||The Perceptions of Politics Scale (POPS) that the researchers use to measure organizational politics vary greatly among different scholars of different countries. Even the inventors, (Kacmer & Ferris) themselves were inconsistent in using the measurement scales. Therefore, the motive and purpose of this study were to construct generalizability, confirmation, validity scale, the predictive relations between the perceptions of organizational politics and job stress, effect of the cross-sample and cross-circumstance so as to apply the model to other sampling population, hoping to faithfully reflecting the variables observed in them.|
This study used the scale of 26 items collected by Dr. Ho Chin-ming et al. using the statistics of 2002 and 2003 measured among 36 organizations of 9 industries in Taiwan. A total of 6,143 questionnaires (return rate of 63.44%) were used. Up to 82.9% of the valid questionnaires were adopted in the final analysis. The data were analyzed by applying statistical methods, including item analysis, reliability (Cronbach’s α), consistency corrected item-total correlation, alpha if item deleted and exploratory factor analysis for data-reduction.
The structural model for POPS was empirically derived and developed. Model development strategy was used to choose the best from four measurement models, all 25 items mapping to one factor (perceptions of organizational politics), five factors retained, three factors retained (19 items), four factor retained (17 items). All were competing models. One measurement model was job stress. First order of hybrid model and second order of hybrid model forecast relationship between the perceptions of organizational politics and job stress. The result of hybrid model factors produced the most parsimonious set of items of acceptable reliability as the best model chosen. Models needed to follow, model specification, parameter estimates ( factors were analyzed and mean values for each item across the factors were calculated ), model identification, t-rule and overall model-fit (e.g., absolute fit indexes: χ2, df, GFI, SRMR, RMSEA & ECVI; relative fit indexes: NFI, NNFI, CFI, IFI & RFI; parsimonious fit indexes: PNFI, PGFI, AIC, CN & Normed χ2) using the available LISREL indicators.
The result of using discriminating validity measure ”supervisor behavior” and “coworker and clique behavior” indicates that they are different constructs. Criterion-related validity measure is also distinct form criterion. Discriminate validity measures three factor and four factor model which are unidimension. The chi-square difference test comparison of three and four factor model is significant. Indicators of AIC, ECVI, CAIC and double cross-validation indicate the first order hybrid model is the best model chosen by the study. Loose replication strategy and tight replication strategy of cross-validation indicate model effect of the cross-sample and cross-circumstance and extension to other sampling population.