This is the final graph from my presentation at the Lawrence Symposium yesterday. My friend Eileen encouraged me to make a post about it as it’s one she feels quite illustrative of the reason for concern.
During the Q&A after my talk, I sorta kinda said I didn’t think the Governor legitimately won the popular vote in 2014. It occurred to me later that probably wasn’t a very ‘politically astute’ thing to say. I hope the Governor does not personal offense. Such thoughts didn’t occur to me at the time because I was talking about how I read the data shown in this chart. I get engrossed in the data analysis and don’t think about anything else.
This chart shows each of the four statewide races, Governor, Senator, Secretary of State, and Attorney General graphed in a cumulative summation analysis. In addition, the cumulative summation analysis of Wichita registered republican voters is provided to give a comparison.
In the cumulative summation models, as the models accumulates more data (think moving left to right across the graph) we expect it to converge to the average quickly and show decreasing differences randomly distributed around the mean. When we see a pattern of the model increasing or decreasing, it indicates that the sequencing of the summation is not random, but has a relationship with the share of the vote achieved.
In the graph above, the % registered republican voters shows a pattern that is not visually distinguishable from random. The four races all show a clearly non-random pattern with striking similarities. All show the pattern of increasing republican share of the vote with increasing votes cast.
These are not independent datasets. The four races all show a similar pattern because they are highly correlated. The pattern itself is the troubling thing.