Chapter 9, Association, Log-linear Analysis and Canonical Correlation Analysis Video Solutions, Statistics for Marketing and Consumer Research

Problem 1

Open the Trust data-set
a. Build a two-way contingency table showing absolute and relative frequencies between the variable q55 (job) and the variable q61 (perceived household wealth)
b. Evaluate the statistical significance of their association using the range of statistics illustrated in this chapter
c. Log-linear analysis can also be applied on two-way contingency tables, look at the role of the interaction between job and household wealth through a hierarchical analysis, looking at the significance of the interaction term
d. What is the difference between the Chi-square statistic computed through CROSSTABS and the one corresponding to the two-way interaction in hierarchical log-linear analysis?

Jameson Kuper

Numerade Educator

07:53

Problem 2

Open the EFS data-set
a. Create a new variable from tenure type (a121) with two categories, rented (a121 from 1 to 4 ) and owned (a121 from 5 to 7 ) - HINT: use the TRANSFORM / RECODE INTO DIFFERENT VARIABLES menu
b. Using log-linear analysis, explore the interactions among the above variables, Government office region (gor) and the number of children considering the first three categories only (childband variable from 1 to 3 )
c. Using hierarchical log-linear analysis, compute effects in the saturated model then estimate a simplified model
d. Check the partial association table - are there lower-order terms that could be discarded?
e. Re-estimate the simplified model with the general log-linear model and find the most parsimonious model which is still acceptable according to the chisquare statistic (HINT: first exclude cases where childband $=4$ using the DATA / SELECT menu)
f. Compare results using alternatively the Poisson and Multinomial distribution assumption. Do estimated effects differ?
g. What are the odds of having a rented rather than owned house in government region 2?

Heather Duong

Numerade Educator

03:46

Problem 3

Open the Trust data-set, run a canonical correlation analysis on the following two sets of variables, the first referring to safety and risk aspects of chicken, the second related to trust in information on chicken safety:
$$
\begin{aligned}
& \mathrm{SET} 1=\text { q12d q21d q21e q27d } / \\
& \mathrm{SET} 2=\text { q43a q43b q43e q43f. }
\end{aligned}
$$
a. What are the values of the canonical correlations?
b. How many canonical correlations should be retained according to the Wilks' Lambda test?
c. Look at the canonical loadings for each set and comment on those for the first canonical correlation
d. Explore the cross-loadings between the first two canonical correlations and comment on them
e. Is the first canonical correlation relevant in explaining variability of the second one?
f. Compute the sample averages of the four canonical scores

James Kiss

Numerade Educator