So, Nate Silver over at 538.com posted this very useful chart:
What does this tell us? The conventional wisdom is that the president's party loses seats in the midterm election. The chart shows us that this has happened quite a bit in the last 60-odd years. But it also is not a very compelling set of data.
What I notice is that since 1982, significant losses in midterms have happened only twice. Both - 1994 and 2006- coincided with a toxic atmosphere where the incumbent president was monstrously unpopular. Both also had structural implications. In both cases, party centrists were turfed out in favor of the other party. In 1994, Southern Democrats lost out to Southern Republicans, even where both were conservative. Brand name, not ideology, won the day. Similarly in 2006, centrist Republicans were turfed out in favor of moderate Democrats. Again, a bad party brand name was crucial.
Going into 2010, then, I would ask (1) how unpopular is Barack Obama compared to Bush and Clinton respectively and (2) how toxic is the "Democratic" brand name compared with the "Republican" brand name. Neither question indicates much similarity between 2010 and 1994 or 2006. In 2006 at this time, Bush was running at 32 favorable, 60 unfavorable. It didn't improve for him, either, in the runup to November. In 1994, Clinton went from being about 50/50 in April 1994 to the low 40s approval, 50+ disapproval by July, and it stayed there till November. Obama, by contrast, is at about 50/50 today. Will that go down? Can't tell for certain, but the nascent economic recovery is a strong arrow pointing the other way. I should note also that in 1982, Gallup showed Reagan at his nadir in popularity with an annual appproval rating of 43%.
Similarly, the generic ballots were much worse for the incumbent party in 1994 and 2006. The generic ballot (D v R in congressional races) also was lopsided in the double-digits in favor of Dems all year long in 2006. Data is harder to come by for 1994, but I have seen a number of archived polls with the generic ballot +10GOP in October 1994. By contrast, today, the generic ballot is also nearly even.
The 2010 numbers are just not particularly toxic numbers for the Dems by comparison with 1994 and 2006.
In fact, one can look at the entire post-WWII series and take issue with whether midterm elections 'usually' cause significant losses. 1974 was the watergate-payback year. 1966, payback for the civil rights act. 1958 and 1946 were recession years. My thesis here is that the incumbent party may have some disadvantages in the midterms (the opposition is more energized and the incumbent party's GOTV efforts will be lesser in an off-year) but that's not historically a very big mover of elections. What in an off-year is that the election becomes a referendum on the president and his party. As it stands in April 2010, the result is likely to be mixed. Obama also has a number of arrows in his quiver that could push the trend more favorably. He's got nothing particularly unpopular that he needs or wants to do for the rest of the year, and the economy is likely to improve, redounding to his credit to some degree.
Wednesday, April 07, 2010
Frank Talk about Midterms
Posted by The Law Talking Guy at 9:12 AM
Subscribe to:
Post Comments (Atom)
7 comments:
I like your analysis of why the exceptions to the pattern of the President's party losing seats in off year elections happened. I also hope you are right about how our current situation looks more like the exceptions to the pattern than the "normal" situations.
But how to you think your analysis would be changed if there were regional variations. What if Obama was especially unpopular in the South and rural areas? If those regions are where the vulnerable Democrats are they might be in a situation that fits 1978 more than 1998.
I don't think though that the fact that we've had two exceptions to the pattern in a row means that the pattern isn't really there.
I don't know what you mean by two exceptions in a row. We have had only two instances of significant losses since 1982, four instances where there were minimal losses or even gains. Since that's going almost halfway back to WWII, it suggests that the pattern is not what it might appear to be.
I can't speak to regional variations because I don't know what kind of data is out there on the subject. What I can and have said is that the polling data in 2010 does not look like 2006 or 1994.
I think the real pattern is not midterm= incumbent party losses but midterm=referendum on presidency + a different electorate than the general election. The result is either little or no change ('62, '70, '86. 90, '98, '02) or a big loss ('46, '58, '74, '82, '94, '06) or something in between ('50, '54, '70, '78). The polling data suggest from the most recent elections that this year looks more like a small change than a big change year.
What is a fascinating trend is that a president basically can never use midterms to increase the size of legislative majorities. The result can be neutral or negative, but not positive (I count 1998 and 2002 as effectively neutral). Presidents have to wait until the general reelection in the quadrennial year to increase legislative majorities. This shows you that there is a pretty strong connection between presidential popularity and congressional elections.
The "since 1982" part is just arbitrary. The argument in favor of a pattern goes back to WWII. Cutting it off at 1982 to support an argument that a pattern does not exist isn't kosher data analysis.
If someone posits a pattern over a 50 year time period and you identify two full and three partial exceptions to the pattern's expectation over 16 observations you can't disprove the original argument by reducing the number of observations.
If you wanted to add other observations (say going back to 1900 or earlier) and show that the pattern doesn't really hold then the people arguing in favor of a pattern of midterm losses would have to justify cutting off at 1946. Which they might not be able to do except by making some reference to WWII being a some kind of natural division between "modern" and "historical" politics.
But you can't just say "we should cut in 1982 instead" and declare victory. It is entirely possible that the pattern still holds most of the time just not lately.
Another way to look at this would be to say that there are really only two unambiguous exceptions since 1946 (1998 and 2002) to the expectation that the President's party loses seats in midterm election. In all other midterm elections observed the President's party lost seats in at least one house.
I am in much stronger agreement with your refined argument that a general election (reelection) is a necessary condition for incumbent Presidents to SIGNIFICANTLY increase their legislative support.
I think there are six exceptions to the "rule" that there are significant losses in midterms: 1962, 1970, 1986, 1990, 1998, and 2002. Six out of 16 is almost more than a third of the total observations. The fact that they are chronologically clumped in the latter period could be an indication of a new trend. There's no other way to observe new trends than this. Nothing arbitrary here.
Extending back in time before 1946 also gives more uniform results.
H S
1942 -4- -6
1938 -80 -6
1934 +9 +10
1930 -50 -8
1926 -8 -7
1922 -75 -6
1918 -19 -6
Note that 1934 was a year when the president was very popular.
I don't think anyone who makes these arguments ever says it will be the case 100% of the time. Social Science is about making probabilistic predictions. If the argument is "the President's party will tend to lose seats in Congress in off year elections" then I think there is sufficient evidence to have to take the argument seriously.
Of course there will be exceptions and I like your analysis as far as explaining some circumstances when exceptions are more likely.
Post a Comment