Statistical fallacy impairs post-publication mood

banksyNo scientific paper is perfect, but a recent result on the affect of mood on colour perception is getting a particularly rough ride post-publication. Thorstenson and colleagues published their paper this summer in Psychological Science, claiming that people who were sad had impaired colour perception along the blue-yellow colour axis but not along the red-green colour axis. Pubpeer – a site where scholars can anonymously discuss papers after publication – has a critique of the paper, which observes that the paper commits a known flaw in its analysis.

The flaw, anonymous comments suggest, is that a difference between the two types of colour perception is claimed, but this isn’t actually tested by the paper – instead it shows that mood significantly affects blue-yellow perception, but does not significantly affect red-green perception. If there is enough evidence that one effect is significant, but not enough evidence for the second being significant, that doesn’t mean that the two effects are different from each other. Analogously, if you can prove that one suspect was present at a crime scene, but can’t prove the other was, that doesn’t mean that you have proved that the two suspects were in different places.

This mistake in analysis  – which is far from unique to this paper – is discussed in a classic 2011 paper by Nieuwenhuis and colleagues: Erroneous analyses of interactions in neuroscience: a problem of significance. At the time of writing the sentiment on Pubpeer is that the paper should be retracted – in effect striking it from the scientific record.

With commentary like this, you can see why Pubpeer has previously been the target of legal action by aggrieved researchers who feel the site unfairly maligns their work.

(h/t to Daniël Lakens and jjodx on twitter)

UPDATE 5/11/15: It’s been retracted

Actually, still no good explanation of ‘that dress’

The last time I almost went blind staring at “that dress” was thanks to Liz Hurley and on this occasion I find myself equally unsatisfied.

I’ll spare you the introduction about the amazing blue/black or white/gold dress. But what’s left me rather disappointed are the numerous ‘science of the dress’ articles that have appeared everywhere and say they’ve explained the effect through colour constancy.

Firstly, this doesn’t explain what we want to know – which is why people differ in their perceptions, and secondly, I don’t think colour constancy is a good explanation on its own.

To explain a little, colour constancy is an effect of the human visual system where colours are perceived as being different depending on their context as the brain adjusts for things like assumed lighting and surroundings. Here’s a good and topical example from XKCD. The dress colours are the same in both pictures but the seem different because the background colour is different.

An important feature of the visual system is that the experience of colour is not a direct result of the wavelength of the light being emitted by the surface. The brain modifies the experiences to try and ensure that things appear the same colour in different lighting because if we just went off wavelength everything would wildly change colour as it moved through a world which is lit unevenly and has different colour light sources.

Visual illusions take advantange of this and there are plenty of examples where you can see that even completely physically identical colours can be perceived as markedly different shades if the image suggests one is in shadow and the other in direct light, for example.

Firstly, this isn’t an explanation of why people differ in perceiving the dress. In fact, all of the ‘science explanations’ have simply recounted how perceived colours can change but not the most important thing which is why people are having two stable but contradictory experiences.

Colour constancy works on everyone with normal colour vision. If you take the panels from the XKCD cartoon, people don’t markedly disagree about what the perceived colours are. The effect of each image is very reliable between individuals.

That’s not the case with the dress. Also, if you say context makes a difference, changing the surroundings of the dress should change the colours. It doesn’t.

Some have argued that individual assumptions about lighting in the picture are what’s making the difference. In other words, the context is the unconscious assumptions people make about lighting in the picture.

But if this is the case, this still isn’t an explanation because it doesn’t tell us why people have different assumptions. Psychologists called these top-down effects or, if we’re going to get Bayesian, perceptual priors.

75% of people in this BuzzFeed poll said they saw white/gold, 25% said they saw blue/black, and a small minority of people say they’ve seen the picture ‘flip’ between the two perceptions. How come?

And there’s actually a good test of the colour constancy or any other other ‘implicit interpretation’ explanation. You should be able to create images that alter the visual system’s assumptions and make perception of the dress reliably flip between white/gold and blue/black, as with the XKCD cartoon.

So, any vision scientists out there who can come up with a good explanation of why people differ in their perceptions? Psychophysicists, have I gone wildly off track?

How muggers size up your walk

The way people move can influence the likelihood of an attack by a stranger. The good news, though, is that altering this can reduce the chances of being targeted.

How you move gives a lot away. Maybe too much, if the wrong person is watching. We think, for instance, that the way people walk can influence the likelihood of an attack by a stranger. But we also think that their walking style can be altered to reduce the chances of being targeted.

A small number of criminals commit most of the crimes, and the crimes they commit are spread unevenly over the population: some unfortunate individuals seem to be picked out repeatedly by those intent on violent assault. Back in the 1980s, two psychologists from New York, Betty Grayson and Morris Stein, set out to find out what criminals look for in potential victims. They filmed short clips of members of the public walking along New York’s streets, and then took those clips to a large East Coast prison. They showed the tapes to 53 violent inmates with convictions for crimes on strangers, ranging from assault to murder, and asked them how easy each person would be to attack.

The prisoners made very different judgements about these notional victims. Some were consistently rated as easier to attack, as an “easy rip-off”. There were some expected differences, in that women were rated as easier to attack than men, on average, and older people as easier targets than the young. But even among those you’d expect to be least easy to assault, the subgroup of young men, there were some individuals who over half the prisoners rated at the top end of the “ease of assault” scale (a 1, 2 or 3, on the 10 point scale).

The researchers then asked professional dancers to analyse the clips using a system called Laban movement analysis – a system used by dancers, actors and others to describe and record human movement in detail. They rated the movements of people identified as victims as subtly less coordinated than those of non-victims.

Although Professors Grayson and Stein identified movement as the critical variable in criminals’ predatory decisions, their study had the obvious flaw that their films contained lots of other potentially relevant information: the clothes the people wore, for example, or the way they held their heads. Two decades later, a research group led by Lucy Johnston of the University of Canterbury, in New Zealand, performed a more robust test of the idea.

The group used a technique called the point light walker. This is a video recording of a person made by attaching lights or reflective markers to their joints while they wear a black body suit. When played back you can see pure movement shown in the way their joints move, without being able to see any of their features or even the limbs that connect their joints.

Research with point light walkers has shown that we can read characteristics from joint motion, such as gender or mood. This makes sense, if you think for a moment of times you’ve recognised a person from a distance, long before you were able to make out their face. Using this technique, the researchers showed that even when all other information was removed, some individuals still get picked out as more likely to be victims of assault than others, meaning these judgements must be based on how they move.

Walk this way

But the most impressive part of Johnston’s investigations came next, when she asked whether it was possible to change the way we walk so as to appear less vulnerable. A first group of volunteers were filmed walking before and after doing a short self defence course. Using the point-light technique, their walking styles were rated by volunteers (not prisoners) for vulnerability. Perhaps surprisingly, the self-defence training didn’t affect the walkers’ ratings.

In a second experiment, recruits were given training in how to walk, specifically focusing on the aspects which the researchers knew affected how vulnerable they appeared: factors affecting the synchrony and energy of their movement. This led to a significant drop in all the recruits’ vulnerability ratings, which was still in place when they were re-tested a month later.

There is school of thought that the brain only exists to control movement. So perhaps we shouldn’t be surprised that how we move can give a lot away. It’s also not surprising that other people are able to read our movements, whether it is in judging whether we will win a music competition, or whether we are bluffing at poker. You see how someone moves before you can see their expression, hear what they are saying or smell them. Movements are the first signs of others’ thoughts, so we’ve evolved to be good (and quick) at reading them.

The point light walker research a great example of a research journey that goes from a statistical observation, through street-level investigations and the use of complex lab techniques, and then applies the hard won knowledge for good: showing how the vulnerable can take steps to reduce their appearance of vulnerability.

My BBC Future column from Tuesday. The original is here. Thanks to Lucy Johnston for answering some of my queries. Sadly, and suprisingly to me, she’s no longer pursuing this line of research.

A taxonomy of ayahuasca hallucinations

A wonderful list categorising hallucinations experienced by the Cashinahua people of Peru after drinking the hallucinogenic brew ayahuasca.

1. Brightly colored, large snakes
2. Jaguars and ocelots
3. Spirits, both of ayahuasca and others
4. Large trees, often falling trees
5. Lakes, frequently filled with anacondas and alligators
6. Cashinahua villages and those of other Indians
7. Traders and their goods
8. Gardens

It was reported by the anthropologist Ken Kensinger in a chapter in the book Hallucinogens and Shamanism.

It reminded me of writer Jorge Luis Borges’ whimsical classification system for animals.

Photographing hallucinations

BMJ Case Reports has a paper that describes two patients with Parkinson’s disease who experienced hallucinations that transferred onto photos they took to try and prove they were real.

This is ‘Patient 1’ from the case report:

Patient 1 was first evaluated at age 66, having been diagnosed with PD [Parkinson’s Disease] at age 58… She complained of daytime and night-time visual hallucinations for the past one year. Most of the time she did not have insight about them. She described seeing three children playing in her neighbour’s yard and a brunette woman sleeping under the covers in one of the beds in her house. She also saw images of different people sitting quietly in her living room. Most of her visual hallucinations subsided in open and brightly lit spaces but were, nevertheless, troublesome. In one instance, she saw a man covered in blood, holding a child and called 911.

Her husband, in an attempt to prove to her that these were hallucinations, took pictures of the neighbour’s yard and the bed in their house. Surprisingly, when shown these photos, the patient continued to identify the same children playing in the yard and the same brunette woman sleeping under the covers. This perception was present every time the patient looked at these photos. Within 6 months of stopping ropinirole and titrating quetiapine to 75 mg every night at bedtime the hallucinations were less severe and shorter in duration, but the patient continued to see them in the photos.


Link to locked article in BMJ Case Reports.

Hallucinating sheet music

Oliver Sacks has just published an article on ‘Hallucinations of musical notation’ in the neurology journal Brain that recounts eight cases of illusory sheet music escaping into the world.

The article makes the interesting point that the hallucinated musical notation is almost always nonsensical – either unreadable or not describing any listenable music – as described in this case study.

Arthur S., a surgeon and amateur pianist, was losing vision from macular degeneration. In 2007, he started ‘seeing’ musical notation for the first time. Its appearance was extremely realistic, the staves and clefs boldly printed on a white background ‘just like a sheet of real music’, and Dr. S. wondered for a moment whether some part of his brain was now generating his own original music. But when he looked more closely, he realized that the score was unreadable and unplayable. It was inordinately complicated, with four or six staves, impossibly complex chords with six or more notes on a single stem, and horizontal rows of multiple flats and sharps. It was, he said, ‘a potpourri of musical notation without any meaning’. He would see a page of this pseudo-music for a few seconds, and then it would suddenly disappear, replaced by another, equally nonsensical page. These hallucinations were sometimes intrusive and might cover a page he was trying to read or a letter he was trying to write.

Though Dr. S. has been unable to read real musical scores for some years, he wonders, as did Mrs. J., whether his lifelong immersion in music and musical scores might have determined the form of his hallucinations.

Sadly, the article is locked behind a paywall. However you can always request it via the #icanhazpdf hashtag on twitter .

Link to locked article on ‘Hallucinations of musical notation’.

A retrospective editing of consciousness

A new study has found that conscious experience can be altered retrospectively, so that experience of visual information can be changed almost half a second later by manipulating where our attention is drawn.

The research, led by cognitive scientist Claire Sergent, involved asking people to stare at a centre point of a screen with two empty circles either side.

At some point, one of the two circles would fill with randomly oriented stripes for just 50ms (one twentieth of a second) and afterwards the participants were asked to say which direction the stripes were pointing in.

Crucially however, each time this happened, one of the two circles would dim either before or after the stripes appeared.

This would happen at different times – from 400ms before the stripes appeared, up to 400ms after the stripes appeared, and the dimmed circle might appear on the matching side to the stripes or on the opposite side.

Dimming one of the circles grabs your attention. It makes you instantly focus more on whichever side of space it happens.

For example, if the left-hand circle dims, it grabs your attention, and if the stripes then appear on the left, you’re more likely to make a correct judgement about which direction they’re pointing because you’re already focused on this area. But if the stripes subsequently appear on the other side, you’re distracted and you do worse.

The key discovery from this experiment was that this also happens if the dimmed circle appears after the stripes. Up to 400ms seconds after.

In other words, you perceive the original visual details that would otherwise have escaped consciousness if your attention is drawn to the area after the picture disappears. It’s like a retrospective editing of consciousness by post-event attention.

This suggests that consciousness isn’t ‘filtered’ sensory information, but an active ‘conclusion’ drawn from information distributed across senses, space and time.

Link to locked scientific study.
Link to open-access commentary from same journal.