To test moderating effects, this research adopted multigroup analysis following previous studies (Chaudhuri, Aboulnasr, and Ligas 2010; Goldsmith, Flynn, and Kim 2010; Porter, Donthu, and Baker 2012; Tantong et al. 2010). Before performing multigroup analysis, we conducted metric invariance analysis. As shown in Table 4, the values of χ2/df (between 1 and 5) and IFI (incremental fit index) and CFI (above 0.8) are acceptable. Metric invariance is achieved when the χ2 differences between unconstrained and constrained models using multigroup confirmatory factor analyses (MGCFAs) are nonsignificant. Changes in χ2 are sensitive to sample size. Cheung and Rensvold (2002) recommended using CFI change as a rule of thumb for metric invariance. When conducting MGCFAs to assess measurement invariance, it has been recommended that changes in CFI values of 0.02 or less are indicative of factor invariance across the groups (Cheung and Rensvold 2002; Limbers, Newman, and Varni 2008; Vandenberg and Lance 2000). The results showed that metric invariance was achieved.
The moderating role of retail store formats was investigated via multigroup analysis. Groups were created for department and discount store groups. In the equal model, this research set all paths of the structural model equally across department and discount store groups. In the free model, this research constrained all paths to be equal across department and discount store groups, except for the link between the antecedents and the OSPI that are potentially affected by the moderator variable. Differences in χ2 values between models determine whether retail store formats act as a moderating variable. A significant decrease of χ2 value from the equal model to a model in which one relationship is set free implies that the moderating variable has a significant impact on the relationship. The result showed that retail store formats moderate the effect of the price value image and the price pleasure image on the OSPI. We further investigated the incremental path coefficient analysis to prove the moderating effect. The result revealed that the within-group path coefficient was lower for the discount store group than the department store group. A significant difference in path coefficient was found for the link between the price value image and the OSPI and between the price pleasure image and the OSPI. The multigroup analyses revealed that the effect of the price value image and the price pleasure image on the OSPI was stronger for department stores than for discount stores, thus supporting H6a and H6d. As the positive effect of the price rewards image on the OSPI was not significant for either the discount or