know the average height of a high school basketball team, we know the average grades of a school class, we know the average speed on the road, and we know if someone has an average fashion sense.  The average is a reference point, but it does not tell us everything.  This is where variance can be useful because it tells us how much a measure deviates from the average.  To begin let's suppose the average height of a male high school basketball player is six feet three inches. If we had a team with players ranging in height from six feet two inches to six feet four inches, then our average would be near six feet three inches.  However, we could also have a team with a player six feet nine inches tall and another player who is five feet nine inches tall.  An analyst would show that, on average, both teams are six feet three inches, but anyone who saw the teams play each other would notice the distinct differences in the teams and whatever advantages and disadvantages that would entail.  The point is that the average measure would be unhelpful in detecting this difference.  It is not incorrect, but another measure would help more.  We would see that the team where the player heights are closer to the average have lower variance, while the team where the heights were farther from the average have greater variance.  This is a quantitative statement and not qualitative.  If we were to apply this way of thinking to the returns in a retirement portfolio, a fund that holds U.S. Treasury bonds generally has lower variance than a fund that holds stocks.

Something I have been curious about when looking at election polling this year is that the variance in polling for the presidential race has been higher in national polls than in state-level polls.  Hypothetically, I can think of better explanations for the opposite being the case (that national polls have lower variance than state polls). More analysis is warranted in this puzzle, but these are my current findings as culminated from the Real Clear Politics collection of polls.  I focus on the variance of Trump's numbers first under the assumption that an incumbent makes a general impression on voters.  I draw out the national data and key swing states WI, AZ, NC, PA, and FL.  These states have been followed for months.  Also, I am including polls taken after June 5, once Biden had secured the delegates to claim the Democrat nomination to polls analyzed on September 6th. 

Inferences from Variance: If I were the leading candidate and I could not be in a blowout, I would want my average to be at or above 50% with low variance.  I would read the variance as how voters move around slightly in different conditions.  If they stick with you, that is a good place to be. If I were the trailing candidate, I would be disappointed with the lower average vote share. However, I might take solace in the fact that there is more variation for my support, because there might be approaches that could help stoke that support.  However, this variance is not consistent across the board.  Trump has much greater variance in Florida than Pennsylvania.  Among the states in this group, the state with the lowest variation for both candidates is North Carolina.  This may be an indicator of fewer swing voters in the Tar Heel State. Returning to my earlier puzzle on possible explanations behind the pattern, it may just be the result of the state's swing status and the low variance.  However, I could expect low variance in non-swing states whether they be Kentucky or New York.  It may be that a few states have movement in their polls that contribute to the national picture that is not seen in these individual states.  However, these states encompass a pretty broad demographic share of the country. To me, this puzzle is even more difficult to grasp because the state polls are generally not held in as high esteem as national polls.  The state polls also have a lower sample size which, by method alone, would lead us to expect greater variance.  It is also possible the results may be a coincidence or a matter of selecting a few states.  However, regardless of the comparison, it is important to note that there is less variation in preferences for these candidates.  One example of a higher relative variance would be Trump’s numbers in Florida.  This may be due to the disproportionate senior population in Florida and reactions to COVID-19.  Census data estimates that 20.5% of the population are seniors in Florida, while seniors account for 16.3% of the population in North Carolina.  Seniors represent 18.2% of the population in Pennsylvania, 17% in Wisconsin, and 17.5% in Arizona. As I have mentioned before, I am very interested in the sentiments of high propensity voting seniors in these swing states.

https://www.census.gov/library/visualizations/2020/comm/map-popest-65-and-older.html Before analyzing election dynamics in the coming months I thought it would be helpful to explore this puzzle and consider how variance can reveal trends we may not recognize when looking at polling averages.  Like all other metrics, it is a mistake to over-focus on one measure. But we get a fuller picture by considering these factors along with other observations.

News:

I am enjoying the chance to share these newsletters with you in the form of the new podcasts and appreciate your continued feedback. You can reply to this email or leave your comments below.  I sincerely enjoy chatting and learning what folks think. Thank you ~ Kevin

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit modlinglobal.substack.com