Skepton⚛mics

Data Doesn’t Rule Out Democratic Backsliding in the US

2020-10-07T00:00:00-04:00

On Twitter, Andrew Little argued that democratic backsliding in the US (i.e., the US becoming a dictatorship) almost certainly won’t happen:

The only countries where this has happened are (1) far poorer than the US, and (2) have not been consolidated democracies for nearly as long. And these are good predictors.

Our baseline expectation should be that major democratic backsliding will almost certainly not happen. https://t.co/L4p5CJXzVC
— Andrew Little ((???)) August 28, 2020

Adam Przeworski says something similar at The Monkey Cage:

The probability that either the incumbent or the opposition would not respect the results of elections falls rapidly as a country experiences more alternations. For the U.S., which holds the world record having gone through twenty-two such events, this probability is 1 in 1.8 million elections.

(Obviously this is in response to Donald Trump refusing to commit to accepting the results of the election.)

My reaction to these claims is skeptical. How can we be so confident given the small number of examples we have? Can we really measure a probability as low as 1 in 1.8 million from a sample of maybe a hundred or so democracies?

Andrew pointed me to the The Democracy and Dictatorship Dataset (Cheibub, Gandhi, and Vreeland 2010) as the basis of his argument, which is conveniently available in the democracyData R package here. Przeworski doesn’t explain where his number comes from, but his published work suggests he ran a logistic regression (Przeworski and Limongi 1997).

After looking closely at the data, I think my skepticism is well-placed. The data can’t support this level of confidence, though backsliding is clearly unlikely.

There are two statistical issues that I think Little and Przeworski missed, and together they drastically lower our confidence. First, they report a point estimate (maximum likelihood) of the probability and ignore the uncertainty surrounding that estimate. We obviously can’t accurately measure a number as low as 1 in 1.8 million in this data, and the uncertainty overwhelmingly goes in the direction of higher probabilities. This is a known problem with rare events data– for example see King and Zeng (2001). Accounting for this uncertainty makes a huge difference to our confidence that we won’t see backsliding.

Second is the issue of effective sample size– they treat each year of democracy as an independent data point, when instead there are probably spatially and temporally autocorrelated errors. (That is, backsliding can occur in clusters in space and time.) That implies a smaller effective sample size, and a smaller effective sample size implies greater uncertainty.

To get started, I looked at the data with a survival analysis. I don’t think this is the best way to do the statistics, but it’s a useful way to get oriented and get a sense of what’s going on.

One thing that jumps out of this data is that older democracies are less likely to revert to dictatorship. For example, new democracies have about a 25% chance of becoming a dictatorship within 10 years. However, for democracies at age 50, 100% of them remained democracies 10 years later. That’s consistent with Little’s and Przeworski’s arguments.

But contrary to Little and Przeworski, this model displays a wide range of uncertainty about the true probability, where older democracies could have up to a 14% chance of backsliding (at the extreme end of the 95% confidence interval, which is the shaded area). The survival analysis approach isn’t the best here, though, because it assumes that each year has a separate probability. The probability of a 50-year-old democracy backsliding is estimated separately from the probability for a 51-year-old democracy, and so forth. The result is that each year has a small sample and the confidence intervals are very large, probably too large for this case.

Since there’s no visible change in the probabilities after democracies are 50 years old, a reasonable approach is to assume that the probability at that point is a constant. In this case we can lump all the years after 50 from different countries together (for a total of 1076 democracy years with no backsliding) and estimate the probability with a binomial model. Trying this approach in R, I get:

## 
##  Exact binomial test
## 
## data:  0 and years_ge50
## number of successes = 0, number of trials = 1076, p-value < 2.2e-16
## alternative hypothesis: true probability of success is not equal to 0.5
## 95 percent confidence interval:
##  0.000000000 0.003422457
## sample estimates:
## probability of success 
##                      0

The probability according to the binomial model is between zero and 0.3%. That’s pretty low, and in line with Andrew’s tweet.

Quick math note: How do you calculate the probability of an event that hasn’t happened before? If you run a standard statistical test, like the binomial model I just ran, it’ll tell you the most likely probability is zero. But a zero probability doesn’t make sense as a prediction– we can’t say with total confidence that something won’t happen just because it hasn’t happened before. To get a better probability for predictions, we need to account for the possibility that the event happens but we haven’t seen it yet. I was surprised to find that this probability is just 1/(n+2), where n is the number of times the event hasn’t happened. This is a result of the rule of succession.

Let’s start applying this. At one extreme, we have the survival analysis approach, with a sample size of 25 (the number of older democracies in the dataset), implying a backsliding probability of 3.7%. At the other extreme is the constant probability approach, with a sample size of 1076, implying a backsliding probability of 0.1%.

While the survival analysis is clearly too conservative, I think the alternative goes too far in the other direction due to its assumption that all democracy years are independent. It’s easy to think of counterexamples: In the US, the president probably influences the probability, but the president usually only changes once every 4 years or less. Economic downturns like the 2008 recession affect many countries at once. Events like the Arab Spring can temporarily affect backsliding probabilities for an entire region.

We end up with a question of effective sample size. Out of 1076 democracy years, how many truly independent samples do we have? The graph below plots the range of probabilities:

I don’t have a good way of deciding between these probabilities except that I think we can rule out the extremes, so I think we’re looking at a probability in the range of 0.2% to 1%. That’s about 10,000 times higher than Przeworski’s probability.

I hate to say it, but we can’t stop here. This is just a prior, which doesn’t account for the details of any given year. If we want to know the risk we face specifically in 2020 we should look at a posterior formed after including the other information we have about 2020.

The salient issue in 2020 is that we have a malevolent president who refuses to commit to accepting the results of the election and a major political party that’s reluctant to rein him in. Intuitively I think we all expect the probability of backsliding to increase in this situation. In terms of Bayes’ theorem, what we want to know is P(Backsliding|MalevolentPresident), where P(Backsliding) is between 0.2% and 1%.

Unfortunately I don’t have enough information about the other terms in the equation to calculate this properly, so I’m not going to provide a number here and claim I know the probability of American backsliding in 2020. However it’s easy to find plausible numbers that lead to backsliding probabilities like 5 or 10% or even higher.

Given the uncertaintly surrounding the numbers, the conclusion from this analysis is less about what we can say and more about what we can’t say. The data we have can’t rule out the possibility that there are rare situations where old democracies are at an elevated risk of backsliding, and 2020 could be one of those occasions.

References

Cheibub, José Antonio, Jennifer Gandhi, and James Raymond Vreeland. 2010. “Democracy and Dictatorship Revisited.” Public Choice 143 (1): 67–101. https://doi.org/10.1007/s11127-009-9491-2.

King, Gary, and Langche Zeng. 2001. “Logistic Regression in Rare Events Data.” Political Analysis 9 (2): 137–63. https://doi.org/10.1093/oxfordjournals.pan.a004868.

Przeworski, Adam, and Fernando Limongi. 1997. “Modernization: Theories and Facts.” World Politics 49 (2): 155–83. https://doi.org/10.1353/wp.1997.0004.

Outbreak Economics

2020-04-07T00:00:00-04:00

Discussions of the policy response to coronavirus outbreaks have been pretty informal, and occasionally I see people claim that the optimal policy response sets R (the coronavirus reproductive number) to 1, which isn’t exactly right. I think it’s worth working through a simple model to see why, and to provide a formal framework for evaluating policy options.

Let’s say we can set public policy once a week, and $Y$ is the output this week (in a wide-ranging sense that includes both traditional economic goods and health). For a simple model, we can say \[ Y(R) = Y_0 - V(R) - p(R) \] where $Y_0$ is the normal output we’d get without any new coronavirus infections, $V$ is the cost of new virus infections, and $p$ is the cost of preventative measures that lower $R$. $Y$, $V$, and $p$ all depend on $R$.

It costs more to acheive lower $R$ values, so $p(R)$ decreases monotonically with $R$. $R$ is controlled indirectly by policymakers, as a result of the preventative measures they choose to take.

Let’s also say that $V$ increases proportionally to the number of current coronavirus cases and the average cost per case. Since this is a weekly cost, and the virus’ serial interval is close to a week, we can say that $V$ is approximately \[ V(R) = R \times cost \times cases \] That is, $V$ is the number of new infections ($R \times cases$) times the cost per infection. (If we wanted to look at longer periods, we’d have to raise $R$ to a power to reflect exponential growth.)

Let’s also assume that we can’t completely eradicate the virus– if we did eradicate it, it would just get reintroduced from another country.

A bit of calculus shows that output is maximized when \[ -\frac{dp}{dR} = \frac{dV}{dR} = cost \times cases \] that is, when the marginal cost of prevention with respect to $R$ equals the marginal cost of infections with respect to $R$. This is shown in figure 1, where I’ve labeled the optimal $R$ value $R^*$.

Figure 1: Optimal R values (R*) for diffent numbers of infections.

This figure clearly shows what some people have missed, that the optimal policy (and the optimal $R$) depends on the number of current infections. If many people are infected, the cost of holding $R$ at 1 and allowing new infections is high, so society can benefit by pushing $R$ lower. If there are few infections, it may make sense to temporarily relax precautions and allow an $R$ above 1.

Conveniently for modelling, a country following this strategy will eventually reach a steady state at $R = 1$ and \[ cost \times cases = -\frac{dp}{dR}(1) \] where the number of new cases remains constant. When a country holds $R$ below 1, the number of cases fall, and in response the country raises $R$. And when a country holds $R$ above 1, the number of cases rises, leading the country to lower $R$.

This lets us investigate the effects of changes of different variables. For example, what happens if coronavirus is less contagious in summer than winter, making it easier to acheive lower $R$ values? Figure 2 shows the affect on the steady state. In this case, the optimal number of infections is higher in the winter than in the summer, even though $R$ is still held at 1 in both (higher marginal cost implies more cases, other things equal).

Figure 2: Seasonal changes in steady state policy responses.

One question I want to try to answer is, how does this end? The optimal steady state holds R at 1, which currently requires a lot of effort, and this line of reasoning says we’re going to be stuck here for a while.

But in the longer term, if we get more efficient at keeping R down, for example by getting good at testing and tracing, then we can ease off on the more costly precautions while still keeping R at 1. A vaccine or a gradual build-up of immune people would have a similar effect by helping to push down R.

Interestingly, if we get better at treating coronavirus, so that the average cost of an infection decreases, then we can tolerate more infections, and thus can ease off precautions temporarily, but we still end up taking the same precautions in the steady state. Even though the average infection cost is lower, eventually that’s offset by the higher number of cases (though it could help by contributing to immunity). An exception occurs if we become extremely good at treating the virus, in which case even at the peak number of cases we might be willing to tolerate more infections, so we never reach the new steady state and we can stay relatively relaxed about infections.

As I was writing this, I found out I was beat to the punch by Martin Eichenbaum, Sergio Rebelo, and Mathias Trabandt, who reach similar conclusions with a fuller and more formal model (and they’re brave enough to plug in some numbers too). Their paper is available here. They also provide citations to a bunch of related research. They even provide their Matlab code, which is probably worth playing with if you’re interested in this. I also found some useful discussions in a health economics textbook, so it’s not a bad idea to take a look at one of those.

Hurricanes and high prices

2017-09-02T21:29:43-04:00

After writing an article in favor of price gouging during hurricanes, Forbes writer Tim Worstall was deluged with naysayers. Among his alleged crimes were not caring for the poor and even wanting babies to die.

But the evidence is clear: Tim Worstall is right, and, if anything, laws against price gouging will cause more babies to die.

There are two main issues to consider. First, the issue of equity-- do anti-price gouging laws promote fairness? The second issue is supply-- do high prices encourage sellers to provide more supplies during natural disasters?

Anti-price gouging fanatics almost always focus on fairness, and assume that low prices will necessarily lead to more of it. But that assumption makes no sense at all, because the low-price rationing system is comically inequitable. Should a man be able to buy the last case of water bottles while the women and children behind him get nothing, just because he happened to get to the store 30 seconds earlier?

Anti-gougers obsess over wealthy people buying all the water, but what if the first person who gets to the store is a rich person? What's to stop this person from buying 5 times as much water if the price is 1/5 of what it would have been if prices were allowed to increase?

And what is with the anti-gouger idolization of first-come, first-served rationing? Is there any reason at all to expect the people first in line to be the people most in need? A rationing system based on high-stakes races to the grocery store sounds more like a Hunger Games novel than a civilized response to hurricanes.

There's just no reason to expect the outcome of high prices to be drastically less fair than the first-come, first-served alternative. It can't be, because the status quo of low prices, lines, and shortages is already so terrible that we can't do much worse.

To add to the equity problems of first-come, first-served rationing, research from W. David Montgomery, Robert A. Baron, and Mary K. Weisskopf finds that the savings from lower prices are offset by the costs of waiting in line (at least for gas). That huge amount of time wasted could instead be spent on, say, assisting with relief or rescue efforts.

What about the supply issue? Advocates of higher prices claim the money will motivate sellers to supply more, while the low price crowd retorts that hurricanes will prevent more supplies from getting in, regardless of price. Unfortunately I can't decide this issue with data because I haven't seen any studies examining the response of suppliers to high prices during hurricanes. But a few years ago liberal nerd Matt Yglesias pointed out that, even if the low price crowd is right, we can still expect higher prices to lead to more goods, because sellers will have an extraordinary incentive to stock up the products before the storm. Without the high prices, sellers have much less incentive to stock up on the relevant goods. Even with the most pessimistic assumptions, higher prices increase the amount of hurricane-related goods available.

We therefore end up with the following scorecard: on equity, a tie between both options; on supply, high prices win. It's safe to say higher prices would improve outcomes.

But that still leaves us with the frustrating outcome on equity. Thankfully we don't have to decide only between first-come, first-served rationing and rationing via high prices. With some creative thinking, there are all sorts of policies that could be devised to address the two problems of equity and supply. For example, we could try to improve the equity of the high-price policy by subsidizing hurricane goods to ensure that they remain affordable. Or, if a perfectly equitable outcome is desired, the government could buy water and ration it out per person. Somehow we've settled on the worst of all possible policies, getting neither fairness nor adequate supplies.

To adapt to climate change, we must implement better public policies to deal with intense weather. As more and more devastating storms hit the coasts, it becomes more important for economists and policy nerds to advocate against price gouging laws and for the many better alternatives.

How political science is like intelligence research

2017-05-04T02:43:19-04:00

[caption id="attachment_244" align="alignnone" width="832"] Scores generated by W-NOMINATE, a method related to factor analysis, help explain Congressional votes on a welfare reform package (from Wikipedia)[/caption]

Suppose you want to know how conservative or liberal your Congressional representative is. Is there a way to do this that doesn’t involve sorting through a small mountain of votes and making an educated guess?

Political scientists have been working on this problem for decades, and have largely solved it. They can take a legislator’s voting record and use it to produce a mathematical estimate of that legislator’s ideology compared to his peers in Congress. And it’s not just votes – they can make similar estimates with campaign contributions, surveys, speeches, and even Twitter followers.

While the particular method political scientists use to make these estimates has varied over the years, conceptually all of them are variations of a technique called factor analysis.

Researchers employ factor analysis to simplify data. In this example, legislators may vote on 500 bills in a session. But to be familiar with our congressperson, we don’t have to keep track of their votes on all 500 bills – if we know his or her political leanings, specifically if the congressperson is conservative or liberal, we can predict many of their votes from that fact alone. Just one number – one “factor” – representing political ideology can summarize the data contained in all 500 votes. Factor analysis looks for these kinds of patterns that allow us to summarize larger sets of data with just a few numbers.

Political scientists weren’t the first to do this kind of analysis. Early models were inspired by work in psychology.

Which brings us to the mini-debate that has sprung up, with Sam Harris and Charles Murray on one side, and PZ Myers and Cosma Shalizi (indirectly) on the other, about IQ.

Like Myers, I’m not a fan of Sam Harris and I refuse to listen to his monotone for 2 hours. And I agree with the Eyes On The Right post that Myers quotes. Where I, at long last, finally disagree is when he dismisses the concept of IQ altogether, citing an article by Cosma Shalizi (in the comments).

Shalizi’s article – read it here – attempts to undermine factor analysis as a whole, as well as it’s specific application to the concept of g, or general intelligence (g is what IQ is supposed to measure). The main workhorses of his argument are two mathematical examples. But of the two, the first one is incorrect and the second one doesn’t prove what he claims.

The first example accidentally assumes what it sets out to prove. Shalizi generated the numbers in his matrix, which represent correlations between test scores, by choosing numbers at random between zero and one. I scratched my head when reading this part. Typically a person running a simulation would generate test scores first, and calculate the correlations directly from the scores. Instead Shalizi first generates the correlations between test scores, then creates the test scores to match the correlations.

This backwardness allows Shalizi to make a mistake he shouldn’t have made. Test scores that aren’t influenced by a common factor should have correlations clustered around zero – that is, the score on one test shouldn’t predict the scores on other tests. Shalizi, however, chooses the correlations by randomly choosing between zero and one, in effect assuming that many of the tests are correlated with each other. The factor analysis then finds a factor in this case because the numbers really do indicate an underlying factor. If Shalizi started by generating factor-less test scores, and calculated the correlations from those test scores (as opposed to randomly generating the correlations), he would quickly discover that this example doesn’t work. Factor analysis may erroneously discover a factor when applied to small amounts of data, but for larger data sets it doesn’t return spurious false positives like this. That’s just not how it works.

The second example is technically correct, but the implications are milder than he suggests. Yes, as demonstrated, IQ may be a summation of many smaller factors, but that doesn’t imply that it’s not a real or useful concept. The NOMINATE ideology scores used by political scientists are also based on many small factors – votes on issues ranging from abortion to taxes to NATO. Yet we all know the difference between conservative and liberal, and given a placement along the left-right axis we can make decent predictions about a person’s views on any number of issues. Clearly the left/right distinction in American politics is not a figment of our imagination, or a statistical artifact, even though it combines many smaller issues into one category. Likewise, though IQ may be the result of many smaller factors (and genetically speaking, I imagine it is), that doesn’t imply it’s not real. It’s correlated – not perfectly correlated, but correlated – with a variety of things, like knowing what the word “vicarious” means, solving MENSA puzzles, and finishing college. These correlations just happen to match one definition of the word intelligence, in the same way that factor analysis of rollcall votes leads to scores that match our understanding of the liberal-conservative spectrum in politics.

Political scientists debate the biases in scores generated by political factor analysis techniques, but there’s basically no debate that the left-right political spectrum is a real thing, and these scores have served as the backbone for a huge amount of research in the last 30 years. As far as I can tell g is in the same place.

My expertise is limited to statistics and social science, so I won’t comment on the debate about to what extent IQ is determined by genes. And I’m skeptical, to put it mildly, of Charles Murray’s policy views. But factor analysis is real.

Disclaimer: Yes, I did read The Bell Curve, about 10 years ago. I think it’s got some rigorous arguments mixed with some really weird arguments. And no, I didn’t encounter any censorship: I rented the book from my local library with no mishap.

Common sense is better than free speech

2017-04-21T03:11:55-04:00

Clichés about free speech are so common in discussions about college speakers that I despair even thinking about it. You can’t swing your arm without hitting at least two cringe-worthy bromides – and both spoken with total condescension and certainty. So to end the despair and to help the self-righteous off their high horses (and to give my arms more room), I’m proposing a radical alternative: let’s look at evidence.

Is unregulated free speech the best way to get the truth? What do the people most serious about pursuing truth think? I suggest starting with scholarly journals, the highly regarded publications that form the backbone of the scientific community. When faced with the decision to publish a paper, do journal editors bark out overused catchphrases about free speech and publish everything? Clearly not. Journals restrict speech radically and extensively. Only a select few authors ever get to publish in the most prestigious journals, and it can take years of effort for them to do so. Editors discriminate so zealously that simply failing to cite the right literature in the bibliography can keep a paper from being published. And far from being a drag, this is the secret to the journals’ success!

Or take another truth-seeking institution – the classroom. When creationists propose that public schools teach creationism alongside evolution, do scientists welcome these proposals as virtuous exercises in “open debate and dialogue”?

Perhaps a third truth-seeking institution will be different: the scientific conference. Do geologists, for example, allow flat-earthers to speak at their conferences because “the solution to bad speech is more speech”? Obviously not.

On the other hand, we’ve all seen the disasters of unmoderated comments sections. Why anyone thinks this kind of wild west free-for-all would be an effective way for college students to learn or debate is beyond me. And due to legal restrictions that prevent public universities from barring bad speakers, “unmoderated comments section” appears to be the current state of college speaking.

To be totally honest I find it hard to care about college speakers, even if the policy is a mess. How much does this actually matter? Very little, in my experience. I suspect this is an overwhelmingly symbolic issue for all sides of the debate. What really annoys me are the allegedly smart people making flat-out unbelievable claims. Are thinking people really supposed to believe that the college speaking tour of Milo Yiannopoulos, the world’s most famous internet troll, is an event that will educate college students? It sounds more like a scene out of the movie Idiocracy. Yet this is an argument I see again and again, without a shred of evidence and without any hint of humility. It’s as if we’ve traveled back to the Middle Ages – what kind of fool would look at evidence when we have the words of the ancient philosophers to rely on? (I will not name the guilty parties to avoid embarrassing anyone. But if they persist I might.)

Anyway, it’s easy to see the solution to this problem. Ideally colleges would be able to enforce policies that prevent self-aggrandizing vile ignoramuses from taking advantage of their lecture halls, in much the same way that a forum moderator would lay out the ground rules for participating in a discussion board. The co-opting of colleges by morons and internet trolls is an embarrassing failure of public policy, not a victory of rationalism.

Update, 4/21/17:
Cass Sunstein writes:

Officials at the University of California at Berkeley this week canceled a scheduled speech by Ann Coulter, the conservative author, on the ground that the school could not “find a safe and suitable venue” for her. ... It should go without saying that at colleges and universities, free speech is indispensable, and interferences with it are deplorable.

My hot take: It should not “go without saying” that trolls have the right to speak at publicly-funded college facilities intended for education. Trolls are not “indispensable” college speakers, and banning them is not “deplorable”. Colleges would do just fine without them. I am very confident of this. There are many colleges Ann Coulter has not visited. Those colleges have not fallen into barbarism.

Ron Paul gets the Fed wrong

2017-04-20T05:24:08-04:00

Former Congressman Ron Paul is claiming that the Federal Reserve is politicized, and if we don’t change our monetary policy by passing the “audit the Fed” bill we can look forward to “rising interest rates, a stock market crash, a decline in the dollar’s value, and a complete loss of confidence in the US economy.”

Sounds frightening! Thankfully, he gets everything wrong.

First, the Fed is careful about raising interest rates. If the stock market or the economy show signs of impending trouble then the Fed will stop raising rates. In fact this has happened over and over for the last 8 years.

And if the stock market did crash the value of the dollar is unlikely to decline. Bad economic times prompt people to save rather than spend their income. That means there are fewer dollars competing for each good and, if anything, prices go down. (This is why inflation has been historically low for the last 8 years as we recover from the Great Recession.)

Finally, there’s little reason for anyone to lose confidence in the US economy even if there is an economic disaster. We’re still the wealthiest nation on the planet. Our government still has a strong track record of paying its bills, which allows it to borrow at low interest rates. We still have all the people and machinery and technology we had before the hypothetical stock market crash. After the Great Recession, the dollar actually gained value internationally because investors viewed the US as a safe bet.

But beyond the purely economic nonsense, Ron Paul’s solution – the audit the Fed bill – wouldn’t fix any of this. While Paul presents the Federal Reserve as a secretive institution in need of sunshine, it already gets plenty of light. Economics writer Catherine Rampell summarizes the facts in an excellent article:

The Federal Reserve Board and its 12 affiliated regional Fed banks already undergo an audit. Multiple layers of audits, in fact, by the Government Accountability Office, the Office of the Inspector General and independent private auditors such as Deloitte. The Fed also releases weekly data about its balance sheet and the minutes from its closely watched Federal Open Market Committee meetings with a three-week delay.

In reality the bill would give legislators more opportunities to pressure the Federal Reserve with no other practical effect. Economists worry political pressure will lead the Fed to give the economy an unsustainable boost during elections, causing inflation. Ron Paul's example of Arthur Banks and Richard Nixon shows this concern is no joke. Under Banks’ watch, inflation rose to double digits. Ironically, while Paul is know for decrying inflation, that's exactly what economists fear this bill will bring.

All of this is why, when surveyed, economic experts strongly disagreed with Ron Paul’s views:

The experts are right: there’s no reason to support this bill.

Donald Trump’s tiny … job deals

2017-01-10T00:00:00-05:00

It turns out the old saying is true: men with small hands have tiny job deals.

On the day that Donald Trump bragged about creating 700 American jobs at a Ford factory, the private sector added about 7,000 jobs without his help. It’s been adding those jobs every day for the last 6 years, at a rate of about 2.5 million per year, as unemployment dropped from 10% to 4.7% under the guidance of Barack Obama’s regular-sized hands.

Given the promised Trump economic revival, why can’t Donald Trump find more impressive deals – deals that affect more than 0.001% of the private sector workforce? Surely in a country of 300 million people Donald Trump can find optimistic companies hiring more than a thousand jobs here, a thousand there. These numbers would be great for the mayor of a small city, not an incoming president.

It’s embarrassing that Trump can only display his gold-plated economic brilliance with these meager examples, but I think we should give him the benefit of the doubt– maybe the source of his impotence is that he hasn’t taken office yet. With the powers of the president, will he be able to match Obama’s record?

The answer is “no”. President Obama had it easy in a sense; there was a lot of room for improvement when he came into office in the middle of the Great Recession. Whatever you think of fiscal stimulus or Obama’s economic policies, it’s obvious that we’ve done most of the improving already. The low-hanging job fruit have been picked and at this point even Donald Trump’s genius won’t be able to harvest many more. Assuming no disasters, the economy will probably continue its slow recovery for another year or so and then level off. In fact you can already see the rate of job growth slowing down over the last 2 years, from 3 million per year to 2 million per year:

But if not more jobs, maybe he’ll at least get better-paying jobs? Hands notwithstanding, Trump still pledges to bring high-paying manufacturing jobs back from the foreign countries they’ve disappeared to. The pledge runs aground on a minor detail, though– foreigners don’t have those jobs. Many socially-deprived economists have been locked in rooms for years studying this issue, and they’ve concluded that manufacturing jobs were stolen not by foreigners, but by machines! And now, even if Trump does get a few manufacturing plants from foreign countries, thanks to the stiff robotic competition the jobs don’t pay as well anymore.

The economics is clear: Donald Trump’s jobs record will be smaller than his predecessor’s. The hands don’t lie.

The Fed understands stagflation

2016-09-26T00:00:00-04:00

Disagreeing with Janet Yellen, who’s built a formidable forecasting record, is always risky. The chance of catching basic errors are slim.

So it’s easy to guess that Jeff Spross’ opinion piece in The Week titled “The Federal Reserve is irrationally obsessed with the 1970s” gets it wrong. Spross argues:

Many Fed policymakers, including doves like Chair Janet Yellen, believe that once the inflation rate gets going, it’s like a runaway train — nearly impossible to stop and demolishing everything in its path. But there’s just one problem. This belief is based on just one badly misunderstood historical example in the United States.

And how is this example badly misunderstood?:

First off, there was a screw-up in the consumer price index (CPI) — probably the most widely used measure of inflation. By the late 1970s, it was overstating the inflation rate by about 2 percentage points — which is a lot. That’s important because unions, which were still pretty strong in the 1970s, had enforced automatic cost-of-living adjustments (COLAs) in a lot of workers’ contracts. Those adjustments relied on the CPI, which created a perverse ratchet effect: As inflation rose, employers had to compensate with pay hikes that overstated inflation. Then prices had to increase to pay for those wages, leading to further inflation.

Whooaaa, let’s slow down and think that through. We’re talking about 2 percentage points, which actually isn’t a lot. We can demonstrate this with a simple example. Let’s say half of workers get automatic cost of living adjustments based on the CPI. (This is being generous. I’m having trouble tracking down the real number, but it’s certainly lower than half.) And let’s say 70% of production costs are wages. (This is a real number.) Let’s also make some simplifying assumptions about the economy: no worker with COLAs gets fired, firms are perfectly competitive, and consumers don’t change what they buy in response to the changing prices. Every single one of these assumptions is wrong in real life, and every single one will make our simple estimate higher than it should be. We are being extremely generous in favor of the COLA ratcheting hypothesis. Now, having made all these simplifying assumptions, when half of workers get their wages raised an extra 2%, we can calculate the amount added to prices:

$\begin{align} raised wages + unraised wages + capital &= (0.7 \times 0.5 \times 1.02) + (0.7 \times 0.5) + 0.3\\ &= 1.007 \end{align}$

The extra wages lead to extra inflation of 0.7% per year. Over a period of 3 years this might lead to 2.1% higher inflation. If only a quarter of workers got COLAs based on the CPI, then the estimate goes down to 1% over 3 years. Even under the most generous assumptions, this is a small ratchet compared to the 14% inflation at the end of the 1970’s.

Alright, so this one is a foul ball. What’s next?

Another issue was the price of oil. The Organization of Petroleum Exporting Countries (OPEC) launched an oil embargo between 1973 and 1974. Then the government of Iran, one of the world’s biggest oil producers, was overthrown in 1979. As a result of both events, oil prices rocketed into the stratosphere during the 1970s. Since oil is a fundamental expense in so much economic activity — commuting, shipping, even chemical engineering — those price jumps rippled into prices throughout the economy. It’s a temporary effect, but it matters.

Economists definitely do not “badly misunderstand” oil prices during the 1970’s– oil plays a leading role in economists’ traditional stagflation lore. Alan Blinder’s book Hard Heads, Soft Hearts has an excellent treatment of this subject. Strike.

Third, labor markets in the 1970s were actually pretty tight, at least early on. Labor force participation wasn’t down — in fact, it was rising due to the advent of the female workforce. So even though unemployment got uncomfortably high after the recessions in 1970 and 1974, wage growth ranged from 5 to 8 percent, often staying well ahead of the rising inflation rate. Those are signs employers are competing for workers, rather than workers competing for jobs.

It’s possible that the labor market was tight in the early 70s, though I’m not convinced by this argument. After the short recession of 1970, I doubt the labor market was any tighter than it is right now. Actually, the situation looks like something that could happen today: interest rates were low to spur recovery from the recent recession, the economy picked up and grew rapidly, generating inflation, the Fed starts to cut back and then the first oil price shock hits and things go downhill from there. We’re missing the oil shock piece but that’s it. Another foul ball.

One more:

Tight labor markets are nowhere in sight: Despite the 4.9 percent unemployment rate, labor market participation is still deeply depressed thanks to the aftermath of the Great Recession. Nominal wage growth is at a mere 2.5 percent, compared to the 4 percent we saw in the economic boomlet of the late 1990s.

This is a controversial statement. If you think a big part of drop in the labor force participation rate (that is, the percent of Americans actively looking for work) is due to the recession, then this statement is correct. If not, then it’s false. In reality we don’t know. Or at least I don’t know, and I’ve been following macroeconomic debates since 2008.

Regardless of labor force participation, 4.9 percent unemployment is low. It’s completely possible that full employment is right around the corner. Due to the lags in our measurements, we could even be at full employment already and just not know it yet. (OK, probably not. But the possibility is there.) When we hit full employment, inflation will start increasing, and raising rates by a small amount could be prudent if there are signs we’re nearing that point.

But let’s put this back into perspective. We’re talking about interest rate increases of 1%, or even 0.25%. The Fed is only experimenting with the fine-tuning knob on the economy right now. If we’re going to have a recession, or a recovery, a 1% higher interest rate isn’t going to matter much. I don’t understand how people get worked up about this.

I think Spross hit the ball at some point in the argument, but I don’t know because I fell asleep. Wake me up when the Fed starts talking about 5% interest rates.

Do 3rd party supporters throw away their votes?

2016-08-26T00:00:00-04:00

According to political betting markets, Gary Johnson, the most popular third-party candidate, has only a 4 in 1,000 chance of winning the election. In all likelihood a vote for Gary Johnson will not elect Gary Johnson— though it may help swing the election from Trump to Hillary (or vice versa, depending on where Johnson’s votes come from). Given that, why would anyone vote for Gary Johson, or any other third party candidate, instead of Trump or Hillary? Voting for third parties must be an irrational and futile gesture, right?

Well, no. This account omits the indirect ways a third party vote can pay off, even when the candidate (invariably) loses.

A third-party candidate’s vote count may determine whether the party can get into debates or onto ballots, which can attract attention and influence the national agenda. Green Party candidate Jill Stein’s views on US drone strikes in the Middle East would likely not be discussed if she didn’t run, or if no one voted for her. The Libertarian Party has attracted new adherents to libertarianism for 40 years by stubbornly fielding candidates in all manner of elections, even though they almost never win.

But still, it would be better to win. Why should anyone even bother with these small-ball benefits when they could help determine the winning candidate instead?

No matter how stark the choice is between the two main candidates in an election, there will always be people in between, and people whose pet issues aren’t represented by either candidate, and radicals who fall off the spectrum entirely. For many of these people, voting for one of the two main parties would be throwing their vote away since they don’t care about the outcome. I can illustrate this nicely on a Nolan chart:

(Original image borrowed from The Advocates for Self-Government)

Libertarians have a good claim to be somewhere in the purple space in between, or off to the side of, the two parties, and therefore indifferent between the two major-party candidates. That two-party apathy probably explains, at least in part, why their third party is so successful. On the other hand, left- or right-wing parties, like the Constitution or the Green parties, face a dilemma. Supporters have to compare the benefits of deciding the outcome of the election, which they may have strong opinions about, vs. changing the agenda.

How does anyone know what to choose in this situation? In individual cases I’m not totally sure how people should be deciding, but I can think of two trends relevant to the decision. The first is that candidates from the two main parties have been gradually moving away from each other on the political spectrum, meaning the outcome of elections have become more important, and therefore voters should place heavier weight on deciding the outcome (as opposed to deciding the agenda). We can see this on graphs from Keith Poole’s Voteview website:

The second trend offsets the first: many states and districts have become less politically competitive, making the outcome of the election a given. Third party voters in these states are unlikely to decide the election and should feel freer to vote for whoever their heart desires. Maxim Lott and John Stossel’s Election Betting Odds website provides a useful map showing the competitiveness of the presidential election in each state. Here’s the current map. Lighter colors mean the race is expected to be closer:

In response to these trends voters in competitive states should become more wary of third parties as time goes on, so I probably wouldn’t vote for the Green Party in North Carolina (even if I agreed with Jill Stein). But in California or New York— why not?

Do third party voters in real life actually make these rational decisions? The statistics required to answer that question are too complicated to attempt in a blog post, and I’m not aware of any study rigorously tackling the subject. But at least anecdotally, it does seem that third party supporters, who are all trying to introduce more options and change the national agenda, are on the right track.

Who are Twitter’s top libertarians?

2015-08-26T00:00:00-04:00

About 70% of Twitter libertarians follow Senator Rand Paul, according to samples I’ve been collecting, making him the top libertarian on Twitter. Next are Ron Paul (about 63%), the Cato Institute (56%), Reason magazine (48%), and Judge Napolitano (46%).

To find Twitter libertarians, I rely on the statistical finding that Twitter users tend to follow accounts that reflect their own ideologies. (See paper 1, paper 2) So I put together a list of 22 popular libertarian Twitter accounts, and downloaded lists of their followers from the Twitter API. If a person follows 3 or more of the accounts, I assume that he or she is a libertarian.¹

From the resulting list of Twitter libertarians— more than 134,000 at last count— I randomly select 1000 to see who they’re following. Interested readers can view the results at willonrails.com/top_libertarians.^{2, 3, 4}

Although I usually focus on the more radical elements of the libertarian movement, the beltway libertarians— that is, the relatively moderate libertarians who work on or write about policy for a living, often in the “beltway” of Washington, D.C.— dominate the top of the list, along with some Fox News hosts and conservative Twitter accounts. Of the more radical, openly anarchist libertarians, Adam Kokesh, known for promoting a “final American Revolution,” is the most popular with 14% of libertarians following his account. Jeffrey Tucker, a much nicer anarchist, comes next at 9%. Stefan Molyneux, an anarchist and Men’s Rights Activist who describes himself as the “greatest philosopher alive,” exerts a smaller influence at about 6%.

I plan to update this every Friday evening, with the hope that other interested researchers and I can use this data to test hypotheses about how ideological groups evolve over time. (But no guarantees!) Are libertarian extremists becoming more or less influential in the movement? How does Twitter popularity change in response to economic or other social conditions? I keep wishing I had some data on this, so now I’m making some.

Technical notes:

The list of accounts is RonPaul, RandPaul, CatoInstitute, reason, ThomasEWoods, jeffreyatucker, StefanMolyneux, bryan_caplan, sfliberty, mises, FreeStateNH, c4ssdotorg, lewrockwell, GovGaryJohnson, RoderickTLong, BobMurphyEcon, LPNational, FreeTalkLive, Antiwarcom, AynRandInst, AynRandOrg, ErikVoorhees

To some extent the selection of these accounts is arbitrary and reflects my interests in certain subgroups in the libertarian movement. A different set of accounts would produce slightly different results.
Since the raw results include a lot of newspapers, Fox News hosts, and other accounts I don’t care about, I had to find a way to automatically remove them. Step 1 was to only include accounts contained in my list of Twitter libertarians. (I made exceptions for John Stossel and the ACLU, who somehow were not in the list.) Even after this, there were still a lot of irrelevant accounts. In the end I imposed a second restriction– that the account must be followed by at least 7 (32%) of the original 22 accounts. Apparently the 22 popular libertarian Twitter accounts follow many of the accounts that I’m interested in, and relatively few non-libertarian accounts.

The result is a list of accounts favored by the libertarian “elite,” along with their popularity among the libertarian masses. It’s not a perfect list– Wikileaks and the ACLU were, sadly, removed, along with some others who should be included, and some accounts that I’m not interested in remain on the list. I’ll probably tinker with this some more to get better results.
The ± numbers are estimates of the 95% confidence interval. I’m deriving these numbers from a formula from D. S. Sivia, “Data Analysis: A Bayesian Tutorial,” p. 22. The estimates may not work well for accounts with few followers.
The R code I’m using to collect and analyze the data can be found on Github.