By the Numbers: Comparing Walker With Goldwater
The Internet and the personal computer have enabled the lay analyst to utilize advanced statistical techniques. This is something of a blessing and a curse. On the one hand, this has paved the way for more rigorous takes on politics: Horse race calls have become more like PECOTA and less like Jimmy the Greek. On the other hand, many of these techniques have substantial limitations, which would often keep an analyst from pushing a piece as far as he or she might like it to go.
So, for example, Mother Jones printed an article last Friday suggesting that Wisconsin Gov. Scott Walker could be the most conservative Republican nominee since Arizona Sen. Barry Goldwater in 1964. Kevin Drum backed up this claim using Stanford Professor Adam Bonica’s DIME project – a mathematical system that uses fundraising and contribution data to assign ideological “scores,” called CFscores, to political candidates.
Without getting into the details (which would require a linear algebra course), CFscores assume that donors generally give money to political candidates who share their ideology. So if someone donates to Massachusetts Sen. Elizabeth Warren, we might safely guess that person and Warren shared similar political values. Similarly, if the same individual also gave money to Florida Rep. Alan Grayson, we could pretty safely assume that Grayson, Warren and the donor are at the very least somewhat ideologically similar. CFScoring looks at millions of such pairings, and uses this basic assumption to place candidates on an ideological spectrum.
While the CFscore is innovative, powerful, and has predictive ability, it also comes with very real limitations that undercut precise conclusions about ideology, especially when long time periods are involved. To be clear, these don’t demonstrate that the scores are wrong, as such. They just illustrate the uncertainties that flow from the metric’s underlying assumptions, which makes using them as exact measurements of ideology, at least as our stereotypical “man on the street” might talk about ideology, more difficult. There are three relevant assumptions here.
First, CFscores often change over the course of a presidential campaign. For example, Mitt Romney’s CFscore was 0.88 before winning the Republican nomination in 2012, but the Romney-Ryan ticket had a CFscore of 1.18 in the general election. This was a substantial shift. The score of .88 put Romney in the left half of the 2012 Republican presidential field, while a score of 1.18 would have placed him firmly in the right half of the field (and pretty close to Walker, whose score is 1.28).
Dr. Bonica would probably argue that the earlier metric is the more accurate one, but there are counters to this position. At best, there is uncertainty as to whether the score before or after the “post-nomination Etch A Sketch” best describes a candidate’s ideology, and (more importantly for our purposes), how the electorate perceives that ideology.
This uncertainty matters quite a lot. If, after winning the nomination, Walker were to shift leftward by about the same amount that Romney and John McCain shifted to the right after winning their nominations, Walker would actually be to the left of the Romney-Ryan ticket. Drum’s analysis assumes either that Walker won’t move, or that such a move would be illusory, but we really don’t know the answer to this.
Second, CFscores assume that donor ideology matches well with candidate ideology, at least as we commonly use the term. Eyeballing the results, the scores seem to match up reasonably well with what we know about candidate ideology. But do they match up well enough to create a rank ordering of candidates over time? If we are going to label someone “the most conservative candidate since __,” the assumed answer is “pretty much.” We are not certain that is the correct answer.
While donors probably tend to support candidates who generally share their ideology, other factors might affect donor decisions – what issues the candidate focuses on the most, the candidate’s public persona and life history, how much a donor simply “likes” a candidate – and all of these preferences are rolled into this rating.
The problem with this particular application is that Drum’s conclusion flows from an application of the CFscore more as a “pure” measure of ideology. Drum points out, for example, that Walker lacks Cruz’s particular style and does not hint at sharply libertarian policies as Rand Paul sometimes does, but that his CFscore is relatively close to Cruz’s and Paul’s. Drum sees this as a sign that Walker is either more willing or able to conceal his far-right positions from the public than some of his more stridently conservative rivals.
But perceptions like those are, at least some degree, already rolled into the CFscore. A better explanation of these similar CFscores might be that the donors who support how Walker cut budgets and busted unions in a blue state also gave to Cruz or Paul for very different reasons: A donation to Cruz might be due to his ideology, while a donation to Walker might be due to his effectiveness (see online references to Walker sitting on a “throne of his enemies’ skulls”). In other words, simply comparing Cruz’s and Walker’s CFscores without understanding how they earned those scores puts us at risk of comparing apples to oranges. That’s not to say Walker isn’t conservative (he is), but rather to say that ideology is complicated and comparing the CFscores of a few presidential contenders does not always tell the whole story about their ideologies.
Finally, and perhaps most importantly, we must always remember that “conservative” and “liberal” are not expressly defined under these metrics. The candidates are simply lined up from most conservative to least, and assigned scores relative to one another. But a score of “.5” doesn’t really tell us much: We don’t know whether the candidate is pro-choice or supports Social Security privatization or where he stands on other issues.
This problem becomes compounded when we attempt to compare candidates across decades, as the definition of “conservative” “moderate” and “liberal” almost certainly shifts over time. CFscores try to compensate for this by building “bridges” between candidates. They compare how donors gave in multiple elections; if a donor gave to Walker and Pat Buchanan, and a different donor gave to Buchanan and Reagan, we would assume all three were ideologically similar.
But this “House that Jack Built” approach has problems that add more uncertainty to the scores. This sort of analysis often works, but can also lead one to conclude that because Buster Douglas defeated Mike Tyson, and Lou Savarese defeated Buster Douglas, that Savarese would defeat Tyson (actual result: Savarese was T.K.O.’d in 38 seconds).
Of course, the large number of pairings analyzed protects against many naïve errors like the one we just described. But consider: You might use a similar analysis to determine whether the 1962 Packers would have defeated the 1985 Bears; some people have tried similar analyses, especially for baseball. You’d miss some important differences though: The defensive ends and tackles on the ’62 Packers were all around 6-foot-3 and 250 pounds. The Bears’ defensive line was more around 6-foot-5 and 270 pounds, without counting William “The Refrigerator” Perry.
The point is that the game changes in ways that make these comparisons difficult: improved knowledge of nutrition, workout routines, and even legal performance-enhancing supplements (creatine, HMB, etc.) occur slowly enough that they don’t disrupt the chain across small time periods. But they become problematic across long time periods, in ways that these sorts of analyses simply can’t control for.
The same thing afflicts political comparisons. A donor in 1980 was contemplating a candidate’s solutions to the Cold War, stagflation, crime, and a 70 percent top tax bracket. None of these are issues in 2016. But these things changed gradually: inflation was no longer an issue in 1984, the Cold War was no longer an issue in 1992, high taxes and crime were no longer issues in 1996. So the issue matrix in 1980 is similar to 1984, and 1984 is similar to 1988 and so forth; we can probably compare across these short gaps. “Similar” doesn’t mean “identical,” however, and it is difficult to conclude that donor priorities in 1980 are anything like those in 1996 (let alone 2016). The small differences add up to big differences over time.
Taken together, the assumptions behind these methods make it almost impossible to compare today’s candidates to Goldwater in any meaningful sense: How do you compare the ideology of a modern candidate to someone who thought battlefield commanders should be allowed to use tactical nuclear weapons without presidential authorization, or who never had to take a position on abortion or gay rights (when he did, Goldwater was actually to the left of most modern GOP candidates)?
But even if we were to concede that Walker would be the most conservative nominee since Goldwater, we’re left wondering, “What does it matter?” Remember, political science also teaches us that Barry Goldwater (like George McGovern) didn’t really lose because he was too extreme; he lost because Lyndon Johnson was in his honeymoon phase and the economy was growing smartly. Bob Dole (CFscore of .6, pre-nomination) was closer to the center (assuming a center of somewhere around zero) than Bill Clinton (CFscore of -.899). George H.W. Bush (CFscore of .795) is as far right of center as Michael Dukakis (-.839) is left.
Perhaps most tellingly, the most extreme nominee under these scores (which only date back to 1980) is Barack Obama, whose pre-nomination score of -1.35 is farther to the left than John McCain’s (.68), Romney’s (.878) or Walker’s (1.28) are to the right. Obama’s post-nomination score (-1.65) is twice as far away from Clinton’s post-nomination score (-.98) than Walker’s is from George W. Bush’s (1.04).
Does this mean that we shouldn’t use statistical techniques like CFscores at all? Of course not. Again, they are useful scores, especially if they are used as heuristic devices. The point is simply that we should use them as political scientists do: with caution, keeping in mind the assumptions that are embedded within the techniques.