Saturday, August 25, 2007

Avoiding Messiness

According to M. Szlazak on the Skeptiko Forum, clinicians are now running away from meta-analyses in favor of single large-scale experiments to test and make a "final resolution" on questions of clinical effectiveness. He cites this paper and this open letter in support of his position.

Frankly, this does not surprise me at all. After all, the desire to have a simple worldview with black and white answers to every question is very tempting for human beings. And when different researchers conduct the same kinds of experiment, they often get contradictory results, or wildly different effect sizes, eg:

So it is not at all surprising that many people, especially a self-selected group of people who tend towards systematizing and simplifying reality, want to use methods that avoid raising difficult questions.

If the psi hypothesis is true, then the beliefs of experimenters can have a strong result on the results of their experiments, even when the experiments are conducted double-blind, most particularly when the overall experiment is statistical in nature and the experimental subjects are extremely complex and non-deterministic. Medical clinical trials certainly fall into this realm. Given the demonstrated reality that the beliefs of experimenters do effect the results of their experiments, it is certainly possible that psi correlations might help account for this (as opposed to experimenter fraud, biased errors, biased conclusions, etc. which materialists believe account for funding effects on experimental outcome).

Given the inevitable messiness of real science and the often varying results of experimental trials, it should be expected that some people want to avoid all those questions being raised by setting "rules" for science along the lines of what Topher Cooper suggested, tongue in cheek:

1) Don't reach a conclusion.
2) Completely ignore the data and just go with your faith.
3) Kill anyone who ever looks like they might ever do a second test of any hypothesis of interest.

But, of course, a committment to truth and reality demands that we face all of the facts, not just those that confirm our world-view. So long live multiple independent labs and experimenters. Long live meta-analysis!


Ryan said...


I think there is a mistake in the last paragraph:

>...reality demands that we fact all of the facts
What strikes me about the debate on the Skeptiko Forums, is that the tone has this underlying sense of being about more than whether the existence phenomenon X (or ψ) is verified by the data.

The conversation often – despite the points being made – has the feel of a religious debate. I think this is much more subtle in the Skeptiko Forums than elsewhere, but I definitely sense that kind of tension.
Whilst your tag line about '...casting doubt on the model of reductionistic materialism', is very useful in giving people a flavour of what this blog is all about, I don't know if that is where the debate is at, these days. If you think my points are just semantic pedantry then ignore them - but is anybody a materialist (as opposed to a physicalist) today? Strong reductionism, also, isn't widely accepted.

At Uni, quite early on, we were pretty much taught that these views are archaic in the light of contemporary physics and current explanatory frameworks. There was virtually no interest in ψ though.

Ersby said...

The graph you posted does not show wildly differing results.

If you take the data and put them in order of size, you will notice that it has a funnel shape converging towards a particular value (except one experiment). In other words, the larger the experiment, the closer the results are to a particular effect size. This is what you'd expect if the experiments were testing for a geniune effect - small experiments have a large deviation in results, large experiments don't.

Kennedy's point is that psi meta-analyses do not have this shape, and this is largely considered in pharmaeuticals to be a sign of methodoligcal bias.

M.C. said...


Do you know of any quantitative comparisons of psi statistics vs. pharmacological statistics?

Frankly I would not be greatly surprised if there were some substantial differences, because psi effects do seem very suceptible to experimentor effects (I would claim DAT / experimentor psi or lack thereof, and you might well claim sloppiness on the part of psi-positive researchers). However, I am not aware of any quantative analyses comparing the two, just verbal opinions from a parapsychologist who perhaps was not able to elicit psi phenomena in the lab and so left the field.

I also would not at all be surprised to see experimenter effects in the data (not just analyses) of medical trials based on study funding (in excess of file drawer considerations). However again I am not aware of any such studies.

Ersby said...


I don't have any qualitative comparisons. Do you?

Do you have anything to back up your suggestion that Kennedy's critique is due to his lack of success? He has a long career in parapsychology and is definitely in the pro-psi camp. I don't think it's wise to dismiss his words so lightly.