- The Black Banners: The Inside Story of 9/11 and the War against al-Qaeda
- Neil Shah
- 3 wds: Captivating, informative, redacted
- The Blank Slate
- Steven Pinker
- 3 wds: Rant, lengthy, correct
- The Trial
- Franz Kafka
- 3 wds: Disheartening, linear, skippable
- I Am A Strange Loop
- Douglas Hofstadter
- 3 wds: Deep, insightful, correct
- Why We Lost The ERA
- Palle Yourgrau
- 3 wds: Thorough, researched, slow
Tweets"The union [say the closures] disproportionately affect Afr Amer kids" -- well do they @LyndseyLayton?? You're the reporter, please tell me!
1 day agoI think we're at 60 votes for background checks if you count letters to constituents http://t.co/gtJ7KNg1sv #TakesSomeBrass
1 day ago
Disclaimer: I’m a monthly donor to IPA
Innovations for Poverty Action — a very cool non-profit that conducts randomized trials of poverty alleviation programs —
has posted two “NYC Taxi Cab” ads to YouTube. The scuttlebutt on Twitter is that these ads are part of IPA’s own randomized A/B test to find out which ad works best at generating donations. Here’s the first ad (and a link to the second):
I love A/B tests of ads and can’t wait until YouTube makes an A/B tool widely available. Good for IPA for taking their organization’s mission and applying it to their own actions. However, I think IPA’s test has a few flaws that provide good lessons for others who want to gain knowledge about their advertising.
1) In order to track which donations came from which ad, IPA set up two different textable (SMS) key words on the same short code. To donate $10 to IPA, you can either text “IPA” (ad 1) or “action” (ad 2) to 80888. The difference of the memorability of those words might lead to differential giving patterns regardless of the ads’ relative effectiveness. If someone doesn’t have their phone in hand when the donation instructions appear on the screen, can they remember the acronym IPA? Maybe if they like the beer…but even then I think an English word is easier to remember, and thus could lead to more donations.
2) To rigorously test one ad vs the other, the ads would have to be sent randomly to the back seats of NYC taxi cabs. Perhaps the taxi ad placing agency, Verifone, has true randomization built into their system, but I worry that Verifone says to clients like IPA “yes, we can randomly place ads” while confusing arbitrary assignment with random assignment. Under arbitrary assignment, ad 1 might end up being seen more on the upper west side, and ad 2 ends up being seen more in the Bronx. NYC has large income and cultural disparities between nearby neighborhoods that could disrupt the experiment.
At the end of the day, my expectation is that issues (1) and (2) only create minor problems, and that if ad 1 were five or ten times more efficient at spurring action than ad 2, IPA would see the difference in their donation results. But that line of reasoning leads me to my final and largest concern:
3) The ads are very similar! I donate to IPA because I want them to test large ideas against each other — such as microfinance vs direct giving. In this way, a few good experiments can really shake up the poverty intervention landscape. With regard to the taxi ads, imagine that you work for IPA that ad 2 does a bit better than ad 1: what would your conclusion be? (Other than people don’t mind the shirt.) Since both ads have a narrator giving a straightforward explanation of the organization’s mission, I’m not sure what I would take from the test. The angle of each ad is slightly different (donor POV vs. scientist POV), but I would have much preferred IPA test two completely different messages or tactics against each other. For instance, staid vs zany, American POV vs African POV, or single-camera vs crowd-sourced. I realize that some of these ideas might be challenging to the creative team, but my position (if I were sitting at the IPA conference room) would be to hold off on the test until Creative had at least one off-beat idea that would contrast with these straightforward ads.
In general, A/B tests of video is a burgeoning field, and I’m glad IPA is leading the way so we can learn both what worked and what could be improved. Hopefully they’ll publish their results!
 Also, to be fair, there aren’t obvious solutions to the first two issues. IPA could spend money on a second short code, but of course the short codes might have variable memorability as well. (And getting deep in the weeds: IPA should have used the “IPA” short code in ad 2, which actually mentions the acronym in the ad’s narration.) For (2), we need more organizations to think like IPA and ask buyers to include true randomization in their ad placements./a
Sarah Kliff (one of my favorite reporters) has a interesting one-off piece on how we perceive the nutritional content of food. Apparently if you only glance at these two candy bars,
then you’ll think the one with the green label is healthier.
Apparently, Reese’s has been onto this type of psychology (or at least branding) for a while — check out this fine print on the back of a Reese’s Pieces:
I’m anxious for self-driving cars to arrive, as I’d love to make better use of commuting and errand time. And while I do expect many aspects of the automated driving experience (highway driving, parking) to become mainstream shortly, this Business Insider article (h/t Jon Kennell) misses the mark. BI identifies three aspects of driving that could cause problems for automated cars:
- Bad maps
This list is extremely superficial and misses the deeper issues. For each of these potential problems, the car has plenty of time to slow down and hand off maneuvering to a person behind the wheel (who likewise would have enough time to put down their cell phone). Google (or another developer of this software) could instruct the car: (1) never self-drive in snow, and (2) look a long distance ahead for road or lane closures (which would take care of both the maps and construction issues).
Where Business Insider goes wrong is that it’s thinking like a human driver. Humans have trouble with snow, sometimes get lost, and–when distracted–miss oncoming construction or traffic cops. Avoiding trouble or an accident in these situations is dead simple for a computer: at worst case, slow down and stop.
Computers, unlike humans, don’t get distracted; rather computers have a lot of trouble with a talent that humans have in abundance — quickly incorporating subtle clues to predict what will happen next. For instance, imagine your a car going left to right in this picture of a pedestrian in the middle of the road:
What do you do? You make sure that the pedestrian sees you (by a quick glance up and a slightly slowing of her gait perhaps), and you continue driving at pretty much the same speed because you believe with very high probability that she will: stop at the double yellow line, let you pass, and then continue walking across the street.
That type of reasoning is very hard for a computer. In fact, there is a good chance that Google–desiring to maintain their perfect automated driving record–will develop software that is too conservative (video example). Maybe when a pedestrian darts halfway across the street, an automated car will presume that the pedestrian will continue moving in the same direction (not a terrible assumption), putting the jaywalker in the direct path of the car. In that case the car will have no choice but to stop suddenly. I fear a very jerky ride.
Another example of a potential inconvenience: on my way home there is a right-on-red turn with a lightly-trafficked cross street. Even when there is a car at the intersection in front of me, I often don’t slow down as I approach the intersection because I know with high probability that by the time I reach the intersection that car won’t be there anymore. Will an automated car be smart enough to do the same? Seems unlikely, at least in the early versions.
These prediction problems are the real issues for self-driving cars. Undoubtedly, that’s why Volkswagon is starting with “auto-pilot” on the highway, which is one of the easiest places to predict what will happen next. (Although even in this case, I wonder what Volkswagen’s software does when it sees a deer grazing on the side of a forested highway road.)
In sum: there is no such thing as a “perfect driver”. We all make trade-offs–balancing an efficient, smooth ride for the possibility that a low-probability event (drunk driver in other lane, irresponsible pedestrian, deer crossing) will result in an accident. Which trade-off points are programmed into automated driving algorithms is a very intriguing set of decisions.
Update, March 9: Ezra Klein wrote Wonkblog post that deals with many of the issues below and the fissures within the Republican Party. I bet he’d find similar fault lines within the Democratic Party as well.
On Twitter yesterday, I had a small debate with two people–Matt Yglesias and Brendan Nyhan–who are smarter than me when it comes to policy. I argued that there’s a chance that copyright and other issues that pit small actors (e.g., start ups, small businesses, creative artists) against large actors (e.g., existing IP holders, large companies, entrenched interests) might create a second ideological dimension in Congress. (The first dimension is the partisan liberal-to-conservative/Democratic-to-Republican spectrum.)
In the mid-20th Century there was a second dimension on race issues in Congress (see the Southern Democrats), but no longer. Nyhan relates his work on a current 2nd dimension this way:
— Brendan Nyhan (@BrendanNyhan) February 28, 2013
My initial tweet linked to an article that argued that pro-reform copyright issues were a natural fit for Republicans given Dem’s alliance with Hollywood. But, since Republicans are primarily funded by big business (who hold a lot of IP), this argument seemed off to me. Rather, perhaps a “small actor” alliance could form between members of both parties.
Admittedly, the chance of this alliance forming and enduring is small. Yglesias rightly mentions the issue of abortion, which took a decade to align itself along the primary political axis, but finally did. (See Achen and Bartels for the political science behind this self-sorting alignment.) If the visceral issue of abortion couldn’t start a second dimension, how can copyright?
Below the evidence that, at least for the moment, small v big actor issues do cut across the two parties. Perhaps if the issue stays in the background, activists and elites will not feel compelled to align their stances with their partisan affiliations. Of course, in that case, there will be few roll call votes to form a statistically relevant 2nd dimension. Respecting those caveats:
- The Leahy-Smith America Invents Act, which at a brief glance, appears to be pro-large actor patent reform. Sponsors were Leahy (D) and Smith (R). Here was the very cross-cutting roll call.
- Jon Tester’s Anti-swipe fee amendment, which pitted big businesses v small businesses:
- Pandora’s Internet Radio Fairness Act of 2012, sponsored by Wyden (D) in the Senate and Chaffetz (R) in the House. It didn’t go anywhere. I couldn’t find an actual roll call vote on this issue; if someone could point me to one, that’d be great.
- Bill McCollum’s (R-FL) pro-small business amendment to the Sonny Bono Copyright Term Extension Act. Democrats were split more than Republicans on this vote; perhaps surprisingly, few Republican’s backed their own colleague’s amendment.
- Berman’s 2007 Patent Reform Bill needed a coalition of Democrats and Republicans to pass the House (with a measly 220 votes).
After briefly reviewing (some of) the evidence, I conclude that there is an underlying second dimension on these small v big actor issues in Congress, but there probably aren’t enough roll call votes for this dimension to pop in the data. Given this current reality, I’ll stand by my position that there is a non-zero chance that this dimension appears in the future. I would just need long odds to bet on it.
Gary King asked the question “What If the Obama Campaign Didn’t Win Him the Election?” and makes the fair point that Obama’s battleground state vote percentage does not look very different than the other states.
On Friday at the Google Political Innovation Summit, a contingent took the side that because we can’t see the effect of campaigns in analyses like King’s, campaign efforts are generally worthless. I want to make the simple point that Obama’s and Romney’s persuasion efforts can cancel each other out, thus making the effects of campaigns hard to see. Examining turnout, rather than support, shows the impact of campaigns, as both Obama and Romney are attempting to turnout their supporters.  Below I replicate King’s exact graph, but with turnout (VEP) as the variable of interest.
The black, battleground states are closer to the 45-degree line than the non-battleground states, demonstrating the effects of campaigns. Turnout in non-battleground states fell 3.9 points from 2012; the same drop in battleground states was only 1.7 percentage points. This battleground state distinction is exactly what we expect to see in between a high-turnout election (2008) and a less exciting election (2012) if campaigns are able to partially offset the lack of enthusiasm.
 The sustained Voter ID effort by the GOP hurts this hypothesis as these tactics had the potential, if the courts hadn’t struck them down, to cancel out some of Obama’s GOTV efforts. I would note that turnout in Kansas, the only state with a new Voter ID law upheld by the courts, dropped 5.4 percentage points — a good deal higher than the non-battleground state average.
Aaron Swartz–hacker, activist, idealist–committed suicide on Friday. I last saw him at Rootscamp, where I was lucky enough to have a group dinner with him and help him get around town. Given the trial hanging over him at the time, he seemed in remarkably good spirits. I wish I had known the truth, and had the opportunity to help. 
I’m glad that eloquent eulogies for Swartz abound (Lessig, Doctorow, Greenwald, official tumblr). There’s an interesting contrast between Greenwald’s remembrance and expert-witness Stamos‘. Greenwald labels Swartz a true hero:
Critically, Swartz didn’t commit himself to these causes merely by talking about them or advocating for them. He repeatedly sacrificed his own interests, even his liberty, in order to defend these values and challenge and subvert the most powerful factions that were their enemies. That’s what makes him, in my view, so consummately heroic.
Whereas, Stamos reviews Swartz’s alleged crimes and comes to the following conclusion:
If I had taken the stand as planned and had been asked by the prosecutor whether Aaron’s actions were “wrong”, I would probably have replied that what Aaron did would better be described as “inconsiderate”. In the same way it is inconsiderate to write a check at the supermarket while a dozen people queue up behind you or to check out every book at the library needed for a History 101 paper.
I disagree with Stamos — Swartz was trying to take down JSTOR’s business model (of charging outrageous fees to universities for academic articles, many of which should have been in the public domain). It’s only because Swartz was caught before he could complete (what I assume was) the mission, that the actions laid out in the indictment can be construed as merely “inconsiderate”. If he had succeeded in making all of those articles freely available, I think the discussion would be colored differently today.
Especially in light of Swartz’s previous run-in with the FBI over PACER/RECAP, I am firmly in the Swartz-as-hero camp. He must have known that the authorities would go after him over JStor, yet (if the indictment is correct):
- When MIT blocked his IP address, he circumvented the system to obtain a new one.
- When MIT then blocked his MAC address, he spoofed a new one.
- When he decided that his computers were too identifiable on the MIT wireless network, he plugged a machine directly into the network.
- And even when he guessed there might be video surveillance, he kept trying to download more and more articles (albeit with a bike helmet over his face when he visited the network cluster).
Those are not the actions of someone who is “inconsiderate.” Rather they are manifestations of a person who is driven, with a deep passion and firm principles, to change the world for the better. 
And he succeeded. In a somewhat ironic twist, Swartz won the larger battle this week, when JStor began offering full articles for free up to the mostly-reasonable limit of 3 articles per week. (That rate would certainly fit my needs as an ex-academic, assuming the articles I requested were part of the set of journals included in the pilot.) 
But this small victory barely dents the sense of grief that we all feel today. The world has lost an inimitable flame that burned brightly, fervently, and too briefly, for the cause of freedom.
Blessed is the match consumed in kindling flame.
Blessed is the flame that burns in the heart’s secret places.
Belssed is the heart with strength to stop its beating for honor’s sake.
Blessed is the match consumed in kindling flame.
 A sentence in Lessig’s post “unable to appeal openly to us for the financial help he needed to fund his defense, at least without risking the ire of a district court judge” makes me think that the legal strategy was for the Internet not to come to Aaron’s aid. Frustrating.
 Undoubtedly, JStor wanted to update its business model on its own terms, rather than have its database dumped on the web. But, to their credit, they did not push for the Feds to bring this case. Clearly the Feds and (according to Swartz’s family) MIT saw the matter in a different light.
This blog might have more hits today because of my mention in this Victory Lab article (which I did not comment for — most of the substance comes from a leaked memo).
Allow me to take advantage of the increased traffic to honor Team Targeting at the DCCC — I have a great (one might even say “eclectic“) team:
One week left!
A few weeks ago, I attended an amazing wedding in Jackson Hole. The ceremony took place on the banks of the Snake River and the bride rode in on a horse-drawn carriage. What impressed me even further is that the couple chose excerpts from A.A. Milne and Taylor Mali as poetry readings; I’m big fans of both authors.
The Mali excerpt was from “How Falling in Love is Like Owning a Dog,” a very clever poem that I hadn’t heard before. (Word to the wise, the rest of this post won’t make sense unless you click that youtube link.)
Needing something clever myself for an entry in the guestbook, I thought I would try my hand at Mali’s analogy with a slight twist:
How Falling In Love Is Like Owning A Cat
One day, unexpectedly, Love shows up on your doorstep ready to be a life partner.Circumstances may be trying, but Love always lands on its feet. Push the wrong buttons and Love scratches. But one caress behind the ears, and the claws retract. Sometimes Love needs to be alone. And sometimes it jumps on you from above — Love is full of surprises. Most importantly, Love always finds its way home.
Happy Marriage, Libby & Dave!
A few weeks ago, I wrote the following nerdy and somewhat-cryptic tweet:
Abromowitz polarization variable: if(yr<96,0,if(is.Inc | app>0,1,-1)) counts as 4 degs of freedom? http://j.mp/SbHyzx
The “degrees of freedom” that I reference is the statistical term, which represents the difference between he wealth of data at your disposal and the complexity of your model. Allow me to explain as best I can:
Let’s say that you want to forecast the 2012 result based on the 16 presidential elections since 1948. In this example, you start with 16 degrees of freedom because those 16 election results could be anything. And in the simplest model, you can retain all 16 of those degrees of freedom:
The simplest model would be to guess that the incumbent party will garner 50% of the vote. (A plausible reason for this guess is the Median Voter Theorem, which says that parties move to the center and votes will generally be split equally.) Because your guess stays at 50% no matter what the 16 presidential elections could possibly be, the 50% model costs you zero degrees of freedom, and you retain all 16 degrees.
But, perhaps a keener idea would be to actually examine the historical trends in an attempt to improve accuracy. In the 16 elections since 1948, the average popular vote percentage for the incumbent party’s nominee is 52.1% — suggesting that the Median Voter Theorem by itself may be missing an inherent, incumbent party advantage. This new model, displayed as the horizontal line in the chart below, costs one degree of freedom for the averaging parameter.
Here’s why that model costs one degree of freedom. Imagine that you only had one election result to work with: the 2008 campaign in which McCain (the incumbent party nominee) won with 46.3% of the two-party vote. Averaging the historical data is trivial: the sole election of 2008 averages to 46.3%. The fact that your modeled forecast exactly equals the historical data means that you have used up all of your degrees freedom left. One data point minus one average equals zero degrees of freedom. In general, each additional data point adds degrees of freedom, and each additional variable to your model subtracts a degree.
A similar thought process controls for two data points. Since one line (defined by a slope and intercept) can always exactly intersect two any two points, the lines’ variables of the slope and intercept cancel out the two data points, and there are no degrees of freedom left. Adding a data point will (usually) cause the line to start missing points, even if it fits the data well — that mis-estimation is the degree of freedom.
In general, if you have lots of degrees of freedom, but your model still estimates the data well, then you’re in good shape. Nate Silver has written extensively on forecasting, saying “A general rule of thumb is that you should have no more than one variable for every 10 or 15 cases in your data set.” This keeps your degrees of freedom from edging toward zero add complexity (e.g., variables) to your model.
Abromowitz’s 2008 forecasting model, which was based on 15 data points, included three variables: presidential approval, the economy (specifically, GDP), and whether an incumbent president is running. Throw in the implicit historical average, and his model’s complexity costs him four degrees of freedom, leaving him with 11. With so few presidential elections to learn from, he was already pushing the limits of the data. Yet, after the 2008 election, he noticed another trend in the data:
The unexpected closeness of all four presidential elections since 1996 suggests that growing partisan polarization is resulting in a decreased advantage for candidates favored by election fundamentals, including first-term incumbents. … In fact, the last four presidential elections have produced the closest victory margins and the smallest inter-election vote swings of any four consecutive elections in the past century.
To account for this trend, he added another variable to his model: polarization. At first blush, this new parameter may appear to cancel out only one degree of freedom from the model (leaving Abromowitz with the same number of DoFs as he had going into 2008 because he gained one data point from Obama v McCain). But, the description of the variable demonstrates more complexity than normal:
For elections since 1996, the polarization variable takes on the value 1 when there is a first-term incumbent running or when the incumbent president has a net approval rating of greater than zero; it takes on the value -1 when there is not a first-term incumbent running and the incumbent president has a net approval rating of less than zero.
It the polarization variable took on a value of 1 post-1996 and 0 before 1996, then that addition would clearly cost Abromowitz one degree of freedom — just as approval and incumbency cost him one degree of freedom. To be specific, what costs Abromowitz the degree of freedom is his ability, retrospectively, to examine the data and atheoretically pick 1996 as the transition point. However, his polarization variable does not just take on one value after 1996. Rather, the variable’s post-1996 value depends on both approval rating and incumbency — his picking of those two criteria eliminate another two degrees of freedom, costing him a total of three. (The question mark in my tweet was very appropriate as I miscounted originally.)
Thus, Abromowitz’s model is now down to (16 – 4 – 3) nine degrees of freedom — in other words, he’s used up nearly half of the degrees available from his data. While I believe that the American electorate has become more polarized, this level of model complexity (relative to the data set) reeks of overfitting to me.