Fitting straight lines when both variables are subject to error

摘要: Abstract If the usual method of least squares is used to fit a straight line observed points when both independent variable and dependent are subject error, true slope will tend be underestimated. We undertook computer-simulation (“Monte Carlo”) study previously published methods which have been designed avoid this bias. None were wholly satisfactory. therefore developed tested large number entirely new methods, each represents some kind compromise between “the regression y on x” (which underestimates slope) x y” overestimates slope). Our computer shows that while no single method, old or new, “best” under all circumstances, judicious choice can increase accuracy decrease bias with estimated. Preliminary rules for selecting most suitable given, several lines further investigation proposed.