Wednesday November 29, 2006
Looks like it could be interesting.
Looks like it could be interesting.
Greg Combs from UT Dallas writes [inter alia]:
Hi Dr. Jackman,
A fellow UTDallas colleague and I are beginning work on a paper looking at Texas House/Senate Roll Call votes using Bayesian statistics, much like your “Most Liberal Senator” article…. I’m very interested in the data set you used for US Senate roll call analysis and the R script to spit it out…
Sure: I’ve added a replication archive for this project to my research pages, which contains the data and lots of R code, which relies on the rollcall class and the ideal function in our pscl R package. There are lots of goodies in the replication archive, including code for making nice graphs etc.
Going back to this project reminded me just how fragile is the inference that Kerry was the “most liberal” senator of 2003. He votes on only 23 of the 62 National Journal “key votes”, and in a liberal direction every time; frankly, its hard to say much about how liberal he is under these circumstances, and the conclusions are surprisingly sensitive to the choice of prior and even how long you run the MCMC algorithm. Basically you are trying to distinguish between three senators (Kerry, Reed, and Sarbanes) with not much information, and they out on the end of the preference distribution: you need a lot of MCMC iterates and a lot of confidence in your model etc to walk away with an authoritative determination as to who is “the most liberal” senator, at least given these data. The analysis based on all 498 non-unanimous roll calls in the 107th Senate (also packaged in the replication archive) is much more robust and makes it clear that Kerry was not the most liberal senator, not even close…
Doing this (re-)analysis prompted me to make some changes to the ideal class in the pscl package. I’ve got to push these up to CRAN; look for version 0.73 of pscl in the next few days (as of writing, 0.72 in on CRAN, but the R code in the replication archive uses 0.73).
Alan Ramsey’s column in this weekend’s Sydney Morning Herald has some excerpts from a censure motion debated in the Australian House of Representatives in late 1995. As Ramsey points out, this was the last sitting day of the House before the March 1996 election — the election that would bring Howard to office on the back of an almighty thumping of Labor (and Keating). Keating’s participation in the censure debate would turn out to be his last, substantive appearance on the floor of parliament. It also marks Howard’s last appearance as Leader of the Opposition.
And I was there. I was an asst prof at U.Chicago, but I took off about two weeks before the end of the quarter so I could have a nice long summer break with Janet in Australia. We went down from Sydney to catch up friends in Canberra. Janet used to work in the Parliament House Press Gallery for the Nine Network, and we also caught up with fellow Queenslander Lenore Taylor (covering national politics for the Australian Financial Review). Lenore got us into the press gallery that day (Nov 30, 1995), to see what we thought would be a typically fireworks-laden Question Time, but instead we got to see the start of the debate on the censure motion. I can remember Keating closing up his huge briefing book, turning to leave the chamber, the beautiful suit he was wearing (Zegna, right?), a hand gesture of dismissal/contempt for the howls of “Bye Bye!” from the Opposition benches, Howard leading the raucous chorus, all of them waving farewell to Keating, grinning from ear to ear. They knew… And of course, wonderful political theatre, and nothing like it in the American system.
One doesn’t get the sense we’re close to a similar situation now. Despite some polling suggesting Labor in a pretty healthy position (at least in two-party preferred terms), Howard hasn’t engendered the anger that Keating did, and indeed, has consistently led on the “better PM” question in the national polls, usually by margins of 25 percentage points or more. “Arrogant, out of touch, 13 years” were the campaign themes that the Coalition hammered in 1996, with much of their attack focussed on Keating himself. It is tough to see that working for Beazley (or his successor?) in 2007 against Howard, at least this far out.
…which could be me, but is actually the mega-famous type designer Erik Spiekermann for Design With Reach.
I use Spiekermann‘s FF MetaÂ® in a lot of my papers etc, which DWR is doing as “Contemporary” in their street numbers product line. I like Neutra better, but its everywhere (at least in Palo Alto/”Eichlerville” where we live).
Spiekermann is the author Stop Stealing Sheep, a great primer for anyone starting to take this stuff seriously (my buddies Mal Enright and Ric Aqua put me onto this when I started to get interested in type…)
Today’s New York Times featured “36 hours in Sydney” (which is kind of weird, since it takes 14 hours to get there from here on the West Coast of the US, so 36 hours hardly seems worth it…)
This density graph is pretty self-explanatory, showing that there are hardly any competitive congressional districts in California. This is based on the Secretary of State’s data from the 2006 election; note that only 45 of California’s 53 districts had both D & R candidates running. The bottom curve shows the seats-votes curve implied by these data under (the simple but untenable) assumption of uniform swing. The actual result doesn’t stray too far from proportionality…
In Crikey.com.au, the in-house psephologist Charles Richardson comments on the U.S. House elections, relying widely on the peculiarly Australian fixation on uniform swing (David Butler casts a long shadow via Malcolm Mackerras…?). Writes Richardson,
To gain the 15 needed for a majority, the Democrats needed a uniform swing of 5.2%. (The boundaries have a slight bias towards the Republicans, since most of them were drawn by Republican legislatures.)
They did a bit better than that. A few seats are still doubtful, but it looks like a total gain of either 28 or 29 seats (the uncertain one is Connecticut’s 2nd district, where the Democrat challenger leads by just 170 votes).
Aggregate votes haven’t been tallied yet, but the two-party swing seems to be between 6 and 7%. A 7% uniform swing would have delivered 26 seats.
Because incumbents are so powerful, swings are even less uniform than in Australia. Five of the six most marginal Republicans are holding on; the gains are disproportionately in seats where incumbents had retired or fallen victim to major scandals. Mark Foley’s seat, for example, recorded a 13% swing.
Nonetheless, the total swing is still a good guide: deviations from uniformity may be greater [in the US than in Australia], but they do roughly cancel out.
Statements like the last one are so vacuous as to be ridiculous, but nonetheless part of what we’ve come to expect from the self-styled “psephologists” working in the Butler/Mackerras tradition. What is “the [uniform] swing”? By construction, “the swing” in the popular vote has to be close to the average of the district-specific swings (say, when the election is held over a large, heterogeneous collection of districts, with little malapportionment, like, well, a contemporary US House election). Hence, to claim that a virtue of uniform swing is “that the deviations tend to cancel out” has the same value as noting that the sum of the deviations around an average are zero (again, true by construction, and hence not especially insightful or helpful).
The more interesting question is how badly uniform swing would perform as a prediction rule, seat-by-seat. In Australia, the answer is not so bad (and hence Malcolm’s pendulums are useful for election night prognostication); in the United States, the variation around the average (aka “uniform”) swing is so large as to render the concept pretty much meaningless. Even in a highly “nationalized” election like 2006, the district-by-district swings are what will determine the election, and the average/”uniform” swing is a fairly meaningless statistical summary, an abstraction lacking the political content it might hold in parliamentary systems a la Australia.
So Neal Beck asks:
so, post last night, do you have and can you send a picture of the current senate lr positions – am curious as to what happens when one crosses out chafee and makes some guesses about tester and webb.
I grabbed the roll call data from Jeff Lewis’ web site using the nifty readKH() function in our pscl R package, and then ran a one-dimensional ideal point model (a la Clinton, Jackman, and Rivers APSR 2004), also implemented in the pscl R package. I then plot the ideal point estimates against Bush vote share in the senators’ respective states in 2004, producing the scatterplot shown below. Each point corresponds to a senator (red for Republicans, blue for Democrats, with senators running for re-election given a heavier shading), with Bush vote on the horizontal axis, and the voting score on the vertical axis (higher is more conservative, lower is more liberal). The gray line is a regression fit, not to be taken too seriously (read on…).
The biggest thing that jumps out at you when do plots like these is the separation by party (there is virtually no overlap between the red and blue dots), but reasonable tracking between a state’s political orientation (at least, in this case, as reflected by Bush 2004 vote share) and legislative behavior.
Anyway, in response to Neal’s query, I’ve highlighted the six Republican senators who have (or seem likely) to lose their seats; they are numbered 1 through 6, the other Republican senators who were re-elected are numbered 7 through 14. Chafee (RI), Santorum (PA) and Allen (VA) seem to be the only Republican losers who are obvious candidates for a “out of step with their state” kind of story (a la Canes-Wrone, Brady and Cogan APSR 2002), lying relatively distant from the regression line. DeWine (OH) seems to have been caught in what was a wipeout for Republicans in OH; neither Talent (MO) nor Burns (MT) appear to have been particularly “out of step” (Burns is projected to lose by a whisker). And keep in mind that we’ve got Republican senators — some perhaps just as apparently “out of step” as Santorum etc — who did not lose their seats (e.g., Kyl, AZ).
It is interesting to speculate on shape of the new Senate (the 110th). Chafee goes, replaced by a Democrat, leaving the Maine women (Snowe and Collins) as the most moderate Republicans. We’ll see just how liberal or moderate the new Democrats are; given their states and the narrow margins with which they are projected to win, it is tough to imagine Webb (D-VA) or Tester (D-MT) being particularly liberal, perhaps more like the relatively conservative Democrats from the plains (Nelson, NE; Conrad and Dorgan from ND) or the other MT senator (Baucus).
A similar exercise for the House would also be interesting (Keith, Jeff…???)
Apparently they are taking all kinds of precautions this year to stop the possibility of exit poll data leaking through the afternoon. Which makes me wonder if my colleague Doug Rivers (at CBS, along with Steve Ansolabehere), or Charles Franklin (at ABC) will be allowed to go the bathroom alone?
From the LA Times:
This time around, the members of the National Election Pool â€” a consortium of five broadcast and cable networks and the Associated Press that commissions exit polls of the major races â€” have decided to sequester two analysts from each news organization in a secret “quarantine room” in New York, where they alone will get access to the first waves of data from precincts around the country.
Stripped of their cellphones and BlackBerrys â€” and even monitored when they use the bathroom â€” the representatives will be able to study the results of the surveys but will not be allowed to communicate them to their newsrooms until 5 p.m. EST. They must sign affidavits guaranteeing that they will not reveal any data before then.
More work needs to be done so as to better understand the effects of same-sex marriage initiatives on (Republican) turnout, so until then I am staying in the skeptical camp, as reported here.
Powered by WordPress
Bad Behavior has blocked 4698 access attempts in the last 7 days.