1989
Life has a stubborn tendency not to follow the course you've charted for it. My own trajectory, painstakingly calculated to place me in an academic orbit, went awry early. In late 1986 I washed up on the shores of the New World, having survived two years of demonstrations, riots, teargas and cosmology at the University of Cape Town, South Africa. I was penniless, abandoned by every funding organization known to man. So I set off on the uncertain life of a free-lancer. For once luck was with me, though, and after four lean months I was writing for Discover magazine, sanguine of the future. But Fortune again intervened, Discover was sold and my financial situation quickly reverted to its previous state. The threat of immediate starvation exacted from me a drastic step: I accepted an offer for an editorial post at Scientific American. And thus I found myself, unwittingly, a physicist on Madison Avenue....
Madison Avenue. The real world. I had heard of such a place ever since my undergraduate days but never expected to set foot in it. As far as I could make out the real world was a country of three-piece suits, high finance, corporate mergers and power lunches. Inhabitants of the real world went to discos, watched Dallas on TV and spent their vacations in Disneyland. At least this is what I have read in tabloids.
The contrast to life in academia could hardly be greater. I do not need to describe the existence of a university scientist in great detail because you are already familiar with it. The makers of Revenge of the Nerds know, as do the millions who have seen it, that all scientists when young are undernourished sociophobes who relate best to a computer terminal through coke-bottle eyeglasses after midnight in a basement laboratory. As they age they put their phobias behind them to become positively pathological, with only one thing on their minds: dissection of the nearest alien. ( E.T., Splash, Lord Greystoke and Iceman provide irrefutable evidence for skeptics.)
And so I arrived on Madison Avenue armed with the knowledge that I was at best useless and at worst dangerous. That I was at least useless was brought home to me after five minutes on the job when I asked one of the stockroom clerks for a blackboard. The fellow was as shocked at the request as I was at the reply: "Impossible."
From my point of view it was natural to think that a blackboard might come in handy. The world of a physicist consists of natural laws--such as conservation of energy and momentum--which govern the behavior of everything in the universe and which cannot be broken by any mortal. The world of a physicist consists of mathematical derivations and proofs, which traditionally carried out with pencil or chalk, go into the logical construction of a theory. And the world of a physicist consists of experimental evidence by which theories are tested, for without experimental evidence all theories are equally good--or equally bad.
The idea that the world is lawful as opposed to lawless is fundamental to a scientist's outlook. The most important message of the natural laws is that some things are impossible (like getting a blackboard). The principle that some things are "no go," leads in turn to the sobering realization that none of us is granted absolute control over the world.
Because the laws determine the rules, scientists learn the habit of following the evidence where it leads, regardless of whether the eventual conclusion is compatible with Marxist philosophy or Reaganomics. This is called scientific method. Such ideas, with their imposition of constraints, are not popular in America, where to "take control of your own life" has become something of a national pasttime. Such ideas run counter to everything Self magazine stands for.
Nevertheless, since Madison Avenue was part of the real world, I had assumed it would obey the same laws of physics as everything else. Here I was mistaken. Madison Avenue, unlike the world of a physicist, consists of advertising departments, circulation departments, promotion departments, production departments and editorial departments (in order of decreasing significance). Advertisers and circulation managers use terms like "base rate," "draw," "sell-through" and "volatility." "Energy" and "momentum" do not figure prominently.
The absence of conservation of energy in a department devoted to filling the pages of Scientific American with liquor ads might not surprise inhabitants of the real world. It did me. The reality survey shows why.
Not long after my arrival, the Editor-in-Chief (or Editor, to use the formal Scientific American term) told me that the majority of the readers were not scientists. The news disturbed me greatly because since the days when I actually read the magazine I had assumed that it was students, academics or researchers who bought Scientific American to learn something outside their fields. No, I was assured that the readers were by and large managers, members of the business community and "busy executives."
The identity of the readership is not an academic question. Management has hoped to launch two new publications, Trends in Computing and Scientific European, which are geared more toward business and industry than to the scientific community. The parent magazine has made similar accomodations with reality. In mid 1988 it inaugurated a new editorial section Science and Business , which eighteen months later was expanded from three pages to ten; the traditional Science and Citizen section--consisting of short news items--is also now twice as large as in years past, presumably on the theory that executives don't have time to read anything longer than the average article in People magazine.
My first thought was that the readership question be simple to answer: make up a subscriber survey and ask the readers who they are. I discovered I wasn't the first person to have such an idea. But the survey I eventually got my hands on was not written by a simpleton. It was written by a market research bureau and market research bureaus do not want to find out who the readers are (which is irrelevant) but how rich they are (which determines whether they can afford a Mercedes).
The survey contains precise questions about income level, length of time an average issue is kept and favorite departments, but anything so open-ended as, "Where do you work?" or, "Are you a scientist?" is absent. Instead, the market research bureau presorts all readers into categories under the heading "type of business." Included are such real-world occupations as real estate, insurance, government, engineering, education services, health services and "other." The reader is also given a choice of real-world job titles: Chairman of the Board, Owner, Partner, Vice President, etc., CEO, General Manager, Professor/Teacher/Other Educator, M.D./D.O./D.D.S and "Other."
The first thing you notice is that is you do not work in in a business you are assumed not to exist. Students and scientists (along with the dodo) are denied ecological niches; the closest slot is "education services." The same pattern follows for job titles. "Professor/Teacher/Educator" is the only pigeon hole for an academic or researcher, although there are no less than six possible slots for executives and managers.
Still the psalmist has said, "Knock and it shall be opened to you." In this survey those with executive or managerial titles accounted for 27 percent .Twenty-seven percent is sizable--it comprises the largest known category. But those answering the description Professor/Teacher/Educator made up about 24 percent and "other" 26 percent. You might guess, depending on your inclinations, that this amorphous 50 percent consisting of educators plus "others" are connected in some way with academia or research.
Inclinations have a subversive tendency to be confirmed. Adding up all the responses to the "type of business" question you find they total only 82 percent--18 percent of the readers vanished from the real world altogether! About 27 percent managed to locate themselves in education services and 11 percent in "other." A missing persons bureau might be set up to find out who all these people are. At the lower end, the number of scientists who slipped through the cracks could be 29 percent, at the higher end 56 percent. A physicist would say that these numbers are consistent with the previous guess.
Is it a coincidence? I think not. Another question on the survey asks for
educational level. Eighty-five percent of Scientific American readers
respond that they have finished college. About 22 percent claim to hold master's
degrees and an additional 40 percent hold doctorates. A strange lot of busy
executive's with Ph.D's.
One might suggest that the management has been pushing the magazine toward business on the basis of a predisposition--not on the basis of evidence. One might still want to ask what a subscriber survey has to do with the conservation of energy. If either thought has occurred to you, you may be less an inhabitant of the real world than you think.
I said earlier that the habit of obeying natural laws forces a scientist to follow the evidence where it leads. The Scientific American survey was at best not designed to find out who the readers were, at worst designed not to find out who the readers were. In either case the design makes in easy to ignore the truth. Many firms hire "answer analysts," people who look for data to support the company's position on a given issue. You want to prove that acid rain is not caused by coal-fired power plants? We'll find you the data. You want to prove that most Scientific American readers are businessmen? I'll design the questionnaire. The answer-analyst approach runs exactly contrary to the scientific spirit, which is to go out, collect data and then decide on a conclusion, not vice versa.
What of my own bias that Scientific American readers are predominantly linked to academia? We all have preconceived notions. A good scientist knows when to throw them out. He cuts the data as many ways as he can to check his answers (I gave you three cuts above), and if the data are insufficient, he goes out and collects more. I am willting to wager that a reasonable survey wil support my conclusions and not those of the marketing division. What is a reasonable survey? Make up the job categories after you have asked the readers what they do, not before. Then if the data prove me wrong, I'm wrong.
Underlying this entire exercise has been a hidden factor: I have been banking on my number sense. If devotion to Law is the primary characteristic that separates physicists from inhabitants of the real world, then not far behind is the physicist's peculiar relationship with numbers. I am not speaking of a facility for arithmetic. Most physicists are probably not as good at arithmetic as most accountants or bank tellers. But what a physicist develops during his training is a sense of when an answer is right and when an answer is wrong--in other words, when a number means something and when it doesn't. Unfortunately, a number sense is itself enough to cancel a physicist's visa to the real world, as I discovered during the great statistics battle I waged against the full might of the Scientific American circulation department.
You must understand that some 80 percent of the magazine's sales are in subscriptions but each month about 100,000 people across the country stop by the local newstand, gawk at Penthouse and buy Scientific American. From month to month the numbers vary; in the last few years the average has been 96,000 but sales range from roughly 70,000 to 120,000. Anyone looking at month-to-month figures would see no obvious pattern; it was just this that troubled the Publisher and members of the circulation department. In particular the Publisher wanted to know why the February 1988 issue sold an extraordinary 122,000 copies. Was there any way to stop such fluctuations and keep sales consistently on the high end?
What causes sales to fluctuate? In the real world cover design has a strong influence on sales. The Circulation Manager Emeritus, had long held that predominantly red covers sell well and blue covers sell poorly. Another idea, attributed to the Publisher, is that the more abstract the cover the better. Because in the last few years the Publisher, not the art director, has had final approval of the cover, the idea resulted in (among other things) a series of abstract covers during the summer and early fall of 1988: a gravitational lens that looked like a blob, an optical illusion that looked like several blobs, a computer-generated crystal that looked like it came out of the final sequence of 2001. Still, doctors recommend.
Accepted wisdom also has it that animal covers are bad bets. In fact, the only animal covers in the last several years were the cheetah on May 1986 and the panda on November 1987. They sold so poorly that animals had been banned from the cover once and for all. Yet another piece of conventional wisdom is that there is a seasonal variation in sales. I can't remember every month that does better or worse than average; ask your nearest circulation manager for the exciting details.
When the volatility question was officially announced each editor was given the 1987-1988 sales figures, xeroxes of all the covers and table-of-contents pages and after several weeks we convened an editorial meeting to discuss the matter. The covers for the two years were posted on the wall. A dozen editors sat down around the executive table and made themselves comfortable; the debate began
. Everyone present had one theory, if not two. Aside from from the blue, red and abstract cover hypotheses, there was the "strong, clean design" hypothesis. Its advocate maintained that "the best covers sold the best and the ugliest covers sold the worst." The Editor proposed that "man-and-machine" covers were the most popular. Yet another editor made the novel suggestion that contents were more important than cover: every issue with a major policy article or a math article or a computer article sold better than average. She bolstered her case by pointing out that at most news stands the cover of Scientific American is buried behind Popular Mechanics.
The discussion went on for one hour and a half. As the other editors argued, I sat there morosely trying to imagine a similar gathering at the editorial offices of Cosmopolitan, from which all covers to emerge are essentially identical. "Every centimeter of cleavage sells 90,000 copies. Blue mascara sells better than brown. Sex after marriage in the upper left hand corner does not do as well as how to carry on an extramarital affair." I couldn't buy it.
It seemed obvious that the factors influencing the sales of a magazine are many: the weather, the cover, the contents, the mood of the buyer, the budget deficit, a chance romantic encounter at the newstand, perhaps the phases of the moon. In fact there must be so many factors influencing sales that it becomes impossible to isolate any one of them and the result is essentially random .
For this reason I knew that even if some slight pattern existed it would be buried in statistical noise and extremely difficult to detect. On the other hand, if the newstand sales were experiencing genuinely random fluctuations, that should be pretty easy to show. Physicists are unscrupulous opportunists, so when I was given the sales figures I did the first thing any physicist (probably even Feynman) would do and made a histogram of the sales. Within about 15 minutes I had my answer: the 22 issues for which I had data fell into the pattern that has become the symbol of randomness, a Gaussian distribution as it is known to physicists, a normal distribution or bell-shaped curve as it is known to everyone else.
That the sales fell into a bell curve was not an entirely foregone conclusion. I also plotted the sales of the international edition and, try as I might, I could not fit them into a Gaussian. From the shape of the distribution I might conjecture that the international edition is largely selling out, but I am reluctant to frame hypotheses. To conspiracy theorists, I do admit that if sales were entirely determined by a single X-factor, say the amount of purple on the cover, and if the amount of purple were fluctuating randomly, sales would then fall into a normal distribution. In this case the sales would be exactly correlated with the amount of purple despite the random fluctuations. (Standardized test scores, I am told, are correlated in this way with family income.) Such a scenario, while logically possible, is farfectched. Even if such an X-factor existed, its random behavior would make spotting it difficult; if two or three random X-factors were driving sales, to sort out the correlations would be, as I've said, hopeless. Simply defining possible X-factors is an exercise in frustration. "Amount of purple" may be a good variable because it is easy to quantify (80%, 93%...). But the editorial suggestions "man-and-machine," "animals" and "abstract," unlike family income, do not fluctuate randomly and cannot be random variables.
Nevertheless, to check some possible correlations was easy enough. When I sat down with the sales figures I made not only a histogram but a chart that divided covers into various categories: predominently blue, predominently red, high-tech, artistic, nature, abstract, animal and so on. Admittedly the categories were subjective but, I thought, not unreasonable. I took the two years' worth of data and put each issue into the appropriate bin, marking whether it had sold better than average or worse than average.
I could see no correlations--roughly half the issues in each category sold better than average and half sold worse than average. Actually I'm lying. Two patterns did emerge: 3 out of 4 "artistic" covers sold worse than average and, to my surprise, red covers sold better than blue covers. Of the 5 red covers 3 sold better than average and of the 9 blue covers 6 sold worse than average. Had I confirmed the Circulation Manager's theory? I'll explain later why I didn't take this very seriously. The lack of a convincing pattern in my chart contributed to my morose silence during the editorial meeting--despite continued commentary on "bold, striking designs" and "man-and-machines."
None of my results could exactly be called earthshaking. What stunned me was that apparently no one on Madison Avenue had ever made a histogram. I soon learned that a histogram is not an obvious thing to draw. For the meeting, the Assistant Art Director had plotted the sales figures for 1987 and 1988 month by month. From his point of view a month-to-month graph was a natural way to look at the world. But it wasn't helpful: the graph zigzagged up and down in no recognizable pattern. In fact 1988 zigged largely where 1987 zagged, allowing you to conclude absolutely nothing.
So I stood up. Since blackboards were still lacking, I had come dressed incognito with a flip chart. "The very number of hypotheses I've heard," I began, "convinces me that my hypothesis is correct--there is no pattern." I explained that the sales figures formed a Gaussian distribution with a mean of about 96,000 and a standard deviation of about 13,000. Because the fluctuations were random, about 5 percent of the points should lie farther than two standard deviations from the mean. The February 1988 issue, which sold 122,000, was just such a fluctuation. The only remaining mystery was how the circulation department had provided sales figures for the November and December 1988 issues. November returns had yet to come in and the December issue had not even appeared. Those points I included out of perversity but marked them "magic."
The presentation appeared to be going smoothly. Midway the Editor turned to the Assistant Art Director and, in a voice perfectly balanced between jest and earnest, quipped, "Institute for Advanced Study school of market analysis." (Later a mathematician friend recommended I let them think that.) After I explained the details of the histogram, a voice was raised.
"You've used absolute sales figures."
"Yes of course."
"What about sell-through?"
I suppose it was inevitable. Throughout the meeting I had noticed that with the exception of Greg Greenwell (who also has a physics background) and me, everyone defined a good or a bad sale in terms of sell-through, which is Madspeak for percentage sale. Each month a certain number of copies are printed, or "drawn," but only a percentage of them are sold. That percentage is sell-through. A good sell-through is considered something like 48 percent.
"What you print is a managerial decision," I answered, "what is bought on the streets is
determined by the consumer. The correct figures to use are not the percentage sales but
the absolute sales."
At the time I did not know it but what I had just said was sufficient to cause excommunication from Madison Avenue. According to Madthink it is an incontrovertible fact that if you raise the draw you raise the sales. I've never been able to get a coherent explanation for the phenomenon; if you put 20 copies out on the newstand and only 10 are selling, it is not obvious that putting out 30 will help matters. I imagine the theory is that if you increase the draw you increase the number of newstands, so more people encounter the magazine.
In any case, to check a possible relationship between sale and draw was easy enough. I merely graphed the sale versus the draw. If you visualize a shot gun fired at a target you have a good picture of the result. There was no correlation whatsoever. Points were splattered all over the place. All attempts to fit a curve to the data were absolutely fruitless.
The problem of basing conclusions on sell-through was particularly amusing (or sad depending on your feelings toward in vivo pharmaceutical testing) when it came to the question of animal covers. When I was making my chart of cover types I noticed one fact which I took very seriously: I did not yet have sales figures for the 1986 cheetah, but the 1987 panda cover had sold better than average. In the world of statistics the First Commandment is thou shalt not do statistics with one data point . I found it extremely odd that animal covers had been banned on the basis of a single experiment. I found it incomprehensible that they had been banned on the basis of a successful experiment.
More than one person noticed the anomaly at the meeting. "The panda sold slightly better than average: 98,000."
"No," came the reply, "only a 38 percent sell-through."
"But this is nuts," I cried (in a rare outburst). "The only reason the sell-through is so low is because the print run was almost 25 percent higher than usual."
You see, once a year Scientific American publishes its famous single-topic issue (perhaps not so famous as Sports Illustrated 's swimsuit issue, but both are based on the principle of overexposure.) As a rule the single-topic issue sells far better than average and so it would be difficult to argue that the special issue is a random fluctuation. For that reason I routinely excluded them in the "calculations" I have described; here then is why two years has 22 data points.
So. For the October 1987 single-topic issue the draw was raised as usual to accomodate the additional sales expected. For reasons known only to God and the circulation department, the draw was not lowered again for November's hapless panda. The sell-through registered low on the Richter scale simply because the print run was too high.
"The moral I take from this story," I said at seminar's conclusion, "is that we should let the art department decide what to put on the cover and not listen to astrological advice from the advertising department."
"The circulation department."
"The circulation department."
Physicists tend not be diplomats, yet another reason why they should be denied visas to the real world. Nevertheless, despite my political naivity--or perhaps because of it--the Editor asked me to make my presentation to the President of the Magazine Division and Publisher and the circulation department. I agreed.
Psychology is strenuously at work here (in particular Result Amplification by Stimulated Emission of Publicity). It is all too easy to see a signal when you want to, especially when the amount of data is small and the ego is large. The traditional solution in science is to go out and collect enough data so that the answer becomes as unmistakable as those golden arches. One might say that more data increases the objectivity of the experiment.
In the real world the solution tends to be the opposite: ignore data you don't like. This approach is undoubtedly the correct one, for the real world becomes a much more subjective and exciting place than the cold, objective world of the physicist. Without astrology, ESP and Shirley MacLaine's past lives, Walden Books might very well go out of business and conversation at Los Angeles cocktail parties would become impossible. Trenchant subjectivity has made the world livable but it has also destroyed any social number sense and has made the concept of number meaningless.
I do not mean to exaggerate. Periodically opinion polls on the likelihood of nuclear war appear in Time or Newsweek . "Do you think that nuclear war in the next 10 years is a) very likely; b) somewhat likely; c) not likely?" There is something reminiscent of Chicken Little here: unless such polls influence strategic policy, the average citizen's opinion on the probability of nuclear war has little if anything to do with the actual probability of nuclear war. The philosophical basis of the surveys is that reality is determined by referendum.
A more extreme reality referendum was a poll taken on the likelihood of nuclear winter. Ninety-two percent of the respondents thought that nuclear winter could take place after a nuclear war. It is intriguing that the laws of physics were not consulted. The 92 percent are close relatives to the salesman who attempted to convince me to buy an optional three-year service contract for a car (one year after they had been introduced) with the argument that "fifty-eight percent of the customers buy them." Rather than carry a joke past its punch line, I leave it as an exercise for the reader to give three reasons why I told the salesman I was not a member of the Pepsi generation.
The distance from my Nissan dealer to Madison Avenue is less than a finger's walk through the Yellow Pages. When numbers become subjective sell-through becomes as good a variable as absolute sales. In science most meaningful numbers are ratios: "A as compared to B." Without the comparison, the quantity under discussion lacks a standard of measurement--a convenient state of affairs in the real world. The New York Times recently reported that President Bush "remains highly popular with the electorate" because of his 61 percent approval rating. As one reader pointed out, another way of putting it would be to say that Bush has the third lowest rating in the last 40 years--every new President since Truman, with the exception of Nixon and Reagan, had a higher rating.
The use of sell-through by Scientific American's circulation department is the real-world approach. Indeed sell-through is the ratio of copies sold to copies drawn, but if the draw changes every month, there is no fixed standard of reference. To say a 38 percent sell-through is bad means nothing to a physicist until you fix the denominator. The use of sell-through finds its roots in the subjectivist interpretation of the world.
At about the same time I developed a "theory" to explain the lack of connection between draw and sales. A golden rule in physics is that when there is only one important number in a problem, that number has to be the answer. When there are two important numbers, the answer must be some simple function of the two of them. You ask a physicist what is the typical time for the decay of an elementary particle and he will say 10®MDSU¯-23®MDNM¯ seconds. Why? Because the only relevant time you can imagine is the size of an elementary particle divided by the speed of light: 10^-23 seconds.
In the newstand sales problem there were only two significant numbers: the mean and the standard deviation. The mean sales must represent something like the demand and the standard deviation must represent the average monthly variation in demand. So one would expect that if you were drawing several standard deviations above the mean, any purported influence of draw on sales would have to disappear. Scientific American is drawn at well over ten standard deviations above the mean sales. There couldn't be any effect. An obvious experimental test suggested itself: lower the draw.
My theories impressed nobody. Even as he handed over the old sales figures the Circulation Manager explained that in my analysis I must take into account the draw, articles run by other magazines, the cover...essentially every phenomenon in the universe. We agreed that it did not make sense to go to all red covers: 50 percent of them would still sell less than average. The Circulation Manager was not alone in his skepticism of objectivity. The presentation to the circulation department, scheduled for the next day, was canceled "for anon" by the President of the Magazine Division and Publisher. According to the Editor, the President was simply too busy and the meeting "might" be rescheduled in the New Year.
The message became louder a few days into 1989 when I received a memo from the Editor. It reiterated his claim that "man-and-his-weird-machine" covers had "extraordinary drawing power" with the public. The Editor had enclosed the sales figures for five man-and-machine covers since 1974. Man-and-machine covers include human-powered hydrofoils, walking machines and other unlikely combinations of sinew and steel. Of the five covers, all but one had sold above average in absolute numbers. My first question was whether the Editor had listed all relevant covers. A little research showed that he had missed at least two--but the sales figures for those also proved to be above average. I was dumbfounded. Six out of seven man-and-machine covers had sold above average. Had the Editor discovered the Holy Grail? A tremor passed through my carefully forged principles.
Here was a classic case of mistaking noise for a signal. When I finally got the sales figures from 1984 onward, I discovered two apparently real trends. Every December sells better than average and every May sells worse than average. The trends are not pronounced enough to destroy the Gaussian distributions I've talked about but in the data on hand there are no exceptions. The Christmas spirit (or airport traffic) probably explains the high December sales, and the low May sales could be associated with the close of the academic year, an admittedly remote possibility since academics do not read the magazine. I see no other evidence for seasonal variation. What does the Christmas spirit have to do with man-and-machine? It turns out that no fewer than three of the seven covers in question were on December issues! At best nothing can be concluded from them; most likely they sold better than average because December always sells better than average. So, three of the data points had to be discarded and that left four--three better than average and one worse than average. I told the Editor that we should try a man-and-machine cover on an issue that was neither December nor May.
My wish came true in March 1989, when Scientific American featured a solar-powered racer with driver on the cover. The Circulation Manager had gone out on a limb, predicting in a lengthy memo that the superior cover, contents and interior illustrations would result in sales of 115,000 copies, far above average. The prediction, if foolhardy, was in the best scientific tradition--easy to prove wrong. And wrong it was. The March issue sold only 89,000 copies, noticably below average. So, we now had five usable data points for man-and-machine: three sold above average and two below average. I would hesitate to claim "extraordinary drawing power" for man-and-machine.
Perhaps the March issue's poor showing was due to its blue cover. I think you can see why I didn't take seriously the fact that in 1987-1988 three out of five red covers sold better than average and six out of nine blue covers sold worse than average: a little more data and everything could change. (It seems almost unsporting to point out that one of the red covers was in December and two of the blue covers were in May. I will be accused of downright protectionism if I remind you that the cheetah was also a May cover.) As for "artistic" covers, they included everything from Renoir to architecture to disleyxia to hunger. A division into "art" and "social problems" would have recategorized the conclusion.
The President of the Magazine Division and Publisher finally arrived. I showed him the five-year Gaussian, explained its significance and told him they should not concentrate on the fluctuations but on the mean. For a golden moment I thought communication had been estabished until one of those present asked what would happen if we could eliminate the tail end of the distribution--those five or six issues that sold the worst. I then made the mistake of answering tautologically: "Then you'd decrease the volatility." "But that's exactly what we want!" The President of the Magazine Division and Publisher shouted.
I did attempt to explain that if I was correct one would never pinpoint the common factor between the six worst sellers. No one was listening. The Publisher offered to get sales figures for four other magazines so I could repeat the exercise, snarled something about rocket scientists doing circulation analysis, grabbed a donut and left. In the aftermath I managed to put to the Associate Publisher the question that had troubled me for months: had anyone on Madison Avenue ever made a histogram? "Kelvin was the fortieth person to have the idea for the refrigerator," he replied.
Events had made clear that a physicist has no business in the real world and I eventually left Madison Avenue. Shortly after I announced my resignation the President of the Magazine Division and Publisher vanished, though I cannot say whether the two events were connected. It is probably too much to hope that even under the new regime Scientific American will lower its draw. In December 1989, some months after I left, a predominantly red animal cover appeared. I claim credit for the breakthrough. However, that birds of paradise spread their plummage in December shows Scientific American still has not learned to design a proper scientific experiment; I'm willing to bet it sold better than average. I'm also willing to bet that someone in the circulation department has seen the figures and has been yet again impressed by the selling power of red covers. Ah, well. As I said, a physicist has no business in the real world. We now return you to your regular programming.