marius | 05-09-2010 12:10Distant Intentionality

A few days ago I had a discussion with a friend who claimed there was more evidence for telepathy than for drugs against AIDS (or HIV transmission probably). That's a bold statement. And, although I didn't have any information to verify it, I immediately felt that must be bullshit: I'm a skeptic. So I responded along those lines, which wasn't pretty. Later I wondered what the evidence actually was and if my reaction was perhaps unfair. So I did a little searching on the web.

The first thing I did was look for the website of The Cochrane Collaboration. They do meta-analyses that both control for publication bias and increase power by virtue of an increased amount of data. So they are more fair and simultaneously increase the ability of finding even a small effect. Unfortunately (for me), they only review medical treatments, and telepathy is not a medical treatment.

They did have a lot of meta-analyses on acupuncture, which also briefly popped up in our conversation. Of all the concluded analyses (35) there were three that showed a slight benefit of acupuncture, namely for treating neck pain, tension type headache and pelvic and back pain in pregnancy. Most reviews state that not enough trials have been done, they were not properly controlled, or the effects can not be distinguished from placebo. Only one says that acupuncture does not help: against epilepsy.

This all surprised me. I do see a small pattern, although that may just be my prejudice against alternative medicine. All positive results were found in treating pain. As far as I know, there is no other way to assess pain than by self-report. I do think that if someone says they're in pain, they probably are in pain. That is, that can be used as a more or less objective measure. However, rating the pain on a scale is susceptible to all sorts of psychological distortions, including placebo effects and other forms of positive thinking. I would also say that, apart from the needles of course, acupuncture has a lot in common with getting a relaxing massage. Being relaxed may alter your perception of pain, though it will not change the number of epilepsy attacks you have. Again, this is the skeptic in me.

So, how about AIDS? There are quite a few reviews of literature on AIDS, with some positive and negative effects and a lot of inconclusive results. Given that this is a much newer field of research than acupuncture or telepathy, I'd say there is more evidence about drugs for AIDS. They definitely don't appear to be less effective than acupuncture, but it's hard to say if they are more effective. One review was rather interesting: comparing the effect of vitamin A and some drug on Mother-to-child-transfer of HIV. People apparently still believe that vitamins work against AIDS. There was a slight effect of vitmamin A though: this was measured among mothers in rural (probably poor) Africa. For me that is just evidence that a normal, balanced diet is healthy, not that you should take additional vitamins.

Is there any evidence for telepathy? First I found a meta-analysis here on pubmed. The people writing it claim there is a small effect but the evidence is shaky and lacks a 'theoretical framework'. (More on that later.) The people doing that research might have been a little biased: the first author published a lot of parapsychology papers. Even so, they ranked all studies according to what -in their experience- were good measures of quality (sound-proofed rooms and all that). Their own experiments were all very high up (in a seperate group) and these showed no overall effect. They did not want to downplay the work of others (or introduce circular reasoning) so they did not exclude the other work from the meta analysis.

I must say they had an interesting way to deal with publication bias: they used unpublished data as well. How can anyone verify the worth of that work without repeating the whole meta-analysis? That defeats the purpose of meta-analyses. And since the authors have a vested interest in showing there is an effect, I don't completely trust the paper.

What was cool is that in the 'Related Citations' there was this study where the researchers first found an effect of experimenter. That is: experiments conducted by a believer showed an effect and those done by a skeptic showed no effect. They were collaborating so there were no other differences between the experiments except for who performed them. They later repeated the effect and found nothing at all.

OK, so my own verdict is that telepathy has not yet been sufficiently disproved, so if people want to, they can still believe in it without ignoring empirical evidence. I also think it is highly unlikely to exist, given that there is no theoretical framework. This means: there is no suggested way by which telepathy might work. To be clear: acupuncture does have a theory, although I would guess there is little to no evidence for qi to be found. Also, until there is a theoretical framework for telepathy, it is hard to test it's existence fairly. And until then, I will keep believing there is no such thing.

No comments

marius | 30-08-2010 19:24Descartes on MTV

The idea has been around for a long time of course, but it's ties in rather well with the somewhat recent embodied cognition school of thought (e.g., Andrew Clark): cognition is situated. What I want to look at now is that all of our cognition, in particular the way the information from our different modalities is processed, has probably evolved for specific reasons, that is, to allow us to do something useful.

In my field this means that vision is good for something. There are lot's of illusions that show that our visual system makes short-cuts that can be exploited. Generally, these short-cuts help make sense of real, ambiguous input however, and they usually work at a low price. I would like to make a very generic statement that is incredibly obvious but not everybody seems to be aware of it: all perception is essentially motor preparation.

Descartes had a lot right in this image: the link between vision and action.

What this means is that all perception should allow us to gather information on the world so that we may act more competently. If, on the other hand, a perceptual modality provides us with information that can not enhance our actions (and chances survival), it is a waste of resources.

When I mentioned the idea that all perception is motor preparation at a summer school last year (when explaining that I got my Master's in motor preparation and now did a PhD project on visual attention), someone said: "And what about movies, we don't need those!" That's true, but I would now like to say that we get excited by action movies or thrillers because they activate a fight-or-flight response. Other types of movies may make us prepare for other actions as well.

Another example is related to the ongoing debate on the role of Saliency (Maps) in vision. It seems to me that saliency can be usefull for animals that can not understand objects. For example, frogs trying to eat any small dark spot moving over their heads. Similarly, when we have no prior knowledge on a scene (when an image is suddenly presented on a monitor, which is a terrible approximation of real-life untill we invent teleporters) or when we have no task to do, saliency may help us orient ourselves.

The position of many vision scientists nowadays seems to be that we always have a task in real life, so that saliency can be safely ignored. If that were the case, I'd say it would be very unlikely to predict any behaviour at all. Rather than the 55% - 60% of fixations that saliency usually can predict: it should not be present in our brains at all. It could be that saliency helps us to decide when to switch tasks. For example, a squirrel gathering nuts has a clear set of actions to perform on a well described set of objects. However, if it simply keeps doing the task, it will die when a fox arrives at the scene and tries to eat it. That fox would be a highly salient, though task-irrelevant stimulus. What the squirrel needs is a way to switch tasks, so it can run away and live.

It makes sense to have a system that detects surprising, but task-irrelevant things or events. Since these by definition can not be pre-defined, a saliency based model is the ideal complement to a task-based model of attention. Incidentally, it's also why MTV is so distracting.

No comments

marius | 28-08-2010 17:27Observations on a vision science conference

Thursday I returned from ECVP 2010, in Lausanne. This swiss city is located on the shores of Lake Geneva and has a view on the outskirts of the really high Alps. The conference was on the campus of EPFL, which is mostly focussed on technical and engineering studies, but houses a relatively new group working on neuroscience as well. It's a good setting for a conference.

On ECVP there's traditionally more than enough food and drinks to loosen up the visitors. This means you get a lot of real and direct interactions. Adding to this is the fact that after more than 20 talks and hundreds of posters each day, people are too tired to be extremely formal. Unfortunately, on the fifth -and last- day (after the big banquet dinner) most people are too tired (or too drunk) to even show up.

We presented some cool work and I think people liked it. I got one idea for an experiment and of course an overview of the field (one experiment I wanted to do, has already been done, and another in a somewhat different form). So, it was a good conference.

There was one weird special symposium on Psychophysics though. It was called: Psychophysics: yesterday, today, and tomorrow, which seemed like a good opportunity to show young researchers that you don't need imaging to get some very cool results. Or even what can be done with psychophysics that can't be done with imaging techniques (and vice versa). However, there were only two speakers that really showed enthousiasm, and one had a very frail voice and the other didn't have any structure in his talk. The other speakers seemed bored with their own work, didn't have any changes of intonation and their talks went nowhere. To top it all off, all speakers were grey old men, so the impression the symposium left was that psychophysics is nearly dead. Considering this was supposed to be a special evening symposium with a lot of time alotted to all speakers it was very disappointing, a chance to show how cool psychophysics can be that the organizers (and speakers) failed to use.

A second point of criticism is that aside from a standard website, the conference had no real web-presence in terms of a twitter channel or conference blogs or some such. Since the main organisation behind the conference, Pion, publishes an open-access journal, i-Perception, one might expect something like that. Perhaps the people in Toulouse, who will organize the conference next year, will introduce some fast on-line experience.

No comments

marius | 28-08-2010 13:44Placebo Pilates

A study mentioned in Ben Goldacre's book Bad Science caught my interest. It is a study by Crum & Langer (2007) about a group of hotel attendants, one half of which is told their work is equivalent to healthy exercise, the other half is told nothing. The people who were told their work is healthy, scored better on a whole set of measures after only 4 weeks: weight, BMI, body-fat percentage, waist-to-hip ratio, systolic blood-pressure and diastolic blood-pressure. The researchers conclude:

"These results support the hypothesis that exercise affects health in part or in whole via the placebo effect."

I like the placebo effect, but this seems a little too good to be true. It has already been mentioned by Ben Goldacre that the researchers did not control for clustering. This might have affected the statistical results.

The researchers did not directly measure excercise, but instead relied on self-report of exercise, both on the job and outside the work-environment. Since self-report is a notoriously unreliable method and probably had a lot more variance than the physiological measures used to assess health, this may explain why no difference in exercise was found. The researchers -to their credit- actually admit that the measures used are not ideal, albeit in a footnote in the discussion:

Future research might benefit from using measures (e.g., pedometers, food journals, or other people’s assessment of diet and activity levels) to enhance control for these variables.

Most interestingly, the researchers did not report any test of differences in the physiological measures between the two groups before the intervention. They simply state there isn't any difference, except in age, but do not say what they tested. They do report that the hotels were matched, so that similar hotels were in both groups. However, they used 7 hotels, so 6 of them could have been matched, leaving 1 hotel to make a difference. The researchers also report that there was a difference in age between the groups, which they accounted for in the analyses. They probably did this by using age as a covariate in the ANOVA's. This would be OK if age only has a main effect, but of course all the results reported are interactions between intervention and time. Age may still explain the effects. When looking at the figures, it also seems like there are differences between the groups before the intervention in weight, BMI, percentage body-fat and waist-to-hip ratio (not in the two blood-pressure measures). If these differences reflect slightly different attitudes toward health before the intervention, this may have affected the results.

There probably still is something to this story, in that telling people that their work has built-in health benefits helps them appreciate that and act accordinly. Still I would like someone to look at the data and show the statistics on pre-intervention differences on all the physiological measures. Since the paper is relatively short it makes me wonder why they left those numbers out

Full list of references -some with links- to be found with Ben Goldacre's chapter on The Placebo Effect can be found on this blog.

No comments

marius | 16-08-2010 21:05Open Science: stimuli

I'v written on Open Science and the sharing of data. Since I've started watching the tweets of science bloggers, I've seen a whole of stuff going on in the open science department. And a few papers on poop. An encouraging example would be this collaborative effort on Alzheimer's disease.

As I said in my previous post: one of the things we will have to organize now is making raw data available for anyone. In some fields this is already happening. Before I opened my twitter account, I had never heard of http://arxiv.org/, which is where a lot of material is shared very early. There are no psychologists there and perhaps a few neuroscientists. So I still have some work to do.

But another question just popped up in my own work. We have some special visual stimuli, that are very hard to get by, but really easy to share (less then a MB of pictures). We would like to keep those stimuli for a while so we have a little edge over other researchers in the field. Now comes the problem: I just made a poster that (amongst other cool stuff) shows all the stimuli. If I make a PDF available for download on my website, then everybody could download our special stimuli.

What to do? In the end we will probably share them anyway. We could wait with it for a while, or simply make the raw images available so everybody has them at optimal resolution.

I've asked my followers for advice on this problem, but so far, there's no answer.

No comments

marius | 30-06-2010 16:02Pylab logarithmic colorbars

The one alternative I know is to create a logarithmic colormap for each dataset. I like this solution better.

#!/usr/env/python

from pylab import *
from numpy import *

samples = 20

data = rand(samples,samples)

ticklabellocs = arange(0.0,1.2,0.2)
ticklabels = []
for tli in ticklabellocs:
  label = u'%0.3f'%tli
  ticklabels.append(label)

ticklabels[0] = '0'

figure(num=1)
X, Y = meshgrid(range(samples),range(samples))
myContour = contourf(X,Y,data**0.5,arange(0.0,1.01,0.01),cmap=cm.hot)

cla()
set_cmap('hot')
imshow(data)

Clines = contour(X, Y, data, ticklabellocs, colors = '#000000', hold='on')

myCbar = colorbar(myContour, ticks=list(ticklabellocs**0.5))
myCbar.ax.set_yticklabels(ticklabels)

xlim((0,samples-1))
ylim((0,samples-1))

show()

One comment

marius | 15-05-2010 14:37Open Science

Through the increasing popularity of Open Access journals, scientific results have become more and more accessible to everyone at a very low cost. Recent Freedom of Information (FOI) trials involving the Climate Research Unit of the University of East Anglia made clear that most people consider scientific data that was gathered using tax money should also be publicly available. Not just under the Freedom of Information act in Great Britain, but everywhere in the world.

There are some initiatives in this direction going on already. The Human Genome project initially planned to release all data publicly, but then changed course to protect privacy. NASA provides access to some data gathered in space. One can buy psychological survey data from ZPID in Germany. There are some researchers and institutes that provide access to their data on their own websites, but this remains a matter of choice.

All access to data is plagued by the problem of choosing the right file format. Most research is done with proprietary (closed source) tools (e.g. Matlab) that encourage users to adopt a file format specific to the tool. So every time those researchers whish to share data, either one needs to find out how to convert the data to a format accessible to the other party or the other party needs to use the specific tool.

When the other party is already using the tool, there is no problem. However, buying the tool just for reading one dataset is likely to be too costly. This will always be the case when the data is made available to the general public. This forces users to find out how to export data to another format, which can be quit a hassle.

One solution I am in favour of is to let students and starting researchers get used to saving and accessing their data in open formats. HDF5 is a good example. A further advantage of such data formats is that there usually are free, open source tools available to work with those formats. A widescale adoptation of using open data formats will make scientific data truly freely accessible.

In the first place, it will enable people downloading such data from individual researchers and institutes to access that data and verify the conclusions drawn by the original researchers. Or perhaps it will allow doing new analyses, providing new insights.

In the second place, if a lot of data is available in truly accessible formats, this may encourage people to gather a lot of that data in large public databases. The data will then no longer be hidden on individual websites and will therefore be found and used by more people. While some replication of scientific findings will always be necessary for verification, the wheel does not need to be reinvented by every research group. This may free some capacity of researchers, which may enable them to do truly innovative research.

Of course, there is also a problem. A lot of researchers do not want to release all data to the public, for two reasons. The first being that in recent years, people have started to resent scientific findings as mostly cooked up to push elitist agenda's. Misunderstanding of basic statistical principles has prompted people to say that climate data shows there has been no rise in temperature in the past 15 years. This is a simply a matter of too much noise and not enough signal, but the public accepts this as proof that there is no climate change. Also, no number of scientific studies seems to be able to change the public view that vaccinations can cause autism. The second reason being that when data is released too soon, rival researchers may publish the results first. Both of these basically harm a researcher's career.

Being open can cause severe problems for researchers. These dangers must be dealt with before researchers will accept that their data needs to be public. Changing the public's view may be difficult to impossible. However, when there are open databases where conributions are registered, so that the source of the data is also publicly available, this may actually prevent data-theft by other researchers.

No comments

marius | 25-01-2010 16:58No Jet

The Jet colormap is the default in Matlab, and while it looks colourfull and can hence probably sell expensive products like Matlab, it also distorts data.

• First, it creates an artificial 'band' of yellow that will only by accident have any meaning in the data.
• Second, there are some 'edges' between other colours, such as between blue and cyan, even when there is supposed to be a smooth transistion between values.
• Third, just near the extreme low (dark blue) there is a small part of the colourmap that seem slightly brighter than the values that are supposedly higher (farther away from the low end). If you wouldn't know better this could look like a small local peak in the data.
• Fourth, if the plot is reduced to it's luminance, either because someone prints your paper on a black and white printer (I estimate 99% of students will do this and lots of people who do not want to waste money or the environment), or if your reader is colourblind, jet no longer has any meaning, since the two extremes now have the same 'colour'. Unfortunately, the edges and bands are still there.

I've illustrated this in the graph here (click for large version):

Other colourmaps also have their problems, but if you like your data, or want to tell the truth; please do not use jet.

No comments

marius | 18-01-2009 18:28FreeBrain

January 3rd 2009 I finally started the Sourceforge project FreeBrain that I'd been thinking about for some time. The purpose is to create a flexible and easy to learn tool for analysing all sorts of signals that neuroscientists measure. The kind of signals we aim at are two-dimensional, so you have some measure over time. This includes EEG, skin-conductance and eye-movements, but excludes fMRI and such.

I have two kinds of reasons to start this project. First, as a student I discovered the disadvantage of using a commercially licensed tool in a large research group; you had to fight for a license. This is very counterproductive of course. I picture the main part of our audience to be undergraduate students in some kind of neuroscience. That's why the tool has to be easy to learn. I also discovered that although we primarily used one piece of software for basic analysis of EEG, more advanced or experimental functions were implemented by each research in a plethora of languages. That's why I personally would like a tool that I can extend to all imaginable applications without a lot of trouble. That's why we use an adaptable plug-in scheme and the open source language Python.

The second kind of reason is that I would like to get some more experience in serious programming. I have never made GUI's, and I have never made software that was directly to be used by others. Another experience related interesting side of this project is that I get to work with some other people on a shared codebase using SVN, which is also new to me.

The first thing we need is a cool slogan and logo...

No comments

marius | 29-04-2008 11:22Variance over measurements

A long time ago I made a proof for a formula that can be used to calculate the variance of two subpopulations without knowing the individual scores. What is needed is the variance, the average scores and the number of scores of all subpopulations, without the actual scores. This was used to decrease the size of a database. PDF-file.

No comments