A. Definition. Multivariate analysis is essentially the statistical process of
simultaneously analyzing multiple independent (or predictor) variables with multiple dependent (outcome or criterion) variables using matrix algebra (most multivariate analyses are correlational). While these analyses have been a part of statistics since the early 1900s, the development of mainframe and microcomputers and subsequent analytical software has made the once tedious calculations fairly simple and very fast.
B. Purpose. Human behavior is very complex. Behaviors, emotions,
cognitions, and attitudes can rarely be described in terms of one or two variables. Furthermore, these traits cannot be measured directly, as say running speed, but must be inferred from constructs which in turn are measured by multiple factors or variables. The researcher must construct his/her view of reality (observations about humans), create multiple measures to assess the constructs, and conduct an appropriate statistical analysis of the data, and correctly interpret the outcomes. Basically, the researcher asks two questions, Do I have anything important? and If so, what do I have? Essentially, the first question asks if the specified model (which is based upon the researchers observations of some aspect of human behavior) adequately addresses the observation, and the second asks what is the best explanation of the relationship between the variables. C. Basic Analysis. Multivariate analysis may either be conducted in a classic experimental design or in non-experimental designs. Whether one is seeking causality or association depends upon the research question, but both are valid in multivariate analysis. Most of the time, a data matrix will be analyzedthe form usually consists of rows representing each subject and columns representing each variable; however, the matrix can also be a correlation matrix, a variance/covariance matrix, or a sum of-squares/cross-product matrix. The matrix is then solved simultaneously through matrix (linear) algebra and yields linear composite scores which are linear combinations of the variables upon which the final solution is based. Most analyses construct composite scores that maximize the variance associated between one set of variables (X set, I.V.s, or predictors) and another set (Y set, D.V.s, criteria, or outcomes). The importance of each variable is determined by its weight (degree of contribution) to its set of variables and is usually indicated by a numerical coefficient. Fortunately, a good understanding of matrix algebra is helpful but not necessary in understanding and using multivariate analytic techniques.