Subscribe via RSS Feed

The Six Year Itch? Whatever.

[ 23 ] November 28, 2012 |

Not too soon, the campaign season will commence again, with a focus on the Congressional elections of 2014.  In the smattering of stories that I’ve read in the past two weeks about this upcoming festival of joy, a term that I was only vaguely aware of keeps popping up, the six year itch.  However loosely defined, six years removed from his (or, presumably someday her) first election, the incumbent President’s party is apparently doomed to suffer atypically huge defeats in these mid term House elections.  On paper, this does not inspire confidence for the Democrats come 2014.  To quote from the Politico article linked above:

The party controlling the White House during a president’s sixth year in office has lost seats in every midterm election but one since 1918, when Woodrow Wilson occupied the Oval Office. And the setbacks typically aren’t small: The average loss in these elections was 30 seats.

Incumbent Presidents tend to suffer losses in damn near every mid term election for whatever reason (see the figure below), so this sentence could be restated as “the party controlling the White House has lost seats in every midterm election but three since 1918 . . .”.  Given the relatively small sample size, this really doesn’t tell us anything.  A better way of looking at the question involves comparing the mean seat loss for the incumbent party in bog standard boring midterm years, and the hypothesized qualitatively different six-year itch years.  During such years, apparently “Anger, exhaustion and frustration tend to set in among voters as presidents approach the last leg of their final term. It happened to Franklin Delano Roosevelt in 1938 when voters recoiled at his New Deal reforms. “  Of course, FDR would be re-elected in 1940, so that anger must have dissipated quickly.



In comparing the means between these two types of midterm elections, we have to settle on a measure of what is, and is not, a six-year itch election.  In terms of consistency, wikipedia lets us down; the brief entry on this topic includes 1974, by which time Ford had replaced Nixon, yet inexplicably overlooks 1998, possibly because it doesn’t fit the model.  In my analysis, I’ve settled on not settling on defining a clear measure.  Instead, I’ve chosen to start with a strict definition, and then progressively loosen the parameters of this six-year itch.  The above figure distinguishes such elections with a solid black border, and additional candidates have a thin border.


The table below compares the average seat loss for an incumbent party in standard (non-SYI) and SYI elections using five different measures.  A strict measure of SYI does what it says on the tin: a President must be in office at the time of the election six years from his first election.  From 1932 (20 total midterm elections), this limits us to five elections: 1938, 1958, 1986, 1998, and 2006.



Using a strict definition, there is no appreciable difference in average seat loss by the incumbent Presidential party in such elections.  The second column adds 1950 to the mix; one might argue that while Truman was not elected President in 1944, he did assume the office less than three months following the January 1945 inauguration thus giving Truman close to a full term in office prior to the 1948 election, but adding 1950 makes little difference.  The third column adds 1974.  Here, one has to argue that the voters either explicitly associated Ford with Nixon’s sins, considered Ford a mere extension of the Nixon years, or simply hadn’t noticed that Nixon was no longer president.  Given the pounding that Republicans experienced in 1974, this moves the means slightly, but still not convincingly.


Not satisfied?  The fourth column measures the SYI by including both 1942 and 1966.  In the case of the former, theoretically, why should FDR suffer from this phenomenon in 1938, but not even worse four years later?  If there is anything to this, then the itch really must have been festering in the minds of the voters in 1942 (as evidenced by the Democrats having lost 45 seats in that election).  1966 can be included for reasons similar to 1974 — LBJ, at least during 1964, was committed to continuing Kennedy policies in most domestic areas, hence his first “term” can be construed as a simple continuation of the Kennedy administration (which, in terms of names and  faces, it largely was).  Democrats suffered 1942 and 1966, so this does push the means even further apart.  Finally, the fifth column merges all this suspect logic by adding 1946.  Only now, do we see real daylight between the average seat loss of ordinary midterm elections and the special SYI elections.  Incidentally, this is also the only version of the five measures of SYI where the difference of the two means approach statistical significance (p=.086), but a) this assumes a one-tailed t-test, the use of which requires solid a-priori theory to suggest both the presence of a relationship and the direction of the estimate, b) these data are not random probability samples, and c) who cares?


Long story short: it doesn’t appear to exist.  There’s nothing really special about a President’s second midterm election that can not be explained by all the reasons why Presidents generally lose seats in any midterm election.  Visually, the only real pattern in the data illustrated by the figure above that is suggestive of the phenomenon is the period between 1952 and 1978, but for this to work one would have to loosen the definition of the measure such that both 1966 and 1974 merit inclusion.  1958, 66, and 74 do look different, but only one (1958) fits a rigorous definition.


What does this mean for 2014?  Nothing.  The Democrats will probably lose seats in 2014, but we don’t need a manufactured non-phenomenon to tell us that.  Alternatively, we can participate in some hard core wishful thinking and ignore oppressive historical precedent and choose to believe that the Democrats can retake the House in 2014 . . .

Comments (23)

Trackback URL | Comments RSS Feed

  1. mpowell says:

    Another way to interpret this data is that when the public gets stuck with an ‘incumbent’ they didn’t vote for the first time around (and if they get elected for that ‘second’ term it’s riding on the coattails of the previous president), they tend to get pissy at the midterms when they realize they don’t like the guy. It makes more sense and the data probably supports it more strongly.

    But pure 2nd term presidents don’t seem to have a problem unless you include FDR (and he’s an exception for other reasons).

  2. c u n d gulag says:

    There are a number of ways the Democrats could increase their majority in the Senate, and make gains in the House (though I’m not sure it’ll be enough to retake it – after 2010, and the redistricting).

    Let’s see what the implementation of Obamacare brings. If it’s positive – WIN!/WIN!
    Let’s see what, if there is any, filibuster reform brings.
    Let’s see what having more Liberals and Progressives in both houses brings. A few more open-throated ones can’t do any harm, after all of the squishy Whoreporatism we’ve seen from Democrats in the past few decades.
    Let’s see what happens after this ‘Fiscal Molehill” BS is resolved.
    Are we out of Afghanistan?
    Let’s see where it economy is.
    And, finally, let’s see how many Teabagging “True Conservatives” decide to primary Republicans in Senate and House races – doing that voodoo, that they do, so well! – in helping get Democrats elected.

    There’s always hope!
    Maybe this isn’t very realistic – but it’s still hope.

    • tonycpsu says:

      The Democrats can probably get closer to parity in the House, but I don’t see any way in the world they can increase their Senate majority.

      The 2014 map is dreadful — far worse than this year’s, and without Obama’s coattails. Even if Rockefeller can hang on in WV (go wingnut primary strike force!) our best pickup opportunity is… McConnell in Kentucky? Meanwhile we have to defend Tim Johnson’s seat in SD (perhaps with Herseth-Sandlin?), Begich in AK, Hagan in NC, Landrieu in LA… It’s brutal. I’ll be happy if we can hold onto a 50+1, frankly.

      • c u n d gulag says:

        Never discount the Teabagging morons, and their ablility to help the Democrats keep Senate seats.
        It’ll be hard to make Landrieu seem reelectable – but I’m sure they’ll find some maniac in LA who’ll make her look positively Byrd-like.

      • tiki says:

        The Obama coattails were pretty limited. Candidates in 2014 will miss the organization/money of a presidential campaign.

    • cpinva says:

      if the GOP does nothing to stem the tide of insanity that has gripped its ranks for the past 20 years (i’m looking at you, newt gingrich!), and takes the position that it’s 2012 losses were the result of not being conservative enough (rather than the reality that their candidates/agenda were/are odious to a majority of the voters), and goes whole hog birther/forced-birth/anti-everyone not white/insane “christian”, then i believe it’s possible for the dems to retake the house, and at least keep its majority in the senate.

      the GOP isn’t getting more popular, and as the next wave of 17 year-olds gains voting age, their (the GOP) base becomes smaller still. the country can only support so many clinically insane people at one time, and i think we’ve reached super-saturation.

      i hope, anyway.

  3. Johnny Sack says:

    There was a great xkcd comic a month or so ago on the problem of induction in election prediction (e.g. We’ve never had a president with a middle name beginning with Q! Etc).

  4. Greg says:

    If the economy picks up once the austerity crisis is resolved and Obamacare implementation is successful, the Democrats will have a good year regardless of historical trends. Did Obama really have coattails in the House this time around? It didn’t really seem like it. And we should pick up Governor seats (eg Maine, Michigan, Ohio, Florida) because it will be four years after a historically bad year. The Senate is troubling, of course, but it was troubling this cycle and the last, and we still held it.

    • catclub says:

      Recent history is encouraging. Reagan and Clinton were pretty popular at six years, while Bush was VERY unpopular. results reflected that.
      Also, economy looking better often seems to help,
      another counterintuitive result I provide at no charge.

    • The Democrats might well have a good year, in terms of tail winds, but what would tailwinds mean?

      Ten House seats would be wildly optimistic, but it still wouldn’t get them Speaker. Pennsylvania and Ohio are just too gerrymandered.

      And with the 2014 Senate map, a good year would be losing two seats.

      If the Democrats have a very good year, and successfully play defense in 2014, they keep themselves within striking distance of 60 Senate seats and a House majority if they also have a very good year in 2016. This time, though, they’re just trying to hold a line.

  5. rea says:

    It’s all very well to talk about what tends to happen in midterm elections, or what tends to happen in second term midterm elections, but that’s always going to be susceptible to being trumped by the specific circumstances of the election. Let the Rs spend the next couple of years bashing minorities and passing remarks about “legitimate rape” and the results of 2014 may be less favorable to them than they might hope . . .

  6. Johnny Sack says:

    @rea: point being that pundits derive facile inductions from banal observations like that. And before his election, it was something you could say anyhow.

  7. Jameson Quinn says:

    What happens in 2014 will depend, as always, on turnout. If people like the ones reading this blog are using the tools we now have to GOTV, we could indeed have an atypical election. Wishful thinking is silly, but optimism of the will is not.

  8. freemansfarm says:

    Er, according to your numbers, and no matter how you define a six year itch election, it always produces more losses for the incumbent presidential party than a “regular” (non syi) midterm election. You say the loss is not statistically significant.

    Well, I’m no statistician, but, as I read it, under definitions one and two, the seat loss increase, comparing “regular” and syi elections, is over ten percent (27.2 compared to 30.4, for an increase of 3.2 seats lost, 3.2 is more than 10 per cent of 27.2: and 27.1 compared to 30, for a gain of 2.9 seats lost, 2.9 is more than 10 percent of 27.1, respectively).Why is ten percent not “significant?” Under definition three, the seat loss percentage differential is over 25 per cent, actually closer to thirty (7.2 seat loss increase from 25.5 to 32.7). With definition four, the incrase is close to 67 per cent (from around 21 to around 35) and in definiton four, the increase is over 100 per cent (from 18 to over 36). Even if I am looking at all this “backward,” and the per cent increase should be computed looking to the increase in comparison to the syi number, the per cent increases still seem “signficant” to me. In one and two, they are still around ten per cent. In three and four, they are still over 20 and 30 per cent, respectively. And in five the increase is over 50 per cent.

    Yet you say only five shows a significant increase, and then wave that off. Why? Take four, for example, are you really saying an increase from 21.7 seats to 35.7 seats is insignificant? Why would that be so? It seems highly significant to me. Fourteen seats in Congress is quite a bit. Sure, there are 435 seats, but most seats are not competitive, with 300 or so of them being a lock for one party or the other, and so, at the margins, even a few seats are precious, and even a few seats lost and gain are important.

    Take even the low end of the scale, definitions one and two. There the losses in “regular” midterms is “only” about three seats less than in syi elections. But, again, every seat, at the margins, in a house with several hundred safe seats and a pretty close balance, is meaningful.

    I’m not trying to be a jerk here. I am really curious as to how and why you can say that the differences, particularly under definitions three and four, are insignificant. And why that, under definition five, the difference is significant, but somehow doesn’t matter (“who cares”). Is it because the sample size is too small? If that is the case, though, then wouldn’t the “correct” thing to say be that we just don’t have enough examples to tell if there is a measurable effect, rather than dismissing the possibility of one? And, even at that, doesn’t the incomplete or preliminary data at least suggest the opposite of the conclusion here, ie that there is no effect (“it doesn’t appear to exist”)?

    • Dave Brockington says:

      To be blunt, the difference between the means aren’t statistically significant simply because they’re not. I ran a one-way ANOVA on the data for each of the five definitions of SYI elections to calculate significance. There are several issues to consider here. One, with an overall N of only 20, only the most bullet proof estimates will be significant. These aren’t. The standard deviation of the means, however defined, are huge: for column 1, the SD for standard elections is 23.7, while for SYI, 25. Even for column 5, they’re 23.9 and 23.3, respectively. It takes an N much larger than 20 to overcome standard deviations that large relative to the estimate to show significance.

      The reason I wave off the “significance” of column 5 is two fold. First, the industry standard for significance, however arbitrary, is .05. The p value for column five’s means is .086. Now one can “break” that rule and go up to .10 for significance, but only when there is a clear and rigorous theoretical reason to expect a significant estimate in the first place in addition to the direction of the estimate. The latter is present here, but the former is not — there’s no theoretical reason to anticipate greater losses during a Year 6 election than a Year 2 election. The second reason is that the rules governing what is, and what is not, a SYI election for column five are so loose as to be meaningless — it’s fitting the data to support a “theory”, rather than taking the data at face value. To my mind, the only accurate measure of SYI is column 1.

      Finally, as I strongly imply, significance is a meaningless concept with these data in any event. I have the entire universe of data going back to 1934 here. Significance technically only applies when we’ve drawn a random probability sample and we’re seeking to rigorously generalise from that sample to the population. However, as it’s become standard to report significance levels no matter how erroneously applied to the data, I did so.

      So what we’re left with in these data is arguing whether or not a difference of three seats matters, but given the wild standard deviations involved and the complete lack of theory to predict this pattern, I would not use these data to argue that 2014 will be qualitatively (let alone quantitatively) different from 2010.

      • freemansfarm says:

        I spoke to a statistician I know, and he claimed that when the sample sizes are so small, that makes it difficult to get “statistically meaningul” results. Something to do with the “p value” being too high. And you seem to be saying the same thing (“with an overall N of only 20, only the most bullet proof estimates will be significant”). As I see it, and as my friend implied, and as I mentioned before, when the data sample is too small, the conservative or “correct” approach for the statisitician is to say just that. IE that he doesn’t have enough data, NOT that whatever phenomena he is supposedly looking for “doesn’t appear to exist.”

        Take a simple example, in April, we want to know if baseball player X is having a good season at the plate. He has ten at bats and three hits, for a batting average of 300, Would anyone say, when asked if he is having a good hitting year, “he doesn’t applear to be?” No, and that would hold true even were batting 100 or 200 or 400. It is simply too early to tell is the right response.

        You seem to be agreeing with all of this now, but, in your original piece, the vibe was not, “well we don’t have the data, and so we can’t say,” but, rather, “nah, the thing doesn’t exist” (as in “Six Year Itch? Whatever…”, “it probably doesn’t exist”…”manufactured non phenonmenon” and so forth). I understand there is a difference between technical, statistical “significance” and how a layman might use that word (“significance”), still, in your main piece, you treated the lack of the former as synonymous with the latter. I don’t think it is.

        Anyway, my friend had a few more comments as well…

        He claimed that your “univariate” approach was problematic. IE that there are other variables, whch when taken into account and controlled for, might bolster or weaken the six year itch argument. Examples of these other variables being the presidential approval rating, the change in the ecomony (however one measures that), and district factors(such as how many incumbents are running for re-election for example, how many are of the president’s party, is there any relationship between presidential approval rating and the decision of incumbents not to run again; what percentage of the vote did the incumbent get in the previous election, etc.)

        I’ll leave it to you to think about those criticisms.

        • dave brockington says:

          Ah, the rarely employed argumentum ad verecundiam.

          FWIW, I did not mean to imply that the lack of statistical significance renders the phenomenon nonexistent — I threw in significance levels nearly as an afterthought. Again, significance levels are technically meaningless and misapplied to these data as they’re the entire universe of data, and not a random probability sample.

          It’s not the lack of significance that makes me believe that the phenomenon doesn’t exist. Rather, I don’t think the phenomenon exists because there’s no convincing theory (no theory at all really) or evidence to suggest that it does. The best, and only accurate measure of SYI is the first column, and the average seat loss only differs by three, but with a very large standard deviation in the constituent data points that make the average. If one were to plot both distributions (normal and SYI) on the same axis, the distributions would overlap more than they don’t. No theory, no evidence, no phenomenon.

          As for the difficulty of finding significant estimates with a small N, your statistician friend and I are saying the same thing. Simply stated, it is “something to do with the p value being too high”. More accurately stated, it’s difficult to find significance with a small N because, unless the difference between the estimates (means in this case) is huge, or the SD of an estimate is very small, the small N robs any statistical power. The opposite is true with large N studies. I presented a model from one of my articles to my students this morning that had an N of 64,000. Everything is “significant” in that multivariate model, even the tiniest estimates (which have little to no substantive effect on the dependent variable) because of the large N.

          It’s likewise a bit misleading to be speaking of the data sample being too small, as it’s not a sample to begin with, but again the universe of data. It’s not that I don’t have enough data to test my hypothesis, which could be ameliorated by a trip to the shop to get more, but I literally have all the data that’s available when it comes to midterm elections back to 1934. The well meaning comparison between a baseball player’s April vs his entire season / career is misleading: it’s not as though I’m going to have 6x the data in five more months. Indeed, if we were to consider the metaphor, with the data presented above as being “April”, it would take an additional 380 years to generate a full season’s worth of data.

          In every case, a multivariate model would be more (or less) convincing, but ultimately, this was a blog post written in some free time, not something I’d send off for review and publication. But then I usually run a simple bivariate relationship estimation to see if there’s anything there to begin with in any project I’m working on. In this case, if there was any theory at all, or if the first column above demonstrated something substantive, I might invest the time to gather the data to play around with a real model. But there’s nothing here. A three seat average difference with a SD of 25 does not tell me that there is a real effect.

          I’m confident that your statistician friend and I would agree on far more than we disagree, but then I’ve had the misfortunate to have to teach this stuff.

  9. Anonymous says:

    Quick calculation of the five SYI elections combined with the prior five TYI elections.

    34-38- +9 and -72 gets 63 lost house seats between both the elections

    54-58- -19 and -49 gets 68 lost house seats between both the elections

    82-86- -26 and -5 gets 31 lost house seats between both the elections

    94-98- -54 and +5 gets 49 lost house seats between both the elections

    02-06- +8 and -31 gets 23 lost house seats between both the elections.

    On average in the combined midterm elections of the first two terms of a presidency the presidents party loses 47 seats (46.8). Obama lost 63 in 2010, so he is already above the average and this would indicate that gains for the dems might be the more likely result than further loss of support. Additionally, if the R’s manage to gain any it will be marginal. The intuitive feel of the map and the rally of never again to 2010 back that up with gut feeling. Nevertheless this is probably less statistically significant than the other stuff, it just seems right that if one party makes massive gains in the two-year itch, maybe the six-year itch should mean less, and aggregated 3 of the combined itchiness are less than the half on our books, 1 equals our start point and 1 modestly exceeds it. The two whose combined numbers are closest to our single 2010 number are the ones farthest removed from us in history. Of course none of them were split by a redistricting, so the whole excercise might be moot. Still, republicans might have maxed anti-Obama house gains in 2010.

  10. Tajuana Stroh says:

    Bed Bugs Chicago, Treatment of Bed Bugs Chicago, Removal of Bed Bugs Chicago Remove those digusting bed bugs now at

  11. I’m impressed, I have to admit. Seldom do I come across a blog that’s both equally educative and entertaining, and let me tell you, you have hit the
    nail on the head. The issue is something that too few people are speaking
    intelligently about. Now i’m very happy that I found this during my hunt for something regarding this.

Leave a Reply

If you want a picture to show with your comment, go get a Gravatar.

  • Switch to our mobile site