Fluoridation defender Ken Perrott is wrong again.
He does not understand epidemiological studies of neurotoxins.

Ken Perrott seems to think the results of the Bashash study should look like those in laboratory experiments where everything can be controlled, even to the extent of having genetically identical cloned animals.  Humans don’t work that way.  You can’t get a population of humans that all have the same IQ score and then vary just their fluoride exposure.

Ken is digging himself deeper and deeper into a hole by maintaining that the large degree of scatter in graphs of fluoride versus IQ in the Bashash study means its findings are of questionable validity.

He doesn’t seem to understand that IQ tests are designed to have scores with a relatively wide degree of variance.  They are commonly normalized so that the population mean is 100 and the standard deviation is 15 points. The standard deviation (SD) is a measure of the degree of variance.  Such IQ tests are also expected to have a roughly normal, or bell-shaped, distribution of scores about the mean.  For any group with a mean score of 100 and an SD of 15, there will be a few children on the extreme “tails” of the distribution with scores up around 130 or down around 70, and most of the rest will be somewhere in between.  That gives an expected overall range for most samples of children of about 60 IQ points.

Furthermore, numerous studies have shown that genetic variation explains most of the variation in IQ scores: more than 80% of the variance.  Epidemiological studies are not able to control for individual variation in genetics, so studies of neurotoxins and other environmental factors that affect IQ will always have a large degree of scatter.

To illustrate what is typical in developmental neurotoxin studies, in terms of degree of scatter of data, here are Neurotoxin study scattergrams of IQ versus lead and mercury.  These studies all found the neurotoxin to have a large and statistically significant effect. The studies were by respected researchers, published in high quality journals, and all concluded that they found clear evidence that the neurotoxin reduced IQ by the estimated amounts.

If Ken continues to argue that the wide scatter in the Bashash study invalidates the conclusion that it found clear evidence of an effect, then he’s going to have to challenge all these other studies of lead and mercury effects on IQ too.  We look forward to hearing back from Ken about fluoride after he has successfully debunked the studies that show that lead and mercury lower the IQ of children.  Then we can all rest easier because we will no longer have to worry about the neurotoxicity of lead and mercury, let alone fluoride.

Ken also needs to look up the difference between a Prediction Interval and a Confidence Interval.  He is incorrectly using Prediction Intervals to assess the confidence in, or the validity of, the findings of the Bashash study.  Confidence Intervals are the proper measures, as used by the authors of the paper.

Here is a concise explanation of the difference:

Prediction Interval is used for predicting the chance that a single observation (a single person in the case of the Bashash study) will have an outcome (IQ score) that falls within the range of the Prediction Interval.

Confidence Interval is used for assessing the probability that the entire population will have the relationship (the dose-response relationship) that falls within the Confidence Interval.

A Prediction Interval might be of interest to a clinician who was asked by their patient:  “Given my individual urine fluoride level, age, socio-economic status, smoking history, IQ, lead, mercury, and other factors, can you predict what the IQ of my child will be at age 4 years, and what is the “margin of error” for your prediction?”  Most clinicians would say this is an impossible question to answer, but if they were to do the calculations based on the Bashash results they would have to use the Prediction Interval to calculate the margin of error, and it would understandably be very wide.

In contrast, the Confidence Interval is what is used in all epidemiological studies to assess the confidence one has that the results of the study reliably predict what the true relationship is in the entire population.  The result of the Bashash study is the predicted dose-response relationship for the entire sample, which is about 6 IQ points lost per 1 mg/L increase in urine F for the Bashash study.  Note that this does represent “all the data”, despite Perrott’s repeated erroneous statements that it does not represent “the data as a whole”.  The multivariate regression analysis by which the dose-response relationship was calculated used all the data.