Cobalt: IELTS Academic Reading practice test

At the Paris exposition of 1855 a small display of dull grey bars was placed beside the crown jewels of France, and attracted a comparable crowd. The bars were aluminium. The emperor was said to reserve a set of aluminium cutlery for the guests he wished to honour most, everybody else being given gold, and when the Washington Monument was finished in 1884 its builders capped it with a pyramid of the metal weighing under three kilograms, which cost roughly what a skilled workman earned in a year. Nothing about the substance itself justified this. Aluminium is not rare.

It is, in fact, the most abundant metal in the earth's crust, making up something like eight per cent of it by weight, which is more than iron. The difficulty is that it is never found in a metallic state. Aluminium binds to oxygen with a tenacity that few other elements can match, and the methods that had served for thousands of years on the ores of copper, tin and iron, which amount to heating them with charcoal until the oxygen leaves, simply do not work on it. Nineteenth-century chemists could separate the metal by driving it out with sodium, but sodium was itself costly, and the yield was a few grams at a time.

The problem was solved twice in the same year, independently, by two men of the same age. In 1886 Charles Hall in the United States and Paul Héroult in France each found that although aluminium oxide will not melt at any workable temperature, it will dissolve in a bath of molten cryolite, and that a current passed through the resulting liquid deposits metal at the bottom of the cell. Two years later Karl Bayer worked out a cheap way of getting pure oxide out of the ore, which is called bauxite. The price of aluminium fell by a factor of thirty within a decade and has never risen again.

What replaced the cost of chemistry was the cost of electricity. Producing a tonne of aluminium by the Hall-Héroult process takes something like fourteen megawatt hours, which is roughly what an average European household uses in four years. The industry therefore went where power was cheapest rather than where the ore was: to the foot of waterfalls in Switzerland and Norway, later to hydroelectric schemes in Canada, Iceland and Siberia, and to any country willing to sell electricity below the price it charged its own citizens. A smelter is best understood not as a factory that makes metal but as a device for turning cheap power into a portable form, which is why one writer called the product congealed electricity.

Cheapness changed what the metal was for. Pure aluminium is too soft to build with, but in 1906 a German metallurgist noticed that a particular alloy he had been testing was harder on Monday than it had been on Saturday, and that the hardening happened by itself at room temperature over a matter of days. The alloy he had stumbled on made the first metal aircraft possible, and the principle behind it underlies most aluminium engineering since. Within fifty years the metal was in aeroplanes, window frames, overhead power lines, cooking foil and the drinks can, an object that uses the lightness of the material to carry itself and its contents at a cost of a fraction of a penny.

There is one respect in which aluminium is unusually well behaved. Melting scrap and casting it again requires about five per cent of the energy needed to win the metal from ore, and the metal does not deteriorate in the process, so a can may become another can indefinitely. The limit is not physical but organisational: alloys made for different purposes contain different additions, and once they are mixed in a furnace the result can only be used for something less demanding than either. Keeping streams of scrap separate is therefore worth more than any improvement in the melting itself.

The costs that remain sit at the start of the chain rather than the end. Refining bauxite leaves a strongly alkaline residue known as red mud, produced in vast quantity and stored behind dams whose failure has caused serious accidents. The smelting cells themselves consume carbon electrodes, which react with the oxygen released and emit carbon dioxide directly, quite apart from whatever the electricity was generated by. Work on electrodes that would not be consumed in this way has been under way for decades and has recently reached demonstration scale, which is the point at which such things have failed before.

The story is often told as one of genius and invention, and the inventions were real enough. But the deeper lesson concerns what makes a material scarce. Aluminium was precious for most of the nineteenth century not because there was little of it, and not because nobody wanted it, but because prising it away from oxygen required more energy than anyone could cheaply supply. When that changed, so did the price, and the crown jewels went back to being the only valuable things in the case.

A remarkable proportion of what passes for knowledge about human beings comes from a single instrument. Someone was asked a question and their answer was written down. Opinion polls, health statistics, customer research, most of psychology and a good deal of economics rest on this procedure, and it is worth stating plainly that it is among the least reliable tools any science uses. Nobody defends it as ideal. What is rarely said is how much of the edifice would have to be pulled down if it were abandoned, and that question deserves an answer before the demolition starts.

The difficulties begin with the assumption that people know the answer. A well-known series of experiments by Richard Nisbett and Timothy Wilson showed that participants would confidently explain choices they had demonstrably made for other reasons: shown four identical items and asked to pick the best, they picked the rightmost far more often than chance allows, and then explained their choice by reference to the quality of the material. The explanation was not a lie. It was a reconstruction, produced after the fact by a mind with no access to the process it was describing. Where a question asks why, the answer may be a theory rather than a report.

The second difficulty is that the question changes the answer. In a study conducted in the 1940s, about half of one group said they would not allow public speeches against democracy, while three quarters of an equivalent group said such speeches should be forbidden. Allowing and forbidding are the same decision, and the difference of a word moved a quarter of the sample. Similar effects follow from the order in which questions are asked, and from the scale offered: people asked how many hours of television they watch report substantially more when the scale runs up to four hours and over than when it runs up to two and a half, because they read the middle of the scale as what is normal and place themselves near it.

Third, people answer in the knowledge that somebody is listening. Voting, exercise, charitable giving, reading and hand-washing are all reported at levels that administrative records show to be impossible; alcohol consumption reported in surveys typically accounts for around half of what is known to have been sold. Techniques exist to get around this, from randomising the response so that no individual answer can be interpreted, to asking how many statements from a list are true rather than which ones, and they work to a degree, at the cost of much larger samples and a good deal of explaining.

The fashionable remedy is to stop asking altogether and watch instead. Purchase data, telephone sensors, search queries and administrative records all record what people did rather than what they say they did, and the enthusiasm for them is understandable. It is also, in my view, running ahead of the evidence. A behavioural record is not self-interpreting: a click may be interest, obligation, mis-taps or a child with a tablet, and inferring a preference from it involves assumptions that are simply less visible than the ones in a questionnaire. Such data also cover only what the instruments happen to capture, which is heavily weighted towards commerce and towards people who own the devices.

There is, besides, a class of questions for which asking is not a second-best method but the only method there is. If the quantity of interest is how much pain a patient is in, whether an employee feels secure, how frightened somebody is of a procedure or how bereaved they still are, then the experience is the object of study and there is nothing behind it to measure instead. Ruth Ilesanmi, who has spent a career on survey methods in public health, makes the point sharply: the critique of self-report confuses an unreliable instrument with an uninformative one. A thermometer that reads two degrees high is not useless; it is useless only to somebody who does not know that it reads two degrees high.

Ilesanmi's programme is accordingly one of repair rather than replacement, and most of it is unglamorous. Ask about specific recent behaviour rather than general frequency, because a person can count last week and cannot average a year. Anchor questions to events the respondent will remember. Use several items rather than one, so that the idiosyncrasies of any particular wording partly cancel. Validate against records wherever records exist, in order to learn the size of the bias rather than to pretend it is absent. And publish the exact wording, always, because a finding reported without the question that produced it is not a finding but a rumour about one.

That last point is where my own impatience lies. The weaknesses of self-report have been documented for eighty years and are taught to undergraduates, yet a newspaper report of a survey almost never carries the wording, the order or the response options, and a surprising number of academic papers do not either. The choice we face is not between asking people and measuring them, since both methods are shot through with assumptions and neither can be done naively. It is between a discipline that keeps quiet about how its numbers were produced and one that does not, and the second kind of science would be more trustworthy even if its instruments were no better.