Antibiotics and Sulphonamides in Pædiatrics

(Pp. 52. 9s.) Hong Kong UnivCrsity Press. London: Oxford University Press, 19%2. THIS book is an introduction to the use of statistical methods in the analysis of snmall samples of data. The author describes, with simple numerical examples, the arithmetic steps in computing averages, measures of dispersion, corielation coefficients, including partial coefficients vith one variable held constant, and regression equations. He shows how to obtain the standard errors of most of the relevant estimates and how to estimate the probability of observed differences between means of samples arising by chance. Comparison of two qualitatively classified distributions using the chi-squared test is the only example given of dealing with qualitative data. Tables of the "t" and chi-squared distributions and of the squares and square roots of integers from 1 to 1,000 are appended. The chapter on frequency distributions, particularly the part dealing with the binomial and Poisson distributions, is too brief to be of value, and the implication that all continuous distributions are "normal" is misleading. The formula given in chapter III (page 11) for the standard deviatioil shows Vni (where n is the number of observations) in the denominator; it is customary, particularly when n is small, to use {/(n-1). If ^fj(n-1) is preferred, theni the alterinative formula on page 12 and the formula for the variance on page 13 will require alteration. In chapter IV it is wrongly suggested that the formula for the standard error of the mean should be modified, when n is less than 30, by dividing the standard deviation by v'(n-1) instead of Vn. In chapter V, where comparisons are made between the means of samples, the differences, involved by the use of %fn rather than Vf (n-1) for the standard deviation and %/-(n-1) rather than V-In for the standard error of the mean, cancel each other; these departures from customary practice could be misleading in other contexts. In the chapter on regression further use is made of standard errors of means to compute the regression coefficient; for no stated reason the standard deviations in this chapter are divided by %' (n-2) which is clearly inconsistent with earlier expressions. On page 38 the formulae given for the standard error of estimate are incorrect. The second formula gives the standard error of the regression coefficient and is, in fact, used as such later in the chapter. The author's decision to exclude the mathematical derivation of well-known formulae was reasonable in such an elementary text, but rather more might have been said about the valid use of the methods. This might have been achieved by including a short chapter on the design of experiments, particularly in view of the author's hope, which many will share, that the use of the methods described will discourage the collection of unnecessarily large numbers of data. E. A. C.