An additional test of non-parametric analysis is conducted to provide additional evidence
that the results provided from the risk ratio capture the sole effects of the compatibility
agreement only. The non-parametric analysis observes the same fatality rates and risk ratio, but
computes them separately for each make and model and then simply compares the number of
models that had lower fatality rates after self-certification to the number of models that had
higher rates. In other words, did significantly more than half the models improve (have lower
rates)? Advantages of this method are: (1) It “controls” for make and model; it is not influenced
by some models having mostly “pre” cases and others have mostly “post” cases. (2) Overall
findings are not overly influenced by one or two high-sales make-models with anomalous
results: all models have equal weight. The disadvantage from this analysis, though, is that it is
less likely to produce significant results, from the same number of cases, than the principal
method; thus, a non-significant finding is not necessarily a negative result, just a caution flag.