The Evolution of "Numerical Generations"

 Phase 1 analyses in both Generation1 and Generation 2 analyses found significant increased risk of neurological developmental delay due to mercury exposure
• significant elevated risks of 1.59 and 1.64
• significant trend (dose-response) with increased risk with rising exposure

However, crude calculations did show significant autism risk using the Generation 2 data
 

Low* exposures: <37.5 micrograms (3 months)

High exposures: >75 micrograms (3 months)

 Odds ratio

*Note: Low exposures defined as <37.5 mcg due to absence of reported data on zero exposure level

  The "Cox Model adjusted" is the proportional hazards logistic regression that forms the basis for all the CDC's risk calculations. It appears to be a standard epidemiological method, but is also a bit of a black box. Regression is complex, but somewhat intuitive; logit regressions are less intuitive and more complex, so the Cox model is a step further.  According to researcher Mark Blaxill, one cannot be certain how they've loaded up the model with variables (the infamous "strata") that reduce statistical significance (personal communication).

Note that in the above graph how neurodevelopmental delay was essentially dropped from the Generation 3 and 4 analysis.

Note again as the invalid Harvard Pilgrim HMO data was added in phase 2 of the 2nd Generation study, speech and language delay dose response was reduced to non-significant.