The intellectual problem of speciation over time -- how can an organism be infertile with its direct ancestor  
 
Steve, the best molecular explanation I've seen so far revolves around stable chromosomal inversions. These are large-ish segments of DNA that are basically intact but pulled out and flipped around in the opposite orientation.  
 
If you have two copies of the same inversion, the sister chromatids can pair up during meiosis without a problem, so that you can generate healthy gametes. If you don't have two copies, you often can't generate healthy gametes b/c recombinations in that region can mess up.  
 
Leaving aside the chicken'n'egg problem of getting the frequency of a given inversion high enough in the population (it could be done if recombination was locally suppressed in the region for a few matings), there is a long term consequence: the more incompatible inversions two partners have, the lower the fertility of their offspring (due to messed up meiotic events). 
 
Over time the idea is that this causes speciation. You could test this by forming an N x N matrix of mating events. In each cell put both the average number of grandchildren and the number of incompatible inversions. You should see a negative correlation between the two values.  
 
Google "inversions speciation" for more. 
 
The bottom line is that speciation is a continuous process that starts with subspecies becoming *less and less* interfertile. It's not an overnight binary thing, and often doesn't become binary for millions of years. That's why ligers and mules are still possible. Heck, horses and donkeys have different numbers of chromosomes but mules still exist.  
 
I have read that in Soviet Russia they did experiments to see if humans and chimps could interbreed. I'd expect success probability to be low, but would not rule it out. Perhaps the most un-PC speculation of all would be to build a haplotype network graph and look for the human subgroup with the smallest number of fixed nucleotide differences from chimp. Obviously all humans are quite different at most loci, represented as a long bridge with say 20 base differences to chimp...but if your Soviet scientists wanted to maximize success... 
 
oh, and actually, before you even did it w/ chimps & humans, you would probably start doing the same haplotype network and inversion type analyses with horses & donkeys and lions & tigers. That'd give you some more quantitative insight into the relationship between recombinational incompatibilities and interfertility.

yeah, isn't loess awesome? That subtle incline towards the end would be imperceptible in a linear regression or by eyeing the cloud.  
 
But it squares with what people have observed about the *ultra* white districts being less familiar with what minorities are actually like, making them more prone to idealization (and hence Obama voters).

200 years is not a trivial interval of time, especially when taking into account the large numbers of Americans who lived between then and now.  
 
Yes. In terms of recent history, 200 years is a very long time. Lots of stuff happened, most people who have ever lived were born during that time, many inventions happened, and so on.  
 
But from an imagined perspective 1000 years in the future, say when the universe population is 1 trillion people, I could imagine people binning the US together with other countries in this kind of manner.  
 
Specifically, I think that future historians will bracket the collapse of the USSR and USA as twin events occurring say 25-40 years apart (1989 and 2014-2029). Once hyperinflation destroys the US economy, we'll look back and say that both were consumed by egalitarianist viruses similar to those that devoured Weimar and Rhodesia.  
 
Needless to say, this is very different from the near-term End Of History view in the early 90s.

Mike, excellent points. I had heard of the Fogbank fiasco...reminds me of how fragile civilization is (in Zimbabwe, they've gone from electricity back to candles in 15 years).  
 
The low-level device engineering -- i.e. the physical contact with the real world -- is sorely needed for those fancy APIs. One example that comes to mind for me is the recent memristor -- without innovation in the low-level device stuff, you'll never be able to build a nice python interface on top of it.  
 
I remember the first time I read Kailath's book on Linear Systems...I was struck by (a) the elegance and (b) the difficulty of implementing the analog feedback in Watt's governor. The first thing I thought of was how tightly coupled the physical system was to the control system, and how different this is from the modern division between "dumb" sensors/actuators and smart control. Relatedly, it took me a while to realize that IIR filter theory comes from constraints on what can be realized in an analog system. 
 
Do you blog anywhere? I like to think we have a pretty good handle here at GNXP on modern CS/genetics/stats, but you seem to have some more old school engineering chops and I think it'd be educational to read your musings.

Interesting comment, mike. This bit jumped out at me, though:  
 
virtually all forms of modern day engineering seem to have atrophied into mere shells of what they once were as analog devices and machinery are replaced by microchips.  
 
There are significant advantages to doing as much as you can in software rather than hardware. Makes you more nimble.

By the way, I've often thought that it would be amusing if someone wrote a virus that installed Firefox, Chrome, Safari, or even IE8 and removed IE6/IE7 as the defaults. They would be heros...probably never have to buy a dinner again in tech circles.  
 
(It might be even possible to uninstall Windows and install OSX on x86 platforms, but that might actually break a lot of people's installs, which would be inconvenient).

In other words, it is an OS for all but 1/10000 of the target audience for Windows, Macs, Linux, etc.  
 
But it's not that uncommon. Macs are at least 10% market share and growing steadily. They are superior for anyone who does not play games. I say this as someone who used to write DirectX games in Windows for fun. The only thing better about Windows (and it is not totally trivial) is that you can resize windows from all four corners and that the maximize button works.  
 
Also -- I was intentionally being a mite provocative, but I am seeing the tea leaves. MS has been alienating both techies *and* the general public. It is notorious for spyware, its PR has been bad, and it is *very* poor for scientific programming and server-side development. Client side development is ok if you spend the time to learn the whole Windows stack, but there are so many advantages to developing in the cloud (especially incremental updates!) that this is a dying programming model.  
 
Let me put it like this: I know of exactly *one* impressive startup (Stackoverflow.com) that uses MS technology. But it's the exception that proves the rule. Similarly, in bioinformatics, only wet lab scientists use Excel for anything but the most trivial of calculations. The places that shape the future of genomics -- NCBI, Whitehead, Wellcome Trust, ILMN -- are either exclusively or preferably into Unix.  
 
Basically, the people who build and program computers determine where the future of the industry is going. And that future is not on Windows.

Jason -- MapReduce, AJAX, GFS, BigTable are all serious breakthroughs. And those are just some of the published ones off the top of my head.  
 
MapReduce and AJAX have literally spawned whole industries based on imitation. Hadoop is a MapReduce clone that became Yahoo's backend and then MS's backend (after the Powerset acquisition). And even nontechnical people understand the impact of AJAX.  
 
Lots of other improvements are subtle but nontrivial. As one example, think about how Goog's indexer seems to always have up to date versions of web pages. That is in part due to Grimes' et al.'s statistical recurrence model which uses ideas from Kalman Filtering to predict when a page will next be updated based on its past history.  
 
They have made too many breakthroughs in machine learning to name, but here's a short list of apps which have some serious shit under the hood: Google Translate, Google Maps/Streetview, Gmail spam filtering, Goog 411... 
 
Bottom line is that Goog is the big dog today in terms of CS innovation. If you know some of the people there, or what's going on inside, there are things they are doing that no one else can do -- both because of the concentration of talent and the scale they're working at. For example, take a look at Barroso and Holzle's recent 108 page writeup on warehouse scale machines:  
 
http://everythingsysadmin.com/2009/05/warehouse-scale-machines-the-d.html 
 
Or Jeff Dean's talk on Goog's backend:  
 
http://glinden.blogspot.com/2008/06/jeff-dean-on-google-infrastructure.html 
 
I believe in giving credit where it is due. Goog is still the most innovative big company in tech. Apple is a very solid second. After them I tend to start thinking of startups rather than MS, IBM, Yahoo, etc. 
 
(With two caveats. First, search.yahoo.com has superior cross-keyword recommendations relative to Goog. Second, Bing's image search is superior to Googe. I've often wondered why someone doesn't just put Orbitz or something like that inline on a search engine frontpage...I guess that day is quickly approaching).

Last thing. MS is the most technically reviled big company around today because they actively fsck web developers. At first MS did this on purpose, but now it's become a habit, forcing web developers to respond in innovative ways. See:  
 
fixoutlook.org 
ie6update.com

Just had a thought -- Macs vs. Windows are kind of like our political climate.  
 
The Democrats are an alliance of the technical overclass with the underclass. The Republicans roughly represent the middle class. And all the momentum is with the Democrats right now, because they appeal to both the unwashed masses *and* the college grads.  
 
In the same way, Macs appeal to the computing underclass (grandma) and the overclass (MIT CS PhD, Bay Area hacker). Windows is the OS of the American sarariman middle class, who knows how to hit Ctrl-Alt-Delete but wouldn't know a thread from a process if their life depended on it. And again, the momentum is only in one direction.

By the way, for anyone interested in *why* Microsoft died intellectually, I recommend this WSJ profile. There is one key sentence:  
 
http://online.wsj.com/article/SB121261241035146237.html?mod=googlewsj 
 
Gradually, Mr. Ballmer made his imprint. He restructured the company to give more decision-making power to executives, and elevated people with general management experience into positions previously held by technology-focused executives. He also worked to settle Microsoft's many lawsuits, taking a more conciliatory line than Mr. Gates typically had, Microsoft executives say. 
 
Boom, that's it right there. Ballmer hired MBA morons like this:  
 
http://www.businessweek.com/bschools/content/jun2007/bs20070607_329811.htm 
 
In my post-MBA job hunt, Microsoft was not the most obvious fit?I'm not a very technical guy. On my first day at Microsoft it took me 30 minutes just to find the latch to open my laptop (though I did successfully find the "on" button pretty quickly). I think that's why my MBA at Kellogg has played such a vital part in my career development.  
 
Yeah, you *think* that he's being self-deprecating. Read the rest and the smile will fall off your face. 
 
Ballmer also killed key projects because they would cannibalize the co's revenues. This is what led to the rise of Google Apps:  
 
 
In one case, two vice presidents clashed over the future of NetDocs, a promising effort to offer software programs such as word processing over the Internet. The issue: Because NetDocs risked cannibalizing sales of Microsoft?s cash-cow Office programs, some executives wanted NetDocs killed. 
 
Messrs. Gates and Ballmer were unable to settle on a plan. First, NetDocs ballooned to a 400-person staff, then it got folded into the Office group in early 2001, where it died. 
 
 
That in a nutshell is why monopolistic organizations almost never innovate themselves away.

The most important point here is whether large monopolistic organizations (such as the military or Bell Labs at its peak) innovate as cost-effectively as startups. I don't think they do.  
 
The most innovative consumer-facing big companies are Goog and Apple. And tons of their stuff comes from startup acquisitions rather than internal development.  
 
That said, there are certain kinds of "big science" that are only feasible for large organizations with big budgets. Sequencing the genome is the canonical example. But then again, just look at the innovative shot in the arm that whole enterprise got once there were *two* horses in the race! Competition is a driver of innovation.  
 
Their major contributions followed after they dominated the search engine market, though: Blogger, Google Maps, Google Books, etc.  
 
First, dominating search required tons of innovation *and* acquisition. Kaltix's block eigenvalue computation and Ori Allon's contextual search were both acquisitions -- and there are dozens more that don't get any publicity.  
 
Second, Blogger was an acquisition of Evan Williams' company. Google Earth was an acquisition of Keyhole. Google Voice was an acquisition of Grand Central. Maybe half of their cool stuff sprung from acquisitions of innovative startups rather than in-house development.  
 
This is not to take anything away from Google, which is probably the most innovative big company around, but most big cos aren't Google. They have layer upon layer of Assistant to the Regional Manager.  
 
Also, to address a few other points... 
 
1. Microsoft and Google are pretty monopolistic, and they've been delivering cool new stuff at low cost  
 
What innovation has MS come out with? Honestly interested here. The cool stuff that is under the MS brand are things like Project Natal and Photosynth, both of which (again) sprung from acquisitions of innovative startups.  
 
2. Overall, though, the pattern is pretty clear -- we haven't invented jackshit for the past 30 years.  
 
See above re: CDMA, etc.  
 
3. we'll just note that most major inventions could not have been born if the inventor had not been protected from competitive market forces  
 
I kind of see where you're coming from with this, but you take the thesis too far. The Industrial Revolution gave the commercial incentives to advance thermodynamics (viz. all the engines). The Information Era gave similar incentives to advance computer science and electrical engineering.  
 
There is a distinction between "isolated from day-to-day market pressue" and "part of a monopoly". The small university research lab doesn't have monopolistic amounts of resources, but they do have time.

Microsoft is way more innovative, on a smaller level, than other related companies (like, e.g., Apple)  
 
Huh? Apple is the first company that married a modern frontend (Aqua) to a hard core backend (Unix/BSD). No one who knows how to code uses Windows over Macs. Check out MIT CS or Bay Area startups some day...the people who are writing the code for the next generation of applications are all on Macs. That means machine learning, physics, genetics, systems programmers, you name it.  
 
With darwinports or Fink you can get all the power of the GNU tools. With Rails or Django you can run the exact same webapp on your local laptop that you'd run on the server. You have the same suite of poweruser tools on the frontend that you do on the backend -- you don't have to context switch. Developing where you deploy has huge advantages.  
 
I pity the data analyst who tries to analyze TB scale data with Excel (!) rather than GNU cut/paste/awk/etc or the web designer who has to deal with IE's CSS bugs [which is all of them].  
 
The only things Windows is superior for are gaming apps. It is an operating system for children and people scared of the command line. But it is now time to put away childish things!  
 
Another good ref:  
 
http://www.paulgraham.com/microsoft.html

As a general rule, overweight women are better served by spending zero time on fashion and 100% time on exercise/diet *until* they're at a good weight. After that point there is some marginal benefit from fashion. 
 
Until that point, they should go with XL t-shirts and jeans. Won't make a difference what they wear...unfortunately for them, they're invisible . Lemmonex learned this the hard way.  
 
PS: Free weights are also a big win -- girls who do squats and lunges (with barbells!) are getting the maximum bang for their exercise minute.

agnostic --  
 
Aside from the hemline vs. male/female ratio relationship (I take no position on that till I look at a time series), there are strong evolutionary reasons to presume that women will become more forward when the male-to-female ratio is low.  
 
Case study #1: Vassar college. Formerly an all girls school. Last I checked, the ratio was about 66/33 female/male.  
 
With predictable consequences.  
 
http://www.miscellanynews.com/2.1578/hot-and-bothered-vassar-women-must-work-to-meet-physical-emotional-needs-on-campus-1.1573390 
 
This column is dedicated to all those single, straight Vassar women out there. I hear your roar. While, in the classroom and in the field, Vassar?s female-empowered environment encourages women to live the confidence and assertiveness typically reserved for men, when it comes to the sexual arena, Vassar?s gender ratio takes on a whole different meaning. 
 
Take your typical Town House party. Since most college students are not of age, and thus not at bars, this party is the social hub of the night. The yard and interior are swarmed with people. 
 
If you?re a guy, probably five out of the 20 girls you have been flirting with are there. You have five potential pursuits. These options infuse you with a sense of confidence. If your game does not work on the first girl, well, at least you have four more chances to get it right. Not to mention the girls are probably horny and available simply by default of the Vassar ratio. Statistically speaking, your chances look good. 
 
If you?re a girl, maybe one or two out of the five guys you?ve been flirting with are there. You have two potential pursuits. But wait, there are three other girls already eyeing one of the guys. You look over at the other one and see him chatting it up with a couple girls. Statistically speaking, the odds do not look good. 
 
What are we to make of this discrepancy? As a woman, I do not feel comfortable trying to appropriate the male experience. What I can say, through observation and conversation with my male friends, is that straight men at Vassar have a much easier time navigating Vassar?s hook-up culture. Let me expand upon this. If you?re a female looking to hook up, there are a number of added obstacles you will probably face simply because of your gender. I believe that the two following scenarios capture a generalized representation of the frustrations of Vassar women.

Meta-analyses are usually crap because they usually combine p-values (often by naive application of Fisher's equation) rather than actually concatenating the underlying datasets (which have all kinds of different rows and columns). This problem is compounded b/c the p-values are often not intercomparable.  
 
That's a bit different from what was done here. Moreover, Lynn's original study was meta-analytic in nature (at least in the sense of aggregating papers).  
 
Still, no doubt Wicherts spun the paper as far more favorable to the nurturist view point than it is. The real upshot is that even after correction, Africans have the lowest IQ among all pops measured in the study...

Just took a look at the paper.  
 
Wicherts' plots show a few things pretty convincingly:  
 
1. The IQs for African subjects are likely to be significantly higher than estimated by Lynn and Vanhanen, yet still the lowest in the world. Wicherts shows this by comparing national IQ to Hanushek's data, Rindermann's data, and the TIMSS and demonstrating that the sub-Saharan African data points are off the regression line (Figs 1,2,3).  
 
2. Lynn seems to have cherry picked the lowest estimates of African IQ scores to obtain a median around 69-70, but the actual median is around 78-82 -- again still the lowest in the world . See Figure 6. Note that an obvious retort might be that Lynn's studies could have been larger. However, Wicherts anticipates this by putting the standard error on the Y axis, which is inversely related to the sample size. I'd like to see a third axis of time added to the plot (to see Flynn type effects) but I think Wicherts makes his case.  
 
Overall I think my initial prediction was right. Lynn is correct about the general trend of IQ being the lowest in the world in sub-Saharan Africa. But for reasons only known to him, he consistently and unnecessarily overstates the case. Did he really need to cite the median around 69 when it's around 78? It just leaves the door open for critics. Anyway, this is a pattern with him (and to a lesser extent Rushton[1]) which is starkly different from someone like Jensen.  
 
All that said, there's no shortage of people lining up to attack Rushton and Lynn and I don't want this to be construed as a typical nurturist jeremiad. My basic POV is that Rushton's rule is up there with Chargaff's rule in terms of getting at a fundamental biological truth which we don't fully understand yet, and that "IQ and the Wealth of Nations" is an important conceptual breakthrough. Even if Rushton/Lynn are angry or racially motivated to an extent, and even if they have overstated their case, nevertheless there is a fairly large baby within all that bathwater no matter how much many (including myself) would like there not to be.  
 
This is basically the same way I think about Lewontin -- sure, he's a communist and persecutor who lied about human differences. But you have to give the devil his due for pioneering the population genetic application of gel electrophoresis. All three guys are extremists who've nevertheless made far-ranging contributions to science.  
 
[1] Razib can speak to this, but there are several aspects in which Rushton's understanding of genetics is rather limited. For any other field and discipline, that wouldn't be such a big deal -- molecular biologists don't know any popgen either. But w.r.t. race/IQ/genetics IMO you really do kind of have to be a polymath, or at least try to be one and be apologetic/honest when you don't know a certain area that well.

Also -- w.r.t. this particular paper, I do like the fact that Wicherts included scatterplots rather than correlations (the former are far more informative). I'll sit down and look it over, but until then -- Wicherts is technically skilled and a rather personally reasonable guy. He's as far from Kamin, Turkheimer, Rose et alia as you can be -- similar to Flynn in that respect.  
 
Lynn on the other hand is not very technical and can be quite sloppy in his work (e.g. his recent post on VDare cited an incorrect racial order for A-levels -- African immigrants actually outscore UK whites on some A-level related measures). But Lynn is definitely on to *something* with IQ and the Wealth of Nations.  
 
So -- while I think the broad gist of Lynn's thought will hold up, I wouldn't be surprised if Wicherts has something here.

If you want my $.02....I've said for a while that definitive settling of cross cultural IQ comparisons will only happen when we can measure intelligence in ratio-scale terms, e.g. in inverse seconds (like computer clock speed) or cubic centimeters (like MRI brain volumes) or some (simple) function thereof.  
 
A multiple regression of pen-and-paper IQ upon all neurophysiological measurements known to correlate with IQ (brain volume, reaction time, etc.) would likely give you something highly predictive of IQ *yet* based entirely on physical measurements.  
 
And physical measurements are of course time, space, and culture invariant. So you can now ask: is the Flynn effect due to genuine differences in (say) cranial volume? Or not? If you've got a strong predictor (say >.9 correlation with IQ), you would almost certainly be able to track the Flynn effect to the determinative columns in the multiple regression matrix.  
 
Solid precedent for this are the brain imaging studies that began regressing IQ upon *multiple* brain volume measurements (of individual regions) rather than a simple scalar aggregate (namely the row sum, representing the entire volume for that individual). Areas known to be linked to cognition from prior information were most highly predictive of IQ.

Agnostic: road rage is about intentional attacks. The accidental death plot need not show a positive correlation.  
 
Note: I seriously doubt that road rage is a statistically significant phenomenon, but these plots do not demonstrate that. What you would want instead are arrest rates for things like vehicular manslaughter.