sign in sign up
<<
Home , Social statistics

180

SOCIAL , SOCIAL SCIENCE AND SOCIAL ENGINEERING

Philip M. Hauser, University of Chicago

The present status and future prospects of also be argued that "explanation" based on sta- social statistics in relation to social science tistical research is also to be preferred to and social engineering is the subject of this alternative forms of explanation in social essay. This subject has been selected for two science. Social science explanation based on reasons. First, it touches upon the roles of social statistics produces "probabilistic" ex- social statistics that engaged Sam Stouffer's planation of a type not as rigorous as the mind, heart and much of his prodigious energy; "deterministic" explanation possible in branches and second, it represents an effort to minimize of the natural sciences. But it constitutes a overlap with what I anticipated would be compre- more rigorous and objective explanation than hensive and excellent papers by Fred Stephan and other types of explanation possible in the social Nat Reyfitz focused more directly upon social sciences, which in the sense discussed by Nagel, statistics per se and on Sam Stouffer's signif- include the "functional," "teleological" and icant contributions to their development. "genetic" forms of explanationN The framework for my discourse was well set The Experimental Method. The most efficient forth over fifty years ago by Richmond Mayo -Smith, and rigorous method for obtaining knowledge that Professor of Political Economy and Social Science science has yet devised is that represented by at Columbia College.1/ After some preliminary the controlled - -a method that neces- discussion he observed: sarily involves measurement.5' That is, to "We get from statistics indications of rela- detect the impact of an experimental variable on tions which maintain themselves with a persis- a dependent variable "before" and "after" meas- tence and constancy that give us an impressive urements are necessarily involved; as is also the sense of the reign of law in the social actions measurement of the experimental variable itself. of men. It is this reign of law which we desire Even though the use of the controlled experiment to see revealed." (p. 15) has relatively limited application in social Mayo -Smith also observed that "We are sur- science, its use is by no inconsequential. rounded by sociological or social problems On the contrary the experimental method employed which urgently demand solution. We cannot wait in social researches has resulted in important for the completed science; we must seek to under- gains in knowledge --as, for example, in small stand the conditions affecting the particular group research.ó/ problem before us. This may be called practical The use of the experimental method even . Everywhere in this domain we find though limited, then, is the first of the ways in statistics a useful instrument of investiga- which we may note social statistics are involved tion. . . ." (p. 16) in social science. In the employment of the The first of these quotations serves admir- experimental method the social variables utilized ably as a concise summary of the role of social in the experiment are quantified and change is statistics in social science; and the second, measured. Moreover, in the conduct of the ex- their role in social engineering. As a pre- periment not only are the various scales of meas- liminary to consideration of social statistics urement and employed but, in relation to social science and social engi- also, the techniques developed by neering it is well to define the term as used in for the design of experiment and statistical in- this paper. "Social statistics" is considered ference. Although the use of the experiment in here as embracing not only the figures which flow social science is limited, the statistical from tabulations of , survey, registration, methods employed in conjunction with the experi- administrative records, and the like but, also, ment may well be viewed as a point where statis- the statistical methodology which is employed in tics on its most sophisticated level contributes social research and the substantive findings of to social science. And the knowledge which flows social research stated in quantified form. from experimental studies in the form of predict- ability and explanatory propositions may appro- SOCIAL STATISTICS AND SOCIAL SCIENCE priately be considered a part, even though as yet a limited part, of the fund of "social statis- Social science like any science has as its tics" as well as, simultaneously, of social objective the achievement of knowledge independ- science. ent of the "caprice" and "willfullness" of man, Statistical Method. Most social science in contrast with knowledge based on the methods research necessarily utilizes methods other than of "tenacity," "authority" and "intuition" as the controlled experiment. These methods best discussed by Morris R. Cohen and Ernest Nagel. achieve the objectives of science, however, to Such knowledge, which can more explicitly be the extent that they approximate the experimental stated as consisting of and ex- method. Thus, the method of the controlled ex- planation, can be achieved only by means of the periment has great indirect utility in social method of science, the method of "reflective science in serving as a yardstick by which all inquiry"3 or research. Statistical methods other methods may be evaluated and their strengths wherever applicable provide more efficient and or weaknesses assessed. Stouffer in his clas effective prediction than alternative social article, "Some Observations on Study Design," science methods. The fact that a quantified pro- has provided an excellent demonstration of the jection or prediction is to be preferred to a manner in which the experiment can serve as a non -quantified one needs no elaboration. It may framework for assaying other methods. 181

The research methods which come closest to to measure many of the phenomena we wish to the controlled experiment in the study of social treat as dependent or independent variables. Our phenomena are the methods of statistics. Math- conceptual frameworks are in a number of areas ematical statistics has produced a wide spectrum ahead of our ability to obtain adequate metrics. of specific techniques for the study of social As a result we are often able to explain only phenomena which in varying degree approximate pitifully small proportions of the total vari- or simulate the experimental method. It may be ances of our dependent variables, not neces- argued that the research methods involved are sarily because the hypothesized relationships are mathematical statistical methods not "social not there but, rather, because we work with such statistics." But generic statistical methods confounded variables that we cannot detect the utilized in social research may appropriately relationships. There is a great need to concen- in this context be considered as included in trate on obtaining better measurement of our the methodological component of "social sta- variables - -to get measurements that more clearly tistics," and the product of research based on match our concepts, as Stephan noted two years these methods is appropriately included as part ago in a paper before this section.&/ of the substantive body of "social statistics." A good example in point is to be found in This is, after all, usage analagous to that of our classification of urban and rural population. the generic term "statistics" itself, which Urbanization has had profound effects on man's possesses both substantive and methodological way of life, and many efforts have been made to connotations. trace its impact by means of the census type of Social science research is becoming increas- measurement of "urban" as an independent vari- ingly quantitative either by means of statis- able. But "urban" as measured in a census tical methods or direct mathematical approaches.) includes persons of third or subsequent urban It is possible, therefore, to say that social descent together with newly arrived in-migrants science is becoming science, in contrast to its from generations of living in the Appalachian 19th and early 20th century philosophical state, highlands or the Mississippi delta. It includes largely by reason of the impact of the statis- persons who have been relatively segregated in tical method. The application of statistical ethnic enclaves within the urban place as well method to social phenomena and the body of pro- as persons of similar origin who have become positions and theory emanating therefrom are completely assimilated. It includes residents playing a major role in the advancement, or of the central city and of the suburb; of persons perhaps more accurately the achievement, of in, as well as outside metropolitan areas and so social science. on. There can be no doubt that the classifica- It must be acknowledged that there are still tion "urban" as we generally use it in social some scholars who feel that social science cannot research is a badly confounded variable that be pursued in the imagery of natural science, and obscures much of the order and regularity we that the experimental method and statistical seek, and that impairs our ability to improve techniques approximating it are not applicable our predictability and explanation. to social phenomena.) But it is probably true There is at the present time considerable that they are a dwindling minority and that the variation among the social sciences in the at- view I am presenting is now the modal position tention and energy devoted to measurement. The in a number of the social science disciplines, psychometrician who devotes several years to e.g., , social psychology, , the development of a single scale certainly sociology, demography; and is gaining ground in stands in sharp and meritorious contrast to the others, e.g., and . social survey operator who develops a 30 -page The specific ways in which social statistics in several weeks - -or even a few has contributed to the emergence of social days. There can be little doubt that the sample science may be simply set forth to serve as a social survey, which has been one of the most ef- frame for pointing to future developments. The fective research instrumentalities developed in social has done much to measure social statistics, is also producing a mass of the phenomena in which the social scientist is spurious information by reason of poor measure- interested; to collect, process and tabulate ment of the type described. social ; to describe and analyze the data As another aspect of the problem of measure- ( "explain "); and to project discerned patterns ment much more work is needed to achieve higher or sequences (trends) into the future ( "pre- order measurement scales in working with social dict"). Much has been achieved over the years, phenomena. Success in this effort would have the as Stephan and Keyfitz at this meeting, and salutary dividend of increasing the power of the others from time to time have shown in respect methods of analysis and inference which may be to each of these areas of activity. employed in social research; and may result, Statistical Bottlenecks. Although much has therefore, in higher orders of predictability been achieved, much, also, remains to be done, and fuller explanation. In general, it is prob- as I am sure the other papers in this session ably true that a large proportion of social will agree. For my part I should like to point scientists remain unaware of the great differ- to a few of the obstacles to the development of ences that obtain in the power of admissible social science which urgently require the con- statistical operations when, for example, a certed efforts of social statisticians. ratio -scale can be utilized in lieu of a nomi- Measurement. The first of these bottlenecks nal scale. constitutes, in my judgment, the greatest con- Error. Many of our statistics still contain temporary obstacle to the development of social relatively large elements of error which can be science. It lies in our inability successfully eliminated. error has, of course, 182

become the easiest component of error to manage. social science types of research problems.2/Such But the various other types of error, as enumer- effort, it may confidently be expected, will in- ated by Deming some years agog can certainly crease and, especially, as more social scientists be better controlled than they yet are, as a achieve high mathematical and statistical com- result of the methods which have petence. The social scientist himself, evolved, of the post -enumerative survey, and of statistician, is both more apt to be able to other developments.1V Data that do not meet as define the problem and to be motivated to find high standards of reliability, validity and the solution than the rjon- social science ori- precision as may be required for our research ented methodologist. objectives stand as obstacles to the advance of New Uses of Computer. Another item that social science. Perhaps the time has come when merits discussion is the new opportunity for a all social statistics, and especially those major inflection point in the advancement of based on a canvass of the population, should social science by means of social statistics routinely carry in addition to a report of afforded by the advent of the computer. Elec- sampling error some indication of the magnitude tronic processing of statistical data has already of other elements of error. Certainly we now produced great dividends both in the handling of possess enough knowledge to do much more of this mass datai and in the statistical analysis of than ever was possible before. social data.12j But only a beginning has been Longitudinal Data. Another bottleneck in made. For thus far, the computer has been used the development of social science which the so- largely routinely as a glorified punch -card cial statistician can help to break is to be tabulator or for the conduct of time -honored found in the disparity between social science forms of analysis. projections and the actual course of events. The The computer has yet to be fully exploited demographer has learned the hard way to label for social statistical analysis of a type not his efforts to foresee the future as "projec- feasible before its appearance. For example, in tions" rather than "predictions." What can be addition to n- dimensional cross tabulations of done to improve social science predictability? data in which the unit of analysis is the person, At least part of the answer, I believe, lies household, age -group, geographic area, etc., in the expansion and elaboration of longitu- tabulations are now possible of patterns and dinal, as opposed to cross sectional, statis- sequences of characteristics - -of profil1 ,of the tics. units, instead of the units themselves.- Such One of the greatest weaknesses of social tabulations would permit utterly new types both science, in general, lies in the drawing of of micro- and macro- analysis that could prove diachronic conclusions from synchronic and, exciting and probably greatly improve the power therefore, inadequate data. Longitudinal con- of social science to predict and explain. To clusions are always dangerous when based on gain the most from this possibility the social cross -section data, and especially so when the sciences must cultivate the training of scholars information is also inadequate in various other who combine statistical with computer competence. ways. Yet most of the data which the social Use of Theory. The increasing role of so- scientist uses for projection or prediction are cial statistics in social science methodologi- of this character. They consist of cross -sec- cally and substantively, has helped push into the tion information drawn from a relatively small background the type of speculative and philo- number of points in time, containing gross and sophical activity usually, but mistakenly in my confounded categories with volatile and often judgment, called "theory" in the social sciences. conflicting components. A good illustration is Theory, in the imagery of science, is neither given by the many attempts to predict fertility "speculation " --the undisciplined drawing of behavior from current statistics. Just as the global generalizations without the benefit of demographer has learned the limitations of cross - empirical research; nor "philosophizing " - -the section birth -rate data and the value of the disciplined drawing of generalizations without cohort approach, so the social scientist, in the benefit of empirical research. Theory in general, with the assistance of the social science may be considered as tentative generali- statistician, must greatly increase his fund of zation in the form of predictability or explana- longitudinal data. In this manner dynamic tion based on empirical research and pointing to patterns and sequences could be discerned that further research. would undoubtedly provide a better basis for At the present time it is correct, I think, projection and prediction. to say that a wide gap still exists between Methods of Analysis. Less of a bottleneck, empirical research and "theory" in social science but, nevertheless constituting some obstacle to to the detriment of both. There can be no doubt the advance of social science is the uneven but that the advancement of social science is development of statistical methods for the being badly retarded by this gap. No one saw analysis of social, as compared with natural, this more clearly than Stouffer who was equally science data. Mathematics and mathematical impatient with "talky- talk" sociology, on the statistics, both, were in large measure produced one hand, and blind on the other. by scholars concerned with physical and biolog- He felt that progress in social science could ical science and with engineering problems. Only be achieved primarily through the interplay of relatively recently, as social science has theory and empirical research either one alone achieved the ability to pose new problems for was sterile. He left some brilliant examples the mathematician or mathematical statistician, of the way in which such interplay could be has concerted effort been devoted to the evolv- effected in his work on migration and norms of ing of techniques specifically oriented to behavior.20 183

The social statistician, having begun the tics. I am sure Martin Gainsburgh's Presiden- task of using methods of empirical research and tial Address on "The Statistics We Live By" will building a fund of empirical data, needs increas- furnish eloquent testimony of the social engi- ingly to turn to the construction of theory. The neering_uçility of statistics on a number of social scientist who specializes in "theory" by fronts.AAJ And it should be added that statis- the same token, needs increasingly to turn to tical developments in respect of decision -mak- the use of empirical research. Social Science ing, in general, have greatly increased the may be expected to achieve important progress as utility of social, as well as other types of the interplay between theory construction and statistics, for action programs. empirical research is increased and, it may be Social engineering needs have in large hoped, as these two functions are increasingly measure generated social statistics. But it is combined in the person of the well- trained also true, of course, that social science has social scientist. had an important role in determining the specific Training. The speed at which many of the form that social statistics have taken. For the specific obstacles to the advancement of social social scientist, although he has not had a com- science can be reduced will, in large measure, pletely free hand, has typically been the de- depend on the extent to which the social scien- signer and operator of projects to collect and tist obtains improved mathematical and statis- disseminate social statistics. The fact that tical training. Although notable improvement social statistics often are the direct product has taken place over the years, the average of efforts to solve social problems is, on the social science student still tends to grumble one hand, the reason for their proliferation; about even the limited mathematical and statis- and, on the other hand, the reason for their tical course work now generally required. In frequent inadequacies from the standpoint of so- many cases it is undoubtedly true that the cial science. Social statistics have in large student just lacks quantitative aptitude - -and measure taken the form they now possess by that is why he is in social science. But many reason of the interplay between social engineer- social science students lack not mathematical ing and social science needs; and it is likely aptitude as much as incentive and motivation that a similar interplay will continue to influ- -to obtain the desired competence. Social ence the course of their development. science faculties could remedy this situation The great bulk of social statistics are relatively quickly, without undue hardship on government collected -- federal, state and local- - students, if more of them would prescribe at in response to needs not only of government but, least some pathways in social science training also, of business, labor, education, civic in which high mathematical and statistical com- enterprises, health and welfare organizations petence are prerequisites. A number of social and a host of other types of organizations and science departments have already done this- -but groups.23' Additional social statistics, the number is still relatively small. Much more generally not as widely available, are generated needs to be done to encourage social science by universities and research organizations; and students to achieve higher mathematical and by various non -governmental activities, largely statistical training. as a by- product of administration. This is not -to say that statistical and Let us first briefly consider government mathematical methods are the only methods by statistics and then turn to the non -governmental. which social science may advance. But there can Government Statistics. Government statis- be little doubt that more and better social sci- tics have greatly expanded since the 1930's when entists cum statisticians would greatly improve the "bloodless revolution" of the New Deal social science output. Certainly there is no placed the federal government into a number of evidence, yet, that we have too many well- trained new fields of activity. Preparation for, and social statisticians. participation in, World War II, and the require- ments of the Cold War have had a similar effect. SOCIAL STATISTICS AND SOCIAL ENGINEERING In consequence, the federal government itself has become the most important and largest single The increased volume and quality of social consumer of its own social statistical product. statistics over the years are more directly The fact that important policy decisions and traced to social engineering, than to social administrative action of the federal government science, needs. For the demand for more sta- are increasingly dependent on the factual picture tistics in virtually every realm, including the provided by social statistics has led to great social, has steadily grown as our society has scrutiny of, and much improvement in, many of become more complex and more interdependent./ the specific statistical activities of the As Mayo -Smith stated in 1910 "we cannot wait for government. The improvements effected in the the completed science" to find solutions for Monthly Report on the Labor Force contribute a urgent social problems. Social statistics, in good case in point. the sense of quantitative information about The status of social statistics produced by problems, are collected primarily to provide a the government was excellently reviewed at a sounder basis than is otherwise available for meeting of this Section ust a year ago by policy formulation and administrative action. In Bowman, Gall and Rubin.2&/ In that paper, as general, there are few realms of social action, well as in Stephan's paper at this session, some in government or out, which are not dependent shortcomings of the federal statistical system on social statistics for intelligence about the were noted and indications given for improvement. problems with which they deal, or which would Stephan makes a goal case for furthering the not be better off if they had adequate statis- improvement of statistics within the established 184 decentralized statiqtical and administrative statistics relating to employment practices, for agency framework.25 But it is also appro- example, are in this process now under the pres- priate to observe that the federal statistical sure of increasing state and federal "fair em- system was never really planned, that it has ployment" activities. relatively few comprehensive reviews, and even In general, the future of non -governmental fewer comprehensive overhaulings. It may well as well as governmental social statistics may be be that it would be in the interest of social considered a bright one in the sense that they engineering and, also, social science, for the may be expected both to be improved and to statistical system as a whole to be not only become increasingly important for social engi- reviewed but, also, overhauled from time to time, neering purposes. It is hardly necessary to although not necessarily every 17 years as pre- mention that one of the developments that jus- scribed for another matter by Tom Jefferson. An tifies this optimistic outlook is to be found occasional hard look and effective reshuffling in the growing professionalization of statis- could conceivably improve the utility of social tical activities, as the growth and programs of statistics for social engineering purposes - -as the American Statistical Association testify. well as increase the statistical output per unit of cost.26 CONCLUDING OBSERVATIONS Non -Governmental Statistics. Social sta- tistics produced by non -governmental organiza- Social statistics, methodologically and tions have also proliferated over the years and substantively, has played a prominent role both especially in the post -war period. Advances in in social science and social engineering. Social the sampling of human populations and in social statistics, in turn, have proliferated largely survey methods, in general, have made it easier as a result of social engineering, especially for non -governmental organizations to do their government, needs. But they have achieved a own fact finding, or to utilize non -profit or relatively high level of quality and utility commercial research agencies for the purpose. largely as the result of the impact of social In consequence, social statistics has become science. part of the mass communication diet of the Social statistics may be expected to a- American people in an unprecedented way, chieve an even more important place than they directly through the public opinion polls, and now possess in social science in the years ahead. indirectly, through the increased dependence of The natural sciences can make greater use of the action agencies on social statistics for, and experimental method and a direct mathematical explanation of, action. approach than can the social sciences. In con- The quality of non -governmental social sta- sequence, although statistical methods are use- tistics varies widely. Some of the statistics, ful in the natural sciences, they assume a much especially those collected by university af- greater importance in the social sciences in filiated organizations such as the Survey that they represent the major means by which Research Center at the University of Michigan social phenomena can be quantified and the and the National Opinion Research Center at the experimental method approximated or simulated. University of Chicago, meet the highest stand- This is likely to remain the situation for many ards. Other batches of social statistics, par- years to come, although direct mathematical ap- ticularly those collected by commercial agen- proaches are increasing in the social sciences. cies under budget pressure, leave much to be Similarly, social statistics is likely to desired. Especially distressing is the failure achieve even greater prominence and utilization of a number of such organizations fully to in social engineering. Our society continues publish the details of the designs, procedures to grow more complex and interdependent, and and forms that they use. As the nation becomes policy and program, therefore, require ever more more and more dependent on social statistics intelligence for sound direction. The most ef- for effective and efficient action, the pres- ficient form of such intelligence yet devised is sures for improved social statistics from all afforded by statistics, including social statis- sources may be expected to increase. Fortu- tics. Moreover, developments in statistics, as nately, most of the important producers of such well as electronics, have materially increased data have become increasingly professionally the and effectiveness of the decision minded and strive to maintain and improve the making process itself. quality of data as may be observed in the activ- It is fitting to close this paper at this ities of the American Association of Public Memorial Session in Honor of Professor Samuel A. Opinion Research.27 Stouffer with acknowledgement of our indebted- Non -governmental as well as government ness to him for the remarkable character of his agencies produce much in the way of social contribution to social statistics both in social statistics as by-products of conducting their science and in social engineering. It is dif- own activites. Some of these data are of high ficult to traverse the field of social statis- quality and are widely circulated as in the case tics over the past thirty years without crossing of the Statistical Bulletin of the Metropolitan Sam's tracks -- directly or indirectly. No one Life Insurance Company. Most of such material, has done more over this period either to advance however, is compiled and used for specific or- the cause, or the science and art, of social ganization purposes and never enters the public statistics. I have not elaborated on his spe- domain. With increasing interdependence and cific contributions here because I have already government interventionism it may be antici- done so elsewhere,2/ and because I anticipated pated that many of the social statistics now the other papers would do so. considered private will become public. The Despite his premature death, Samuel A. 185

Stouffer is still in a position to exert great Hosteller, Stochastic Models for Learning, New influence on the future of social statistics. York: John Wiley & Sons, 1955; H.A. Simon, Today, about thirty years after the completion Administrative Behavior, New York: Macmillan, of his own statistical training which he a- 1947; Leo A. Goodman, "The Use and Validity of chieved relatively late in his career, he may a Prediction Instrument. U. The Validation of yet be viewed as a prototype of the well- trained Prediction," American Journal of Sociology, Vol. social scientist. It is a sad commentary that LVIII (1953), p. 510; "Measures of Association the vast majority of social scientists, includ- far Cross Classification" (with William Kruskal), ing even the majority of recent Ph.D.'s, have Journal of the American Statistical Association, not matched his training in methods of research. Vol. LIV (1959), p. 123; "Snowball Sampling," In this thought lies an additional well- deserved Annals of , Vol. XXXII tribute to Sam; and a basis for sombre reflec- (1961), p. 148. tion for the rest of us about how much remains 15/ The career of Professor Leo A. Goodman to be done to produce competent students really of the Department of Sociology and Statistics at equipped to advance both social science and so- the University of Chicago is a good case in cial engineering by means of greatly improved point. social statistics. 16/ The use of UNIVAC by the U.S. Bureau of the Census for tabulation of the 1960 FOOTNOTES of Population and Housing is a case in point. See Richard A. Hornseth, "Programming the Popula- 1/ Richmond Mayo- Smith, Statistics and tion Census," Proceedings of the Social Sta- Sociology, New York: The Columbia University tistics Section, 1959, Washington: American Statis- Press, 1910, Ch. 1. tical Association, 1960, pp. 200ff. 2/ Morris R. Cohen and Ernest Nagel, An In- 17/ E.g., James S. Coleman, "The Use of Com- troduction to Logic and the Scientific Method, puters in the Study of Social Structure: Inter- New York: Harcourt, Brace & Co., 1934, Ch. X. action in a 3-Person Group," Proceedings of the 3/ Ibid., pp. 195ff. Social Statistics Section, 1959, Washington: 4/ Ernest Nagel, The Structure of Science, American Statistical Association, 1960, p. 42. New York: Harcourt, Brace & World, Inc., 1961, 18/ This possibility was proposed by my Ch 2, 14. colleague, Professor 0.D. Duncan, in discussions 5/ Cohen and Nagel, op.cit., XIII; E. about next steps in the development of tabula- Bright Wilson, Jr., An Introduction to Scien- tions of social data. The idea merits intensive tific Research, New York: McGraw Hill, 1952, exploration. Ch. 4. 19/ Philip M. Hauser and Otis Dudley Duncan, 6/ Harold H. Kelley and John W. Thibaut, op.cit., p. 15ff. "Experimental Studies of Group Problem Solving 20/ Samuel A. Stouffer, "An Analysis of Con- and Process," in Gardner Lindzey, Handbook of flicting Social Norms," American Sociological Social Psychology, Cambridge, Mass.: Addison - Review, Vol. XIV (1949), 707 -717. Wesley Publishing Co., 1954, Vol. II, Ch. 21. 21 Philip M. Hauser and William R. Leonard See bibliography of Ch. 21 for further refer- (eds.), Government Statistics for Business Use, ences. New York: John Wiley & Sons, 1956, Ch. 1. Samuel A. Stouffer, "Some Observations 22 Martin Gainsburgh, Presidential on Study Design," American Journal of Sociology, Address, American Statistical Association, 1961. Vol. LV (1950), pp. 355 -361. Will appear in March 1962 issue of Journal of 8/ Paul F. Lazarsfeld and Morris Rosenberg American Statistical Association. (eds.), The Language of Social Research, Glencoe, Bureau of the Census, Statistical Ill.: The Free Press, 1955; Paul F. Lazarsfeld Abstract of the United States, 1961, Washington: (ed.), Mathematical Thinking in the Social Government Printing Office, 1961 (issued Sciences, Glencoe, Ill.: The Free Press, 1954. annually). See especially "Bibliography of Herbert Blumer, "Sociological Analysis Sources," pp. 957 -994. and the 'Variable,'" American Sociological 24/ R.T. Bowman, Alexander Gall, and Israel Review, Vol. XXI (1956), pp. 683 -690. Rubin, "Social Statistics: Present Conditions, 10/ Frederick F. Stephan, "Relations of Some Future Needs and Prospects," in Proceedings of Social Science Concepts to Statistical Data," the Social Statistics Section, 1960, Washington: Proceedings of the Social Statistics Section, American Statistical Association, 1961, pp. 74- 1959, Washington: American Statistical Associa- 81. tion, 1960, pp. 170 -171. 25/ Frederick F. Stephan, "Samuel A. 11/ W. Edwards Deming, "On Errors in Stouffer and the Progress of Social Statistics," Surveys;'American Sociological Review, Vol. IX to appear in Proceedings of the Social Statistics (1944), pp. 359 -369. Section, 1961 of the American Statistical 12/ Philip M. Hauser and Otis Dudley Duncan, Association. "The Data and the Methods," in Philip M. Hauser 26/ Philip M. Hauser, "Non -random Observa- and Otis Dudley Duncan (eds.), The Study of tions on Statistics from Washington," The Federal Population: An Inventory and Appraisal, Chicago: Statistics Users' Conference, Washington: Fourth University of Chicago Press, 1959, Ch. 3. See Annual Meeting, September 1960, pp. 65 -74. bibliography, especially references to work of See Public Opinion Quarterly, the Morris Hansen, William Hurwitz, Leon Pritzker journal of the American Association of Public and A. Ross Eckler. Opinion Research. 13/ E.g., P.K. Whelpton, Cohort Fertility, 28 "Samuel Andrew Stouffer, 1900 -1960," Princeton: Princeton University Press, 1954. The American Journal of Sociology, LXVI (1961), 14/ E.g., Robert R. Bush and Frederick pp. 364 -365.