Nightjar: IELTS Academic Reading practice test

Every autumn, in the fields and roadsides of eastern North America, a generation of monarch butterflies emerges that is unlike the generations before it. Its members do not mate. Instead they turn south, and over the following two months they travel as much as four thousand kilometres to a dozen patches of high forest in the mountains west of Mexico City, arriving in November. There they settle on the trunks and branches of the oyamel firs in such numbers that the boughs bend under them, and there they remain, barely moving, until the days lengthen in February.

What makes the journey remarkable is not its length but its authorship. The butterflies that fly south in October are the great-grandchildren of the ones that flew south the year before. The migratory generation lives for as much as eight months, sees out the winter and starts north in the spring, but it gets only as far as the southern United States before it lays its eggs and dies. Three or four short-lived summer generations then work their way northward, each living a few weeks, until the cycle produces another autumn generation in the north. No individual butterfly makes the round trip, and none has ever met an insect that has. Whatever they are following, they did not learn it from anybody.

Part of the answer is a sun compass. A monarch held in a flight simulator and shown an artificial sun will orient south-west, and will shift its heading predictably if the sun is moved, which shows that it is reading a position rather than a brightness. Using the sun to hold a bearing, however, requires knowing the time of day, since the sun moves across the sky, and the clock that performs this correction turns out not to be in the insect's head. Painting the antennae with opaque varnish destroys the animal's ability to hold a course; painting them with clear varnish does not. The timing mechanism sits in the antennae themselves. Monarchs also appear to carry a magnetic sense that responds to the angle at which the field meets the earth, which would allow them to keep a heading on the many days when no sun is visible at all.

Where they were going remained unknown to science until surprisingly recently. A Canadian zoologist spent forty years recruiting volunteers to stick numbered paper tags on the wings of butterflies and report the recoveries, and the tags kept pointing south-west into Mexico without ever arriving anywhere. The colonies were finally located in January 1975, by a correspondent who had answered an advertisement the zoologist placed in a Mexican newspaper and who spent two years searching the mountains on the strength of it. What he found was a forest carrying tens of millions of insects in a few hectares. The site works because it is cold without being lethal: the firs hold the temperature just low enough to shut the butterflies' metabolism down, and their canopy keeps the frost off.

The other half of the monarch's life is a plant. The caterpillars will eat nothing but milkweed, and milkweed defends itself with a group of bitter compounds that stop the hearts of most animals that swallow them. The caterpillar not only tolerates these poisons but stores them, so that the adult butterfly is unpalatable to birds, which is what its conspicuous orange advertises. The relationship is exact and it is also a vulnerability, because a butterfly that can lay its eggs on only one kind of plant is a butterfly whose range is drawn by that plant.

That vulnerability has become the centre of an argument. The area of Mexican forest occupied each winter, which is the standard measure of the population, has fallen a long way since the 1990s, and the explanation usually offered is the loss of milkweed from the farmland of the American Midwest, where herbicide-tolerant crops allowed the weed to be cleared from between the rows. Others point to illegal logging at the wintering sites, or to pesticides, or to the weather encountered en route. There is also a puzzle in the numbers: counts made by volunteers across the summer breeding range have not fallen nearly as steeply as the winter measurements, which has led several researchers to argue that the losses that matter most are those suffered during the autumn migration rather than during the summer.

Conservation has run ahead of the science, occasionally in the wrong direction. Planting milkweed in gardens has become a popular response, and in the north it is straightforwardly helpful. In the far south of the United States, however, the species most often sold is a tropical milkweed that does not die back in winter, and monarchs that find it growing in December tend to stay and breed instead of continuing south. Those that do so carry much heavier burdens of a protozoan parasite, which builds up on foliage that is never cleared away. A well-meant garden can therefore hold back the very behaviour it was planted to support.

It is worth being clear about what is at risk. The monarch as a species is in no danger at all; it is found in Australia, in Spain, on Pacific islands, and it is doing well in most of them. What is threatened is a behaviour, a four-thousand-kilometre round trip conducted by an animal that cannot be taught it, and there is no obvious category in conservation law for such a thing. We are used to protecting organisms. Protecting something an organism does, which exists only in the relationship between an insect, a weed and a particular hillside, is a problem we have barely begun to think about.

There is something odd about the market for forecasts. Almost everyone who buys one is aware that it may be wrong, and almost everyone who sells one is aware that the record of such things is poor. And yet the supply of confident statements about next year keeps growing, and so does the demand. Part of the explanation is that a forecast is not only a prediction; it is also a way of authorising a decision that somebody has to take anyway, and a decision taken against a published number is easier to defend afterwards than a decision taken against a feeling.

The most thorough attempt to find out how good such statements actually are ran for two decades. Philip Tetlock collected nearly thirty thousand specific, checkable predictions from several hundred people who were paid to comment on political and economic affairs, and then waited to see what happened. The average performance was poor. Taken as a group, the experts did little better than chance on questions with three possible outcomes, and were beaten by simple rules that assumed next year would look much like this one. The finding that attracted most attention was a secondary one: within the group, the forecasters with the greatest public profile performed worst, which suggests that whatever earns an invitation to speak is not accuracy.

What separated the better forecasters from the rest was not their politics or their field but their style of thinking. Tetlock borrowed a distinction from an old essay about the hedgehog that knows one big thing and the fox that knows many small ones. The hedgehogs had a governing theory and read every new development as further evidence for it; they were confident, quotable and wrong more often. The foxes held several partial explanations at once, were uncomfortable in front of a camera, and scored better. The difference was not knowledge. It was the willingness to let an account be complicated.

A decade later a government research agency turned this into a competition. Thousands of volunteers were recruited to answer questions about world events, their answers scored for accuracy and calibration, and the best performers identified and grouped into teams. These volunteers, who had no access to secret material, outperformed the control groups by a wide margin and, by the organisers' account, matched or beat professional analysts who did have such access. What the best of them did was describable and, more importantly, teachable: they broke a vague question into parts that could be estimated separately, they began from a base rate, asking how often things of this kind happen in general before considering anything specific about this case, and they updated their estimates in small increments rather than swinging between positions.

Not everyone accepts that the tournament settles very much. The political scientist Marek Oduya points out that a question cannot be scored unless it resolves, and a question cannot resolve within the life of a research programme unless it is fairly short-term and fairly precise. The questions that dominate real decisions are neither. Whether a government will still be standing in ten years, whether a technology will become cheap enough to matter, whether a region will remain habitable: these do not resolve on schedule, and being well calibrated about the price of oil in six months is no evidence at all of being well calibrated about any of them. Oduya's complaint is not that the superforecasters are not good. It is that they are good at the tractable questions, which are not the ones we are anxious about.

He is largely right, and the response is to be honest about what kind of uncertainty is in play. Some of it is simply noise: the outcome depends on events that are genuinely random and no amount of study will remove them. Some of it is ignorance about the mechanism, which more research can in principle reduce. And some of it is of a third kind that has no analogue in the physical sciences, because in human affairs the forecast becomes part of the situation. A prediction that a bank will fail can help to make it fail; a prediction that a policy will be unpopular can determine whether it is attempted. A weather forecast does not change the weather, which is one reason weather forecasting has improved so much more than the other sort.

If that is right, the useful reforms are modest and unglamorous. Where a single number is not knowable, the honest product is a set of scenarios, each internally coherent, together with an account of what would have to be true for each. The decision that follows should then be chosen for how well it survives across several of them rather than how well it performs in the most likely one, which is the difference between a robust plan and an optimal one. And whatever is predicted should be written down with an explicit probability attached and revisited later, because a prediction without a number cannot be scored, and an organisation that never scores itself will go on believing that it saw the last crisis coming.

None of this delivers what is actually wanted, which is to know what will happen. My own view is that the real value of forecasting lies somewhere else, and that the scoring is a means rather than an end. Forcing a belief into the form of a probability about a dated event exposes the assumptions it is resting on, several of which usually turn out to be doing more work than their owner realised. The people who came out of the tournament best were not the ones who could see further than the rest. They were the ones who could say most precisely what they were assuming, and who noticed soonest when one of those assumptions stopped being true.