This Digest presents a discussion of the assumptions of multiple regression that is tailored to the practicing researcher. The focus is on the assumptions of multiple regression that are not robust to violation, and that researchers can deal with if violated. Assumptions of normality, linearity, reliability of measurement, and homoscedasticity are…

This paper gives concise descriptions of a robust location statistic, the remedian of P. Rousseeuw and G. Bassett (1990) and a robust measure of dispersion, the "Sn" of P. Rousseeuw and C. Croux (1993). The use of Sn in least absolute errors regression (L1) is discussed, and BASIC programs for both statistics are provided. The remedian…

Data from records of 99 patients were used to classify cardiac patients as to whether they were likely or unlikely to experience a subsequent morbid event after admission to a hospital. Both a linear discriminant function and a logistic regression equation were developed using a set of nine predictor variables that were chosen on the basis of…

Missing data occur in virtually every study. This paper reviews some of the various strategies for addressing this problem. The paper also provides instructional detail on two accessible ways of estimating missing data, both using the Statistical Package for the Social Sciences for Windows: (1) substitution of missing values with the variable mean…

The efficacy of two improvement-over-chance or "I" effect sizes, derived from predictive discriminant analysis (PDA) and logistic regression analysis (LRA), were investigated for two-group univariate mean comparisons. Data were generated under selected levels of population separation, variance pattern, sample size, and distribution…

The assumption that is most important to the hypothesis testing procedure of multiple linear regression is the assumption that the residuals are normally distributed, but this assumption is not always tenable given the realities of some data sets. When normal distribution of the residuals is not met, an alternative method can be initiated. As an…

Three potential applications of stepwise procedures in canonical analysis and several alternative stepping decision rules are described. Results of an empirical investigation of the procedures indicated that more parsimonious approaches to maintaining variables held up better under cross-validation. (Author/JKS)

The first eigenvalue of a correlation matrix indicates the maximum amount of the variance of the variables which can be accounted for with a linear model by a single underlying factor. The first eigenvalue measures the primary cluster in the matrix, its number of variables and average correlation. (Author/RL)

A Bayesian solution is presented for the Johnson-Neyman problem (whether or not the distance between two regression lines is statistically significant over a finite interval of the independent variable). (Author/CTM)

Two commentaries describe some shortcomings of a recent discussion of the significance testing of R-squared by C. J. Huberty and upward bias in the statistic. Both propose some modifications. A response by Huberty acknowledges the importance of the exchange of ideas in the field of data analysis. (SLD)

Used the jackknife method coupled with a linear regression model to illustrate uncertainty in the overall scores used to rank colleges by "U.S. News and World Report." Results indicate that the top 50 schools in each ranking can be shown to group into 2 or 3 bands. (SLD)

Describes a study on Internet marketing strategies of business-to-consumer (B2C) firms in Singapore. Results of a survey and hierarchical regression analyses indicate that strategies to attract customers and to relate to customers have significant positive relationships to online brand equity, which is positively related to financial growth.…

The effects of two different methods of centering, in comparison with the use of raw scores, on the parameter estimates of random coefficient models were studied. Analyses show that centering around the group mean amounts to fitting a different model than centering around the grand mean or using raw scores. (SLD)

The impact of scaling changes on the size of conditional relationships in meta-analysis is explored by examining a number of parameters. The most nearly appropriate parameters for researchers to aggregate appear to be unstandardized regression coefficients or partial correlation coefficients. (SLD)

Discusses problems researchers face when they want to describe relationship between several predictors and criterion variable. Considers ways of addressing problem of contribution of each predictor depending on which other predictors are in regression equation. Focuses on parallel information for each variable, examining initial and final…