社会调查设计与数据分析（孙永强）001-信息系统研究行为学派.pptVIP

反射与组成测量 (Petter et al. 2007) Reflective Formative Causal direction Latent ? Observable Observable ? Latent Interchangeability Yes No Co-vary Suggested Not suggested Reputation of WHU’s SIM Reputation of WHU Highly Rated Listed in best Recom. SIM Repu. EMS Repu. … It is an issue of how to measure it rather than what the construct should be. * 一阶与二阶变量 Reputation of WHU SIM Repu. EMSRepu. … Second-order construct First-order construct Items Second-order construct as a strategy for balancing simplicity and comprehensiveness. The relationship between first- and second-order constructs can be reflective or formative. Selection between first- and second-order construct: abstraction level and research objective (e.g., social capital). * 测量与结构模型 IV DV Measurement Model Measurement Model Structural Model Measurement Model Structural Model Relationships Latent - observable Latent - Latent Testing issues Reliability Convergent validity Discriminant validity Path coefficients R-square Fit index * 探索性与验证性因子分析 V1 V2 V1 V2 EFA CFA Latent-observable relationship Unspecified Specified Software SPSS PLS, LISREL * 基于方差与协方差的结构方程模型 Model Structure Match IV DV Data Structure PLS-SEM (Partial Least Squares): Pursue the max. R-square explained. Make the DV more predictable. CB-SEM (Covariance-based): Pursue the match between data and model. Make the model more reliable. * 工具选择标准 Reasons for using PLS-SEM Reasons for using CB-SEM Exploratory study Confirmatory study Small sample Large sample (CB-SEM assumption) Formative measures Reflective measures Formative measures with two emitting paths (specification issues). Abnormal distribution Normal distribution (CB-SEM assumption) Need to use latent variable scores Require a global goodness-of-fit If CB-SEM is applicable, taking CB-SEM as the priority (exploratory – confirmatory goal is questionable) – more rigor in algorithm. PLS-SEM can be a good approximation of CB-SEM even when the study is confirmatory in n