• How to Bullshit With Statistics

    As you probably know, President Obama put in place higher mileage standards for new cars before he left office. President Trump, naturally, wants to get rid of them. This requires a report demonstrating that the Obama standards would be bad, so he ordered OMB Director Mick Mulvaney to produce one.¹ Mick saluted and scurried off to carry out his boss’s order with the help of the Department of Transportation, which is conveniently run by Elaine Chao, the wife of Senate majority leader Mitch McConnell.

    Today the EPA made public an … um … assessment of the report that it wrote a couple of months ago. The EPA’s assessment “privately challenged the Trump administration’s rationale for freezing Obama-era mileage standards,” according to the Associated Press, and that made me curious to see it. Eventually I did, and now I’m going to share a small part of it with you. Fair warning, though: even for chart nerds, this is pretty tedious. But I promise that it’s worth it, for reasons that I hope will become clear.


    First off, you have to estimate the effect of higher mileage standards on the total size of the automobile fleet. But the Trump version makes no sense:

    The As-Received² model estimates that the Augural³ standards will reduce the year-over-year annual increase sales of new vehicles by approximately 8,000 vehicles on average between CY2021 and CY2032. However, during the same period, the As-Received model estimates that the used fleet will grow by an average of 512,000 vehicles per year, far exceeding the decrease in new vehicle sales. It’s hard to imagine any real-world scenario under which over 60 additional used vehicles are retained for each new vehicle that the sales model predicts will be unsold as a result of the higher new vehicle prices.

    And here’s the chart:

    As you can see, there’s really only a small difference between the red line and the dashed black line. Who would even notice that it goes in the wrong direction and is therefore projecting that a higher vehicle cost will produce a bigger fleet? But it does, and this is “not consistent with the basic principle of supply and demand,” as the EPA says blandly.

    The Rebound Effect

    If gas mileage gets better, cars will use less gasoline. Hooray! However, there’s also something called the rebound effect, which means that since cars are using less gasoline, people will tend to drive a little more because it’s cheaper to do so. The result is that although improving gas mileage has a positive effect by reducing the amount of gas used per mile, it also has a small negative effect because people drive more miles. Sometimes life is like that.

    Somehow, though, the OMB/DOT folks got this 180 degrees backward. Here’s the chart:

    Notice that the dashed lines are all below zero, which means that OMB is projecting lower VMT (vehicle miles driven) from the Obama standards, even though the Obama standards produce higher gas mileage and therefore should produce a positive rebound. As a consequence, this would also project a negative rebound from repealing the Obama standards, which is totally wrong.

    Yet More Rebound

    It gets even worse. The rebound effect is fairly small, so what if we just assumed it was zero?

    With no rebound, we would not expect to see any change in total VMT….However, even with 0 percent rebound, the As-Received model does produce total VMT values that are influenced by stringency level. See Figure 6, below. We believe that this zero-rebound VMT growth is an artifact of the disconnect between the sales model, scrappage model and mileage accumulation schedules described with Issue #1. And while this problem is not directly related to the model’s calculation of the rebound effect, it points to the importance of carefully considering how the various elements are integrated when making changes or additions to a model.

    This time, just to make their point clear, EPA places a gigantic red X through the incorrect numbers:

    This should be a straight line, not a curve. If there’s no rebound effect, how can the rebound effect produce anything other than a result of zero all the way across?

    What It All Means

    There’s more, but I’ll stop here. Even for chart enthusiasts, enough is enough.

    In any case, the point here is not that the OMB/DOT charts are wrong. The point is that they’re totally disconnected from reality. Basically, somebody whomped up a spreadsheet and then played around with the numbers until they got ones they liked. The fact that it implied gravity was negative or the sky was green never occurred to them. To them, this wasn’t an exercise in creating a model of the real world, it was an exercise in fiddling with spreadsheet cells until the results were ones the boss would like.

    And who cares anyway? It’s not like anyone in the press is ever going to look carefully at this stuff. Even the ones who know better will describe it as a “tense technical battle between experts at two separate government agencies” because they’re not allowed to say that the OMB and DOT just bullshitted some numbers and then called it a day. But that’s what happened, and that’s why it was worth wading through a bit of this. This is how the Trump administration works. It’s how the Republican Party works these days. These guys make Karl Rove look like Einstein.

    ¹I assume it was Mick Mulvaney, anyway. I suppose it might have been someone else.

    ²“As Received” means the version EPA received from OMB.

    ³Don’t even ask. For some reason, the Trump report uses “augural” to refer to its proposed future standards. God only knows why.

  • “Revenge,” the Telenovela, Still Drawing Big Audiences

    Cheriss May/NurPhoto via ZUMA

    One of the impressive things about Donald Trump is his creativity in taking revenge on people. I mean, who else would have gotten the idea of using the presidential pardon power this way? And yet, it’s right there in the Constitution. He could write “Donald Trump Jr. is pardoned for everything” on a White House napkin and that’s all it takes.

    Likewise, the president does have the power to remove someone’s security clearance, but I don’t think that even Richard Nixon ever came up with the idea of using it to get back at any of his enemies. But Trump has. It might not be the biggest deal in the world, but it’s a telling little humiliation for someone who’s accustomed to knowing all the little behind-the-scenes secrets.

    Or, on the world scene, has any president refused to sign a pro forma communique with allies just because one of them said something he found slightly vexing?

    Then again, what goes around comes around. Has any presidential aide ever roamed the White House routinely recording conversations in hopes of being able to use them for some kind of petty revenge against the president? Not that I know of.

    It’s quite the telenovela we have going on here. But how will it do during sweeps week this November?

  • Lunchtime Photo

    This is a May Pride peach, but I don’t feel like waiting until next May to put it up. So have a peach!

    June 22, 2018 — Irvine, California
  • Can We Please Stop With the Dimwitted Robot Car Owns?

    Sure, this Waymo car is fine when it's driving on a "street" in "California" where robots have more rights than humans. But what if this was Texas and someone suddenly pulled out an AR-15 and just started shooting it up for no reason? THEN WHAT???Andrej Sokolow/DPA via ZUMA

    Atrios is my usual foil for arguments about whether self-driving cars will ever work, and today he highlights this devastating critique from the Governors Highway Safety Association via CNN:

    Jaywalking could become a critical issue. Pedestrians and pranksters, knowing that the cars are programmed to yield to any in their path, could bring traffic to a halt. Outfitting the cars with facial recognition technology could help identify violators, but that raises its own tricky issues.

    Yeah, I guess that could be a problem. Of course, pranksters could do this right now, but for some reason they don’t. And if they did, I guess the robot cars could … go around them? Or something?

    Anyway, the list of things robot cars will never, ever be able to do seems endless. What if it snows?!? What if the lane markers are hard to see? What if it has to drive onto a ferrry boat? What if a police officer gives it instructions? WHAT IF SOME BRILLIANT PHILOSOPHER COMES UP WITH A DIABOLIC VERSION OF THE TROLLEY PROBLEM AND DOWNLOADS IT AS A VIRUS, BRINGING THE ENTIRE ROBOT CAR POPULATION TO A SUDDEN, CONFUSED, STOP???

    Come on, folks. This reminds of the early aughts when people were insisting that the entire apparatus of the NSA was probably worthless because the terrorists could just use burner phones, like they did on 24. The thing is, the NSA probably saw that episode too. They probably had a few thoughts on the matter.

    Ditto for self-driving cars. Anything that pops into your head without any real trouble is probably something that Waymo and Ford and Daimler have thought about too. Are these things all easy to solve? I’d guess not, or else we’d have full self-driving cars already. But they aren’t insurmountable either.

    There’s no question that this is difficult technology, and that it will go through a lot of failures on its way to mass-market acceptance. But please, can we stop with the dimwitted claims that there are all sorts of obvious problems that the robot car developers haven’t even thought of? Believe me: if you’ve thought of them, so have they.

    FOR THE RECORD: My current prediction for full-on, no excuses self-driving cars is now 2022. Limited versions (shuttles, geographically restricted cars, etc.) may come earlier.

  • Employment Is Pretty Close to Reaching Its Peak

    This is apropos of nothing in particular. I’m just futzing around with economic figures to see if anything looks interesting. Here’s the employment rate for prime-age (25-54) men and women in the United States:

    The employment-population reached its peak in early 2000. It dipped during the 2001 recession and plummeted during the 2009 recession, but has been steadily improving ever since 2012. It’s now at about 97 percent of its level at the peak of the dotcom boom.

    There’s no dotcom boom on the horizon right now, so my guess is that the employment population ratio can go up another percentage point and that’s about it. That’s approximately another million jobs before it peaks again, which should take a year or so. By the first half of 2019, the economy probably will have topped out and will be ready to start declining again.

    Unless, of course, Republicans decide to artificially boost the economy yet again in hopes of squeezing out just enough activity to keep them in office a little while longer. But if they do, they better be pretty sure of their timing. After all, the bigger the artificial boost, the harder the artificial fall.

  • The Establishment Still Has Its Claws on the Democratic Party

    At least we can all be happy that No Labels has a zero percent success rate so far.538/ABC News

    Is there a red tide taking over the Democratic Party? 538 did the tedious work of actually finding out. The key statistic is not how many centrists have won their primaries or how many firebrand progressives have won theirs. The key statistic is: what happens when one of them runs against the other?

    The organization with the best endorsement record in Democratic primaries remains the Democratic Party itself….In races where a party-endorsed candidate ran against a progressive-group-endorsed candidate (excluding any races where a candidate was endorsed by both sides), the party-endorsed candidate won 89 percent of the time.

    I’d be curious about how this compares to past election cycles. I’d also be cautious about making too much of this. Almost by definition, party-endorsed candidates tend to have more money and more experience, while progressive challengers are new to the game and have fewer supporters. You’d hardly expect them to do as well as mainstream candidates.

    That said, 538 also reports that quite a few #Resistance-endorsed candidates have won their primaries, though in some cases they’re just the same candidates endorsed by the establishment. (Or sometimes they’ve “won” hopeless races where the establishment didn’t even put up a primary candidate.) So the ultimate test for the #Resistance—the general election—is still very much up for graps.

  • Sure, Robots Can Feel Sorry For You

    PLOS One

    I see that Sherry Turkle is at it again:

    Years ago I spoke with a 16-year-old girl who was considering the idea of having a computer companion in the future….But there is something she may have been too young to understand — or, like a lot of us — prone to forget when we talk to machines. These robots can perform empathy in a conversation about your friend, your mother, your child or your lover, but they have no experience of any of these relationships. Machines have not known the arc of a human life. They feel nothing of the human loss or love we describe to them. Their conversations about life occupy the realm of the as-if.

    Yet through our interactions with these machines, we seem to ignore this fact; we act as though the emotional ties we form with them will be reciprocal, and real, as though there is a right kind of emotional tie that can be formed with objects that have no emotions at all. In our manufacturing and marketing of these machines, we encourage children to develop an emotional tie that is sure to lead to an empathic dead end….Children will lose the ability to have empathy if they relate too consistently with objects that cannot form empathic ties.

    Meanwhile, in the real world:

    Eighty-nine volunteers were asked to help improve a robot’s interactions by completing two tasks with it: creating a weekly schedule and answering such questions as “Do you rather like pizza or pasta?” The tasks with the robot, named Nao, were actually part of a ploy, however. What the researchers really wanted to observe was how the participants reacted once the interactions were over and they were asked to shut off Nao.

    “No! Please do not switch me off! I am scared that it will not brighten up again!,” Nao said to about half of the participants….Of the 43 people who heard Nao beg to stay online, 13 chose to listen and did not turn him off, according to the study. Some merciful participants said they felt sorry for Nao and his fear of the void.

    I realize there’s little point in people like Turkle and me arguing with each other. Neither of us knows what will happen in the future and neither of us knows what kinds of advances we’ll make in artificial intelligence. In the end, if we were doing this in person, we’d probably end up in some version of Are too! Is not! until we got tired of each other.

    And yet! Consider the assumptions Turkle is making. First, she’s assuming that human empathy is based on certain experiences that no robot will ever have. But why not? There’s nothing magic about human emotion. It’s all neurons and cortisol and dopamine and so forth, just like everything else in the human brain. If we feel like it, we can program analogues of human neurochemistry into an artificial intelligence and then send it out into the world to have all sorts of emotional experiences. After we’ve done that a few thousand times, we’ll have a template we can copy into any new robot if we feel like it. We wouldn’t bother with this if we were building a robot to assemble cars, but we would if we were building a robot to take care of a child.

    How real would this seem to a person interacting with a robot? Well, those 89 volunteers in the second story spent a grand total of a few minutes with Nao engaged in tasks that built no emotional connection at all. And yet, a surprising number of them thought that Nao’s simulated fearfulness was real enough that they refused to toggle off the power switch. Presumably, all it took was a crude bit of manipulation of Nao’s voice to get this reaction.

    So will robots develop empathy? I don’t see why not. Will they at least develop simulated empathy as good as the real kind? Plenty of humans do that successully—actors, sociopaths used car salesmen, etc.—so again, I don’t see why not. In fact, here’s my prediction: artificial intelligence will eventually do everything better than humans do it. That includes the development and expression of emotions.

    Humans are both less and more impressive than most of us think. Many people take for granted our immense ability to manipulate the world and create complex societies using nothing more than a 3-pound lump of nerve tissue working at chemical speeds with only a few watts of energy. That’s impressive! At the same time, though, three pounds of nerve tissue and a few watts of power just isn’t an awful lot. Much of what we do is, for lack of a better word, little more than crude simulation and simpleminded heuristics: we’re just faking things we’ve seen before. This is pretty obvious if you give it a moment’s thought, and not just for infants. And crude or not, most of the time this is better than the alternative. When we’re not simulating, we’re all too often just reacting blindly to a rush of neurotransmitters that are themselves reacting blindly to some perceived effect on the body.

    So yeah: robots will develop empathy. They’ll do it because we’ll program them to do it. Will it be “real”? Of course it will. It will be as real as the chemical version we all rely on now. Will we like it? We’ll love it. And our robots will love us back.

  • Lunchtime Photo

    These are the Needle Mountains, a subrange of the San Juan Mountains, which are themselves a part of the Rocky Mountains. They’re about an hour north of Durango, which is where Marian and I spent this past weekend visiting a friend.

    As it turned out, it wasn’t much of a photography trip thanks to the haze left over from a month of wildfires in July, but I got a few good pictures. I also spent an excellent half-day with Frank Comisar of Scenic Aperture, who took me to a few places I never would have seen on my own and taught me a number of techniques I wasn’t familiar with. I found him just by wandering into his gallery and being blown away by the quality of his work. I highly recommend a visit if you’re ever in Durango.

    August 11, 2018 — San Juan National Forest, Colorado
  • New Report Says China Shock Badly Hurt Wages of Low-Income High School Grads

    Tyler Cowen points us to yet another paper examining the impact of the China shock during the 2000s. First, the authors decompose the effect of the China trade into both direct employment and employment from supply chain channels. It looks something like this when you combine their two charts into one:

    If you look solely at the effects of direct competition (on the left), it looks like employment goes down when a region is exposed to competition from China. But when you add in supply-chain effects (on the right), total employment in areas exposed to the China trade goes up. However, the effect on wages is not so happy a story:

    College-educated workers almost uniformly benefited from living in places that were exposed to the China trade. However, non-college workers were just the opposite. With the exception of a few workers with incomes between the 50th and 75th percentile (i.e., individual incomes around $40-50,000), everyone lacking a college degreee suffered an income loss. The very poorest third suffered a huge income loss, losing between 10 percent and 30 percent of their initial incomes.

    As always with this kind of thing, it’s above my pay grade. It’s a serious study, but I can’t judge whether it’s correct. If it is correct, however, the bottom line is that the China shock didn’t affect employment very much but did badly hurt the wages of low-income high-school grads.

  • Nothing Much Happening Except For Possible Global Economic Collapse

    Hum de hum. Nothing much going on this this morning. I guess I’ll go check in on whether Omarosa has said anything new. And hey, how about that true freshman quarterback phenom at Southern Cal? It looks like JT might be the real deal.