The brain is hidden inside a bony case, and until recently the only way to see a living brain at work was to wait for something to go wrong. Every method neuroscientists use has strengths and blind spots, and knowing them is the best protection against the flood of claims that begin "scientists found that the brain..."

Learning from damage

The oldest method is the lesion study: find people with damage to a particular area and see what they can no longer do. In 1874 the German neurologist Carl Wernicke described patients with damage toward the back of the left temporal lobe. Unlike patients with damage to Broca's area, who struggled to produce speech, Wernicke's patients spoke fluently but with little meaning, and they had trouble understanding others.

Put together, these cases form a double dissociation: damage to one area harms the first ability but not the second, and damage to another area does the reverse. That pattern is much stronger evidence that two abilities rely on different machinery than a single case could ever be. Lesion studies have limits, though. Injuries rarely respect neat boundaries, they differ from person to person, and the brain reorganizes after damage, so what remains may not show what was lost.

Listening to electricity

In 1924 the German psychiatrist Hans Berger recorded the first human electroencephalogram, or EEG, from electrodes on a person's scalp. He was cautious and waited until 1929 to publish. The EEG picks up the combined electrical activity of millions of neurons firing together, and it shows rhythms such as the alpha waves that appear when a person relaxes with closed eyes. EEG's great strength is timing: it tracks changes within milliseconds. Its weakness is location. Signals blur as they pass through skull and scalp, so it is hard to tell exactly where they began. EEG remains essential for diagnosing epilepsy and studying sleep, and it is cheap and portable.

Watching blood flow

Since the early 1990s the star of brain research has been functional magnetic resonance imaging, or fMRI. Active neurons use energy, and within a few seconds the local blood supply surges to meet the demand. fMRI measures changes in blood oxygen near active neurons, because oxygen-rich and oxygen-poor blood respond differently to a strong magnetic field. It can pinpoint activity to within a few millimeters, far better than EEG, and it is safe enough to scan the same healthy volunteer many times.

But fMRI is an indirect, slow measure. The blood response lags several seconds behind the neural activity that caused it, and each small cube in the image contains hundreds of thousands of neurons. There are statistical traps too. A scan divides the brain into tens of thousands of cubes and tests each one, so some will appear active by pure chance. In 2009 researchers led by Craig Bennett made the point memorably. They scanned a dead Atlantic salmon while "showing" it photographs and, using common but loose methods, found a cluster of apparent activity in its brain. The lesson was that uncorrected statistics can find activity in pure noise. Standards improved sharply afterward. More recent work has shown that studies linking brain scans to traits such as personality often need thousands of participants, not dozens, to give reliable results.

Another trap is in reasoning rather than math. Suppose a region is active when people feel fear. It does not follow that activity there means fear, because the same region may also respond to surprise, novelty or pain. Reverse inference, concluding that a mental state is present just because a region is active, is tempting and often wrong.

Testing causes

Imaging shows what happens together, not what causes what. To test causes, researchers intervene. Transcranial magnetic stimulation, or TMS, uses a magnetic pulse to briefly disrupt or excite a region; if disrupting an area slows a task, that area probably helps with it. In animals, a technique called optogenetics, developed in the mid 2000s, makes chosen neurons respond to light so they can be switched on and off with remarkable precision.

Worked example: the glowing headline

A news story announces: "Brain scans prove voters love Candidate X." The study scanned twenty volunteers as they watched campaign ads and found that a reward-related region was more active during Candidate X's ad than during others. How should a careful reader respond?

First, sample size: twenty people is small, and effects found in small samples often shrink or vanish on repetition. Second, statistics: were the results corrected for testing thousands of brain locations, or could this be a salmon-style fluke? Third, and most important, this is reverse inference. The region may respond to bright colors, music or simple surprise, not love. Fourth, fMRI measured blood flow over seconds, not feelings. A better headline might be "One region responded more to one ad in a small study." That is less exciting, but it is what the evidence supports. Knowing how a method works lets a citizen judge the claim instead of trusting the glow of the picture.