factoranalysis(因素分析).docVIP

Factor Analysis (因素分析) This is something new to you. It is also very useful if you use survey as your tool for the thesis. So get ready … Factor analysis (FA) is a variable reduction technique. Very often we have a lot of variables and we need to know if they are all useful for some analysis purpose (such as regression). But do we need them all? Is it possible that some of them may actually be grouped together and be represented by one variable? For example, if you have the test scores of language (ex. Chinese and English) and analytical ability (ex. math and science), could it be possible that these scores may refer to your IQ? Or different kind of IQ? Factor Analysis will help to identify this question. Factor Analysis: To extract J latent variables (潛在變數, Y1, Y2, …, YJ) from K observed variables (X1, X2, …, Xk), JK. The latent variables are usually not easily observed, and have to be obtained by the observable X variables. The model is Xi=fi1Y1 + fi2Y2 + …+ fiJYJ + ei ( looks like a regression model, right?) Where fij: pattern loading (組型負荷量), a regression coefficient. ei: the unique value for the ith variable Xi. If we draw the relationship, it looks like this: IMPORTANT: We assume that all the factors are independent, or they are orthogonal to each other. There are two kinds of FA: Exploratory FA (EFA): used to see if we can reduce the number of variables to a small sets of variables. Confirmatory FA (CFA): We should first have some theoretical relationship to support the existence of the latent variables Ys and some observed variables Xs that we can actually measure, and the Xs are used to evaluate the Ys. This is very often the case with survey research. We use questions (Xs) and ask people to get their values on the Xs and then use CFA to confirm that the Xs do support the measure of Ys. We will see examples like this later. Example: An IQ measure (WISC). See the file “factor.sav”. What are the factors? Analyze/Data Reducti