# Posts Tagged probability and statistics

### My Trouble with Bayes

In past consulting work I’ve wrestled with subjective probability values derived from expert opinion. Subjective probability is an interpretation of probability based on a degree of belief (i.e., hypothetical willingness to bet on a position) as opposed a value derived from measured frequencies of occurrences (related posts: Belief in Probability, More Philosophy for Engineers). Subjective probability is of interest when failure data is sparse or nonexistent, as was the data on catastrophic loss of a space shuttle due to seal failure. Bayesianism is one form of inductive logic aimed at refining subjective beliefs based on Bayes Theorem and the idea of rational coherence of beliefs. A NASA handbook explains Bayesian inference as the process of obtaining a conclusion based on evidence,  “Information about a hypothesis beyond the observable empirical data about that hypothesis is included in the inference.” Easier said than done, for reasons listed below.

Bayes Theorem itself is uncontroversial. It is a mathematical expression relating the probability of A given that B is true to the probability of B given that A is true and the individual probabilities of A and B:

P(A|B) = P(B|A) x P(A) / P(B)

If we’re trying to confirm a hypothesis (H) based on evidence (E), we can substitute H and E for A and B:

P(H|E) = P(E|H) x P(H) / P(E)

To be rationally coherent, you’re not allowed to believe the probability of heads to be .6 while believing the probability of tails to be .5; the sum of chances of all possible outcomes must sum to exactly one. Further, for Bayesians, the logical coherence just mentioned (i.e., avoidance of Dutch book arguments) must hold across time (synchronic coherence) such that once new evidence E on a hypothesis H is found, your believed probability for H given E should equal your prior conditional probability for H given E.

Plenty of good sources explain Bayesian epistemology and practice far better than I could do here. Bayesianism is controversial in science and engineering circles, for some good reasons. Bayesianism’s critics refer to it as a religion. This is unfair. Bayesianism is, however, like most religions, a belief system. My concern for this post is the problems with Bayesianism that I personally encounter in risk analyses. Adherents might rightly claim that problems I encounter with Bayes stem from poor implementation rather than from flaws in the underlying program. Good horse, bad jockey? Perhaps.

Problem 1. Subjectively objective
Bayesianism is an interesting mix of subjectivity and objectivity. It imposes no constraints on the subject of belief and very few constraints on the prior probability values. Hypothesis confirmation, for a Bayesian, is inherently quantitative, but initial hypotheses probabilities and the evaluation of evidence is purely subjective. For Bayesians, evidence E confirms or disconfirms hypothesis H only after we establish how probable H was in the first place. That is, we start with a prior probability for H. After the evidence, confirmation has occurred if the probability of H given E is higher than the prior probability of H, i.e., P(H|E) > P(H). Conversely, E disconfirms H when P(H|E) < P(H). These equations and their math leave business executives impressed with the rigor of objective calculation while directing their attention away from the subjectivity of both the hypothesis and its initial prior.

2. Rational formulation of the prior
Problem 2 follows from the above. Paranoid, crackpot hypotheses can still maintain perfect probabilistic coherence. Excluding crackpots, rational thinkers – more accurately, those with whom we agree – still may have an extremely difficult time distilling their beliefs, observations and observed facts of the world into a prior.

3. Conditionalization and old evidence
This is on everyone’s short list of problems with Bayes. In the simplest interpretation of Bayes, old evidence has zero confirming power. If evidence E was on the books long ago and it suddenly comes to light that H entails E, no change in the value of H follows. This seems odd – to most outsiders anyway. This problem gives rise to the game where we are expected to pretend we never knew about E and then judge how surprising (confirming) E would have been to H had we not know about it. As with the general matter of maintaining logical coherence required for the Bayesian program, it is extremely difficult to detach your knowledge of E from the rest of your knowing about the world. In engineering problem solving, discovering that H implies E is very common.

4. Equating increased probability with hypothesis confirmation.
My having once met Hillary Clinton arguably increases the probability that I may someday be her running mate; but few would agree that it is confirming evidence that I will do so. See Hempel’s raven paradox.

5. Stubborn stains in the priors
Bayesians, often citing success in the business of establishing and adjusting insurance premiums, report that the initial subjectivity (discussed in 1, above) fades away as evidence accumulates. They call this washing-out of priors. The frequentist might respond that with sufficient evidence your belief becomes irrelevant. With historical data (i.e., abundant evidence) they can calculate P of an unwanted event in a frequentist way: P = 1-e to the power -RT, roughly, P=RT for small products of exposure time T and failure rate R (exponential distribution). When our ability to find new evidence is limited, i.e., for modeling unprecedented failures, the prior does not get washed out.

6. The catch-all hypothesis
The denominator of Bayes Theorem, P(E), in practice, must be calculated as the sum of the probability of the evidence given the hypothesis plus the probability of the evidence given not the hypothesis:

P(E) = [P(E|H) x p(H)] + [P(E|~H) x P(~H)]

But ~H (“not H”) is not itself a valid hypothesis. It is a family of hypotheses likely containing what Donald Rumsfeld famously called unknown unknowns. Thus calculating the denominator P(E) forces you to pretend you’ve considered all contributors to ~H. So Bayesians can be lured into a state of false choice. The famous example of such a false choice in the history of science is Newton’s particle theory of light vs. Huygens’ wave theory of light. Hint: they are both wrong.

7. Deference to the loudmouth
This problem is related to no. 1 above, but has a much more corporate, organizational component. It can’t be blamed on Bayesianism but nevertheless plagues Bayesian implementations within teams. In the group formulation of any subjective probability, normal corporate dynamics govern the outcome. The most senior or deepest-voiced actor in the room drives all assignments of subjective probability. Social influence rules and the wisdom of the crowd succumbs to a consensus building exercise, precisely where consensus is unwanted. Seidenfeld, Kadane and Schervish begin “On the Shared Preferences of Two Bayesian Decision Makers” with the scholarly observation that an outstanding challenge for Bayesian decision theory is to extend its norms of rationality from individuals to groups. Their paper might have been illustrated with the famous photo of the exploding Challenger space shuttle. Bayesianism’s tolerance of subjective probabilities combined with organizational dynamics and the shyness of engineers can be a recipe for disaster of the Challenger sort.

All opinions welcome.

### A New Era of Risk Management?

The quality of risk management has mostly fallen for the past few decades. There are signs of change for the better.

Risk management is a broad field; many kinds of risk must be managed. Risk is usually defined in terms of probability and cost of a potential loss. Risk management, then, is the identification, assessment and prioritization of risks and the application of resources to reduce the probability and/or cost of the loss.

The earliest and most accessible example of risk management is insurance, first documented in about 1770 BC in the Code of Hammurabi (e.g., rules 23, 24, and 48). The Code addresses both risk mitigation, through threats and penalties, and minimizing loss to victims, through risk pooling and insurance payouts.

Insurance was the first example of risk management getting serious about risk assessment. Both the frequentist and quantified subjective risk measurement approaches (see recent posts on belief in probability) emerged from actuarial science developed by the insurance industry.

Risk assessment, through its close relatives, decision analysis and operations research, got another boost from World War II. Big names like Alan Turing, John Von Neumann, Ian Fleming (later James Bond author) and teams at MIT, Columbia University and Bletchley Park put quantitative risk analyses of several flavors on the map.

Today, “risk management” applies to security guard services, portfolio management, terrorism and more. Oddly, much of what is called risk management involves no risk assessment at all, and is therefore inconsistent with the above definition of risk management, paraphrased from Wikipedia.

Most risk assessment involves quantification of some sort. Actuarial science and the probabilistic risk analyses used in aircraft design are probably the “hardest” of the hard risk measurement approaches, Here, “hard” means the numbers used in the analyses come from measurements of real world values like auto accidents, lightning strikes, cancer rates, and the historical failure rates of computer chips, valves and motors. “Softer” analyses, still mathematically rigorous, involve quantified subjective judgments in tools like Monte Carlo analyses and Bayesian belief networks. As the code breakers and submarine hunters of WWII found, trained experts using calibrated expert opinions can surprise everyone, even themselves.

A much softer, yet still quantified (barely), approach to risk management using expert opinion is the risk matrix familiar to most people: on a scale of 1 to 4, rate the following risks…, etc. It’s been shown to be truly worse than useless in many cases, for a variety of reasons by many researchers. Yet it remains the core of risk analysis in many areas of business and government, across many types of risk (reputation, credit, project, financial and safety). Finally, some of what is called risk management involves no quantification, ordering, or classifying. Call it expert intuition or qualitative audit.

These soft categories of risk management most arouse the ire of independent and small-firm risk analysts. Common criticisms by these analysts include:

1. “Risk management” has become jargonized and often involves no real risk analysis.
2. Quantification of risk in some spheres is plagued by garbage-in-garbage-out. Frequency-based models are taken as gospel, and believed merely because they look scientific (e.g., Fukushima).
3. Quantified/frequentist risk analyses are not used in cases where historical data and a sound basis for them actually exists (e.g., pharmaceutical manufacture).
4. Big consultancies used their existing relationships to sell unsound (fluff) risk methods, squeezing out analysts with sound methods (accused of Arthur Anderson, McKinsey, Bain, KPMG).
5. Quantitative risk analyses of subjective type commonly don’t involve training or calibration of those giving expert opinions, thereby resulting in incoherent (in the Bayesian sense) belief systems.
6. Groupthink and bad management override rational input into risk assessment (subprime mortgage, space shuttle Challenger).
7. Risk management is equated with regulatory compliance (banking operations, hospital medicine, pharmaceuticals, side-effect of Sarbanes-Oxley).
8. Some professionals refuse to accept any formal approach to risk management (medical practitioners and hospitals).

While these criticisms may involve some degree of sour grapes, they have considerable merit in my view, and partially explain the decline in quality of risk management. I’ve worked in risk analysis involving uranium processing, nuclear weapons handling, commercial and military aviation, pharmaceutical manufacture, closed-circuit scuba design, and mountaineering. If the above complaints are valid in these circles – and they are –  it’s easy to believe they plague areas where softer risk methods reign.

Several books and scores of papers specifically address the problems of simple risk-score matrices, often dressed up in fancy clothes to look rigorous. The approach has been shown to have dangerous flaws by many analysts and scholars, e.g., Tony Cox, Sam SavageDouglas Hubbard, and Laura-Diana Radu. Cox shows examples where risk matrices assign higher qualitative ratings to quantitatively smaller risks. He shows that risks with negatively correlated frequencies and severities can result in risk-matrix decisions that are worse than random decisions. Also, such methods are obviously very prone to range compression errors. Most interestingly, in my experience, the stratification (highly likely, somewhat likely, moderately likely, etc.) inherent in risk matrices assume common interpretation of terms across a group. Many tests (e.g., Kahneman & Tversky and Budescu, Broomell, Por) show that large differences in the way people understand such phrases dramatically affect their judgments of risk. Thus risk matrices create the illusion of communication and agreement where neither are present.

Nevertheless, the risk matrix has been institutionalized. It is embraced by government (MIL-STD-882), standards bodies (ISO 31000), and professional societies (Project Management Institute (PMI), ISACA/COBIT). Hubbard’s opponents argue that if risk matrices are so bad, why do so many people use them – an odd argument, to say the least. ISO 31000, in my view, isn’t a complete write-off. In places, it rationally addresses risk as something that can be managed through reduction of likelihood, reduction of consequences, risk sharing, and risk transfer. But elsewhere it redefines risk as mere uncertainty, thereby reintroducing the positive/negative risk mess created by economist Frank Knight a century ago. Worse, from my perspective, like the guidelines of PMI and ISACA, it gives credence to structure in the guise of knowledge and to process posing as strategy. In short, it sets up a lot of wickets which, once navigated, give a sense that risk has been managed when in fact it may have been merely discussed.

A small benefit of the subprime mortgage meltdown of 2008 was that it became obvious that the financial risk management revolution of the 1990s was a farce, exposing a need for deep structural changes. I don’t follow financial risk analysis closely enough to know whether that’s happened. But the negative example made public by the housing collapse has created enough anxiety in other disciplines to cause some welcome reappraisals.

There is surprising and welcome activity in nuclear energy. Several organizations involved in nuclear power generation have acknowledged that we’ve lost competency in this area, and have recently identified paths to address the challenges. The Nuclear Energy Institute recently noted that while Fukushima is seen as evidence that probabilistic risk analysis (PRA) doesn’t work, if Japan had actually embraced PRA, the high risk of tsunami-induced disaster would have been immediately apparent. Late last year the Nuclear Energy Institute submitted two drafts to the U.S. Nuclear Regulatory Commission addressing lost ground in PRA and identifying a substantive path forward: Reclaiming the Promise of Risk-Informed Decision-Making and Restoring Risk-Informed Regulation. These documents acknowledge that the promise of PRA has been stunted by distrust of the method, focus on compliance instead of science, external audits by unqualified teams, and the above-mentioned Fukushima fallacy.

Likewise, the FDA, often criticized for over-regulating and over-reach – confusing efficacy with safety – has shown improvement in recent years. It has revised its decades-old process validation guidance to focus more on verification, scientific evidence and risk analysis tools rather than validation and documentation. The FDA’s ICH Q9 (Quality Risk Management) guidelines discuss risk, risk analysis and risk management in terms familiar to practitioners of “hard” risk analysis, even covering fault tree analysis (the “hardest” form of PRA) in some detail. The ASTM E2500 standard moves these concepts further forward. Similarly, the FDA’s recent guidelines on mobile health devices seem to accept that the FDA’s reach should not exceed its grasp in the domain of smart phones loaded with health apps. Reading between the lines, I take it that after years of fostering the notion that risk management equals regulatory compliance, the FDA realized that it must push drug safety far down into the ranks of the drug makers in the same way the FAA did with aircraft makers (with obvious success) in the late 1960s. Fostering a culture of safety rather than one of compliance distributes the work of providing safety and reduces the need for regulators to anticipate every possible failure of every step of every process in every drug firm.

This is real progress. There may yet be hope for financial risk management.

### Belief in Probability – Part 2

Last time I started with my friend Willie’s bold claim that he doesn’t believe in probability; then I gave a short history of probability. I observed that defining probability is a controversial matter, split between objective and subjective interpretations. About the only thing these interpretations agree on is that probability values range from zero to one, where P = 1 means certainty. When you learn probability and statistics in school, you are getting the frequentist interpretation, which is considered objective. Frequentism relies on directly equating observed frequencies with probabilities. In this model, the probability of an event exactly equals the limit of the relative frequency of that outcome in an infinitely large number of trials.

The problem with this interpretation in practice – in medicine, engineering, and gambling machines – isn’t merely the impossibility of an infinite number of trials. A few million trials might be enough. Running trials works for dice but not for earthquakes and space shuttles. It also has problems with things like cancer, where plenty of frequency data exists. Frequentism requires placing an individual specimen into a relevant population or reference class. Doing this is easy for dice, harder for humans. A study says that as a white males of my age I face a 7% probability of having a stroke in the next 10 years. That’s based on my membership in the reference class of white males. If I restrict that set to white men who don’t smoke, it drops to 4%. If I account for good systolic blood pressure, no family history of atrial fibrillation or ventricular hypertrophy, it drops another percent or so.

Ultimately, if I limit my population to a set of one (just me) and apply the belief that every effect has a cause (i.e., some real-world chunk of blockage causes an artery to rupture), you can conclude that my probability of having a stroke can only be one of two values – zero or one.

Frequentism, as seen by its opponents, too closely ties probabilities to observed frequencies. They note that the limit-of-relative-frequency concept relies on induction, which might mean it’s not so objective after all. Further, those frequencies are unknowable in many real-world cases. Still further, finding an individual’s correct reference class is messy, possibly downright subjective. Finally, no frequency data exists for earthquakes that haven’t happened yet. All that seems to do some real damage to frequentism’s utility score.

The subjective interpretations of probability propose fixes to some of frequentism’s problems. The most common subjective interpretation is Bayesianism, which itself comes in several flavors. All subjective interpretations see probability as a degree of belief in a specific outcome, as held by a rational person. Think of it as a fair bet with odds. The odds you’re willing to accept for a bet on your race horse exactly equals your degree of belief in that horse’s ability to win. If your filly were in the same race an infinite number of times, you’d expect to break even, based on those odds, whether you bet on her or against her.

Subjective interpretations rely on logical coherence and belief. The core of Bayesianism, for example, is that beliefs must 1) originate with a numerical probability estimate, 2) adhere to the rules of probability calculation, and 3) follow an exact rule for updating belief estimates based on new evidence. The second rule deals with the common core of probability math used in all interpretations. These include things like how to add and multiply probabilities and Bayes theorem, not to be confused with Bayesianism, the belief system. Bayes theorem is an uncontroversial equation relating the probability of A given B to the probability of A and the probability of B. The third rule of Bayesianism is similarly computational, addressing how belief is updated after new evidence. The details aren’t needed here. Note that while Bayesianism is generally considered subjective, it is still computationally exacting.

The obvious problem with all subjective interpretations, particularly as applied to engineering problems, is that they rely, at least initially, on expert opinion. Life and death rides on the choice of experts and the value of their opinions. As Richard Feynman noted in his minority report on the Challenger, official rank plays too large a part in the choice of experts, and the higher (and less technical) the rank, the more optimistic the probability estimates.

The engineering risk analysis technique most consistent with the frequentist (objective) interpretation of probability is fault tree analysis. Other risk analysis techniques, some embodied in mature software products, are based on Bayesian (subjective) philosophy.

When Willie said he didn’t believe in probability, he may have meant several things. I’ll try to track him down and ask him, but I doubt the incident stuck in his mind as it did mine. If he meant that he doesn’t believe that probability was useful in system design, he had a rational belief; but I disagree with it. I doubt he meant that though.

Willie may have been leaning toward the ties between probability and redundancy in system design. Probability is the calculus by which redundancy is allocated to redundant systems. Willie may think that redundancy doesn’t yield the expected increase in safety because having more equipment means more things than can fail. This argument fails to face that, ideally speaking, a redundant path does double the chance having a component failure, but squares the probability of system failure. That’s a good thing, since squaring a number less than one makes it smaller. In other words, the benefit in reducing the chance of system failure vastly exceeds the deficit of having more components to repair. If that was his point, I disagree in principle, but accept that redundancy is no excuse for lack of component design excellence.

He may also think system designers can be overly confident of the exponential increase in modeled probability of system reliability that stems from redundancy. That increase in reliability is only valid if the redundancy creates no common mode failures and no latent (undetected for unknown time intervals) failures of  redundant paths that aren’t currently operating. If that’s his point, then we agree completely. This is an area where pairing the experience and design expertise of someone like Willie with rigorous risk analysis using fault trees yields great systems.

Unlike Willie, Challenger-era NASA gave no official statement on its belief in probability. Feynman’s report points to NASA’s use of numeric probabilities for specific component failure modes. The Rogers Commission report says that NASA management talked about degrees of probability. From this we might guess that NASA believed in probability and its use in measuring risk. On the other hand, the Rogers Commission report also gives examples of NASA’s disbelief in probability’s usefulness. For example, the report’s Technical Management section states that, “NASA has rejected the use of probability on the basis that such techniques are insufficient to assure that adequate safety margins can be applied to protect the lives of the crew.”

Regardless of what NASA’s beliefs about porbability, it’s clear that NASA didn’t use fault tree analysis for the space shuttle program prior to the Challenger disaster. Nor did it use Bayesian inference methods, any hybrid probability model, or any consideration of probability beyond opinions about failures of  critical items. Feynman was livid about this. A Bayesian (subjective, but computational) approach would have at least forced NASA to make it subjective judgments explicit and would have produced a rational model of its judgments. Post-Challenger Bayesian analyses, including one by NASA, varied widely, but all indicated unacceptable risk. NASA has since adopted risk management approaches more consistent with those used in commercial and military aircraft design.

An obvious question arises when you think about using a frequentist model on nearly one-of-a-kind vehicles. How accurate can any frequency data be for something as infrequent as a shuttle flight? Accurate enough, in my view. If you see the shuttle as monolithic and indivisible, the data is too sparse; but not if you view it as a system of components, most of which, like o-ring seals, have close analogs in common use, with known failure rates.

The FAA mandated probabilistic risk analyses of the frequentist variety (effectively mandating fault trees) in 1968. Since then flying has become safe, by any measure. In no other endeavor has mankind made such an inherently dangerous activity so safe. Aviation safety progressed through many innovations, redundant systems being high on the list. Probability is the means by which you allocate redundancy. You can’t get great aircraft systems without designers like Willie. Nor can you get them without probability. Believe it or not.

### Belief in Probability – Part 1

Years ago in a meeting on design of a complex, redundant system for a commercial jet, I referred to probabilities of various component failures. In front of this group of seasoned engineers, a highly respected, senior member of the team interjected, “I don’t believe in probability.” His proclamation stopped me cold. My first thought was what kind a backward brute would say something like that, especially in the context of aircraft design. But Willie was no brute. In fact he is a legend in electro-hydro-mechanical system design circles; and he deserves that status. For decades, millions of fearless fliers have touched down on the runway, unaware that Willie’s expertise played a large part in their safe arrival. So what can we make of Willie’s stated disbelief in probability?

Friends and I have been discussing risk science a lot lately – diverse aspects of it including the Challenger disaster, pharmaceutical manufacture in China, and black swans in financial markets. I want to write a few posts on risk science, as a personal log, and for whomever else might be interested. Risk science relies on several different understandings of risk, which in turn rely on the concept of probability. So before getting to risk, I’m going to jot down some thoughts on probability. These thoughts involve no computation or equations, but they do shed some light on Willie’s mindset. First a bit of background.

Oddly, the meaning of the word probability involves philosophy much more than it does math, so Willie’s use of belief might be justified. People mean very different things when they say probability. The chance of rolling a 7 is conceptually very different from the chance of an earthquake in Missouri this year. Probability is hard to define accurately. A look at its history shows why.

Mathematical theories of probability only first appeared in the late 17th century. This is puzzling, since gambling had existed for thousands of years. Gambling was enough of a problem in the ancient world that the Egyptian pharaohs, Roman emperors and Achaemenid satraps outlawed it. Such legislation had little effect on the urge to deal the cards or roll the dice. Enforcement was sporadic and halfhearted. Yet gamblers failed to develop probability theories. Historian Ian Hacking  (The Emergence of Probability) observes, “Someone with only the most modest knowledge of probability mathematics could have won himself the whole of Gaul in a week.”

Why so much interest with so little understanding? In European and middle eastern history, it seems that neither Platonism (determinism derived from ideal forms) nor the Judeo/Christian/Islamic traditions (determinism through God’s will) had much sympathy for knowledge of chance. Chance was something to which knowledge could not apply. Chance meant uncertainty, and uncertainty was the absence of knowledge. Knowledge of chance didn’t seem to make sense. Plus, chance was the tool of immoral and dishonest gamblers.

The term probability is tied to the modern understanding of evidence. In medieval times, and well into the renaissance, probability literally referred to the level of authority –  typically tied to the nobility –  of a witness in a court case. A probable opinion was one given by a reputable witness. So a testimony could be highly probable but very incorrect, even false.

Through empiricism, central to the scientific method, the notion of diagnosis (inference of a condition from key indicators) emerged in the 17th century. Diagnosis allowed nature to be the reputable authority, rather than a person of status. For example, the symptom of skin spots could testify, with various degrees of probability, that measles had caused it. This goes back to the notion of induction and inference from the best explanation of evidence, which I discussed in past posts. Pascal, Fermat and Huygens brought probability into the respectable world of science.

But outside of science, probability and statistics still remained second class citizens right up to the 20th century. You used these tools when you didn’t have an exact set of accurate facts. Recognition of the predictive value of probability and statistics finally emerged when governments realized that death records had uses beyond preserving history, and when insurance companies figured out how to price premiums competitively.

Also around the turn of  the 20th century, it became clear that in many realms – thermodynamics and quantum mechanics for example – probability would take center stage against determinism. Scientists began to see that some – perhaps most – aspects of reality were fundamentally probabilistic in nature, not deterministic. This was a tough pill for many to swallow, even Albert Einstein. Einstein famously argued with Niels Bohr, saying, “God does not play dice.” Einstein believed that some hidden variable would eventually emerge to explain why one of two identical atoms would decay while the other did not. A century later, Bohr is still winning that argument.

What we mean when we say probability today may seem uncontroversial – until you stake lives on it. Then it gets weird, and definitions become important. Defining probability is a wickedly contentious matter, because wildly conflicting conceptions of probability exist.  They can be roughly divided into the objective and subjective interpretations. In the next post I’ll focus on the frequentist interpretation, which is objective, and the subjectivist interpretations as a group. I’ll look at the impact of accepting – or believing in – each of these on the design of things like airliners and space shuttles from the perspectives of Willie, Richard Feynman, and NASA. Then I’ll defend my own views on when and where to hold various beliefs about probability.

Autobrake diagram courtesy of Biggles Software.

### Intuitive Probabilities

Meet Vic. Vic enjoys a form of music that features heavily distorted guitars, slow growling vocals, atonality, frequent tempo changes, and what is called “blast beat” drumming in the music business. His favorite death metal bands are Slayer, Leviticus, Dark Tranquility, Arch Enemy, Behemoth, Kreator, Venom, and Necrophagist.

Vic has strong views on theology and cosmology. Which is more likely?

1. Vic is a Christian
2. Vic is a Satanist

I’ve taught courses on probabilistic risk analysis over the years, and have found that very intelligent engineers, much more experienced than I, often find probability extremely unintuitive. Especially when very large (or very small) numbers are involved. Other aspects of probability and statistics are unintuitive for other interesting reasons. More on those later.

The matter of Vic’s belief system involves several possible biases and unintuitive aspects of statistics. While pondering the issue of Vic’s beliefs, you can enjoy Slayer’s Raining Blood. Then check out my take on judging Vic’s beliefs below the embedded YouTube video – which, by the way, demonstrates all of the attributes of death metal listed above.

.

Vic is almost certainly a Christian. Any other conclusion would involve the so-called base-rate fallacy, where the secondary, specific facts (affinity for death metal) somehow obscure the primary, base-rate relative frequency of Christians versus Satanists. The Vatican claims over one billion Catholics, and most US Christians are not Catholic. Even with papal exaggeration, we can guess that there are well over a billion Christians on earth. I know hundreds if not thousands of them. I don’t know any Satanists personally, and don’t know of any public figures who are (there is conflicting evidence on Marilyn Manson). A quick Google search suggests a range of numbers of Satanists in the world, the largest of which is under 100,000. Further, I don’t ever remember seeing a single Satanist meeting facility, even in San Francisco. A web search also reveals a good number of conspicuously Christian death metal bands, including Leviticus, named above. Without getting into the details of Bayes Theorem, it is probably obvious that the relative frequencies of Christians against Satanists governs the outcome. And judging Vic by his appearance is likely very unreliable.

South Park Community Presbyterian Church

### Kaczynski, Gore, and Cool Headed Logicians

Yesterday I was talking to Robert Scoble about context-aware computing and we ended up on the topic of computer analysis of text. I’ve done some work in this area over the years for ancient text author attribution, cheating detection in college and professional essay exam scenarios, and for sentiment and mood analysis. A technique common to authorship studies is statistical stylometry, which aims to quantify linguistic style. Subtle but persistent differences between text written by different authors, even when writing about the same topic or in the same genre often emerge from statistical analysis of their writings.

Robert was surprised to hear that Ted Kaczynski, the Unabomber, was caught because of linguistic analysis, not done by computer, but by Kaczynski’s brother and sister-in-law. Contrary to stories circulating in the world of computational linguistics and semantics, computer analysis played no part in getting a search warrant or prosecuting Kaczynski. It could have, but Kaczynski plead guilty before his case went to trial. The FBI did hire James Fitzgerald, a forensic linguist, to compare Kaczynski’s writings to the Unabomber’s manifesto, and Fitzgerald’s testimony was used in the trial.

Analysis of text has uses beyond author attribution. Google’s indexing and search engine relies heavily on discovering the topic and contents of text. Sentiment analysis tries to guess how customers like a product based on their tweets and posts about it. But algorithmic sentiment analysis is horribly unreliable in its present state, failing to distinguish positive and negative sentiments the vast majority of the time. Social media monitoring tools have a long way to go.

The problem stems from the fact that human speech and writing are highly evolved and complex. Sarcasm is common, and relies on context to reveal that you’re saying the opposite of what you mean. Subcultures have wildly different usage for overloaded terms. Retirees  rarely use “toxic” and “sick” as compliments like college students do. Even merely unwinding phrases to determine the referent of a negator is difficult for computers. Sentiment analysis and topic identification rely on nouns and verbs, which are only sometimes useful in authorship studies. Consider the following sentences:

1) The twentieth century has not been kind to the constant human striving for a sense of purpose in life.

2) The Industrial Revolution and its consequences have been a disaster for the human race.

The structure, topic, and sentiment of these sentences is similar. The first is a quote from Al Gore’s 2006 Earth in the Balance. The second is the opening statement of Kaczynski’s 1995 Unabomber manifesto, “Industrial Society and its Future.” Even using the total corpus of works by Gore and Kaczynski, it would be difficult to guess which author wrote each sentence. However, compare the following paragraphs, one from each of these authors:

1) Modern industrial civilization, as presently organized, is colliding violently with our planet’s ecological system. The ferocity of its assault on the earth is breathtaking, and the horrific consequences are occurring so quickly as to defy our capacity to recognize them, comprehend their global implications, and organize an appropriate and timely response. Isolated pockets of resistance fighters who have experienced this juggernaut at first hand have begun to fight back in inspiring but, in the final analysis, woefully inadequate ways.

2) It is not necessary for the sake of nature to set up some chimerical utopia or any new kind of social order. Nature takes care of itself: It was a spontaneous creation that existed long before any human society, and for countless centuries, many different kinds of human societies coexisted with nature without doing it an excessive amount of damage. Only with the Industrial Revolution did the effect of human society on nature become really devastating.

Again, the topic, mood, and structure are similar. Who wrote which? Lexical analysis immediately identifies paragraph 1 as Gore and paragraph 2 as Kaczynski. Gore uses the word “juggernaut” twice in Earth in the Balance and once in The Assault on Reason. Kaczynski never uses the word in any published writing. Fitzgerald (“Using a forensic linguistic approach to track the Unabomber”, 2004) identified “chimerical,” along with “cool-headed logician” to be Kaczynski signatures.

Don’t make too much – as some of Gore’s critics do – of the similarity between those two paragraphs. Both writers have advanced degrees from prestigious universities, they share an interest in technology and environment, and are roughly the same age. Reading further in the manifesto reveals a great difference in attitudes. Though algorithms would have a hard time with the following paragraph, few human readers would identify the following excerpt with Gore (this paragraph caught my I eye because of its apparent reference to Thomas Kuhn, discussed a lot here recently – both were professors at UC Berkeley):

Modern leftist philosophers tend to dismiss reason, science, objective reality and to insist that everything is culturally relative. It is true that one can ask serious questions about the foundations of scientific knowledge and about how, if at all, the concept of objective reality can be defined. But it is obvious that modern leftist philosophers are not simply cool-headed logicians systematically analyzing the foundations of knowledge.

David Kaczynski, Ted’s brother, describes his awful realization about similarity between his brother’s language and that used in the recently published manifesto:

“When Linda and I returned from our Paris vacation, the Washington Post published the Unabomber’s manifesto. After I read the first few pages, my jaw literally dropped. One particular phrase disturbed me. It said modern philosophers were not ‘cool-headed logicians.’ Ted had once said I was not a ‘cool-headed logician’, and I had never heard anyone else use that phrase.”

And on that basis, David went to the FBI, who arrested Ted in his cabin. It’s rare that you’re lucky enough to find such highly distinctive terms in authorship studies though. In my statistical stylometry work, I looked for unique or uncommon 2- to 8-word phrases (“rare pairs“, etc.) used only by two people in a larger population, and detected unwanted collaboration by that means. Most of my analysis, and that of experts far more skilled in this field than I, is not concerned with content. Much of it centers on function-word statistics – usage of pronouns, prepositions and conjunctions. Richness of vocabulary, rate of introduction of new words, and vocabulary frequency distribution also come into play. Some recent, sophisticated techniques look at characteristics of zipped text (which obviously does include content), and use markov chains, principal component analysis and support vector machines.

Statistical stylometry has been put to impressive use with startling and unpopular results. For over a century people have been attempting to determine whether Shakespeare wrote everything attributed to him, or whether Francis Bacon helped.  More recently D. I. Homes showed rather conclusively using hierarchical cluster analysis that the Book of Mormon and Book of Abraham both arose from the prophetic voice of Joseph Smith himself. Mosteller and Wallace differentiated, using function word frequency distributions, the writing of Hamilton and Madison in the Federalist Papers. They have also shown clear literary fingerprints in the writings of Jane Austen, Arthur Conan Doyle, Charles Dickens, Rudyard Kipling and Jack London. And for real fun, look into New Testament authorship studies.

Computer analysis of text is still in its infancy. I look forward to new techniques and new applications for them. Despite false starts and some exaggerated claims, this is good stuff. Given the chance, it certainly would have nailed the Unabomber. Maybe it can even determine what viewers really think of a movie.

### Why Rating Systems Sometimes Work

Goodfilms is a Melbourne based startup that aims to do a better job of recommending movies to you. Their system uses your social network, e.g., Facebook, to show you what your friends are watching, along with two attributes of films, which you rate on a 10 scale (1 to 5 stars in half-star increments). It doesn’t appear that they include a personalized recommendation system based on collaborative filtering or similar.

In today’s Goodfilms blog post, Why Ratings Systems Don’t Work, the authors point to an XKCD cartoon identifying one of the many problems with collecting ratings from users.

The Goodfilms team says the problem with averaged rating values is that they attempt to distil an entire product down to a scalar value; that is, a number along a scale from 1 to some maximum imaginable goodness. They also suggest that histograms aren’t useful, asking how seeing the distribution of ratings for a film might possibly help you judge whether you’d like it.

Goodfilms demonstrates the point using three futuristic films, Blade Runner, Starship Troopers, and Fifth Element. The Goodfilms data shows bimodal distributions for all three films; the lowest number of ratings for each film is 2, 3, or 4 stars with 1 star and 5 stars having more votes.

Goodfilms goes on to say that their system gives you better guidance. Their film-quality visualization – rather than a star bar-chart and histogram – is  a two axis scatter plot of the two attributes you rate for films on their site – quality and rewatchability, how much you’d like to watch that film again.

An astute engineer or economist might note that Goodfilms assumes quality and rewatchability to be independent variables, but they clearly are not. The relationship between the two attributes is complex and may vary greatly between film watchers. Regardless of the details of how those two variables interact, they are not independent; few viewers would rate something low in quality and high in rewatchability.

But even if these attributes were independent of each other, films have many other attributes that might be more telling – length, realism, character development, skin exposure, originality, clarity of intent, provocation, explosion count, and an endless list of others. Even if you included 100 such variables (and had a magic visualization tool for such data), you might not capture the sentiment of a crowd of viewers about the film, let alone be able to decide whether you would like it based on that data. Now if you had some deep knowledge of how you, as an individual, compare, in aesthetics, values and mental process, to your Facebook friends and to a larger population of viewers – then we’d really know something, but that kind of analysis is still some distance out.

Goodfilms is correct in concluding that rating systems have their perils; but their solution, while perhaps a step in the right direction, is naive. The problem with rating systems is not that they don’t capture enough attributes of the rated product or in their presentation of results. The problem lies in soft things. Rating systems tend to deal more with attributes of products than with attributes of raters of those products. Recommendation systems don’t account for social influence well at all. And there’s the matter of actual preferences versus stated preference; we sometimes lie about what we like, even to ourselves.

Social influence, as I’ve noted in past posts, is profound, yet its sources can be difficult to isolate. In rating systems, knowledge of how peers or a broader population have rated what you’re about to rate strongly influence the outcome of ratings. Experiments by Salganik and others on this (discussed in this post) are truly mind boggling, showing that weak information about group sentiment not only exaggerates preferences but greatly destabilizes the system.

The Goodfilms data shows bimodal distributions for all three films. The 1 star and 5 star vote count is higher than the minimum count of the 2, 3, and 4 star rating counts. Interestingly, this is much less true for Imdb’s data. So what’s the difference? Goodfilms’ rating counts for these movies range from about 900 to 1800. Imdb has hundreds of thousands of votes for these films.

As described in a previous post (Wisdom and Madness of the Yelp Crowd), many ratings sites for various products have bimodal distributions when rating count is low, but more normally distributed votes as the count increases. It may be that the first people who rate feel the need to exaggerate their preferences to be heard. Any sentiment above middle might gets cast as 5 star, otherwise it’s 1 star. As more votes are cast, one of these extremes becomes dominant and attracts voters. Now just one vote in a crowd, those who rate later aren’t compelled to be extreme, yet are influenced by their knowledge of how others voted. This still results in exaggeration of group preferences (data is left or right skewed) through the psychological pressure to conform, but eliminates the bimodal distribution seen in the early phase of rating for a given product. There is also a tendency at Imdb for a film to be rated higher when it’s new than a year later. Bias originating in suggestion from experts surely plays a role in this too; advertising works.

In the Imdb data, we see a tiny bit bimodality. The number of “1” ratings is only slightly higher that the number of “2” ratings (1-10 scale). Based on Imdb data, all three movies are all better than average – “average” being not 5.5 (halfway between 1 and 10) but either 6.2, the mean Imdb rating, or 6.4, if you prefer the median.

Imdb publishes the breakdown of ratings based on gender and age (Blade Runner, Starship Troopers, Fifth Element). Starship Troopers has considerably more variation between ratings of those under 18 and those over 30 than do the other two films. Blade Runner is liked more by older audiences than younger ones. That those two facts aren’t surprising suggests that we should be able to do better than recommending products based only on what our friends like (unless you will like something because your friends like it) or based on simple collaborative filtering algorithms (you’ll like it because others who like what you like liked it).

###### Imdb rating count vs. rating for 3 movies

So far, attempts to predict preferences across categories – furniture you’ll like based on your music preferences – have been rather disastrous. But movie rating systems actually do work. Yes, there are a few gray sheep, who lack preference similarity with the rest of users, but compared to many things, movies are very predictable  – if you adjust for rating bias.  Without knowledge that Imdb ratings are biased toward good and toward new, you high think a film with an average rating of 6 is better than average, but it isn’t, according to the Imdb community. They rate high.

Algorithms can handle that minor obstacle, even when the bias toward high varies between raters. With minor tweaks of textbook filtering algorithms, I’ve gotten movie predictions to be accurate within about half a star of actual. I tested this by using the movielens database and removing one rating from each users’ data and then making predictions for the missing movie for each user, then averaging the difference between predicted and actual values. Movie preferences are very predictable. You’re likely to give a film the same rating whether you saw it yesterday or today. And you’re likely to continue liking things liked by those whose taste was similar to yours in the past.

Restaurants are slightly less predictable, but still pretty good. Yesterday the restaurant was empty and you went for an early dinner. Today, you might get seated next to a loud retirement party and get a bad waiter. Same food, but your experience would color your subjective evaluation of food quality and come out in your rating.

Predicting who you should date or whether you’d like an autumn vacation in Paris is going to require a much different approach. Predicting that you’d like Paris based on movie tastes is ludicrous. There’s no reason to expect that to work other than Silicon Valley’s exuberant AI hype. That sort of prediction capability is probably within reach. But it will require a combination of smart filtering techniques (imputation-boosting, dimensionality reduction, hybrid clustering), taxonomy-driven computation, and a whole lot more context.

Context?  – you ask. How does my GPS position affect my dating preferences? Well that one should be obvious. On the dating survey, you said you love ballet, but you were in a bowling alley four nights last week. You might want to sign up for the mixed league bowling. But what about dining preferences? To really see where this is going you need to expand your definition of context (I’m guessing Robert Scoble and Shel Israel have such an expanded view of context based on the draft TOC for their upcoming Age of Context).