case control studies.

Proponents of coloured overlays and lens do not argue that they are treating dyslexia itself. Most acknowledge that dyslexia is primarily a language based disorder. Nevertheless, they argue that if visual stress is also present, it is an additional barrier to the acquisition of reading. The assumption is that visual stress reduces the motivation to practice, resulting in a widening gap between good and poor readers.
If this is the case, we might expect to find visual stress over-represented in the population with dyslexia. For example, a person with borderline phonological skills who might otherwise not become dyslexic could be 'tipped over the threshold' as a result of a putative visual processing disorder. We would also expect to find that reading impaired individuals who have the additional handicap of visual stress are more severely affected and resistant to conventional teaching methods.
The most commonly used methodology for assessing visual stress as a risk factor is the case control-study, in which a group of poor readers (the cases) and a matched group of average and above average readers (the controls) are screened for the variable - visual stress.

Like cars, case control studies that are not carefully set up
veer off course resulting in bias 

First, a note about bias in the statistical sense of the word. Although it may sound like an accusation that is not necessarily the case.   The analogy I prefer is to driving a car:  if the steering is not set up correctly or if the tyres are under-inflated on one side, the car may veer off, without any bias on the part of the driver. Case-control studies are like that. If they are not carefully set up they can veer one way or the other and give rise to misleading conclusions.

A failure to recognise bias in case-control studies has resulted in spurious associations (and masked true associations). For example, some years ago, a case-control study revealed a strong negative correlation between tuberculous lesions and evidence of cancer at autopsy. The authors argued that there was an 'incompatibility antagonism' between these two conditions and treatment of terminal cancer patients with tuberculin was enthusiastically started. Subsequently, it was discovered that the case control-study suffered from a bias in the control group that invalidated the results. Great care has to be taken in selecting both case and control groups to ensure that they are drawn from the same population. This example of bias resulted in an unhappy outcome for patients who received an ineffective treatment and one of the researchers who lost a prestigious appointment at Harvard. He was also forced to publish a retraction in the journal Science.
Studies performed in a clinical setting are particularly problematic, because the factors that bring patients to the clinical setting may be related to the disease factor of interest. This means that studies comparing controls recruited from a classroom setting and reading impaired individuals from dyslexia clubs or specialist clinics may be systematically biased because the latter group may have already been made aware he potential beneficial effects of coloured overlays. So called selection bias.
Another source of bias is the expectation of the screeners. Many of the tests for visual stress are highly subjective and dependent on the interaction between the screener and the subject and may be influenced by the expectations of both parties involved. In order to avoid so called ascertainment bias, screeners should ideally be masked to the reading status of the individuals being tested. While this may be less important for computer based screening methods they are still dependent on the interaction between subjects and the screener so it is unlikely that they are completely free of bias.
So what about case control studies to compare the prevalence of visual stress among poor readers and good readers? Well, there are a few studies out there and most are hampered by obvious sources and bias and small sample sizes. I have included them all in the 'forest plot' below. A feature common to each one is small sample size which means that the confidence intervals are very wide and could be consistent with visual stress being more common in good readers OR<1 or more common among poor readers OR>1. It would be a mistake to attempt meta-analysis of these studies which are heterogeneous, at different risks of bias and use different testing methodologies.
I will be discussing these studies in further posts. In the meantime, it is striking that the study with the lowest risk of bias found no difference in the prevalence of visual stress among average readers and poor readers OR 0.64 95% CI 0.17-2.38.

Forest plot of odds ratios of case-control studies examining the relationship between visual stress and dyslexia. <1 means visual stress more prevalent in controls. >1 mean VS more prevalent in dyslexics. In all cases, 95% confidence intervals are so wide as to allow both possibilities. The study at the lowest risk of bias, with screeners masked to the reading status of subjects as well as cases and controls drawn from the same population, has an odds ratio less than 1.

1.         Kruk R, Sumbler K, Willows D. Visual processing characteristics of children with Meares-Irlen syndrome. Ophthalmic Physiol Opt J Br Coll Ophthalmic Opt Optom. 2008 Jan;28(1):35–46.

2.         Kriss I, Evans BJW. The relationship between dyslexia and Meares-Irlen Syndrome. J Res Read. 2005 Aug;28(3):350–64.

3.         Singleton C, Henderson L-M. Computerized screening for visual stress in children with dyslexia. Dyslexia. 2007 May;13(2):130–51.

4.         Henderson LM, Tsogka N, Snowling MJ. Questioning the benefits that coloured overlays can have for reading in students with and without dyslexia. J Res Spec Educ Needs. 2013 Jan;13(1):57–65.