Microsoft Word 2007 ichrie conference Proceedings Final-Final 06-06-07. doc

Test for measurement invariance
.
Before comparing key paths between groups, a measurement invariance 
test was conducted to assess whether a measurement model is invariant across groups. If measurement invariance 
was not supported, findings based on the measurement are uncertain or mistaken (Steenkamp and Baumgartner, 
1998). A non-restricted model, in which the factor loadings of the four constructs were relaxed to vary across 
groups, was first assessed, and then the equality of the factor loadings between the two groups was assessed.
 
For gender groups, a non-restricted CFA model yielded a reasonable fit to the data (
χ
2 
= 76.024, 
df
= 42, 
RMSEA = 0.033, CFI = 0.997, NFI = 0.993). A chi-square difference test is the general way of testing measurement 
invariance across groups, and the measurement model is invariant when the chi-square does not show a significant 
difference (Yoo, 2002). Thus, the significance of the chi-square difference between the non-restricted model and the 
full metric invariance model was tested for gender groups. Since the chi-square difference between the non-
restricted model and the full metric invariance model was not significant (
Δχ
2 
(5) = 10.144, p>.01), full metric 
invariance was supported. That is, factor loadings were equivalent across male and female groups. Accordingly, 
this full metric invariance model for gender was used in subsequent analyses. For age groups, a non-restricted CFA 
model also fits the data well (
χ
2 
= 57.808, 
df
= 42, RMSEA = 0.036, CFI = 0.999, NFI = 0.995). There was no 
significant chi-square difference between the non-restricted model and the full metric invariance model (
Δχ
2 
(5) = 
3.809, p>.01), supporting full metric invariance. This full metric invariance model minimizes the effect of probable 
variation across groups in measurement structure (Bagozzi and Yi, 1989). Therefore, a full metric invariance model 
for age was used to run the structural equation modeling and in further analyses.