| Abstract: | In this study the author examines the general problem of estimating the influence of social conditions on mortality rates and suggests that the methods of multiple regression analysis extensively used for this purpose cannot always be relied on to give the kind of results required. The argument is illustrated by reference to the work of Woolf and Waterhouse (J. Hyg., Camb., vol. XLIV, no. 2, 1945) on infant mortality in the county boroughs. The Confluence Analysis of Frisch, applied to their results suggests that the statistical series used by these authors are multicollinear, and use of Hotelling's method of Principal Components suggests that these five series represent not five but only two fundamental influences. One of these, almost equally strongly loaded in three of these series is identified with the ‘general social conditions'— a complex of bad housing and low money-incomes—which these and other writers have studied. This leads to the suggestion that the use of multiple regression analysis in attempts to estimate the separate influences of the individual constituents of this complex are likely to be unprofitable and misleading. It is suggested that the available information relating to social conditions in various districts can be most usefully employed in the form of a complex index which might be constructed by means of some form of Factor Analysis. |
