信息系统研究方法结构方程模型120130402IntructiontoSEMLiuLuning

Seminar on Research Methods in Information Systems,Introduction to Structural Equation Modeling,6/14/2019,1,刘鲁宁 哈尔滨工业大学,Todays Agenda,Structural Equation Modeling Methods Two-Step Approach to SEM,6/14/2019,2,Structural Equation Modeling Methods (based on Gefen et al. 2000, Rigdon, 1998, and Raykov and Marcoulides, 2006),6/14/2019,3,6/14/2019,4,Structural Equation Modeling,What is Structural Equation Modeling (SEM)? SEM is also called causal modeling, latent variable structural equation (LVSE) modeling, and analysis of covariance structures 因果模型、潜变量结构方程、协方差结构分析 It is a family of methods for representing, estimating, and testing a theoretical network of (mostly) linear relations between variables 基于理论的变量之间的线性关系 Those variables may be either observable or directly unobservable, and may only be measured imperfectly. 变量可能是可观测的、不可直接观测的、不完美度量的 SEM is a generalization of both regression and factor analysis, and subsumes most linear modeling methods as “special cases.“ SEM是回归和因子分析的集合体，是线性建模方法的特例,Structural Equation Modeling,The main idea behind SEM SEM proceeds by assessing whether a sample covariance or correlation matrix is consistent with a hypothetical matrix implied by the model specified by the user. 样本的协方差或相关系数矩阵 Vs 用户指定模型的假设矩阵 The inputs to SEM are either raw data or sample moments computed from the data, and a model to be evaluated. SEM的输入是原始数据、样本数据、模型 The sample moments will include either variances and covariances or correlations. 样本数据包括方差、协方差、相关系数 The model consists of a network of proposed equations, with some parameters fixed to particular values and others “free to be estimated.”一个特定的方程组、对某些参数进行赋值,6/14/2019,5,Structural Equation Modeling,The main output of SEM Estimates of the designated model parameters参数估计 For the dependent variables, estimates of the proportion of variance explained, often called squared multiple correlations (or SMCs), which are akin to the R2 statistic in regression复相关系数 Overall goodness-of-fit statistics, which assess the overall consistency between the specified model and the data拟合优度 Diagnostic statistics, which aid in pinpointing the sources of any fit problems. 诊断统计,6/14/2019,6,6/14/2019,7,Benefits of Structural Equation Modeling,SEM offers convenient and flexible modeling The most important feature of SEM is that it is designed for working with multiple, related equations simultaneously. 允许多个相关方程的同时运算 It allows reciprocal relationships to be modeled and allows the disturbances for different equations to be either correlated or uncorrelated. 允许相互关联关系方程的联立 The methodology also allows researchers to compare the performance of a model across multiple populations. 允许一个模型多次被测量,Benefits of Structural Equation Modeling,Explicit modeling of measurement error SEM allows researchers to explicitly recognize the imperfect nature of their measures.允许测量的不完美 When measuring abstract psychological variables such as “attitude” or “intention,” researchers accept that the observed value is the combination of true value plus measurement error. 当测量抽象的心理变量时，观测值是真实值加上测量误差项,6/14/2019,8,Benefits of Structural Equation Modeling,Resolution of multicollinearity解决多重共线性问题 Researchers often include multiple items in a questionnaire in an attempt to guarantee that key variables are tapped appropriately. These multiple measures tend to be highly correlated. Including these multiple measures as predictors in the same regression equation leads to the familiar consequences of multicollinearity, including biased parameter estimates and inflated standard errors. In SEM, multicollinearity is no longer an issue!,6/14/2019,9,Benefits of Structural Equation Modeling,Evocative graphical language图形化的语言 The development of SEM as a statistical method has been accompanied by the development of an evocative graphical language. This language provides a convenient and powerful way to present complex relationships to others not familiar with SEM.,6/14/2019,10,Benefits of Structural Equation Modeling,6/14/2019,11,The TAM model,Benefits of Structural Equation Modeling,6/14/2019,12,The ERP assimilation model,How to Use SEM for Research,Conceptualization of research model Proposal interesting and important research questions.提出研究问题 Identify an overarching theory or theories.明确理论和假设 Identify dependent variable(s).明确因变量 Identify independent variables.明确自变量 Understand that the functional form of the relationships between IVs and DVs are assumed to linear in SEM.明确因变量和自变量之间的线性函数关系,6/14/2019,13,How to Use SEM for Research,Choose measures for variables The measurement items must be valid, meaning that they are conceptually associated with the variable or construct.有效的，度量能代表变量 The measurement items must be reliable, meaning that they must be relatively free from error. 可信的，测试结果的一致性，不受误差干扰 The measurement items must be unidimensional, meaning that they all load to the same variable or construct.度量是一维的 Convergent validity (correlation with each other)一个构念的多重指标彼此间聚合 Discriminant validity (correlation with others)一个构念和另一个构念之间彼此区别开来,6/14/20