leftovereggs

This plot is so frustrating. 1. Using a line plot here is a bad choice. It implies there is some continuity or progression across the X axis. The states are sorted in descending order creating the appearance of some kind of progression. 2. Of the 6 lines on each panel only 2 are necessary and they should be replaced with a single bar. There are 2 sets of 3 lines showing predicted, actual, and previous election vote share for Republicans and Democrats (R share is R/(D+R)). So, the Democrat share is just the mirror image of the Republican share. One set of 3 lines is totally redundant. You don't actually need a line to show prior election share and the 0.85 * prior election share as the latter is a linear transform of the former. Using the (bad) "line plot logic" of the original plot, you only need the yellow and dashed red line showing the predicted value and the actual. Using better logic, you really only a need a bar representing the difference between the predicted and actual by state. 3. The thing the plot's creator thinks is suspicious is how close the predicted line is to the actual line. That's the whole point of the plot. And yet, this conclusion is magnified by the choice to have a Y-axis that scales from 20 to 80 percentage points. When the important thing you want to show is a few percentage points, it's frustrating to have those differences of a few points illustrated on a cramped and crowded panel, not very tall to begin with, that spans 60 percentage points. 4. The plot gives you no reason to think this is even suspicious. Again: The creator wants you to look at this plot and think "Election fraud" but there is no reason to reach that conclusion from this chart. You'd have to show that the differences from the "15% reduced vote share" are unusual/suspicious or more likely with voter fraud, and there is no reason to think that.