A quantitative investigation of types of statistical graphics and maps being used for presentation of geochemical data has been made for the 1984 volumes of Cleveland (1984) in a recent survey of usage in 57journals in natural, mathematical, and social sciences. Between 18% and 35% of 1463
graphs and maps in these journals contain at least one item of error or poor presentation, indicating the need for substantial improvement in both standards of preparation and refereeing of diagrams, and we suggest guidelines for both authors and referees.KEY WORDS: mathematical geology, statisics, statistical graphics, geochemistry.
INTRODUCTIONGraphic design as a means for communication of statistical information became established by the 1800s for a number of applications, mainly portraying timeseries. Beniger and Robyn (1978) and Tilling (1975) give accounts of these early developments. In geology, graphic data display began relatively late, despite early interest in application of quantitative methods by men of influence like Charles Lyell and John Playfair (Merriam, 1981 However, graphic methods for studying variation of rock composition (frequency distributions, bivariate scatter plots, and triangular diagrams of proportional contribution of three end-members) were established firmly in general use by 1928 with publication of Bowen's classic work The Evolution of the Igneous Rocks. As we shall show, these graphic techniques still remain the most widely used methods even today.Increasing interest over the last few years by the statistical community in the development of improved methods for graphic presentation and analysis of data is evident. This has led to publication of a number of books and articles on appropriate methodology for presentation of statistical information in the form of graphs (e.g. Chambers et al., 1983;Cleveland and McGill, 1984;Everitt, 1978; Feinberg, 1979;Fisher, 1983;Gnanadesikan, 1977;Graedel and McGill, 1982; Schmid, 1983;Tufte, 1983;Tukey and Tukey, 1981;Wainer, 1984;Wainer and Thissen, 1981) or maps (e.g., Bertin, 1980;Cleveland et al., 1982;Dickinson, 1973;Rhind et al., 1973;Wainer and Francolini, 1980). Interest has focused recently on how such information currently is used and presented, both to the general public in newpapers and magazines (Tufte, 1983;Wainer, 1984) and in scientific journals (Cleveland, 1984a).Graphics in the popular press, perhaps not surprisingly, are full of examples of omitted information and (often deliberate) graphical distortion (see Tufte, 1983, andWainer, 1984, for some appalling examples). Our often-complacent attitude that "as trained scientists we are bound to do it right" is shaken with Cleveland's (1984a) analysis of 377 graphs in 37 articles and 212 reports of volume 207 (1980) of Science that found 30% of these graphs to contain at least one error. Cleveland chose this particular joumal for detailed study as it contains papers covering a wide variety of scientific disciplines, and he conjectured that: "It is likely that Science does not have higher erro...
