The projections HowWillAmericaVote will publish later today, which we derive purely from polling, will not be predictive of today's outcome because the polling is inaccurate. This is fine because our objective is to analyze polling data, not predict election outcomes. Election outcomes do however play an important role in validating the principal that polling reflects public opinion. An election’s outcome serves as a single basis point by which a pollster’s result can be assessed. At this moment in time, our analysis shows that current polling (in the Democratic Primary) is not reflective of public opinion and we expect this trend to continue, most notably in Ohio.
Ohio and Michigan are both demographically and geographically similar; based on this fact, it would logically follow that their voting patterns would generally be similar. Let's look at the pattern Michigan has established. Below are our Michigan polling projections with an additional column illustrating the calculated margin between each candidate:
According to the polls Clinton was expected to win, and win decisively. Obviously we know that didn't happen as per the results below:
There is a fair amount of data in the tables above and it's probably data you've already seen. The important concept being illustrated is the degree to which each demographic was inaccurate and the fact that Bernie actually won among whites and males.
If we look at Ohio's projections, they are very similar to those from Michigan. The sample correlation coefficient (r) between the two margin columns is .7933; the possible values range from -1 to 1 with a value of 1 implying a perfect relationship and a value of -1 an inverse relationship. The demographic polling projections between Ohio and Michigan are related.
In Ohio, Bernie again loses each demographic by a considerable margin based on polling, but the data is generally more optimistic by comparison.
Based on polling data from Ohio, Bernie should lose white voters by 13% and males by less than a percent. The simple question to ask is why would Bernie suddenly lose the two demographics he has most consistently won? What has changed to warrant this conclusion? Unfortunately nothing has changed as pollsters are again making the same failed conclusions they made in Michigan.
For some additional context on Bernie's support among males and whites, the table below was compiled with the demographic margins for each state in which an entrance or exit poll was conducted; as with the other tables in this article, positive numbers denote a Bernie lead:
The table above depicts a fairly strong geographic association between Bernie's support among whites and males; Bernie has not lost males or whites in a northern state. This is an observation, a trailing indicator based on past performance, but its another layer of demographic evidence to suggest that the polling in Ohio does not reflect public opinion. We feel confident in this pattern of polling inaccuracy as nothing has fundamentally changed after Michigan. The concept or practice of polling has not been invalidated, but the models on which recent polling has been predicated have proven to be consistently wrong.
There will undoubtedly be great shock later this evening when polling fails yet again and hopefully this time those in positions of influence will learn from their past mistakes. The central lesson to be learned is the importance of demographic weighting and its affect on topline results. There are three outcomes for a topline result; (a) they can be accurate with accurate sub-samples, (b) they can appear accurate with inaccurate sub-samples, or (c) they can be inaccurate which is always due to inaccurate sub-samples. There should be no distinction between (b) and (c); polls are either accurate, or inaccurate, which we uphold as part of our ranking process.