One of the fun things about the WPRI/Providence Journal poll is the way the two media organizations represent the poll with interactive graphics. I generally tweet my criticisms when the polling comes out, and this year, except for one minor hiccup, I’m pretty impressed with how WPRI displayed the polling. The key thing is that they chose to have a consistent y-axis value on all their graphs. In this case, because all polls that use percentages are naturally out of 100%, the y-axis goes from 0 to 100%; and I can’t stress this enough, it’s that way on all their graphs. For instance here’s how WPRI represented who respondents thought would win the Democratic primary for governor based on age group (click for higher resolution):
The information is clear and leaves you with the correct context: no age group had a majority in agreement on which candidate was going to win, and there are a relatively high number of undecideds in each category. When you scroll over a bar, it lightens and displays the x-axis item’s label and the exact percentage.
Now here is how The Journal represented that same question based on who the respondent was supporting:
First thing to notice is the shifting y-axis. The intervals range from 10 to 15, and even 7.5 — which is simply bad. The reason why is obvious, the y-axis is shifting in relation to the highest response. While that might be a good thing if you are representing quantities, what’s being represented here are percents, and in this case, if you quickly a clicking through the results here, you are going to either 1) assume 100% throughout the graphs, or 2) assume the top level is whatever it was for the first tab you saw (which was “Total” and was set at 40). That means you are either going to assume a much lower level of support or a much higher level based on whatever your first assumption was.
The problem is really with the subtlety of the change in intervals – since there are always five values represented, you only visually notice the change on ones where the extra decimal point is added.
There are other issues as well. For some reason The Journal chose to assign colors to each candidate, probably to differentiate them. It’s not an uncommon thing to do with bar graphs and works fine if the graphs are moving around, but since in the interactive the x-axis is always the same no matter what crosstab you select (and each graph is in the exact same spot on the screen), it doesn’t really matter here. In two-person races like the treasurer’s race and the secretary of state’s race, it appears they used shades of blue; but they put “not sure” in red which they use for Raimondo here.
Finally, there’s the tabs. Two tabs labelled “Not Sure” and tabs that say “Yes” and “No” without any explanation. While you can infer that the “Not Sure” tabs are in relation to what they follow (party affiliation and candidate support), the “Yes” and “No” tabs seem inexplicable until you click on them and discover it’s an answer to whether the respondent is a member of a union or not.
I hope The Journal will look to improve how it represents data visually, because it’s the paper of record for the state, and it’s important it displays data well. Also, it slapped a 3-D pie chart on its front page!
Looking for bad data displays? I recommend this round-up of 15 terrible bar charts from the UK’s Liberal Democrats, and the subreddit “Data Is Ugly” has a good example every other day.