To the example of Study Suggestions 1 and 2, I added another category-politician. Following are the illustrative data:
Using a computer program, do a logistic regression with effect coding, assigning -1 to the last category (politician). Call for the printing of the covariance matrix of the b's, C. If you are using a program that reports only the correlation matrix of the b's (e.g., SPSS), follow procedures I explained in this chapter and in Chapter 11 to transform it to a covariance matrix.
(a) What is the regression equation?
(b) Does the inclusion of the source of the message lead to a statistically significant improvement of the model, over a model composed of the intercept only?
(c) Using relevant information from the regression equation, calculate the effect of the last category (i.e., Politician).
(d) Using information from (a) and (c), calculate the odds ratios of (1) Economist to Labor Leader, (2) Economist to Politician, and (3) Labor Leader to Politician.
(e) Assume that dummy coding was used and the Politician category was assigned O. Use relevant information from (a) and (c) to arrive at the regression coefficients for the vectors in which Economist and Labor Leader categories were identified (i.e., assigned 1 's).
(f) From the computer output, what is C (the covariance matrix of the b's) for the coded vectors?
(g) Augment C reported under (0 to obtain C*.
(h) Using relevant information from C*, test the difference between (1) b for Economist and b for Politician, and (2) b for Labor Leader and b for Politician. As I explained in the chapter, I use the term "b" for the category assigned -1 (politician, in the present example), although it is not part of the regression equation.
(i) If you were to run the analysis with dummy coding, assigning 0 to Politician, to what results would those obtained under (h) be equal?