Author Archives: sanhei

Clinical trials of other drugs in other species to predict a drug’s effect in humans

Suppose we want to know whether a drug is safe or effective for humans, but do not have data on what it does in humans, only on its effects in mice, rats, rhesus macaques and chimpanzees. In general, we can predict the effect of the drug on humans better with the animal data than without it. Information on “nearby” realisations of a random variable (effect of the drug) helps predict the realisation we are interested in. The method should weight nearby observations more than observations further away when predicting. For example, if the drug has a positive effect in animals, then predicts a positive effect in humans, and the larger the effect in animals, the greater the predicted effect in humans.

A limitation of weighting is that it does not take into account the slope of the effect when moving from further observations to nearer. For example, a very large effect of the drug in mice and rats but a small effect in macaques and chimpanzees predicts the same effect in humans as a small effect in rodents and a large one in monkeys and apes, if the weighted average effect across animals is the same in both cases. However, intuitively, the first case should have a smaller predicted effect in humans than the second, because moving to animals more similar to humans, the effect becomes smaller in the first case but larger in the second. The idea is similar to a proportional integral-derivative (PID) controller in engineering.

The slope of the effect of the drug is extra information that increases the predictive power of the method if the assumption that the similarity of effects decreases in genetic distance holds. Of course, if this assumption fails in the data, then imposing it may result in bias.

Assumptions may be imposed on the method using constrained estimation. One constraint is the monotonicity of the effect in some measure of distance between observations. The method may allow for varying weights by adding interaction terms (e.g., the effect of the drug times genetic similarity). The interaction terms unfortunately require more data to estimate.

Extraneous information about the slope of the effect helps justify the constraints and reduces the need for adding interaction terms, thus decreases the data requirement. An example of such extra information is whether the effects of other drugs that have been tested in these animals as well as humans were monotone in genetic distance. Using information about these other drugs imposes the assumption that the slopes of the effects of different drugs are similar. The similarity of the slopes should intuitively depend on the chemical similarity of the drugs, with more distant drugs having more different profiles of effects across animals.

The similarity of species in terms of the effects drugs have on them need not correspond to genetic similarity or the closeness of any other observable characteristic of these organisms, although often these similarities are similar. The similarity of interest is how similar the effects of the drug are across these species. Estimating this similarity based on the similarity of other drugs across these animals may also be done by a weighted regression, perhaps with constraints or added interaction terms. More power for the estimation may be obtained from simultaneous estimation of the drug-effect-similarity of the species and the effect of the drug in humans. An analogy is demand and supply estimation in industrial organisation where observations about each side of the market give information about the other side. Another analogy is duality in mathematics, in this case between the drug-effect-similarity of the species and the given drug’s similarity of effects across these species.

The similarity of drugs in terms of their effects on each species need not correspond to chemical similarity, although it often does. The similarity of interest for the drugs is how similar their effects are in humans, and also in other species.

The inputs into the joint estimation of drug similarity, species similarity and the effect of the given drug in humans are the genetic similarity of the species, the chemical similarity of the drugs and the effects for all drug-species pairs that have been tested. In the matrix where the rows are the drugs and the columns the species, we are interested in filling in the cell in the row “drug of interest” and the column “human”. The values in all the other cells are informative about this cell. In other words, there is a benefit from filling in these other cells of the matrix.

Given the duality of drugs and species in the drug effect matrix, there is information to be gained from running clinical trials of chemically similar human-use-approved drugs in species in which the drug of interest has been tested but the chemically similar ones have not. The information is directly about the drug-effect-similarity of these species to humans, which indirectly helps predict the effect of the drug of interest in humans from the effects of it in other species. In summary, testing other drugs in other species is informative about what a given drug does in humans. Adapting methods from supply and demand estimation, or otherwise combining all the data in a principled theoretical framework, may increase the information gain from these other clinical trials.

Extending the reasoning, each (species, drug) pair has some unknown similarity to the (human, drug of interest) pair. A weighted method to predict the effect in the (human, drug of interest) pair may gain power from constraints that the similarity of different (species, drug) pairs increases in the genetic closeness of the species and the chemical closeness of the drugs.

Define Y_{sd} as the effect of drug d in species s. Define X_{si} as the observable characteristic (gene) i of species s. Define X_{dj} as the observable characteristic (chemical property) j of drug d. The simplest method is to regress Y_{sd} on all the X_{si} and X_{dj} and use the coefficients to predict the Y_{sd} of the (human, drug of interest) pair. If there are many characteristics i and j and few observations Y_{sd}, then variable selection or regularisation is needed. Constraints may be imposed, like X_{si}=X_i for all s and X_{dj}=X_j for all d.

Fused LASSO (least absolute shrinkage and selection operator), clustered LASSO and prior LASSO seem related to the above method.

Leader turnover due to organisation performance is underestimated

Berry and Fowler (2021) “Leadership or luck? Randomization inference for leader effects in politics, business, and sports” in Science Advances propose a method they call RIFLE for testing the null hypothesis that leaders have no effect on organisation performance. The method is robust to serial correlation in outcomes and leaders, but not to endogenous leader turnover, as Berry and Fowler honestly point out. The endogeneity is that the organisation’s performance influences the probability that the leader is replaced (economic growth causes voters to keep a politician in office, losing games causes a team to replace its coach).

To test whether such endogeneity is a significant problem for their results, Berry and Fowler regress the turnover probability on various measures of organisational performance. They find small effects, but this underestimates the endogeneity problem, because Berry and Fowler use linear regression, forcing the effect of performance on turnover to be monotone and linear.

If leader turnover is increased by both success (get a better job elsewhere if the organisation performs well, so quit voluntarily) and failure (fired for the organisation’s bad performance), then the relationship between turnover and performance is U-shaped. Average leaders keep their jobs, bad and good ones transition elsewhere. This is related to the Peter Principle that an employee is promoted to her or his level of incompetence. A linear regression finds a near-zero effect of performance on turnover in this case even if the true effect is large. How close the regression coefficient is to zero depends on how symmetric the effects of good and bad performance on leader transition are, not how large these effects are.

The problem for the RIFLE method of Berry and Fowler is that the small apparent effect of organisation performance on leader turnover from OLS regression misses the endogeneity in leader transitions. Such endogeneity biases RIFLE, as Berry and Fowler admit in their paper.

The endogeneity may explain why Berry and Fowler find stronger leader effects in sports (coaches in various US sports) than in business (CEOs) and politics (mayors, governors, heads of government). A sports coach may experience more asymmetry in the transition probabilities for good and bad performance than a politician. For example, if the teams fire coaches after bad performance much more frequently than poach coaches from well-performing competing teams, then the effect of performance on turnover is close to monotone: bad performance causes firing. OLS discovers this monotone effect. On the other hand, if politicians move with equal likelihood after exceptionally good and bad performance of the administrative units they lead, then linear regression finds no effect of performance on turnover. This misses the bias in RIFLE, which without the bias might show a large leader effect in politics also.

The unreasonably large effect of governors on crime (the governor effect explains 18-20% of the variation in both property and violent crime) and the difference between the zero effect of mayors on crime and the large effect of governors that Berry and Fowler find makes me suspect something is wrong with that particular analysis in their paper. In a checks-and-balances system, the governor should not have that large of influence on the state’s crime. A mayor works more closely with the local police, so would be expected to have more influence on crime.

Reduce temptation by blocking images

Web shops try to tempt customers into unnecessary and even harmful purchases, including grocery and food ordering sites which promote unhealthy meals. The temptation can be reduced by blocking images on shopping websites. I find it useful when ordering food. Similarly, Facebook and news sites try to tempt viewers with clickbait and ads. To reduce my time-wasting, I make the clickbait less attractive by blocking images. The pictures in most news stories do not contribute any information – a story about a firm has a photo of the main building or logo of the firm or the face of its CEO, a “world leaders react to x” story has pictures of said leaders.

The blocking may require a browser extension (“block images”) and each browser and version has a little different steps for this.

In Chromium on 20 Jan 2021, no extension is needed:

1) click the three vertical dots at the top right,

2) click Settings to go to chrome://settings/,

3) scroll down to Site settings, click it,

4) scroll down to Images, click it.

5) Click the Add button to the right of the Block heading. A dialog pops up to enter a web address.

6) Copy the url of the site on which you want to block pictures, for example https://webshop.com into the Site field.

If seeing the images is necessary for some reason, then re-enable images on the website: follow steps 1-4 above, then click the three vertical dots under the Add button under the Block heading. A menu of three options pops up. Click the Allow option.

Alternatively, you may block all images on all websites and then allow only specific sites to show images. For this, follow steps 1-4 above, then click the blue button to the right of the Allow all (recommended) heading. Then click the Add button next to Allow. A dialog pops up to enter a web address. Copy the url of the site on which you want to block pictures, for example https://webshop.com into the Site field.

Dilution effect explained by signalling

Signalling confidence in one’s arguments explains the dilution effect in marketing and persuasion. The dilution effect is that the audience averages the strength of a persuader’s arguments instead of adding the strengths. More arguments in favour of a position should intuitively increase the confidence in the correctness of this position, but empirically, adding weak arguments reduces people’s belief, which is why drug advertisements on US late-night TV list mild side effects in addition to serious ones. The target audience of these ads worries less about side effects when the ad mentions more slight problems with the drug, although additional side effects, whether weak or strong, should make the drug worse.

A persuader who believes her first argument to be strong enough to convince everyone does not waste valuable time to add other arguments. Listeners evaluate arguments partly by the confidence they believe the speaker has in these claims. This is rational Bayesian updating because a speaker’s conviction in the correctness of what she says is positively correlated with the actual validity of the claims.

A countervailing effect is that a speaker with many arguments has spent significant time studying the issue, so knows more precisely what the correct action is. If the listeners believe the bias of the persuader to be small or against the action that the arguments favour, then the audience should rationally believe a better-informed speaker more.

An effect in the same direction as dilution is that a speaker with many arguments in favour of a choice strongly prefers the listeners to choose it, i.e. is more biased. Then the listeners should respond less to the persuader’s effort. In the limit when the speaker’s only goal is always for the audience to comply, at any time cost of persuasion, then the listeners should ignore the speaker because a constant signal carries no information.

Modelling

Start with the standard model of signalling by information provision and then add countersignalling.

The listeners choose either to do what the persuader wants or not. The persuader receives a benefit B if the listeners comply, otherwise receives zero.

The persuader always presents her first argument, otherwise reveals that she has no arguments, which ends the game with the listeners not doing what the persuader wants. The persuader chooses whether to spend time at cost c>0, c<B to present her second argument, which may be strong or weak. The persuader knows the strength of the second argument but the listeners only have the common prior belief that the probability of a strong second argument is p0. If the second argument is strong, then the persuader is confident, otherwise not.

If the persuader does not present the second argument, then the listeners receive an exogenous private signal in {1,0} about the persuader’s confidence, e.g. via her subconscious body language. The probabilities of the signals are Pr(1|confident) =Pr(0|not) =q >1/2. If the persuader presents the second argument, then the listeners learn the confidence with certainty and can ignore any signals about it. Denote by p1 the updated probability that the audience puts on the second argument being strong.

If the speaker presents a strong second argument, then p1=1, if the speaker presents a weak argument, then p1=0, if the speaker presents no second argument, then after signal 1, the audience updates their belief to p1(1) =p0*q/(p0*q +(1-p0)*(1-q)) >p0 and after signal 0, to p1(0) =p0*(1-q)/(p0*(1-q) +(1-p0)*q) <p0.

The listeners prefer to comply (take action a=1) when the second argument of the persuader is strong, otherwise prefer not to do what the persuader wants (action a=0). At the prior belief p0, the listeners prefer not to comply. Therefore a persuader with a strong second argument chooses max{B*1-c, q*B*1 +(1-q)*B*0} and presents the argument iff (1-q)*B >c. A persuader with a weak argument chooses max{B*0-c, (1-q)*B*1 +q*B*0}, always not to present the argument. If a confident persuader chooses not to present the argument, then the listeners use the exogenous signal, otherwise use the choice of presentation to infer the type of the persuader.

One extension is that presenting the argument still leaves some doubt about its strength.

Another extension has many argument strength levels, so each type of persuader sometimes presents the second argument, sometimes not.

In this standard model, if the second argument is presented, then always by the confident type. As is intuitive, the second argument increases the belief of the listeners that the persuader is right. Adding countersignalling partly reverses the intuition – a very confident type of the persuader knows that the first argument already reveals her great confidence, so the listeners do what the very confident persuader wants. The very confident type never presents the second argument, so if the confident type chooses to present it, then the extra argument reduces the belief of the audience in the correctness of the persuader. However, compared to the least confident type who also never presents the second argument, the confident type’s second argument increases the belief of the listeners.

University incentives not to punish cheating

Universities have incentives not to expel cheaters, because these students would stop paying fees. By extension, there is a motive to avoid punishing academic dishonesty in any way that increases the chance of the student dropping out, like giving a cheater a failing grade. The university then gives a similar incentive not to punish academic dishonesty to its departments, who then pass on these incentives to faculty. In every university I have been in, it is far easier for a faculty member to do nothing about cheating – it requires no work, as opposed to a lot of bureaucracy documenting the cheating, imposing the punishment, dealing with the appeals, etc. There is no punishment for a faculty member who fails to report cheating, but to be fair, also no punishment for a false accusation of cheating, or an accusation that does not have sufficient evidence or gets overturned on appeal.

Even if a faculty member tries to punish academic dishonesty, the cheating student appeals to the university hierarchy and the higher-ups overturn the punishment. If not the first level of the hierarchy, then one of the higher levels. Thus even an inherently honest professor who for psychological reasons would be willing to spend the time to document academic dishonesty if it led to the cheater getting punished does not do so, because in the end there is no punishment.

A solution is to change university incentives: the students who are expelled because of cheating must pay their fees in full for their entire course of studies. A problem is that this leads to legal challenges because the student is not getting the education service but must pay for it. One solution is to require the tuition paid up front, non-refundable on expulsion for cheating. However, in this case students have to take a loan (which may be prevented by credit constraints or risk aversion) and may instead choose universities that do not charge them up front.

The university-side incentives also seem problematic if tuition is fully paid for expelled cheaters, because the university could save on the teaching costs by kicking out all the students on fabricated charges and keep the money. The long-term reputation cost for the university prevents such rip-offs.

A milder way to improve the university incentives is to require the cheater to re-take the course. This may delay the time at which the cheater can take follow-up courses that have the re-taken course as a prerequisite. The resulting delay in graduation may require the cheater to pay extra fees for the additional time, but the extra payment of course depends on the specific regulations of the university.

Virulence of a disease may cause vaccines to be effective

My uninformed speculation: vaccines may be so effective against Covid-19 (90-95% vs flu vaccine 70%) for the same reason why Covid-19 is so infectious – it binds strongly to biochemicals in the organism. If high affinity to the angiotensin-converting enzyme 2 on the surfaces of lung cells is positively correlated with strong binding to antibodies and immune cells, then the immune system, once triggered, removes the viral particles faster for those respiratory viruses that infect cells more easily. Strong binding and the consequent intense immune triggering may also be the reason for the life-threatening immune overreaction (cytokine storm) to the novel coronavirus.
This hypothesis could be tested on a cross-sectional dataset of viral diseases using some measure of the infectiousness of a disease, the effectiveness of a vaccine against it and the frequency of immune overreaction to it.
Infectiousness may be measured by ID50: what number of microbes makes half the organisms exposed to this number sick. This measure depends on the state of the organisms studied. For example, if people’s immune system is weaker in the winter on average, then ID50 measured in the winter is lower than in the summer.
Vaccine effectiveness is typically measured in percent – what fraction of vaccinated people are protected, in the sense that they do not catch the disease in circumstances in which unvaccinated people catch it. This measure of may depend on what the exposure to the disease is. For example, if a large enough dose of the microbe makes everyone sick, vaccinated or no, then exposure to this dose shows zero effect of the vaccine. Similarly, if a small enough dose fails to infect anyone, then the vaccine effect seems zero, but at least the lack of infections among the unvaccinated shows that no information about vaccine efficacy can be obtained from this exposure test.
Immune overreaction needs to be confidently ascribable to the disease studied for it to be a relevant measure for testing the theory about the connection between virulence and vaccine efficacy.

Preventing cheating is hopeless in online learning

Technology makes cheating easy even in in-person exams with invigilators next to the test-taker. For example, in-ear wireless headphones not visible externally can play a loop recording of the most important concepts of the tested material. A development of this idea is to use a hidden camera in the test-takers glasses or pen to send the exam contents to a helper who looks up the answers and transmits the spoken solutions via the headphones. Without a helper, sophisticated programming is needed: the image of the exam from the hidden camera is sent to a text-recognition (OCR) program, which pipes it to a web search or an online solver such as Wolfram Alpha, then uses a text-to-speech program to speak the results into the headphones.

A small screen on the inside of the glasses would be visible to a nearby invigilator, so is a risky way to transmit solutions. A small projector in the glasses could in theory display a cheat sheet right into the eye. The reflection from the eye would be small and difficult to detect even looking into the eyes of the test-taker, which are mostly pointed down at the exam.

If the testing is remote, then the test-taker could manipulate the cameras through which the invigilators watch, so that images of cheat sheets are replaced with the background and the sound of helpers saying answers is removed. The sound is easy to remove with a microphone near the mouth of the helper, the input of which is subtracted from the input of the computer webcam. A more sophisticated array of microphones feeding sound into small speakers near the web camera’s microphone can be used to subtract a particular voice from the web camera’s stereo microphone’s input. The technology is the same as in noise-cancelling headphones.

Replacing parts of images is doable even if the camera and its software are provided by the examiners and completely non-manipulable. The invigilators’ camera can be pointed at a screen which displays an already-edited video of the test-taker. The editing is fast enough to make it nearly real-time. The idea of the edited video is the same as in old crime movies where a photo of an empty room is stuck in front of a stationary security camera. Then the guard sees the empty room on the monitor no matter what actually goes on in the room.

There is probably a way to make part of the scene invisible to a camera even with 19th century technology, namely the Pepper’s Ghost illusion with a two-way mirror. The edges of the mirror have to be hidden somehow.

All public statues should be removed

There is no benefit to spending taxpayer money on creating or sustaining personality cults. The same goes for all public art – the current (local) government should not decide on which people to popularise. No significant market failure exists in physical art objects. The government thus does not need to intervene in the market for statues (copying digital art is another matter). Private individuals can put almost any statues and art on their own property as part of free speech.

The materials of which the statues are made could be used for something beneficial instead, like public housing for the poorest members of society. Clearly the government’s goal in erecting statues is to provide circus to the public in order to get re-elected, not to benefit society.

If the influential people whom the statues depict were asked whether the person or the idea matters more, my guess is that almost all would emphasise the idea. Most would ask the resources to be spent on more reasonable things than statues of them.

If the goal of a statue is to signal the importance of the ideas of the person depicted, then there are more efficient ways for this signalling. For example, a scholarship, a charity or a public library in the name of the person.

Partial cleaning may make surfaces look dirtier

The reason why incomplete cleaning may increase the visual perception of dirt is by increasing the contrast between the patches of thicker grime and the normal colour by removing a uniform covering of thinner dirt. If something is uniformly grimy, then the colour of the covering dirt may be perceived as the thing’s normal hue. Cleaning may remove approximately the same thickness of dirt from all points on the surface. If some patches initially have a thicker layer, then these remain the colour of the dirt after the cleaning, but other areas may be fully cleaned and revert to the original look of the surface. The human visual system mostly perceives contrast, not the absolute wavelength of the reflected light, as various optical illusions demonstrate. Higher contrast between the thicker patches of grime and the rest of the surface then enhances the perception of dirtiness.

Bar-coding videos to prevent faking

To prevent clips from being cut out of a video or inserted, add a non-repeating sequence of bar codes onto either the whole frame or the main object of the video, such as a person talking. The bar code can use subtle „watermark” shading that does not interfere with viewing – it only needs to be readable by computer. The sequence of bar codes can be recreated at a later time if the algorithm is known, so if a clip is cut out of the video or added, then the sequence no longer matches the replication. Altering the video frame by frame also changes the bar code, although the forger can bypass this security feature by reading the original bar code, removing it before retouching and adding it back later. Still, these extra steps make faking the video somewhat more difficult. The main security feature is that the length of the video cannot be changed without altering the sequence of bar codes, which is easily detected.

The maker of the video may generate the bar codes cryptographically using a private key. This enables confirming the source of the video, for example in a copyright dispute.

Probably the idea of bar-coding videos has already been implemented, because watermarks and time stamps on photos have long been used. The main novelty relative to treating each frame as a photo is to link the bar codes to each other over time.