that the variables are unstandardized. While conventionally analyses such as ordinary
least square regression attempt to minimize differences between observed and expected
individual cases, SEM aims to minimize differences between observed and expected
covariance matrices. In other words, SEM, based on the covariance statistic, attempts "to
understand patterns of correlations among a set of variables and to explain as much of
their variances" (Kline, 1998, pp. 10-11).
Second, SEM allows separating measurement error from hypothesized relationship
among constructs that cannot be operated by conventional analyses. For example, a
regression coefficient is composed of two elements: structural coefficient between the
independent and dependent variable and the reliability for the predictor variable. By
distinguishing a structural model from a measurement model, SEM can examine the
relationship among constructs that are not influenced by measurement errors (Newcomb
& Bentler, 1988).
Third, SEM allows incorporation of latent variables into analyses unlike
conventional analyses that focus only on observed variables. Since SEM is not limited to
relations among observed variables, it gives researchers more flexibility to study any
combination of relations.
Finally, SEM allows researchers to estimate very complicated multivariate
relationships, while conventional analyses cannot accommodate multiple indicators of the
same construct due to potential problems such as multicollinearity. Even conventional
multivariate techniques that deal with multiple dependent variables cannot examine
relations simultaneously and only examine a single relationship at a time. However, SEM