Thomas Simpson and the arithmetic mean (Q1071744): Difference between revisions

Let the error of observation X about a ''true value'' to be measured have a triangular distribution on \([-e,e]\). Let \(\bar X\) be the sample mean of n observations on X, and \(X_ 1\) be distributed as X and independently of \(\bar X.\) T. H. Simpson (1757) in his Miscellaneous tracts... (London) is giving a statistical argument for using the arithmetic mean of several observations \((\bar X)\) rather than a single observation \((X_ 1)\), in astronomical measurements, derived on expression for Pr\(\{| \bar X| >| X_ 1|\}\). Simpson's approach was tantamount to putting \(Y=1-\{| X_ 1| /e\}\) and \(Z=1-\{| \bar X| /e\}\), so \(0\leq Y\), \(Z\leq 1\), and evaluating Pr\(\{Z<Y\}\), after finding an explicit expression for Pr\(\{Z\leq z\}\) for arbitrary n. The author points out Simpson's underlying use of continuous densities, their distribution functions, and a good understanding of the theorem of total probability leading to the expression Pr\(\{Z<Y\}= \int^{1}_{0}(\int^{y}_{0}f_ Z(z)dz)f_ Y(y)dy,\) where \(f_ Y(y)=2y\), \(0\leq y\leq 1\), all of which were advanced for their time. Simpson's contribution has been misunderstood hitherto.