SUMMARY COMMENTS 223 OC8SIDIAN SIIDIES IN 1984 Fred Stross Twenty years have elapsed since the publication of the first papers showing that the sources of obsidian glass could be determined by physico-chenical methods. Since that time, the quality of the analytical determinations has i ved, leading to real confidence in the source assignments; provenience detezminations have been correlated with dating procedures, thus complementing arhaeological infonration, and criteria have been suggested that allow those interested to assess the reliability of the deterninations. Data banks repre- seating detailed compositions of obsidian sources are being established in many region and cost effective procedures are beimng developed which enable the user to reduce the cost of determinning provenience. Attenpts to use computer tech- niques to improve interpretation of results, or further to reduce the cost, continue to be made. Finally, other physical measurements have been developed which have contributed to the infonration desired, such as the measurement of obsidian hdyration to determine the time at which the obsidian was last wrked. My specific conts on the papers in this volume will be restricted primarily to those dealing with obsidian source analysis, since obsidian hydration analysis is covered in detail in the following paper by Meighan. Hughes' paper fittingly sets the tone by enphasizing that producing high quality data is mrre important than using sophisticated statistical, computerized methods for their intepretation. The statistics needed to make valid proven- ience assignmnts on the basis of high quality data are relatively simple; if the data are not sufficiently discriminating to make such assignments, this fact is normally quite obvious without the use of canplex techniques. I think it was very worthwhile for Hues to have gone to the trouble of pointing out in detail how treacherous it can be for the non-expert in statistics to use advanced methods and how, in the few amrbiguous cases, misratch by computer is at least as likely as mitch on the basis of simpler calculations. Nelson's article reflects careful work, and his data generally compare very well with data obtained by other workers on some of the same sources. They provide a useful addition to the data bank on Great Basin obsidian already avail- able in the literature. However, given the reservations expressed by Nelson on page 29 of his paper, I am unsure why the author used discriminant analysis to distinguish between sources. In my estimation, computers are useful in obsidian analysis for cataloging data and for performing routine analytical calculating and evaluating functions, but in the interpretive techniques that have become so fashionable in this type of study, the pitfalls for many practitioners are far more significant than the benefits they are likely to gain, as noted previously by Hughes. Hampel's paper is much to the point, concisely and competently discussing some problem in x-ray fluorescence analysis that are most commonly ignored i obsidian analytical studies. Bettinger, Delacorte and Jackson explore the utility of visual sourcing. They make the point that in a limited geographic area, some obsidian sources can 224 be recognized by visual inspection, aided by low power magnification. Given their success, one might consider recomending visual inspection as a first step, rough classification in sorting out a group of otherwise unsorted artifacts before subjecting them to formal geochemical analysis. Discussion Source data. It seems obvious (though not generally appreciated) that it is highly desirable to present data on source composition in absolute concentra- tion units (i.e. parts per million or weight percent) and to report the abundances of the largest possible number of independently variable elements to allow users of different analytical methods some latitude in the use of the data bank. Well defined error limits should be recorded for the results, so that other workers may estimate the reliability of the data. Source composition data require much more complete, accurate and sensitive infornmtion than artifactual data, since the validity of the artifact-to-source assignments rests upon the reliability of the source composition data. Another advantage of high quality source infonmation is the reduction in cost that may be achieved if detailed source data are available. It often is possible to reduce the number of elements needed for source assignment by an ability to focus on diagnostic elements, and to use more economical methods for determining them than used for the mrxre complete source analysis. Generating high quality source data makes it possible to match analytical results on exca- vated artifacts against published source composition data, thus avoiding the cost and duplication of effort of determining source compositions in an area more than once. Cbmputer analysis. Since elemnt composition data provide distinguishable patterns, such arrays of data seem eminently suitable for pattern recognition applications by computers. Although a host of papers have appeared which describe computer use for this purpose, the results of these projects have at times been disappointing. The first classification of artifact data often involves only a few elements, so that the greatest potential of computing techniques in this connection is not utilized. In fact, the labor of setting up and using the often complex programs may be greater than judicious use of a hand calculator. Worst of all, many workers are tempted to regard statistical computer methods as a substitute for careful analytical work. As Hughes has emphasized, the application of pattern recognition capabilities of the computer to inadequate analytical work may lead not only to incorrect results, but it also is difficult to recog- nize this defect because tha apparent rigor of the procedure engenders a false sense of security in both the unwary analyst and the reader. Obviously, the ccaputer has its place in this work for many of the calculation procedures, but its benefits in pattern recognition applications in artifact sourcing have been overestimated; it is very pertinent that sane of these shortcomings have been pointed out in this volume.