The Second Phase of Realignment: 1976-2012
Shifts in American political geography (“realignments”), which I wrote about on Thursday, can be viewed at a glance using the following diagr...
The Realignment Myth
With a candidate as strange as Donald Trump, it is tempting to speculate that the usual red-state and blue-state assignments may not hold. Trump is probably not a leader of change in the Republican Party, but rather the visible manifestation of a realignment within the party that has been brewing for years. Will national voting patterns change too?
Probably not. State-by-state party preferences mostly change in a gradual manner. Today I will show that the last major national rearrangement happened in the years surrounding 1964 – when President Lyndon Johnson signed the Civil Rights Act.
It is not easy to tell simply by examining election results when a realignment has occurred. For example, from 1988 to 1992, Democrats went from winning 10 states (Dukakis) to 32 states (Clinton). However, that was principally driven by a nationwide shift in Democratic-Republican vote margins.
To detect when the electoral map has been rearranged, I will use the correlation coefficient, a measure is not affected by national swings in opinion. The maximum possible value for correlation coefficient is +1.0, which indicates perfect proportionality with only an offset. (A value of zero indicates no relationship, and a value of -1.0 indicates a perfect anticorrelation.) In this definition, an across-the-board shift is not a “realignment” but just a strong year for one party.
In the case of 1988 to 1992, the correlation is +0.90:
In this example the average difference was nearly 13 percentage points, which is not a real realignment. On top of that was a state-by-state standard deviation of 7 percentage points. That 7-point deviation measures the extent to which states are genuinely realigning.
This is pretty typical of consecutive Presidential elections. Here is a graph of election-to-election correlation coefficients since 1964:The biggest change here is the election of 1964, when national patterns of support for Johnson v. Goldwater were totally uncorrelated with patterns of support for Kennedy v. Nixon. The sharp change coincides with President Johnson’s signing in July 1964 of the Civil Rights Act. From 1960 to 1964, the average state Democratic-Republican margin moved 19 points toward Democrats…but the standard deviation of additional change was a whopping 25 percentage points. Now that’s a realignment!
A little more re-sorting of partisan voters took place over the next three elections, interrupted only by the election of 1976, in which Carter temporarily regained southern states for Democrats. In 1980, the major realignment became complete.
Since 1980, the pattern of partisanship has changed only a little bit from election to election. Note that the graph above includes current 2016 polls of the Clinton v. Trump race; the correlation with 2012 is +0.90. So far, no realignment is apparent.
Changes accumulate over time, as shown by this set of comparisons of different years with the 2012 election:Again, the overall pattern is a sudden change from 1960 to 1964, and gradual change occurring since then. In this graph, you can see the transient effect of southerner Jimmy Carter quite well as a notch in the graph for 1976 and 1980.
Incidentally, correlations are useful for comparing 2016 polls with 2012 election results. Current polls show a high number of undecided voters, and both Clinton and Trump are underperforming compared with 2012. You can see this by the fact that overall, polling data falls along a shallower line than the red diagonal, which indicates equal performance.
In every state in which Trump is ahead, Romney won in 2012. And in every state in which Clinton is ahead, Obama won in 2012 (except for Arizona, where the current median polling margin is quite narrow, at Clinton +1.5%). For now, the 2012 map basically applies. It is highly premature to claim that either candidate is doing differently from his/her party four years ago. Though I must say…Utah, wow.
If you want to work with the data yourself, it is tabulated here.
Update: spreadsheet and graphs corrected to fix a small error in the comparisons with 1960.
I don’t think I’ve ever seen the Southern Strategy quantified quite so clearly. Well done!
There’s something else this chart reveals that surprises me, and that concerns the post-Watergate anomaly of 1976. That’s the first presidential election I actually remember, and I’ve always thought of it as a real oddball, with a map that looks like it was lifted out of the late 19th century.
But this analysis reveals that there was a sudden change in geographic patterns from 1972 to 1976, but not a sudden change afterward: the 1980 map had a high correlation coefficient with 1976, masked mostly by the way Reagan was winning everything. Carter temporarily interrupted the conversion of the South to a Republican stronghold, but Reagan mostly just made things more Republican everywhere, and the move to a solid Republican South after that was much more gradual.
Nice work, but I’m a little confused.
1988-1992 sure looks like a realignment if you stare at the EV maps. California went from reliably Republican to reliably Democratic in 1992, as did several Northeast states.
The 1988 map looks like it’s from another era, while the 1992 map seems familiar (modulo MT and Southern states like GA, TN which voted for B. Clinton).
According to your metric, is 1988->1992 a realignment or an adiabatic shift? If the latter, then the question becomes are we due for a similar shift maybe with AZ,
and (gasp!) UT?
I think the point is that the 1988-1992 shift was a nationwide shift toward the Democrats, not a change in regional patterns. The move toward the red state/blue state map we all know was already under way during the 1980s; it was just masked in the election results by the Republicans winning a majority almost everywhere. And the most Democratic states in 1988 were also generally the most Democratic states in 1992.
It only appears to be a realignment because EV maps don’t distinguish between low margin and high margin wins (i.e. California (+ 3.57 R) appears to be just as red as Utah (+ 34.17 R)). However, if you looked at the voting margins in 1984 and 1980, you could see that 1988 was already trending more to the blue side of the spectrum. In 1980 and 1984 California had voted red at + 16. 1988 reduced that margin to 3.57.
The correlation coefficients don’t care whether the majority voted in a red or blue, they only care about the margin between them. This can also be seen in the margins in 1988 for some of the Northeastern states (PA +2.3 R, VT +3.5, MD +2.9, etc.). Once those states were fairly close on their voting margins, it wouldn’t take much change to push them over into voting majority blue. This would appear to result in a big EV change, but not necessarily be too surprising from a margin viewpoint.
Even in 1976 when TX was blue, CA was red. Other than the Goldwater blip of ’64 CA was red throughout the postwar decades until 1992. Switching of order~100 EV’s from R to D in 1992 is a remarkable change in presidential politics.
…It looks to me as if, with the exception of Utah, what’s happening now is a continuation of patterns that have been going on under the skin for a while now. The coastal South and the Southwest are gradually getting more Democratic; Appalachia and the whiter interior South are rapidly going deep, deep red; the upper Midwest is probably getting more Republican, but slowly.
And I’m wondering if New England, especially in the suburbs, might be trending slightly more Republican after a peak of Democratic support around 2006. It’s not enough to flip states yet except for possibly NH.
There is a difference between groups of states realigning and candidates doing better or worse based on the contours of the race. Sam’s point seems to be that had Bush-Dukakis gone the opposite way–if Dukakis had won, say, 30 states instead of 10–the map wouldn’t look much different than it did in 1992.
The Democrats got 49 EVs in 1980, 13 in 1984, and 111 in 1988. Since then they’ve never gotten fewer than 250 EVs. Given that history, the maps are going to look very different before and after 1992, realignment or not. From 1992 on the Democrats have always won every state they won in 1980, 1984, or 1988, except for Georgia (1980) and West Virginia (1980 and 1988).
Amitabh: look at the differences between TX and CA and the national vote over that span of time. In 76, CA was 4 points more Republican than the nation as a whole, while TX was 2 points more Democratic. By 1984, even the CA goes red, it’s now 2 points bluer than the country as a whole, and 4 points bluer in 1988, again despite going red. In ’92 it’s 10 points bluer. The trend had been going on for several elections, but the state only flips when the Dems do well nationwide. TX, of course, goes in the other direction, becoming 9 points redder than the nation as a whole in the ’92 election.
Nice quantitative analysis capturing the shift in national mood. The importance of correlation between swing states may be more important than all states though, i.e. the election will be decided by small shifts in the swing states.
What happens when you do this for other statewide offices—senator and governor? Obviously far more noise, but it might make 1976 look like even more of an outlier; the realignment might have been firmer and more durable at the state level.
I idly tried to see what would happen if I restricted the correlation coefficient calculations to only use “close” states, as alluded to in the comments. In particular, I decided to compare only states that were within 6% in the previous election to their results in the next one.
As expected, the data became sparse, with an average of 12 states meeting this criterion per election (7-8 in the last three elections). In 1972, only one state did. The resulting election-to-next-election correlation coefficients were:
0.58 0.97 0.68 0.73 0.52 0.50 0.39 0.65 -0.09 0.49 -0.26 N/A 0.03 -0.57
(2016 2012 2008 2004 …)
So, this is much more all over the place, as expected from restricting to a small, deliberately hopefully-swingy data set. Standard deviation of the coefficients becomes 0.45, compared to 0.29 for the full data set (not a very meaningful measure, but quantifies the “all over the place” a little!)
I’m pretty sure this is not useful, but in case anyone was curious!
By the way, Sam, not that it makes any difference to the correlation, but on the “correlations” sheet, your margins for 2000-2012 are in decimal (0.1, etc), while for all other years, they’re in percent (10, etc). (I guess it comes from different formats on other pages.)
Interesting. A simpler description of a realignment focusing on the swing states could be just crossing the barrier of 0 in the D-R margin, i.e. flipping the state from D to R or vice versa. Using Sam’s data to count how many times such ‘swings’ have occurred in some chosen ‘swing’ states, I get the following counts for DR ‘flips’:
7 OH, FL
5 WI, NH, NC, IA, CO
4 PA, NV, NM
3 VA, ME
2 NJ, MN, MI, AZ, UT
Utah shows that even a deep red state has changed affiliation (1964 and 1968). If previous ‘mutability’ of states is any indication, and we require at least 2 previous ‘flips’ since 1960, then OH, FL, WI, NH, NC, IA, CO, PA, NV, NM, VA, ME, UT might be the target states for Trump. Imposing a second filter of a range within -5 and 5 in the state D-R margin in the 2012 election in these states, i.e. expecting a change less than 5 points compared to 2012, leaves only VA, OH, NC, and FL. It does seem like a very steep climb till we get to President Trump.
Interesting analysis given all of the hype about Trump scrambling the map. That doesn’t seem to be happening. The Cook Report seems to support this. There’s a link to a tool that allows you to change the demographics and see what might happen in the fall.
The thing that does slightly mystify me is, if the vote in 1980 was 90% correlated with 1976, why aren’t their individual correlations with 2012 more similar? Probably if I worked through the math I’d see it…
Here is how it could have occurred. The 1976-to-1980 correlation of r=+0.89 means that r^2=0.79 of the variance in 1980’s results can be explained by 1976’s results. Thus 0.21 (i.e. 21%) of the variance in 1980 is unexplained by 1976’s results.
Now think about 1980-to-2012. That correlation (r=0.63) means that 1-r^2, or 60%, of 2012’s variance is unexplained by 1980’s results.
At this point, as much of 60%+21%, or 81%, of 2012’s results is potentially unexplained by 1976’s results. The actual amount of unexplained variance, which is r=+0.43 for 1976 to 2012, is 1-0.43^2, or 81%. So it is possible.
I figured the way variances add in quadrature probably had something to do with it.