Some exposition of the causal inference analysis employed in this paper is appropriate to clarify the techniques employed for the benefit of readers who might be less familiar with the application of modern statistical techniques to causal inference in direct refutation of the “causalophobic” reputation for which classical statistics is so well known [68]. The major short-coming of classical regression techniques in causal assignment has been the potential for unobserved variables to confound the analysis, and in particular the potential for the exposure of interest to be more common amongst the tested group than amongst the control group, a fallacy which results in a comparison of “apples with oranges”. Inverse probability weighting of an observational or ecological population can be applied to the regression model so that the exposure in the two groups becomes equivalent thereby constructing so-called “pseudo-randomized” equally exposed populations, from which causal conclusions can indeed be drawn. This is equally true for continuous variables as was used in the present study. This methodology is further strengthened by the use of robust regression techniques as was employed herein.

Another major way in which causality can be assessed is by the use of “Expected Values”, known as “E-Values” [69]. This is best illustrated by a classical illustration from recent medical history which has been eloquently recounted by one of the foremost causal inferential statisticians of our time Judea Pearl [70]. Prior to the 1950′s lung cancer was a very rare disease. However, with the rise of tobacco and then cigarette smoking lung cancer became common. It was realized before long that cigarette smoking posed a considerable exposure risk with a relative risk elevation amongst smokers of nine-fold. However, the debate was protracted and conflicted since many of the world’s leading scientists—including leading statisticians—smoked tobacco. The debate centred upon whether the observed effect might be confounded by unobserved variables—such as a pre-disposing genetic risk—which might confound the observed relationship. Eventually a genetic risk was indeed identified in 2016 with the identification of a gene which doubled the lung cancer risk. The gridlock was resolved when Robins showed that it was inconceivable that an unobserved confounder variable—which has to be related to both the exposure and the outcome—would be sufficiently powerful to account for all of the observed tobacco effect.

The demonstration in the present study then of many elevated minimum E-Values strongly suggests that the association between cannabis use and Down syndrome is likely to be causal. For example, the present results compares favourably with many modern reports which often have typical E-Values of about 1.25 [48]. However, since both inverse probability weighting and E-values have a number of underlying theoretical assumptions formal demonstration of the causal nature of the link must await further experimental studies. Given the strong evidence presented in the present report such studies should be pursued with high and urgent priority.

Moreover, it is further noted that the present work fulfills all the qualitative Hill criteria of causality as well [71]. Strength of association, consistency amongst studies [6,13,64,65], specificity, temporal sequence (see geotemporospatial analysis), coherence with known data, biological plausibility, dose–response relationships, analogy with similar situations elsewhere [6,13,64,65] and experimental confirmation are all very adequately fulfilled by the epidemiological analyses presented herein and published experimental work as summarized in the Section 4.2 above. These findings may be summarized in the following Table 8.