Relevant and even prescient commentary on news, politics and the economy.

Election Forecasting

Polls vs Polls plus.
Rule number 1 of forecasting is do not quarrel with
Rule number 538 is not ever.

So here I go. I am going to start with the fivethirtyeight Senate forecast(s). (s) because there are three and an “pick a model” icon to toggle them. I like the “lite” just polls forecast. I like it because it estimates a 70% probability of a Democratic majority, while the “Classic” “polls, fundraising, past voting patterns and more” model gives them only a 67% chance and the “Deluxe” “we add experts’ ratings to the classic” gives them only 62%.

Which model is optimal ? The Deluxe model gives the best fit with past elections — this must be true because it nests the other two. It would be true also if the deluxe model gave worse forecasts. Generally, the problem is over-fitting if one estimates many parameters. The Deluxe model adds few new parameters (I guess only one but I won’t check). An argument against it has to be not the standard watch out for overfitting argument (It will also outperform using the Akaike Information Criterion).

So how can I argue against it (and then go on to argue against the classic model)) ? Basically, I will argue that things have changed, so past performance is not a reliable indicator of future performance. Some changes are obvious — many more polls are conducted than used to be. Everyone, even experts, knows about averaging polls and how much better it works than looking at them and trying to judge. There is extreme turmoil.

OK now something along the line of evidence. I am going to present data with states in alphabetical order (and below selected so the figure is almost legible).

Here are deluxe forecasts

Here are lite (polls only) forecasts

You can see (if you have excellent eyes — fewer but larger numbers after the jump) that the deluxe forecasts are systematically more favorable to Republicans than the lite forecasts. This would be very unlikely if all of polls, money raised, past voting, and experts’ ratings gave unbiased estimates. The logic of addiing more useful variables is that it increases the precision of the estimates not that it eliminates bias. If there has been a shift of support from Republicans to Democrats, then forecasts based on past voting will be biased in favor of Republicans. It still often makes sense to include data on past voting, because it reduces the variance due to random sampling of the forecast. There are two arguments against — one is that the error in polls has low (not zero) correlation from state to state (one part of it is the change in true public opinion from polling day to election day) so, while each state forecast’s mean squared error is reduced by adding past voting patters as an explanatory variable, the nationwide calculations are worseened.

Another argument is (see above) there are many more polls than their used to be. This makes polling averages better forecasts (not as much as it would if each poll had an genuine independent sampling error but still a lot). If one decided on weights optimizing using data on old elections, one would put too low a weigh on the polling average. I think this happened.

The big change comes when experts’ ratings are added. Here one thing is that experts’ ratings are given by category
Solid R, Likely R, leans R, tossup, Leans D, likely D, Solid D. Now lets pretend that the “experts” have learned that the best strategy is to average polls, do a “lite” calculation then classify based on estimated probabilities. Such “experts” would add no useful information and would remove information. Also they would outperform the other “experts” just as Nate Silver systematically outperformed the previously recognized experts.

The key word here is “learned”. I speculated about a change over time from experts trying to infer without relying totally on polls to experts presenting polling averages as judgment calls.

This is a kind of herding. There is a difference between the strategy which enables me to give my best forecast (lowest mean squared error) and that whic make my forecast the most useful contribution to an average. The best strategy for Robert Waldmann would be to just cut and past fivethirtyeight (see above). My effort to improve on their three forecasts by emphasizing one which they don’t headline above the one which they present as the default is an effort to add something useful. Just following them is probably the best strategy to avoid embarrassment. But challenging them might be useful.

My wild guess is that experts have learned to average polls, then use the average to assign races to categories (eliminating useful information) then change a few of the ratings so it isn’t obvious what they did (and so their ratings aren’t identical to those of another expert). If that’s true, then their ratings used to contain useful information and don’t anymore.

The Cold Warriors

I have no expertise in this field. This post will not be cluttered with links, because I will write from memory and not link to anything. I suppose in a way, this post is a slap in the face of Tom Nichols, who is a subset of the topic, is supposed to be an expert on the topic, and is the author of “The Death of Expertise“. I will attempt to explain how his errors are due to envy and neurosis.

Honestly, my trigger was lest nasty (and less based on envy). Someone asked in a Tweet what is the consensus on the old domino theory which lead to US involvement in the war in Vietnam (which is also called “the American war” by the Vietnamese). I will put my anti Nichols spite after the jump (note I advertised his book). His alleged field of expertise is preventive war. There, that’s another advertisement. Actually I think I will just post a separate post sniping at him.

OK so the Domino theory.

The logic was as follows. In 1938, France, the British Empire, and Czechoslovakia could have stopped Hitler. But all he demanded was the Sudetenland which was predominately inhabited by ethnic Germans. Neville Chamberlain insisted on reaching an agreement. Benes and Daladier had to go along, so the chance to defeat Nazism with heavy but not immense losses was lost.

Heeeyyy wait a minute, wasn’t I supposed to be talking about the 1960s not 1938 ? Yes, but the first problem is that there are influential people in the USA for whom all years are 1938 (note I use the present tense — they are still around and are very dangerous).

The first key methodological assumption of the Domino theory is that all years are 1938 and all negotiations are held in Munich. The second is that Neville Chamberlain made every possible error, so, as long as one did the opposite of what he would have done, everything will be fine. The rest is commentary.

I now invite historians, experts in international relations and political scientists to contest my analysis (knowing that not very many read Angry Bear).

The domino theory maintained that the USA had to stop the Communists in Vietnam or else they would move on to Cambodia, Laos, Thailand, Burma, East Pakistan, and India. The logic was exactly (and only) that it would have been better to fight Hitler at the old fortified border between Germany and Czechoslovakia than to let him take the Sudetenland, then the rest of Czechoslovakia, and then fight him in Poland. Notably, Hitler was surprised when France and Britain declared war on September 1 1939. The theory was that, restraint, compromise, or the most dreaded retreat would be perceived as weakness and make further aggression inevitable. One detail was overlooked. Hitler was one person, Khrushchev, Mao and Ho Chi Minh were three different people. The USSR had advanced weapons, the PRC had huge armies, North Vietnam had no fear of either and knew how to play one off the other.

Then Khrushchev was overthrown by the Red Army. The communist Soviet Union had not reached the advanced stage of Communist development which made a Communist military dictatorship possible later in Poland, so the generals gave power to a troika (sleigh pulled by 3 horses). The first among equals was Leonid Brezhnev. The USA still faced 3 adversaries lead by men incapable of pity. Brezhnev was incapable of pity or any other higher mental functioning. He liked clowns. The one key qualification for being Khrushchev’s second in command was being a total idiot (preferably lacking in ambition) and therefore being no threat. From then on, the analogy should have been negotiating with Rudolf Hesse in Munich (Hesse was similarly chosen for his total idiocy).

Notably one of the challenges for the US war effort in South Vietnam was the fact that the Communist Pathet Lao effectively controlled Laos and Communist friendly (and superhumanly vain) Prince Norodom Sihanouk) controlled Cambodia. Notably this is a problem for the domino theory. The dominoes which were supposed to be knocked down by the fall of South Vietnam had it already fallen. Their impact on Thailand was fairly minor (it might not have seemed that way to the Thai communists who fought and died in the jungle, but there were never many of them and almost no one noticed when they gave up and made peace (I forget the date)).

In contrast, US firm resoluteness in Vietnam made a large fraction of the world (and a substantial minority in the USA) hate the US government. It is also estimated to have caused 3 million deaths (from surveys decades later asking people if they had lost relatives).

During the resolute effort against the Hitler like world Communist movement, the USSR and the PRC fought a border war. They became each others’ most bitter enemies — the USA was not even number one on their enemies list. Soon after the final victory of the Vietnamese Communists, there was a brief war between Vietnam and the PRC. The enemy was the enemy of the enemy. The Soviet Chinese war occurred some time during the US war in Vietnam. It should have changed everything. But to completely reverse US policy, US policymakers would have to admit that they had made a mistake, and that is not possible.

The after aftermath is that Communist China became more capitalist than the USA and the USSR collapsed. Impressively right up to the collapse, US hawks insisted that there was a high risk of Soviet conquest of the world. Also impressively the people who clearly demonstrated that they were clueless gained status from the collapse, because it (coincidentally) occurred while Ronald Reagan was in the White House. Oddly, some sincere people including Max Boot and Anne Applebaum took seriously Reagan’s claim to be a principled supporter of freedom around the world. I am not much older than they are and remember the distinction between acceptable authoritarians and unacceptable totalitarians (in other words our sons of bitches and sons of bitches who weren’t ours). I remember the mockery of Carter’s human rights campaign. I remember the US alliance with Argentine fascist mass murderers in opposition to the fascists non mass murdering Sandinistas (currently in power to remind us of the utter worthlessness of the domino theory).

So how can we assess the scientific standing of the domino theory ? How does it compare with the Ptolemaic model of the solar system, the phlogiston theory of burning, the caloric theory of heat, and the four humors theory of health and disease ? Digressions after the jump.

Naming Forts

It appears possible that the US military will cease to honor traitors and will change the names of bases named after Confederate generals. This raises the question of what new names to give them. This is one of the topics on which I have the very least expertise, so I will make my suggestions.

1) Fort York. Named after Sergeant Alvin York who, when he was corporal York during World War I, personally captured 132 German soldiers. I like the idea of naming a fort after a sergeant. Also I just learned that, when drafted, York initially was a conscientious objector before being convinced to the distinct advantage of the 132 German soldiers and ot the disadvantage of the 25 he killed when leading the attack on the German machine gun nest.

Only risk. Gaffe prone President Biden might slip up in the decidation ceremony and inadvertently plagiarize “Now is the Winter of our discontent maid glorious Summer by this noble son of York” *I still remember when Neil Kinnock’s ancestors mysteriously became Biden’s ancestors back in 1988).

2) Fort Bradley
Come on, station GIs in a fort named after the GI’s general.

3) Fort Howard, named after General Oliver Otis Howard head of the Freedman’s bureau and founder of Howard University. NO compromise with treason.

4) Fort Walker named after the only female Medal of Honore recipient Mary Edwards Walker MD. I’ll drink to that.

5) Fort Anderson named after James Anderson Jr who threw himself on a hand grenade in Cam Lo in 1967

They also served who died in pointless wars. We owe them gratitude along with infinite apologies. Infinite.

6) Fort Baldonado named after Jose Rodriguez Baldonado who doesn’t even have a Wikipedia article.

7) Fort Montgomery. Clearly there might be some need for disambiguation. I am writing as someone raised in Montgomery County Maryland hearing stories about the Montgomery Bus Boycot. I am thinking of lieutenant Jack C Montgomery, more or less the sergeant York of World War II.

I propose renaming Fort Rucker Alabama Fort Montgomery.

As a gesture at national unity play “Sweet Home Alabama” when dedicating it (hoping that people notice the closing line “My, Montgomery’s got the answer” which should have been completely clear in the context of the 1960s also “the governor boo boo boo” should have been fairly clear.

8) Fort Hayashi. I am thinking of Joe Hayashi, but it is OK if people think of Shizuya Hayashi. The name can honor two Medal of Honor recipients with one fort.

Hydroxychloroquine After Action Report

I was a vehement advocate of prescribing hydroxychloroquine (HCQ) off label while waiting for the results of clinical trials. I wasn’t all that much embarrassed to agree with Donald Trump for once. Now I feel obliged to note that my guess was totally wrong. I thought that the (uncertain) expected benefits were greater than the (relatively well known) costs.

The cost is that HCQ affects the heart beat prolonging the QT period (from when the atrium begins to contract to when the ventrical repolarizes and is read to go again). This can cause arrhythmia especially in people who already have heart problems. I understood that one might argue that all people with Covid 19 have heart problems but didn’t consider that argument decisive (I probably should have).

The positive expected value of the uncertain benefits was based on strong in vitro evidence that HCQ blocks SARS Cov2 infection of human cells in culture. (this is a publication in the world’s top general science journal).

Already in early May, there was evidence that any effect of HCQ on the rate of elimination of the virus must be small. In this controlled trial conducted in China, the null of no effect is not rejected. Much more importantly, the point estimates of the effects over time are all almost exactly zero. I considered the matter settled (although the painfully disappointed authors tried to argue for HCQ and that their study was not conclusive).

There are now four large retrospective studies all of which suggest no benefit from HCQ and two of which suggest it causes increased risk of death. I am going to discuss the two studies most recently reported.

One is a very large study (fairly big data goes to the hospital) published yesterday in The Lancet. In this study patients who received HCQ had a significantly higher death rate with a hazard of dying 1.335 times as high. The estimate comes from a proportional hazard model with a non parametric baseline probability and takes into account many risk factors including crucially initial disease severity. It is also important that only patients who were treated within 48 hours of diagnosis were considered.

I am, of course, dismayed by this result. I am also puzzled, because it is quite different from the result obtained in a smaller retrospective study published in JAMA

I think the practical lessons are that it seems unwise to give Covid 19 patients HCQ. Also maybe Robert Waldmann should be more humble. After the jump, I will discuss the two studies in some detail and propose an explanation of the difference in results.

The Amateur Epidemiologist II

I am interested in critiquing my understanding of the simplest SIR epidemiological model and also praising a critique of an effort to extend the model and guide policy developed by some very smart economic theorists.

First the useful point is that this post by Noah Smith is brilliant. As is typical, Smith argues that the useful implications economic models depend on strong assumptions so economic theory isn’t very useful. He praises simple empirical work instead.

I will discuss Smith contra Acemoglu, Chernozhukov, Werning and Whinston and Smith pro Sergio Correia, Stephan Luck and Emil Verner after the jump, but really Smith is better at presenting Smith than I am.

It made me wonder. In the simplest model herd immunity stops an epidemic when 1-1/R0 of people have been infected. R0 as I recently learned and everyone now knows is the number of people who would catch a pathogen from one infected person if no one had any resistence. Over time people develop resistence so Rt < R0. If 1-1/R0 of people are resistent, then Rt =1. A bit later Rt<1 so each infected person will lead to a geometrical decreasing series of expected infections so total infections would be 1-1/R0 plus a (small) constant over the number infected at that critical time t. The SIR has susceptible, infected and resistent. The idea is that if one has not been exposed one is vulnerable. If one becomes infected, one carries and sheds the pathogen for a while and then one recovers. After one recovers one is immune and won't get it again. The key assumption in the model is that for every infected people R0 people are exposed (and infected if not immune) and that those people are chosen at random out of the entire population. It is necessary to assume that spread is equally likely from Mr A to Ms B if they share a house or live on opposite sides of the country. This is a silly assumption and the model is the old model used to teach kids and not, I'm sure, current research. It is also the model always used to guide public policy decisions (see me contra benchmark models ) In population biology and evolutionary biology the silly assumption is called "pan mictic" in economics it is called "random matching". The assumption is made very often because doing without it can get one stuck in really hard math. I would like to put a few minutes of effort into trying to figure out if the random matching assumption affects the level of infection needed for herd immunity (of course everyone knows it matters a lot). Below I will always assume R0 =3. Model 2 the population is actually divided into N equal subpopulations and there is no spread from one to the other. The disease starts with one case in one sub population. It will spread until a few more than two thirds of that population has been infected. Spread will stop when 1/(3N) of the whole population is infected. the relaxation of random matching assumption reduceces the incidence needed for herd ommunity by the factor N. This works for any N. Model 3 very like model 2. Half of people have innate immunity to the virus. People transmit the virus to on average 6 other people (on average 3 have innate immunity). the virus will spread until 5 of 6 are immune. that means (5/6)-(1/2) = 1/3 must acquire immunity (by getting infected). So 1/3 not 2/3. OK can we be sure that the number who will get onfected is less than 2/3 ? Consider Model 4. people live one to a square of an invisible chess board (which is a really big square) they transmit the pathogen to those with whoù they share an edge. R0 = 3 (I get it from 1 neighbor and early in the epidemic give it to my other 3 not yet infected neighbors). How many people get infected ? All of them Katy. The currently infected are always in the border zone between the resistent and the vulnerable. So R0 = 3 implies herd immunity will stop the spread at some level which ranges from 1/(3N) for N as big as I like, to 100%. R0=3 and a priori reasoning without arbitrary assumptions which we know are false and make for convenience tells us nothing at all. Without some assumption about mixing, matching, and population structure, the core SIR assumptions have no implications. Maybe economists and epidemiologists have more in common than we thought.

Antiviral Rumors

Tired: Remdesivir
Wired: Merimopodib
Inspired: Both

Merimopodib (of which I just read for the first time) is an inhibitor of an enzyme used to make Guanosine. Viruses need a lot of Guanosine (and other nucleosides) to reproduce, so it is an antiviral. It can be taken orally and there is a known safe dose.

A preprint asserts that a combination of Remdesivir and Merimopodib completely blocks SARS-CoV-2 replication in vitro.

Here is the abstract

The IMPDH inhibitor merimepodib provided in combination with the adenosine analogue remdesivir reduces SARS-CoV-2 replication to undetectable levels in vitro [version 1; peer review: awaiting peer review]
Natalya Bukreyeva, Rachel A. Sattler, Emily K. Mantlo1, Timothy Wanninger, John T. Manning, Cheng Huang, Slobodan Paessler, Jerome B. Zeldis

Home Browse The IMPDH inhibitor merimepodib provided in combination with the adenosine…









The IMPDH inhibitor merimepodib provided in combination with the adenosine analogue remdesivir reduces SARS-CoV-2 replication to undetectable levels in vitro [version 1; peer review: awaiting peer review]
Natalya Bukreyeva1, Rachel A. Sattler1, Emily K. Mantlo1, Timothy Wanninger1, John T. Manning, Cheng Huang1, Slobodan Paessler1, Jerome B. Zeldis2

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is the novel coronavirus responsible for the ongoing COVID-19 pandemic, which has resulted in over 2.5 million confirmed cases and 170,000 deaths worldwide as of late April 2020. The pandemic currently presents major public health and economic burdens worldwide. No vaccines or therapeutics have been approved for use to treat COVID-19 cases in the United States despite the growing disease burden, thus creating an urgent need for effective treatments. The adenosine analogue remdesivir (REM) has recently been investigated as a potential treatment option, and has shown some activity in limiting SARS-CoV-2 replication. We previously reported that the IMPDH inhibitor merimepodib (MMPD) provides a dose-dependent suppression of SARS-CoV-2 replication in vitro. Here, we report that a 4-hour pre-treatment of Vero cells with 2.5µM MMPD reduces the infectious titer of SARS-CoV-2 more effectively than REM at the same concentration. Additionally, pre-treatment of Vero cells with both REM and MMPD in combination reduces the infectious titer of SARS-CoV-2 to values below the detectable limit of our TCID50 assay. This result was achieved with concentrations as small as 1.25 µM MMPD and 2.5 µM REM. At concentrations of each agent as low as 0.31 µM, significant reduction of viral production occurred. This study provides evidence that REM and MMPD administered in combination might be an effective treatment for COVID-19 cases.

Remdesivir VIII

There is a severe Remdesivir shortage

On March 2 2020, I warned you that this was going to happen.

I did not warn about the opaque and arbitrary Trump administration policy, because the Trump administration is always “worse than you imagine possible even taking into account the fact that it is worse than you imagine possible” Brad DeLong 2003 or so referring to the last Republican presidency.

When are Americans going to notice the pattern ?

The Amateur Epidemiologist

I frequently read a debate about whether, when assessing anti covid 19 performance, one should look at deaths per capita or deaths on days since 1000 deaths. Like everything involving Americans, this has become a pro v contra Trump debate — clearly he wants deaths per capita (and the absolute number of tests performed).

The arguments are as follows. for number of deaths on time since a certain number was reached, it is argued that all countries are at the negligible fraction of people are resistant (naturally exponential growth) stage, so the relevant variable is rate of growth of cases (or deaths). So cases now divided by cases a week ago and not by population.

The counterargument is that, come on it’s obvious.

I think that it is natural to expect a transition from roughly the same growth (no matter what population is) to cases (very roughly) proportional to population. All of this is during the neglible fraction resistant phase.

I am going to set up a straw man and knock him down with a silly super super simple model. So the straw man is that it is reasonable to assume that if two countries have the same number of cases at time t, then they will have similar numbers later. The silly model is that people live on a giant chess board (1000 squares on a side) and infect people who share an edge. This gives R_0 between 2 and 3. So say start with two cases, one in each country. Straw man says there should be the same number of cases in each country in each subsequent period.

OK now country one is the upper right quadrant and country two is the rest of the board. Strra man predicts the same number of cases. Or what if all is the same but I draw the border so country 2 is the lower left quadrant and country 1 is the rest. Again the same number.

So straw man concludes that there are never any cases in the lower right or upper left. This can’t be right.

Now I will discuss a model which is slightly less silly. Assume most transmission is local so the infected and the infector are in the same country. Assume people are infectious for one period and that, during tht time, each infected person infects n nearby people. Also assume lower rate of distant infection, so an infected person infects someone chosen at random in the whole world with probability m less than 1 less than n.

This distant infection seeds a new outbreak with a new patient 1.

Assume that at t=1, each country has the same number of infected people.

There are countries indexed by i and caseload x_it.
x_(it+1) = n x_(it) + (sum_j x_j)m(population_i/ (sum_j population_j))

If m is much less than n, then, at first the rate of growth in all countries is roughly n. But eventually x_(it) becomes proportional to population_i .

The reason is that, in each country, there is the same number of people infected in the outbreaks that had already started at time 1. However, the number of new outbreaks is proportionatl to population (from someone chosen at random in the whole world). So the (expected) number of people infected in outbreaks which started after t=1 is proportional to population.

As t goes to infinity, the fraction of infected people infected in the outbreaks which had already started at t=1 goes to zero. So in the medium run (after a lot of long distance transmission but before there is a significant fraction of resistant people) the infection rates per capita converge.

OK the bit about initial growth is similar conditional on similar numbers infected at t=1 sure fits the data (where t=1 is t when the number of infections passes say 1000). Thus people could talk about “days behind Italy” and accurately predict the number of cases (and not change how many days behind different countries are).

But on the other hand, after a while, similar countries have rates roughly proportional to population. So, for example, the number of cases in the USA is similar to the number of cases in Europe.

The alternative is to claim this figure illustrates a pure coincidence.