Abstract |
Initial study of an analytical method for determining the amount of some substance frequently leads to a calibration curve on which the response of a test indicator are plotted against carefully measured amounts of the substance. In practice, however, such a curve is used to estimate an unknown amount of substance from an observed value of the indicator variable - exactly converse to the way in which it is prepared. Berkson has shown that this estimation is valid when the relationship between substance (x) and indicator (y) is linear, provided that in obtaining the calibratiin line an attempt is made to measure x to certain predetermined amounts. The estimation of x from y is not invalidated by the fact that the original measurement of x may have been subject to error. The purpose of the note is to call attention to confidence limits available for such an estimate. (Author) |