Jon cited Nate Silver on Friday for the following proposition:
Looking at new Gallup data on partisan identification by age, Silver tried mapping it against the question: "who was President when you turned 18?" As it turns out, "the popularity -- or lack thereof -- of the President when the voter turned 18 would seem to have a lot of explanatory power for how their politics turned out later on".
Silver’s conclusion is based on an incredibly helpful publication from Gallup, which illustrates the net Democratic party identification based upon age. It concludes that the Democrats’ are the strongest among the baby boomers and (to a greater degree) Generation “Y”.
Silver doesn’t subject his hypothesis to any statistical testing, which is anomalous for him. Instead he breaks the Gallup chart down by Presidency, and provides a qualitative narrative of which Presidents were popular, which were unpopular, and to what degree the Democrats’ partisan advantage correlates with those Presidencies.
At the outset there are some problems with this. Silver’s years don’t match up with the given Presidencies. If you look at the chart closely, George W. Bush’s Presidency is longer than Clinton’s, Reagan’s, or Eisenhower’s – and it still doesn’t cover all the datapoints at the end of Gallup’s series. But Bush’s Presidency certainly felt longer than any of those, and the changes are minimal, so I guess Silver gets a pass here.
The descriptive data itself are more problematic. Silver alternates between (i) generalized descriptions of Presidencies over an eight year tenure, (ii) how Presidents are viewed today, and (iii) (his actual hypothesis) how President were viewed when the people turned 18. For an example of each:
(i) “Reagan, a highly successful President who was popular throughout most of his term and may be even more popular today, is associated with a considerably above-average amount of Republican support.”
(ii) “Johnson, whose complicated time in office is generally regarded today as having been an above-average Presidency, is associated with generally above-average levels of Democratic support.”
(iii) “Kennedy, who was very popular throughout his brief tenure in the White House, is associated with above-average levels of Democratic support. (You can almost see the spike in popularity among 64- and 65- year olds, who would have been about 17 when Kennedy took office.)”
All of these descriptions are problematic – Reagan was highly unpopular in the second and third years of his Presidency, when Republicans suffered the sixth-worst midterm election for any party since World War II – but this results in only a one-point increase in partisan identification for Democrats. Johnson’s Presidency may be viewed as successful today by some, but from May 1966 through 1968 his approval ratings never rose above 50%. Oddly, this period correlates with a jump in Democratic identification; likewise as Nixon becomes less popular the Democrats’ advantage declines. If (iii) is true, one would expect a downward spike for Ford, whose approval ratings were mired in the 30s and 40s for most of his very short term.
Subjected to statistical rigor, the popularity of the incumbent President when a voter turns eighteen has very little predictive value. If you take Gallup’s data regarding the Democratic affiliation of each age range of voters, and you regress it against the average Presidential approval rating for the year those voters turned 18 (for Republican Presidencies, use 100-[approval] so we’re always looking at how the Democratic point of view on a Presidency is viewed), and you get an r-square of .04 and a variable that just misses statistical significance at the 95% confidence interval.
To put it differently, if we chart the Democratic party ID advantage by the year that a voter turned 18, we get a chart that looks identical to Silver’s. If we superimpose the average annual approval rating for the year that voter turned 18 over that, and then add trendlines for each dataset, we get a chart that looks like this (I’ve divided the approval rating by 5 just to get the scale to line up; since this is a linear transformation it doesn’t matter):
If the correlation that Silver hypothesizes were present, the trendlines would move in tandem – as the approval of a Democratic (or disapproval of a Republican) President would correlate nicely with a spike in Democratic identification. As we already know from our r-square, this doesn’t really happen.
But, as you expand the age range, the correlation improves. If you average the Presidential approval for when the voter was 18 and 19, and regress it against the Democratic identification advantage, the correlation is about .08, and the variable is fully significant. As you keep adding on years, the correlation grows stronger, until you are averaging the Presidential approval ratings from when a voter was 18 until she was 32 (at which point the r-square is about .282 and the t-stat is 5.22. Every point increase in average Democratic Presidential approval rating over those years results in a .3% increase in Democratic affiliation. After that, adding years to the spread decreases the model’s predictive abilities.
If we revisit our chart, we come up with something that looks like this:
Note that now our trendlines are moving together nicely, with peaks and valleys roughly correlating.
Then, if we move the bottom end of the range backward, the model continues to improve. In other words, if we look at average approval rating where the voter is 17-32, 16-32, 15-32, etc., the model continues to improve. It reaches maximum efficiency at an r-square of .66. I’m ultimately skeptical of the strength of this model, because we know that things like race, gender, income, and parental partisan affiliation have an impact on a person’s id as well. It seems like this is explaining too much. Nevertheless, it seems that – as we may intuit – partisan identification is not determined precisely at age 18, but rather is a process that occurs over several years of a young voter’s life.
This explains one of the mysteries of Silver’s narrative. He writes that “Finally, we get to Truman and Roosevelt, where things seem to break down a bit. Truman is regarded quite favorably by historians today but was unpopular for much of his tenure; he is associated with average-to-slightly-below levels of Democratic support. [Note that here Silver is talking about contemporaneous views of the President versus modern views]. The numbers then bounce around a bit for FDR, perhaps because there aren't all that many people in their mid-80s and so the sample sizes are small.”
The Gallup data explain that the sample size never gets below 650, which is plenty of voters to tamp down statistical noise. What is probably going on is that a voter who turned 18 in the early 1940s came of political age in the late 1930s, when FDR was not particularly popular. They then went through a peak in FDR’s popularity during World War II, a trough in Democratic popularity through the Truman years, and a peak in Republican popularity under Eisenhower. Unsurprisingly, voters here are mildly, though not enthusiastically, Democratic (I don’t know that the data here are any noisier than for those who turned 18 in the 1950s and 1960s).
Ultimately, Silver’s main thesis is intact: Bush is likely to haunt Republicans for a generation, as the back-to-back successful Clinton Presidency and failed Bush Presidency have likely set younger voters on a pro-Democratic path that will endure. Of course, even voters who turned 10 in 1996 have not yet finished their partisan development, so if Obama’s Presidency takes a wrong turn, Republicans will have an opportunity to bend the trendline in Democratic ID back their direction.
Finally, we should note that “party identification” does not necessarily equal “voting.” If this were the case, Kentucky would be one of the most solidly Democratic states in the country. Likewise, the 45-59 demographic should have been a good one for Obama, but it wasn’t. Ultimately issues, candidates and campaigns matter, which could truly be the Republicans’ saving grace in the longrun.