“Hello old friend.”

The United flight from Sydney comes in around about 10.30am every morning. If the 28s are being used SFO arrivals, we get an overflight or close to it: the flights from Hawaii, Auckland or Sydney typically arrive via the Woodside VORTAC (OSI) to intercept the 28 localizer, almost always on a vector that tracks right over Stanford and Palo Alto or Menlo Park.

Screenshot from PlaneFinderHD; click on the thumbnail.

A reader asked about updating the graph of income growth by presidential administration. The reader asked for median income growth (by income category, by party of the presidency), rather than means. Here you go.

The story looks pretty much the same, with the differences between Democrats and Republicans even larger, for all but the very top of the income distribution. At the 95-th percentile, median real income growth is 1.4% per year under Democrats, and 1.7% per year under Republicans; using means, the corresponding numbers are 1.9% and 1.6%.

So maybe the answer to the question in the title is “the rich” (if you prefer medians over means as point estimates — symmetric linear loss versus quadratic loss, if you know what I mean).

Using all non-unanimous roll calls cast in the 112th Senate, I find Olympia Snowe (R-ME) to come in as the 3rd most liberal Republican in the Senate. See the plots linked above or here (PDF). Snowe has her Maine colleague Susan Collins and next door neighbor Scott Brown (R-MA) to her left. Details below.

It is a little more interesting to think about what kind of Senator might replace Snowe. The picture below plots ideal points from the 112th Senate against Obama two-party vote share in 2008. Maine went 58-42 for Obama in 2008 — right along side states like WA and MI — and it is rare to find Republican senators from states this “blue”: e.g., Kirk (R-IL) and Brown (R-MA).

Should the Democrats pick up the seat, we might expect a Senator racking up a voting history a la Levin (D-MI), Stabenow (D-MI) or Cantwell (D-WA) or Murray (D-WA), right in the middle of the Democratic pack. I’ve seen some commentary looking at Democratic House members from ME taking the seat. See the voteview take (which utilizes a joint scaling of House and Senate members) and a perspective from Sydney, Australia.

If a Republican wins, I think its tough to say. Johnson (R-WI) is racking up quite a conservative voting history, given how “blue” WI was in 2008. My best bet would be that a Republican replacement would be voting quite similarly to Brown and Collins, right at the far left end of the Republican caucus, but who knows…?

It is straightforward to use the output of the MCMC algorithm in ideal to induce a posterior density (or actually, a posterior mass function) over (a) the order statistic of a legislator’s ideal point; (b) the identity of the legislator occupying a particular rank. All of this assumes that ideal points sit on one dimension (a pretty good assumption in contemporary U.S. roll call data).

The pictures below (click on thumbnails) show the story for the 112th Senate thus far. The model attaches most probability to Snowe occupying rank 56 (about 80%). She is by the most likely contender for rank 56. Brown and Collins are harder to distinguish, but both sit just to the left of Snowe.

I posted some analysis of the role social conservatives are playing in the Republican primary contest on the “Model Politics” blog at YouGov.com.

Will Super Tuesday be the last gasp for Santorum (38% in MI notwithstanding), before the Great Healing starts, and Romney starts to talking to a general election audience?

In any event, the key outputs — a table and a graph — appear below.

I also have a Sweave replication document available (here’s to transparency, and all that).

JPG thumbnail to PDF in new window:

Thought of the day re the Labor leadership fight…an interesting “feedback” effect of the polling on the issue.

With several national polls showing Rudd unambiguously preferred to Gillard as Labor leader, and by large margins – and the Labor caucus set to stick with Gillard – the die is cast for the next election.

That is, Gillard cements her reputation as an “unpopular” (at best) or even “illegitimate” leader. The electorate – or enough of it – hardens in its view that Labor isn’t listening … And returns the favor.

The next election was always going to be horrible for Labor, right up there with historic wipeouts a la 75, 96 etc. Sticking with a leader who is now widely known to be not “the people’s choice” helps lock in that destiny.

In turn, expect the vitriol against Abbott to continue and amplify. It’s about all Labor has now as an argument, and about the one thing the party can agree on.

This announcement came from the PolMeth list yesterday:

We are pleased to announce that “Frontiers in the Analysis of Causal
Mechanisms,” the first academic conference on statistical analysis of
causal mechanisms, will take place on March 24th in Cambridge,
Massachusetts.

The “first.” Really?

To which a clever friend replied:

Well, it’s happening in Cambridge, it must be first. I’m not sure what your point is.

I just found this cute re-rendering of the CRAN Task Views page (click on the thumbnail below).

The elephant for Bayesian inference is (of course!) a cap doff towards Basu and the circus elephant problem. The particular view of the elephant (!) does perhaps also highlight the fact that Bayesian inference is all about characterizing a posterior density…hee hee.

No prizes for guessing what is with the bell under the entry for probability distributions.

Perhaps the most garish web site I visit, the Sydney Morning Herald. I like their iPad app, but this has long been awful and could be getting worse.

Screen-grab, click for PNG in new window:

Seth Hill requests, I deliver: a look at the House roll call this morning, final passage (taking up the conference report) on HR3630, the Middle Class Tax Relief and Job Creation Act.

It’s something of a mess. Legislators’ ideal points (estimated on the basis of the previous roll calls of the 112th House) have very little predictive power. This is less true among Republicans (see the 2nd graph below), but the Yeas and Nays for this one are all over the place among the Democrats.

Thumbnails below will open JPGs in new windows:

I’m guessing the culprit here is the fact that there is a ton of content jammed onto the bill: extension of payroll tax reductions, continuation of unemployment benefits, self-employment assistance, Medicare extensions, TANF (Temporary Assistance for Needy Families), changes to Federal employee retirement contributions, and provisions concerning radio spectrum auctions. See Roll Call’s take here.

Larry Bartels’ book Unequal Democracy (Amazon) has many amazing findings, but one that always catches my eye is Figure 2-1.

That graph plots annual levels of income growth in the United States at various quantiles of the income distribution, grouped and averaged by party of the president. Under Democratic administrations, there is very little variation in average levels of income growth across the income distribution; under Republican administrations, poorer people see less income growth (on average) than richer people. At the very top of the income distribution (95th percentile) there is essentially no difference in average levels of income growth across party of the president. For just about everyone else, income growth is higher (on average) under Democratic administrations than under Republican administrations.

Like me, students find this intriguing, even enthralling. The finding leads to obvious questions like (1) “why would anyone ever vote for Republican presidential candidates”?; (2) “why don’t voters and the parties figure this out?” (3) “what the devil is the mechanism? (it is surely spurious, right?)”; and even (4) “is this a product of some outliers or some other oddity in the data?”

More generally, the idea that presidents might have this much sway over the economy goes against so much conventional wisdom (academic and lay): presidents aren’t generally thought to have this much sway over the macro-economy (but — my own view — perhaps they do have sway over the distributional consequences of macro-economic growth, via tax policy, spending priorities, etc).

One of my students had a shot at replicating the finding. I did too. In the end we asked Larry to direct us to the data. He’s working on an update that will appear soon.

In the meantime, I can’t resist sharing some graphs we made out of the data Larry sent through, which is quite simply, Table F-1 from the U.S. Census Bureau’s Historical Income tables.

Here is the update of Figure 2-1 from Larry’s book (click for larger version):

The dark lines connect the percentile average growth rates. Each plotted point is a yearly, real income change (percentage point), for the indicated percentile of the income distribution, by president. The four or five extra years we’ve had since Larry made his Figure 2-1 haven’t changed anything much; the pattern that Larry spends much of Ch 2 in his book analyzing is still there.

What is interesting is the amount of variation around the averages. Larry’s modeling (reported in Ch 2 of his book) takes all of that into account, but it is nonetheless quite striking when you see it graphically. In any given year, income growth bounces around a lot. Thus, presidents don’t exactly control the economy like a puppet-master; but, looking over the post-WW2 era, we’ve still got the finding that on average, almost everyone does better under a Democratic president. Go figure…or go read Ch 2 of Larry’s book for more.

The other thing I’ve done is highlight the one annual observation we have for the Obama administration (2010 income levels over 2009 incomes); we’ll get data 2011 over 2010 in March of 2012, and — using Larry’s classification scheme (a one year lag, of sorts) — 2009 income growth relative to 2008 is assigned to the Bush administration.

The darker dots on the left of the graph are the 2010 over 2009 changes. These all lie significantly below the Democratic average (unsurprising, given the slow pace of economic recovery in the US). Moreover, the hit to lower income households is quite dramatic, symptomatic of stubbornly high levels of unemployment in the US.

Which leads me to also wonder about using these dis-aggregated numbers in some election forecasting work. So much election forecasting work uses aggregate economic numbers. I wonder how an income growth number computed for middle incomes (a) differs from the aggregate growth number; (b) fares in an election forecasting model?