Monthly Archives: July 2006

Tar Baby

Leave a reply

I don’t like my Governor, Mitt Romney. And I think people who use the word niggardly just because its linguistic roots happen to be independent of the word nigger are obnoxious insensitive pedants. That said, I’m sad to see that there are some members of the black community who feel the term tar-baby is offensive to their community. The folklore wherein the tar-baby appears is an excellent story full of wisdom. It would be a shame to lose the lessons it teaches. One of which just happens to be how the weak minority can bring down the clever and powerful.

Mast Year, Network Failure, and Information Cascades

3 Replies

Tree’s don’t get around much, but they still engage extremely syncrhronized behaviors. From time to time all the trees of a given species though out a region will decide to throw a party. These are known as mast years. In these years all the trees in the region will produce vastly more seeds than in other years. It’s an orgy! The distribution of seed production/year is highly skewed with the majority of seeds being produced in these mast years.

I’ve been thinking about power failures, in particularly electrical power failures. Random failures in the power grid pop up all the time, but with surprising regularity large swaths of the power grid fail. I suspect that if you had a plot of the # of customers-days of various failures you’d get a highly skew’d distribution. We know a fair amount of why these grid failures happen. The grid isn’t a grid, it’s a scalefree network. If it were more like a grid then it would be more robust; but a grid is expensive compaired to a scale free network. The grid failures arise because a random failure hits some reasonably key component and then the rest of the grid fails as the problem cascades thru the network.

For example last summer, or the summer before, we had a power grid failure across the megalopolis on the east coast of the North America. The network was running at capacity that hot day when something near Ohio failed. As the load shifted the safety triggers on other components decided that they should resign from the network – to protect themselves. Each resignation accelerated the cascade and soon a hundred million people were without power. I found that interesting at the time because it makes a link between the issues of pure go-it-alone self interested capitalism and the issues of collective good. We have been playing out a recent enthusiasm for handing public goods over to private actors here in the US. These private actors have trouble successfully coordinating the building of enough excess capacity and reliablity into their networks. As the network failures become more likely the individual actors, seeing that their capital equipment is more at risk, tend to shift their safety triggers down; or at least i presume they would.

This year we had a example that’s worse, in it’s way, of a power grid failure. The grid in Queen’s New York failed. This time it appears the the safety triggers were set too high. Again during record load a component failed; but this time as the failures cascaded other components stayed loyal to the network with the result that rather than resign they committed sucide. Which is way bad because to reboot the system they have to pull new cables to replace the ones that burnt out.

Both those models are, to be clear, entirely speculative. But I’d love to know if after the first failure the guys in Queens went around and readjusted thier safety triggers.

The mass years, presumably, are information cascades thru some communication channel the species members have stumbled upon. I bet that when they figure it out they will discover that larger groves of trees play a role in triggering a successfull cascade.

Trees, like other members of the ecology, are embedded in an web of inter-species relationships. Observers have noticed that the mass years throw quite a ripple thru that web.  The squirrels get fat when oaks have a mass year.  They have lots of offspring.  The orgy cascades. The population bubbles and the next year it starves. This  pattern is actually good for the oaks; who would like to get their seeds past those pests.  During the abundant year many seeds get past the squirrels. The following year every acorn is found by now desperate squirrels.  By the third year most of the squirrels have died and the oak can again get a lot of acorns past those pests.

I bet there are similar patterns in the supply chain web after each of these power failures. For example I bet there comes season a bit after a large grid failure when you can get a generator really cheap from a vendor who was fat and happy just a season ago.

Polarization and Paralysis

Leave a reply

Here’s another interesting point from Polarized America.

When your designing your governance scheme one of the levers you can adjust is how much consensus is required before it’s possible to make major changes to the rules. For example here in the US it’s is very tedious to change the constitution. Another example is the Senate’s rules that make it impossible for a contentious issues to pass with a slim majority. There are lots and lots of these schemes; for example all the checks and balances built into the system.

So it’s no surprise that if the nation becomes polarized then the Congress becomes is paralyzed. That’s how the system was designed and it’s one of the patterns the authors of Polarized America illustrate that with data.

So then what happens? A few things. The two sides in the argument trash around looking for other means to achieve their goals. This model has something to say about the president’s repeated efforts (largely successful) to expand the power of the executive branch. This thrashing around attempting to find alternate ways get control of the government’s power is inherently dangerous because they skirt the boundaries of what is legal. The frustration of polarization creates an emotional climate where the political actors can self justify falling off the edge.

Because many of the programs that are designed to temper the concentration of wealth (i.e. programs that redistribute wealth) like the minimum wage, social services, education funding, health care funding, are not indexed to inflation this paralysis has the side effect of eroding their effect. Since this time around the primary poles of the division are about wealth that reinforces the polarization.

One notable thing about the models underlying Polarized America is the counter intuitive result that when you look at the actual votes in congress the social conservative dimension is a extremely weak predictor compared to the economic one. That’s counter intuitive because most of the rhetoric about American politics is about ethical and moral issues; e.g. stem cells, minor rights (race, gay, women), and the degree of separation between secular and religious institutions.

That too can be explained by this the realization that the architecture of our government means that a polarized you can’t make major changes.

The irony here is that the architecture is probably protecting the right from getting tossed out on its ear. The data is clear. The electorate has broad deep support for the redistribution programs that temper the corrosive effect of concentrated wealth. They also are an extremely tolerant bunch with little interest in the socal-right’s conservative agenda. The architecture has allowed the right to avoid the blame for eroding the redistribution schemes of economic liberals, and prevented them from the most socially conservative acts.

Conventional Wisdom or Propaganda?

Leave a reply

It is conventional to sing the praises of private schools while dismissing public schools as yet another example of Government’s ineffective nature. That rhetoric is part of the Republican campaign to undermine voter confidence in state based solutions. It’s a natural outgrowth of the Republican party’s real goal: to lower taxes and increase wealth disparity. You can’t achieve that goal if most of the public likes the services the government provides; i.e. health, transportation, safety, education.

So it’s with some amusement that I read this item from the New York Times.


WASHINGTON, July 14 – The Education Department reported on Friday that children in public schools generally performed as well or better in reading and mathematics than comparable children in private schools. The exception was in eighth-grade reading, where the private school counterparts fared better.

The report, which compared fourth- and eighth-grade reading and math scores in 2003 from nearly 7,000 public schools and ore than 530 private schools, found that fourth graders attending public school did significantly better in math than comparable fourth graders in private schools. Additionally, it found that students in conservative Christian schools lagged significantly behind their counterparts in public schools on eighth-grade math.

Conventional wisdom is wrong it is just Republican Propaganda. Public schools may not be as good as we’d like, but private schools aren’t any better.

Blog Payolla!

Leave a reply

A fun article in a recent the New Yorker about a radio station in New York City mentions in passing that Eliot Spitzer the NY attorney general is currently investigating a bunch of radio stations for taking payola. I’m kind of glad, if surpised, that there are still consumer protection laws on the books about that.

Meanwhile bloggers can now take payola, see over here.

This article about blogger payola is wonderfully ironic once you notice the amazing assortment of advertising techniques in use by it’s web site.

Polarized America

Leave a reply

I’ve been awaiting this book for months; and I finally got a copy. Polarized America: The Dance of Ideology and Unequal Riches by Nolan McCarty, Keith T. Poole, and Howard Rosenthal. I paid full price, which is both totally out of character and an indication of how very important I think this book is, but you don’t have to.

This book’s big picture is that very key statistics have all move near lock step very rapidly over the last 25 years in: income distribution, political polarization, percentage foriegn born, declining local government services, and others. These are connected in a complex dance.

Here’s something I didn’t realize. While the Republican party has been moving rapidly right; the Democratic party has on the one hand abandoned their advocacy of general welfare platforms and shifted toward a what they notably refer to as issues of ascription. These are direct decendents of the civil rights movement. These are issues about ascriptive characteristics (race, gender, sexual preference) of individuals. We should reclaim the general welfare issues!

Availablity

Leave a reply


Here’s some more fun stuff gleaned from the book I’m reading.

One of the canonical reasons for poor judgement is known as availablity; i.e. we think and work using the tools that are easily at hand. If a tool is beyond reach we are far less likely to apply that tool. There are lot of examples. The folks that plan for floods, for example, rarely plan for floods other than those they have experienced. If your asked a mess of people to estimate how many words start with the letter r, like road, v.s. how many end with the letter r, like tour they will get the wrong answer, begins with, because your noodle is set up to recall words by their first rather than their last letter. I love the variant of that example; people will guess that abstract nouns, like love, hate, truth, are occur more commonly than concrete knowns, like door, street, bicycle. Apparently people have an easier time recalling abstract nouns v.s. concrete ones. Who knows why, possibly because the abstract nouns cover broad swaths of experiance.

In reading about availablity, and the other insights outlined in this book, I find myself thinking about how they get mixed into the ways people market products, debate issues, manage planning, and even how you could create carnival games out of them. One reason a component manufature has to give way a huge number of free samples is to assure that his components are highly available at the moment when some random designer is reaching out to pick something up to solve his problem. That’s one reason Open Source works; it adopts a r-Strategy.

This example of availably is particularly thought provoking. A reasonably standard formal way to estimate risk is to create a fault tree. You can view lots of examples at Google Images. The process of making a fault tree is lot like the process people use in making any plan. You attempt to collect all the contengencies and then sort out their likelyhood and dependencies. For example in thinking about why the car might not start you might include dead battery and left headlights on by mistake. Professional planning and fault tree work involves getting experts to help. You might do that by doing surveys and investigations of various car starting falures; or you might interview experts (say are mechanics) and have them help generate contengencies and probablities of those contengencies.

Because of availablity asking the experts does not work.

Say you have methodically assembled a large and nominally complete fault tree and because you doubt you’ve thought of everything you include a node for “other causes.” At that point your fault tree covers a 100 percent of the failure scenerios. You go to the experts and ask them to help by assigning 100 points (i.e. percentage chance) across the perimeter of the tree.

Why doesn’t that work?

It doesn’t work because studies show that you get radically different answers from the experts if you check their work by varying the fault tree you present. If you have a tree with 10 nodes and you collapse 2 or 3 into the “other causes” node the experts will then raise their estimates for all the remaining causes. Because the nodes you merged into “other causes” are no longer easilly available.

Of course what’s in that “other causes” node is the long tail of contengencies.

Learning at the Knee of a Random Number Generator

Leave a reply

Here’s a thought provoking intersection between behaviorism and statistical thinking; plucked out of this book.

Here’s a very naive model of behaviorism: If the animal behaves well we reward him, and if he behaves poorly we punish him. This primitive version of behaviorism is fraught with problems; but it will do for this discussion.

A very naive statistical model of behavior breaks into two parts; with average behavior and random behavior around that. Animal trainers are experts in leveraging that random bit. If you want to train a goldfish to swim clockwise then before you feed him then you wait for the random clockwise turn before feeding.

Of course animal training is a two way street. Pets expend a great deal of effort attempting to train their owners to feed them. It’s fun to walk in front of the tanks at a large pet store wearing the same color shirt as the staff an watch the fish attempt to trigger your into feeding them.

The hope of training is that you shift the average; but that takes time and in the meanwhile random variations will generate occasional good and bad performances relative to the mean.

So here’s the rub. If the animal behaves in an exceptionally good or bad manner it is likely that in following period his behaviors will return to the previous average behavior. In the jargon of statistics this is an example of regression to the mean.

Regression to the mean is enough to train a bad behavior in the trainer! Consider this scenario: the animal behaves well, the trainer rewards him, and then in the following training rounds the animal is certain to regress back toward his average behavior. The trainer learns from that the reward triggered the regression. The trainer learns not to reward. Which is bogus.

But it get’s worse. Consider this scenario: the animal behaves badly, the trainer punishes the behavior, and then in the following rounds the animal’s behavior randomly regresses back to the mean. The trainer learns that punishment precedes behavior improvements and; not because the animal is learning anything but because the statistics say so. The statistics alone are enough to train the trainer to punish bad behavior but not reward good behavior.

This is very bad!  It’s a fundamental insight of sophisticated behaviorism that punishment is far less effective than rewards; so much so that punishment often doesn’t work at all! Only a stupid or a captive animal will put up with training based on punishment. This problem is redoubled because if you punish the animals become more cautious; which reduces the not just the random variance (which you need) but also suppresses the smart active searching you want the animals doing. Punishment doesn’t work on smart animals and when you can get away with it makes the animal become stupid.

What a mess! Notice that you don’t need an animal to teach the trainer this bogus behavior pattern of never reward, but do punish. If you had the trainer work to train a random number generator he’d learn the same lesson. There really isn’t a more stupid beast than a random number generator. Which I think goes a long way toward explaining why fans of punishment often describe the animals they are trying to train as stupid.