We introduce a method that eliminates the specification error and spurious relationships in regression. In addition, we introduce a test of strong causality. Furthermore, hypothesis testing (inference) becomes almost unneeded. Moreover, this method virtually resolves error problems such as heteroscedasticity, autocorrelation, non-stationarity and endogeneity.