You are on page 1of 2

Method of Reduction-to-Major-Axis (RMA). Discussion and References.

Scott Long, Principal Geochemist AMEC Mining & Metals: Mining Consulting Group

Using the linear regression function available in most commercial software spreadsheets such as EXCEL will not provide an unbiased fit of the data between two sets of assay results. The reason for this is that the software assumes one variable is independent, and the other is dependent. For example, the independent variable might be a setting on an instrument, and the dependent variable might be a measurement of output. The underlying assumption is that the independent variable is precisely known while the output is not. This is not the case for pairs of assay results, where both sets of results are independent. One can approximate an unbiased fit by using the linear regression fit twice, assuming first one, and then the other result of the pair is the independent variable. This will give two lines with different trends that cross (after algebraically solving one of the linear equations so that both are in the form y = mx + b, i.e. one equation is restated as x = y/m - b/m). An estimate of the linear trend of the data can then be obtained by calculating the bisectrix (the line that falls halfway between) of the two linear regressions. However, the bisectrix is not the optimum fit of the data if the variance of one set of results is different than the other. In other words, the trend of the data should take into consideration that one set of the paired results may be more erratic than the other. The method of reduction to major axis calculates the slope of the trend as: m = [standard deviation of y] / [standard deviation of x] The intercept b is calculated by solving the equation y = mx + b, by substituting the mean of x and the mean of y into the equation. The error on the slope em is calculated as:

em = m * sqrt[(1-r2)/N]
The error on the intercept eb is calculated as:

eb = y* sqrt{[(1-r2)/N]*[2+( x/mean(x))2*(1+r)]
where is the standard deviation, N is the number of pairs, and r is the correlation coefficient.

References Agterberg, F.P., 1974, Geomathematics; Developments in Geomathematics 1, Elsevier Scientific Publ. Co., Amsterdam, 596p. Davis, J.C., 1986, Statistics and data analysis in geology (2nd ed.); John Wiley and Sons Inc., New York, 646 p. Sinclair, A.J., 1999, Evaluation of errors in paired analytical data by a linear model, Explor. Mining Geol. V 7, Nos . 1,2, pp 167-173.

You might also like