Last year I wrote a post about what your beer says about your politics in response to a survey done by a marketing firm. I shelved it, mostly because I had a lot more to say about the accompanying chart but not enough time to write it. This year, they published the same survey, but this time with wine and spirits instead of beer. I published my draft comments on the beer chart yesterday. Today, I submit to you my thoughts on the wine/spirits post. Some of my comments are very similar (I don’t like the “indexed” Republican/Democrat score). Some are new (lack of geographical or income cross tabs). If you combine the two posts, you get very nearly my entire critique.
I suppose, as someone said in the comments section at the Washington Post blog, GovBeat, this chart showing potential relationships between certain liquor brands and partisan voting was probably done in fun. But seeing as how it combines my two favorite things: politics and booze…with a smattering of two of my semi-favorite topics: marketing and visual presentation of statistics, I can’t pass up the opportunity to comment.
The first thing to note is that whether a drink falls on the right or left side of the Y-axis is an “indexed” score, subtracting “Democrat Score” from the “Republican score.”
I think this is just poor analysis because it removes good information from the resulting chart. Just as (18 minus 12) and (6 minus 0) both equal 6, there could be many ways that both Woodbridge and Beringer end up at +6 (slightly Republican) on this scale. It could be the case, for example that many people both Republicans and Democrats drink Woodbridge, but more Republicans than Democrats; or it could be the case that no Democrats voted Woodbridge but lots of moderate Republicans did.
Which leads me to another point. There’s no information on how the Republican or Democratic scores were computed. Was it a simple self-report (“Are you a Republican or Democrat,” “Are you a registered Republican or registered Democrat?”) Or were the participants asked a series of questions that would place them on one side or the other? Or was it a graduated question (“On a scale of -10 to +10, where -10 is totally Democratic and +10 is Totally Republican, how Republican/Democratic are you?”) Without knowing this, a lot of potential information is removed from the chart…or is it least obfuscated by the multitude of paths through which various scores could be achieved.
There are similar faults on the x-variable, voter turnout. How was “voter turnout” computed, a simple “How often do you vote?” Likert scale? Or a more nuanced series of questions about recent elections and whether or not an individual voted in it or not (e.g., “Did you vote in the 2013 elections?” “Did you vote in the 2013 primaries?” Did you vote in the 2012 elections” “Did you vote in the 2012 primaries?”…”Did you vote in the 2010 midterms?” “Did you participate in any special elections including recalls?” etc.)
The bigger problem is that, although in a scientific sense correlation does not imply causation, in a more general sense, if a human sees a pattern, it does assume a causality. The WaPo blogger, Reid Wilson, makes this natural implication explicit when he says,
The results are fascinating: Analyzing voting habits of those who imbibe, Dube found that 14 of the top 15 brands that indicate someone is most likely to vote are wines.
If you see someone at your New Years party tonight drinking Kendall-Jackson or Robert Mondavi wines, that person is highly likely to vote, and they’re likely to vote Republican. Someone who savors a Chateau Ste. Michelle Merlot, one of Washington State’s top producers, or Smoking Loon, they’re likely to cast ballots for Democrats.
Or maybe it was Jennifer Dube of National Research. Wilson says she is the one that analyzed the data, but I cannot find any report. When I click the link that says “Read the full report” I just get the chart.
At any rate, sure “likely” but “how likely?”
Let me muddy the waters. Rich people in blue states tend to vote Democrat. Rich people in red states tend to vote Republican. And rich people tend to drink wine at higher rates than poor people. So if you just saw a picture of someone buying Chateau St. Michelle and you were tasked with predicting their vote, you might also want to know if they live in Oregon or Tennessee.
Of course it could be the case that someone buying Chateau St. Michelle in Tennessee is making an inconoclastic purchase as a means of signaling being separate from their fellow citizens. It might mean that they are the rarest of birds: a rich Tennessean who votes for Democrats.
We just need more data to really determine “What our drinks say about our politics.”
However, here are some things we can determine from this chart, if I understand the methodology at all.
Democrats drink more. There are more bubbles and larger bubbles on the Democratic side…and the range on the Democratic side is bigger (about 85 compared to about 55). I’ll quickly add that this was not true on the beer chart I posted yesterday. The beer chart ran from -10 to +60 and there were more and larger bubbles on the Republican side.
Higher voter turnout corresponds with more drinking…a fact that, if true, I would say can be accounted for by the fact that wealthier people vote more and wealthier people drink more (because they can afford it) and because Mad 20/20 doesn’t show up on the scale. In other words, while a scale that includes Woodbridge and Arbor Mist isn’t what I would call “discerning,” it is still truncated. This analysis is also corraborated by the beer chart from yesterday. The beer chart was normalized so that “low voter turnount” was represented by figures below 100 and “high voter turnout” by figures above. That is also true of this chart; however, while the beer chart ran from 70 to 130 (30 points below 100 and 30 above) this one ranges from 75 to 145 (25 below 100 and 45 above it). If you combine this chart with the beer chart, you get a nice correspondence with two pieces of information we know: that as people become wealthier they (1) drink less beer and more wine and spirits and (2) vote more often. This is just one more indicator that the real story here is that this chart is not measuring a relationship between booze and voting, but disguising a relationship between income and voting on one side and income and booze on the other.
There’s also, generally more diversity on the left. This corresponds with the “bigger tent” currently housed under the Democratic banner. The left currently houses what would be separate Green, Labor, Socialist, Communist, Christian Democrat, and Socialist Democratic voters in a legitimate multi-party system. The modern Republican party currently houses some diversity as well including Libertarian and Nationalist parties, but not so much as the left. I’m not sure why that would show up in drink preference, but it is an interesting coincidence, if nothing else. This does not show up on the beer chart, except that it looks like there is less clustering on the left, but this could be an artifact of how many and which beers were measured.
There might be a story in the brands. Eyeballing the brands, I think we can see the financial story being told in terms of voter turnout. Tanqueray, Clos Du Bois, Bombay Sapphire, Korbel are all above 120 on a scale from 75-145. At the bottom are Don Julio, Jagermeister, and the Smirnoff brand flavored vodkas. To be sure, there are premium brands near the bottom (Belvedere) and mass market brands near the top (Turning Leaf)…so I could be seeing a story that isn’t really there. Again, I’d like to see more data and then break it apart into different graphs.
Conclusion: This chart tells us almost nothing that we could not determine by looking directly at income data directly. I wouldn’t want to predict a persons’ voting behavior based off their drink choice from the info gleaned from this chart. If I had the actual crosstabs, maybe. If I saw, for example, that a significant majority (+80%, maybe?) of my consumers tended to vote one way then I might be more or less prone to making my political positions known. Or, if I saw that a lot of my brands tended to skew one way, I might be more prone to extending my marketing to the other side, maybe with new products that look more like the Democratically skewed products on the list (diversifying my audience might pay off if we believed that Democratic leadership tended to make Democratic voters more confident or more spendthrift consumers and vice versa). But this same effect could probably be better achieved simply by offering a range of budget–>super premium drink selections.