What do you think about James Cantor’s research on pedophilia?

[Мурзик]

А можна сспрасить - чиво еето за фуита тут происходит?

 

[Ksenia]

It will be the complete mess after that -

Если Вы пережили насилие. Что с этим делать? - Преодоление-X

Что представляет собой сексуальное насилие?

overcoming-x.ru

I think all kinds of abuse create mess (emotional, physical), so sexual abuse isn't special.

 

[Зоя_Вексельштейн]

Hi!

Can you elaborare why ANOVA isn't appropriate statistical tool? I think it is used in other neuroscientific papers. In fact Cantor used 2 methods of analysis. The first one is Voxel Based Morphometry, and the second method is manual parcellation of brain structures and after that application of ANOVA. Regarding VBM with Family Wise Correction it is pretty standard tool in the neuroscience. Also Cantor wrote that paper with a lot of coauthors, a lot of members of the CAMH and the University of Toronto participated in that project.

In short, Jensen's inequality. Simple summary of Jensen's inequality: "a function of an average is not an average of a function, and the difference increases with disorder".

In a basic summary, - Cantor is generalizing from an average (whether it be white matter content or left-handedness) by applying a function to it (in his case mostly ANOVA, but he uses others too) in order to ponder and generalize on very complex emergent phenomena, in this case paraphilias.

This is using a mean-field technique (I doubt he knows what that even means) where it is NOT applicable at all. Mean-field techniques involve generalizing an average interaction and applying this generalization to the ensemble.

This fails in thinking about complex phenomena because distortions from mean field techniques will necessarily and always occur in the presence of nonlinearities. Mean-field techniques were initially used for simple ferromagnetic Ising lattices where there are no non-linear responses. Nothing like the brain.

Using his method (looking at averages of parts and applying functions to them to think about things) only is possible if and only if there is no dependence between one interaction and another with whatever we are looking at. Of course, nothing of relevance in the brain concerns a single part of it in isolation. So it follows that this type of averaging is not applicable if there are nonlinearities in the effect of the interactions. Of which in the brain there are many. More than we conceivably understand.

It is extremely easy to disprove his work with this alone, - we know hordes of people with very low white matter who do not have paraphilias that is indicative of this.

In fact Cantor used 2 methods of analysis. The first one is Voxel Based Morphometry, and the second method is manual parcellation of brain structures and after that application of ANOVA.

But the error is the same regardless of technique he is using.

 

[Ksenia]

In a basic summary, - Cantor is generalizing from an average (whether it be white matter content or left-handedness) by applying a function to it (in his case mostly ANOVA, but he uses others too) in order to ponder and generalize on very complex emergent phenomena, in this case paraphilias.

I think you refer to Jensen enquality as a metaphor, because in fact Cantor didn't apply function to the avarage mathematically.

This is using a mean-field technique (I doubt he knows what that even means) where it is NOT applicable at all. Local properties fail in complex systems because of dimensionality.

Yeah, it is true, Cantor didn't claim that varition of the volume of the white matter explains all variation of paraphilic attraction, correlation isn't 1.

Using his method (looking at averages of parts and applying functions to them to think about things) only is possible if and only if there is no dependence between one interaction and another. Well BIG SURPRISE - nothing of relevance in the brain concerns a single part of it in isolation. And, of course, this type of averaging is not applicable if there are nonlinearities in the effect of the interactions. Of which in the brain there are many.

There are a lot of nonlinear models that are used in the neuroscience. For example, support vetor machines and neuronal nets are used for guessing of sex of the person based on his\her MRI scan, but his paper isn't paper about prediction of paraphilias based on scan with nonlinear models, his paper is about exploration of some relationships between brain structure and pedophilia, so standard statistical tools are ok for such purpose. In fact Cantor said during his ATSA lecture that his work can't be used for clinical practice.

 

[Зоя_Вексельштейн]

I think you refer to Jensen enquality as a metaphor, because in fact Cantor didn't apply function to the avarage mathematically.


Yeah, it is true, Cantor didn't claim that varition of the volume of the white matter explains all variation of paraphilic attraction, correlation isn't 1.


There are a lot of nonlinear models that are used in the neuroscience. For example, support vetor machines and neuronal nets are used for guessing of sex of the person based on his\her MRI scan, but his paper isn't paper about prediction of paraphilias based on scan with nonlinear models, his paper is about clarification some relationships between brain structure and pedophilia, so standard statistical tools are ok for such purpose. In fact Cantor said during his ATSA lecture that his work can't be used for clinical practice.

"His paper is about clarification some relationships between brain structure and pedophilia. [...] In fact Cantor said during his ATSA lecture that his work can't be used for clinical practice."

Well not really. It is hardly a clarification as there is nothing in that paper that actually examines any relationship between brain structure and pedophilia. Nothing. Worse than nothing.

I too can find some relationship with 500 people (about the no. of people he used) with a super low p-value of some property in the brain and some highly complex emergent phenomena. I would only be fooled by randomness.

That's exactly what happened with the transporter gene 5-HTTLPR and depression. There were whole bodies of literature on 5-HTTLPR and depression that had super low p-values and replicated with thousands of people. Only now they're all discarded.

If we consider something "basic" (at least relatively to white matter volume or density), let's say polygenicity, - in generalizing about a high-order -effect, the standard now is to require AT LEAST 34,000 people in the sample (pg. 3 of that linked paper). The standard now in genetics is to use hundreds of thousands of people in thinking about polygenicity for higher-order effects like depression.

So his work on white matter, left-handedness and all sorts are not a clarifications at all, - but highly glorified false positives.

 

[Ksenia]

I too can find some relationship with 500 people (about the no. of people he used) with a super low p-value of some property in the brain and some highly complex emergent phenomena. It would all be lies though. I would only be fooled by randomness.

If such 500 people are the random sample of pedophiles and teleophiles then you can find some significant difference, but Cantor's sample isn't random, it is convenience sample (pedophiles were refered to CAMH by court order).

 

[Ksenia]

If we consider something "basic" (at least relatively to white matter volume or density), let's say polygenicity, - in generalizing about a high-order -effect, the standard now is to require AT LEAST 34,000 people in the sample

34 000 sample is robust sample for studying SNPs, DNA polymorphisms, there is another standard in the neuroscience. You need big samples in the DNA stuidies due to multiple correction.

 

[Зоя_Вексельштейн]

If such 500 people are the random sample of pedophiles and teleophiles than you can find some significant difference, but Cantor's sample isn't random, it is convient sample (pedophiles were refered to CAMH by court order).

Are you referencing this study - Cerebral white matter deficiencies in pedophilic men - PubMed ?

Because in that one the sample size is even worse at 127. And it is biased convenience sample too, that makes it worse (people with low white matter and no paraphilias are excluded, for example). Generally speaking with small sample sizes like that even if there were no violations in Jensen's inequality I would be on guard as p-values with small sample sizes are highly skewed and volatile. But there is a Jensen inequality violation regardless.

34 000 sample is robust sample for studying SNPs, DNA polymorphisms, there is another standard in the neuroscience.

It should be much, much, higher, no?

 

[Ksenia]

It should be much, much, higher, no?

No, it is smaller, becaue you do more statistical tests in the DNA studies, so you need very strict correction for multiple testing. There are less data in the MRI studies, so there are fewer statistical tests in the MRI studies. MRI studies are not based on the comparison of individual neurons, they are based on the comparison of voxesl or the whole bunches of voxels.