Figure 5. Factor loadings related to C
Table 6 shows the results of the multiple regression analysis on ZY1, which was generated from the Y group of explained variables using the variable increasing and decreasing method with synthetic variables ZA1, ZA2, ZB1, ZB2, ZC1, and ZC2, which were generated from the categories of explanatory variable question groups. Looking at Table 6, the standardized partial regression coefficients show that ZB1 was the highest at .38 (significant at the 1% level). Defined as ‘how officers deal with and explain issues to customers at the counter’, ZB1 was the most important variable for the explained variable (ZY1). Next, ZC1 and ZA1 had positive effects on the objective variable, in that order (significant at the 1% level). Since each Variance Inflation Factor: VIF was under 2.00, we consider that independence among the selected explanatory variables was maintained to a certain degree.
We then re-examined each questionnaire item in the Y group, which contained the explained variables. A correlation analysis of ZY1 with YQ1, YQ2, and YQ3 showed respective correlation coefficients of 0.882, 0.917, and 0.691 (all significant at the 1% level). Excluding YQ3, which showed a low correlation with ZY1iii, a new principal component analysis was performed on the Y group, while a multiple regression analysis using the first principal component of the calculated composite variables (ZY2_1 and ZY2_2) was conducted in the same way as before (see Table 7). Here, the adjusted R2 increased to .51, while the β of ZB1 increased to .45. In both Models 1 and 2, the most important variables were staff responses and explanations, thus supporting the results of a previous study conducted in Malaysia (Mansor and Razali, 2010). In Model 2, each VIF was under 2.00. Therefore, we consider that independence among the selected explanatory variables was maintained to a certain degree.
4. Discussion and Conclusion
This study distributed questionnaires to customers of the Kurose branch office of the Higashihiroshima City government after they had finished their business at the office counter. The survey was conducted on six business days (from August 24–31, 2020) and resulted in a total of 240 responses. Based on the remaining survey flyers (110 of 510), once the research period concluded, the participation rate was approximately 60%. The sample for the principal component analysis and multiples regression analysis was composed of the 205 valid questionnaires collected from August 25–31, for five business days.
The questionnaire was designed to assess factors (explanatory variables) and items related to customer satisfaction with the counter services provided at the branch office. There were also some open-ended questions about the survey research itself and local city policies. Questionnaire items were distributed across four categories (A, B, C, and Y), then individually condensed into pairs of two principal components via principal component analysis. Next, two principal components from each category were submitted to a multiple regression analysis as explanatory variables. This two-stage analysis (referred to as a selective principal component regression analysis) was implemented based on two models. Among the explanatory variables (i.e., A–C), the regression analysis showed that in group B ‘software related to officers’ human factors’ was the most important for the dependent Y variables, followed by groups C and A (adjusted R2 value of .51; Model 2). The β of ZB1 was .45. According to both Models 1 and 2, the most important variables were staff responses and explanations (group B), thus supporting the results of a previous study conducted in Malaysia (Mansor and Razali, 2010).
After the research detailed in this paper had concluded, an online questionnaire survey was conducted among the users of the ward office in Osaka City (Author, 2020). The principal component scores for the three factor categories mentioned above (A–C) were also statistically significant variables in the last multiple regression analysis using the dataset of the Osaka City survey. There were differences in the results between this study’s examination of the Kurose branch office and the study on the wards offices of Osaka City. The latter indicated that C (service delivery quality) was more important than B (customer service provided by the counter staff). The differences between the results of the two studies may be due to differences in the services provided by the branch offices and the ward offices of government-designated cities, as well as differences related to the attributes of the customers of the offices. Comparing the two studies alone may not provide an accurate picture of the factors that create this difference, and we would like to explore the relevant factors through a survey in another municipality and a follow-up survey in the same municipality.
This study focused on three categories, including hardware, customer service provided by the counter officer, and service delivery quality (groups A–C). Future research may need to implement additional aspects related to other customer satisfaction factors, thus establishing a more explanatory model. Items that were not subjected to the principal component analysis within a given category because their correlation coefficients with the composite variables (comprising question items belonging to the Y category; that is, the results category) were less than 0.4. More appropriate question items may produce different results or lead to a model with a higher coefficient of determination (R2 value). We would like to explore models and questionnaires that can better grasp customer satisfaction at government offices by conducting a follow-up survey in the same Kurose branch office. We would additionally like to conduct a first-time survey in the main office building of Higashihiroshima City government, which is in the Saijo area.