Is comfirmation bias essential to anti-fluoride “research?”

By Ken Perrott 13/04/2015

Anti-fluoride propagandists like Declan Waugh and Paul Connett avidly scan the scientific literature looking for anything they can present as evidence for harmful effects of community water fluoridation (CWF). Sometimes they will even do their own “research”  using published and on-line health data looking for any correlations with CWF, or even just with fluoride levels in drinking water.

Several years ago an activist going under the nom de plume “Fugio” posted images showing correlations of mental retardation, adult tooth loss and ADHD with the incidence of CWF in the US. These images are simply the result of “research” driven by confirmation bias and data dredging.They prove nothing. Correlation is not proof of a cause. And no effort was made to see if other factors could give better correlations.

I go through Fugio’s examples below – partly because I noticed one of their images surfacing recently on an anti-fluoridation Facebook page as “proof” that CWF causes tooth loss. But also because they are just more examples of the type of limited exploratory analysis used in two recently published papers – Peckham et al., (2015) (discussed in my article Paper claiming water fluoridation linked to hypothyroidism slammed by experts) and Malin and Till (2015) (discussed in my articles More poor-quality research promoted by anti-fluoride activistsADHD linked to elevation not fluoridation and Poor peer-review – a case study).


This figure is essentially the same as that reported by Malin & Till (2015). In fact, I wonder if Fugio (who posted December 2012) is the unattributed source of Malin & Till’s hypothesis. Fugio chose the ADHD data for 2007 and fluoridation data for 2006 whereas Malin and Till (2015) concentrated mainly on fluoridation data for 1992 which had the highest correlation with ADHD figures.

I won’t discuss this further here – my earlier article ADHD linked to elevation not fluoridation shows there are a number of other factors which correlate with ADHD prevalence just as well or better than CWF incidence does and should have at least been considered as confounding if not the main factors. I found a model using mean elevation, home ownership and poverty only (no CWF included) explained about 48% of the variation whereas their model using CWF and mean income explained only 22-31% of the variation. And when these confounder factors were considered the correlation of ADHD with CWF was not statistically significant.

In other words we could do a far better job of predicting ADHD prevalence without involving CWF.

Water Fluoridation and Adult Tooth Loss

Fugio posted a figure showing a correlation of adult tooth loss with CWF incidence in 2008. It was statistically significant explaining 11% of the variation. But quite a few other factors display better correlations with adult tooth loss. For example, the data for smoking by itself explains 66% of the variation (see figures below).


Checking out correlations with a range of factors I found a model involving only smoking and longitude  explaining  about 74% of the variation. The contribution from CWF was not significant statistically – it added nothing to this model.

Water Fluoridation and Mental Retardation

Fugio found a better relationship between CWF in 1992 and mental retardation in 1993 – a correlation explaining 19% of the variation. Apparently the concept of “mental retardation” was later abandoned as there do not appear to be any more recent statistics.

But again, if Fugio had not stopped there he/she would have found a number of other factors with better correlations. I give an example in the figure where state educational level (% Bachelors Degree in 1993) explained 50% if the variation. This correlation is negative as we might expect.


 Again I used multiple regression analysis to derive a model involving educational level (% with Bachelors degree in 1993), poverty in 1993 and mean state elevation which explained 69% of the variation. No statistically significant contribution from CWF occurred.


I am not suggesting here that the factors I identified have a causal effect. Simply that they give better correlations  than CWF. These and similar confounding factors should have been considered by Fugio and Malin and Till (2015).

My purpose is to show that this sort of exploratory analysis of easily available data can easily produce results for anti-fluoride activists who are searching for some “sciency” looking arguments to back up  their position. Provided they don’t look too deeply, stop while they are ahead and refuse to consider the influence of other factors.

Unfortunately poor peer review by some journals is allowing publication of work that is no better than this. Peckham et al (2015) did nothing to check out other factors except gender in their correlations of hypothyroidism with CWF. The glaring omission was of course dietary iodine which is known to have a causative link with hypothyroidism. (I could not find US data for hypothyroidism so was unable to check out Peckham et al’s hypothesis for the US.) Malin and Till (2015) included only socioeconomic status (as indicated by income) in their analysis despite the fact that ADHD is known to be related to a number of factors like smoking and alcohol intake.

As I keep saying, when it comes to understanding the scientific literature it really is a matter of “reader beware.” It’s easy to find papers supporting one’s pet obsession if you are not critical and sensible with your literature searches. And it is important not to take at face value the claims of activists who clearly rely on confirmation bias when they explore the literature.