Transcript 11/19/2015
Psychology 202a Advanced Psychological Statistics November 19, 2015 The Plan for Today • • • • • Visualizing ANOVA (continued) ANOVA as a special case of regression Post hoc comparisons Contrasts Orthogonal contrasts and contrast coding Visualizing ANOVA • • • • Parallel boxplots Bar plots of means Line graphs of means Rules for choice of bar plots or line graphs – Grouping variable’s level of measurement technically should be interval or ratio for line graph – Often violated (without particularly dire consequences) Reviewing ANOVA in SAS the easy way • The “class” statement. • “class” tells SAS “Figure out how to code this classification variable so that we can handle it using regression.” • Let’s see how such coding works. • Examples in SAS An outlier test • Sometimes we can use creative coding to do clever things in regression. • Recall that our regression of Peabody on Raven had one observation that disturbed us. • Create a dummy code for that observation. (Example in SAS.) • Also known as the “externally Studentized residual” Results of dummy coding Group D1 D2 Massed Practice 1 0 Spaced Practice 0 1 No Practice 0 0 Other forms of coding • Any coding system that uses two variables to identify the three groups will produce the same ANOVA • This idea will turn out to be profoundly useful • Example: effects coding Example of effects coding Group D1 D2 Massed Practice 1 0 Spaced Practice 0 1 No Practice -1 -1 Asking more detailed questions • So far, we haven’t really learned anything interesting about these means. • Post hoc procedures – Illustration in SAS Asking more detailed questions • When possible, if we can plan our questions in advance, we will be more likely to find effects. A priori contrasts • A contrast is a question about a linear combination of means. • Example: H0 : Massed Spaced 2 None 0 • Shorthand notation: 1/2 1/2 -1 • Equivalent: 1 1 -2 • Another question that might interest us is 1 -1 0. Contrasts (continued) • Once a contrast is specified, its sum of squares is calculated: k SS contrast c i M i i 1k 2 2 c i n i 1 i • Contrasts always have 1 df, so the sum of squares is a mean square. • Division by the error mean square provides an F statistic that tests the contrast.