This article develops a procedure for path analysis of ordinal variables using only ordinal statistics and presents an ordinal path analysis of data previously analyzed by Sewell and Armer (1966a). Substantively, it shows that when ordinal statistics and path analysis are used, neighborhood context has very important direct and indirect effects on college plans. This is true when neighborhood socioeconomic status is trichotomous or dichotomous. Methodologically, it demonstrates that path analysis techniques can be successfully applied to ordinal data using ordinal statistics rather than assuming equalinterval scales and applying interval statistics, or using dummy variables. Possibly, small effect of neighborhood context reported by Sewell and Armer is a consequence of their ignoring a curvilinear relationship between neighborhood context and college plans, their assuming interval data, and their use of a linear multiple correlation model when in fact their data did not conform to underlying assumptions. This article develops a procedure for path analysis of ordinal variables using only ordinal statistics and presents an ordinal path analysis of data previously analyzed by Sewell and Armer (1966a) in their article, Neighborhood Context and College Plans.' In it, they tested hypothesis that: the socioeconomic status of high school district-since it presumably reflects shared norms and aspirations of its members would have an important effect on educational aspirations of its youth over and above that of family socioeconomic status or individual ability (1966a: 162). Sewell and Armer first treated their variables as attributes. They showed that neighborhood socioeconomic status has a positive zero-order effect on college plans of high school seniors. (Their data-set described universe of seniors attending Milwaukee high schools during 1957-58 academic year.) Then they presented cross-tabulations between each test factor (sex, intelligence, and socioeconomic status) and independent and dependent variables. In order to determine whether original relationship between neighborhood and college plans still held, they tested it by controlling for test factors singly and in combination. Finally, they assumed interval measurement and carried out a multiple correlation analysis. They entered test factors first and then independent variable neighborhood context-to determine magnitude of variance in college plans it independently explains. The researchers did not carry out a causal analysis of interrelations between neighborhood context and 3 test factors. They felt their research hypothesis did not require it (1966b:708), and that precise causal order of variables was unobtainable ( 1966b:709). Given their analytic procedures and symmetric measures of correlation, they found that: . . neighborhood status results in an absolute increase in explained variance of college plans of 1.8 percent beyond effects of sex, socioeconomic status, and intelligence [ 199 ] * Revised version of a paper presented at annual meeting of Eastern Sociological Society, New York, April 1971. The author is indebted to Paul F. Lazarsfeld for suggesting that he study this problem, and to Robert Somers, Neil Henry, Robert HaLuser, and numerous colleagues for their helpful comments on earlier drafts. Preparation of this article was facilitated by general research grants from Academic Senate of University of California, Santa Barbara, and by grants of computer time from National Institute of Mental Health. 1 For earlier methodological and substantive comments on Sewell and Armer's analysis see Boyle (1966), Michael (1966), Sewell and Armer (1966b), Turner (1966). This content downloaded from 207.46.13.183 on Thu, 21 Apr 2016 06:29:34 UTC All use subject to http://about.jstor.org/terms 200 / SOCIAL FORCES / vol. 51, dec. 1972 ( 1 966a: 167). Contrariwise, results reported below based on a reanalysis of very same data, but using ordinal statistics and path analysis, indicate that neighborhood context has very important direct and indirect effects on college plans.
Read full abstract