Nick Brown looks at study on sexual harassment and 'CFS'

Amw66 said:
This feeds into the ACE agenda that is percolating down into school/ social work/ police.
The Bath safeguarding seminar will no doubt build on this too.
Cue CBT sessions trawling for past memories and creating potential for future PTSD
Click to expand...
Welcome back in the 70s, it seems. I thought we had overcome the "childhood trauma model" as an explanation for any problems.
 
What's the difference of sexual harassment and sexual abuse of a child? Wouldn't the child feel it as abusive either way?

And if a parent makes compliments about the look, is that sexual harassment? Isn't it more important what the child feels like while it's happening?
 
I don't have the brain tthink about about this at present, and it's too way too technical for me anyway, but just had one thought about Hip's virus post:
Hip said:
Well, if the sexual harassment involved kissing, this is a means to transmit respiratory viruses, and nearly all the viruses linked to ME/CFS are respiratory viruses (can be transmitted via the respiratory route).
Click to expand...

Being close enough to breathe air freshly expelled by an infected person can be enough to be at risk of resp. viruses carried in saliva aerosols and a kiss isn't necessary, surely? I don't know how far that applies to viruses commonly associated with developing ME.

I'm thinking being close would often be a part of a traumatic experience - whose to say what's what. (Even if I thought the theory in the paper was any good which it's not)
 
Last edited:
If they are looking at childhood trauma as a cause for disease, there is no way that harassment could lead to disease when actual abuse does not. It is the psychotherapy thing of looking for the "trauma". It will always be there to be found as most teenagers get sexual comments from their peers, male and female.

This is completely different from being raped every night for years or being left to rummage for food in empty cupboards because of alcoholic parents. If that sort of survivor does not get ill, then why should anyone less?

The only answer would be a weakness among some folk who succumb to CFS so we are back to it being a disease of self indulgence. Now that matches the "it hurts so I won't try it" theory of deconditioning by the BPSers so this is just another of their stupid papers with unacceptable assumptions about us.
 
Hi - I'm the author of the blog post in question here. (I'm not sure how I prove that, but JohnTheJack believes me, as does my Twitter account with the same username, ).

I'm busy for most of today and just managed to get a couple of minutes to write this post on my phone. I will try and post about the TEC thing when I get back to my laptop, either today or tomorrow. I'm also happy to answer any questions you might have about the post.

Cheers,
Nick
 
Skycloud said:
Being close enough to breathe air freshly expelled by an infected person can be enough to be at risk of resp. viruses carried in saliva aerosols and a kiss isn't necessary, surely? I don't know how far that applies to viruses commonly associated with developing ME.
Click to expand...

It depends a lot on the particular virus, and also the phase of the infection. During the acute phase of a highly contagious viral infection like say influenza A or norovirus, just ordinary social contact is usually enough to transmit the virus, as global influenza pandemics illustrate.

However, for viral infections that persist in the body (such as herpesviruses and coxsackievirus B), once the acute phase is over and the virus goes into latency, the propensity to transmission is much lower, as there are far fewer viral particles in the saliva.

In the case of Epstein-Barr virus, even though 90% of the adult population have this virus in their body (usually in a latent state), most children and younger teenagers are not infected with EBV (in the developed world at least), despite the fact they are surrounded by adults who are EBV-infected. It's not until the late teens or early 20s that EBV is typically first picked up, sometimes by kissing a girlfriend or boyfriend (hence the name "kissing disease" for glandular fever / mononucleosis).


Some years ago when I first read the studies finding a higher prevalence of child sexual abuse among ME/CFS patients, my first thought was that this might be explained by the transmission of viruses to the poor abused child.

That's not to rule out the possibility that childhood physiological trauma may permanently affect things like cortisol levels and immunity (this study found differences in cortisol in adults abused as a child). But the way I see it, in these studies which try to examine the long-term health effects of such psychological trauma, you cannot control for the possibility that child sexual abuse might also involve the transmission of infections to a young child, as well as cause a psychological trauma.
 
Last edited:
Reading the blog very quickly I think the regression explanation very interesting but I wonder if there is a more fundamental flaw with the quality of the data.

I had thought there were serious concerns about the accuracy of retrospective surveys on things like abuse. I think I've read that the data is just really unreliable (I suspect @Woolie could say more). I think it was something about a tendency for people who were struggling with illness or emotional issues tend to think back for possible emotional trauma more than others.

From the blog:

The authors define sexual harassment as "being submitted to sexual acts without physical contact" and sexual abuse as "having undergone sexual acts involving physical contact". I'm not remotely qualified in these matters, but it seems to me that with these definitions, "sexual abuse" would probably be expected to lead to more problems in later functioning than "sexual harassment". Indeed, I find it difficult to imagine how one could be subjected to "abuse with contact" while not also being subjected to "abuse without contact", more or less by the nature of "sexual acts". (I apologise if this whole topic makes you feel uneasy. It certainly makes me feel that way.)
Click to expand...

I wonder if it is quite easy to remember occasions of sexual harassment (because there is a lot of it about) where as its harder to come up with sexual abuse. This brings a potential issue with their model which is I don't see it being predictive because a lot of healthy people have also suffered such issues and is that reflected in the data? So is there model saying that given someone has a diagnosis of ME/CFS then this model will predict the likely function but again this feels wrong because ME is a variable condition where there can be quite big swings in function at different times.

I guess I'm saying I don't understand how their model fits onto the real world rather than a small snapshot of data. Having gone back to ML recently reminds me how easy it is to find non-existent effects with bad analysis or biased data.
 
OK, some notes on the TEC measure:

- The description of the TEC in the de Venter et al. article is somewhat different from that in the Nijenhuis et al. article, which de Venter et al. cite when introducing the measure
- The Nijenhuis et al. article's description doesn't correspond, in several ways, with the questionnaire, which you can download from http://www.enijenhuis.nl/tec/.
- However, it seems that there is a "secret sauce" to the scoring. On the page that I just linked to in the previous point, there is a button marked "Scoring & Interpretation", which gives you a ZIP file. In there you'll find a file called "TEC manual scoring form.pdf", which shows how to take the raw numbers that the user supplies and give the 0-4 (times three) scoring that the articles describe.

This is all rather unsatisfactory, but I don't think it's the biggest problem. Far more important is that 126/155 people answered 0 for sexual harassment and 141/155 answered 0 for sexual abuse; plus, almost everyone who had a non-zero score for abuse will have had a non-zero score for harassment. These two predictors are likely to be highly correlated (I'm guessing .7), which means that they are kind of measuring the same thing. And when you throw both of them into a regression, one or the other will stick out, which is likely what gave the above-one regression coefficients.
 
sTeamTraen said:
These two predictors are likely to be highly correlated (I'm guessing .7), which means that they are kind of measuring the same thing. And when you throw both of them into a regression, one or the other will stick out, which is likely what gave the above-one regression coefficients.
Click to expand...


I thought the predictor variables should be independent?

If I develop a model from data I would expect to have test and verification data sets to check the model that is learned so that gives a better idea that the model isn't just a feature of the particular training vectors that are used. With regression in this paper this idea seems to be ignored (and in other papers more generally). So the regression model basically says the model fits that given data set but its not too surprising that they can get some sort of fit. What surprises me is the assumption that any fit on a data set says anything about the relationships themselves?
 
Adrian said:
I thought the predictor variables should be independent?
Click to expand...
Indeed. Of course, predictors are never independent (r = .0000000), and there is also what Lykken and Meehl famously called the "crud factor", aka "everything is correlated with everything else", at least to some degree. But when intercorrelations are less than .15-.20 or so that needn't cause a problem. And there are some situations where fairly strongly correlated predictors can have distinct meanings. But you have to argue very carefully for those, otherwise the most parsimonious default assumption is that it's an artefact.

Adrian said:
If I develop a model from data I would expect to have test and verification data sets to check the model that is learned so that gives a better idea that the model isn't just a feature of the particular training vectors that are used. With regression in this paper this idea seems to be ignored (and in other papers more generally). So the regression model basically says the model fits that given data set but its not too surprising that they can get some sort of fit. What surprises me is the assumption that any fit on a data set says anything about the relationships themselves?
Click to expand...
That's certainly a danger, and researchers who were trained 20-30 years ago do not always appreciate it. But this model is so crazy (at least on paper) that we don't even need to get to that point.

Note that the authors only provide the individual regression coefficients, not the overall fraction of variance explained by the regression (which would be expressed as R^2). This is a very basic deficiency of the reporting and should have been picked up by the reviewers. I'm guessing that R^2 is tiny, simply because so many people answered 0 on some or all of the measures.

The fact that the regression coefficients for sexual abuse (.83 and -.53) have such high p values (.20 and .55, respectively), despite the fairly decent sample size, is also a cause for concern. With n=155, zero-order correlations of .83 and -.53 would have given p values below .00001. So the standard errors are hugely inflated, which is a classic consequence of highly correlated predictors.
 
sTeamTraen said:
This is all rather unsatisfactory, but I don't think it's the biggest problem. Far more important is that 126/155 people answered 0 for sexual harassment and 141/155 answered 0 for sexual abuse; plus, almost everyone who had a non-zero score for abuse will have had a non-zero score for harassment. These two predictors are likely to be highly correlated (I'm guessing .7), which means that they are kind of measuring the same thing. And when you throw both of them into a regression, one or the other will stick out, which is likely what gave the above-one regression coefficients.
Click to expand...

Welcome :)

Don't have the brains to go into the finer details (cognitive symptoms) - but this bit stands out to me. That can't be right, would be only 18.7% ?

How is childhood defined? If it's up to 18 year olds, that number should be much higher, as sexual harassment unfortunally is very common. This survey finds 43 % of middle school students?
https://www.news-medical.net/news/2...nt-among-adolescent-students-study-finds.aspx

I would suspect they have a problem of under-reporting?
 
inox said:
Don't have the brains to go into the finer details (cognitive symptoms) - but this bit stands out to me. That can't be right, would be only 18.7% ?

How is childhood defined? If it's up to 18 year olds, that number should be much higher, as sexual harassment unfortunally is very common. This survey finds 43 % of middle school students?
https://www.news-medical.net/news/2...nt-among-adolescent-students-study-finds.aspx

I would suspect they have a problem of under-reporting?
Click to expand...

I think this may be a question of language --- the authors of the TEC (from the Netherlands) and of the de Venter et al. study (2017) are all, I think, native speakers of Dutch, and maybe "harassment" is the closest English word they found to whatever they had in mind in Dutch. The TEC definition of sexual harassment ("being submitted to sexual acts without physical contact") seems to me to be more restricted than what we might consider those terms to mean in English. To me, the TEC definition means stuff like having someone expose themselves to you, whereas if you think about how we define sexual harassment in (say) the workplace, that would include for example sexualised talk without anyone having to unzip anything.
 
Last edited:
sTeamTraen said:
Indeed. Of course, predictors are never independent (r = .0000000), and there is also what Lykken and Meehl famously called the "crud factor", aka "everything is correlated with everything else", at least to some degree.
Click to expand...

I guess the point in my mind was around structural reasons to believe variables may not be independent. This seemed to be what you were saying around variables of sexual harassment and abuse - I assume because harassment is often a precursor to abuse. So in these cases it seems like those producing a model should be aware of the potential issues and think through the structure of the model and the problem they are trying to solve. I think its one of the reasons its useful to have both stats (or ML in my world) experts working with those who really understand the problem space so that care can be taken over things like the meaning of variables and the likely underlying structural relationships to other variables.

Welcome to the forum as well.
 
Adrian said:
I guess the point in my mind was around structural reasons to believe variables may not be independent. This seemed to be what you were saying around variables of sexual harassment and abuse - I assume because harassment is often a precursor to abuse. So in these cases it seems like those producing a model should be aware of the potential issues and think through the structure of the model and the problem they are trying to solve. I think its one of the reasons its useful to have both stats (or ML in my world) experts working with those who really understand the problem space so that care can be taken over things like the meaning of variables and the likely underlying structural relationships to other variables.
Click to expand...
Oh, absolutely. It's not clear to me from the limited amount I've read whether the TEC is really a suitable measure to be using in a survey (as opposed to individual clinical) context. It might be of interest to a clinician with a patient to know that harassment took place but abuse didn't, or whatever, but to use that relation across a larger sample to draw inferences about a population could be a problem. The Nijenhuis et al. article from 2002 that looked at the properties of the TEC didn't report the correlations among the subscales, perhaps partly because it seems to have been (re-)scored differently from how de Venter et al. did it.
 
Slightly OT general questions for @sTeamTraen - I expect the answers will be along the lines of 'there's no easy answer.'

How do you decide what to do now with studies like these? Their results seem implausible in a couple of ways but the researchers are resistant to independent scrutiny of their data. It's a relatively minor paper that, while it could still have influence, is unlikely to attract the sort of sustained interest that was needed to get the release of PACE trial data. Is it worth writing a letter to the journal? Making a comment on pubpeer? Is there any way of encouraging greater scrutiny of these researcher's work in the future (other than people just making a mental note)? Is there some hope blogging about it encourages others to take a look?

It seems like there are loads of duff papers around that would take huge effort to get meaningful corrections to, yet leaving them in the literature can seriously distort the
scientific consensus on important issues.

I've been amazed by how bad the systems of science are at encouraging (requiring) debate. I guess that I was naive to expect anything else, as it seems that the people at the top of most systems have skeletons in their closet (getting to the top often involves being a bit of a shit) and want to avoid making it easy for others to challenge them.
 
You must log in or register to reply here.
Back
Top Bottom