Exploratory Factor Analysis with STATA 15

-stata初心者向け、初めてEFA（探索的因子分析をするとき）

−参考になる動画↓

STATA15で初めてEFAを行った時の記録です。

For factor analysis(前提条件）
Needed cases:
　　[Rule of 10] 10 cases for every item
　　[Rule of 100] number of respondents should be larger of a) or b).
　　　　a)5 x # of variables
　　　　b)100
Least number of variables for EFA= <3
Normality? —not always but?

＞＞ここからEFA(Exploratory Factor Analysis: 探索的因子分析

1）データをimportする

TOPバーからimport

import delimited /Users/*****/Desktop/***/data/*****.csv

2)いらないvariableを指定する

　drop v17

3）どんなdataかを見る
　describe
　summarize( Variable | Obs Mean Std. Dev. Min Max）

4）相関(correlation)を見る
　correlate

5）Test for Basic Assumptions(初めての時はこれをしないとエラーが出る）　　　　　　　　　　　　　　　　　　　　　　　　　

　findit factortest

それから
　factortest ae1-tc4 

  1）Bartlett’s Test of Sphericity (with large Ns tend to be statistically sig)
　　-compares the (our) correlation matrix to an identity matrix.
　　[identity matrix] is a correlation matrix with 1.0 on the principal diagonal 　　　　　　　　　　and zeros in all other correlations. (Null)
　　-p-value >0.05 (significant): indicates sufficient intercorrelations to 　　     　　conduct the factor analysis
  2)Kaiser-Meyer-Olkin(KMO) measure of sampling adequacy
　　-(Generally) indicates whether or not the variables are able to be 　　　　grouped into smaller set of underlying factors.
　　-Answers the question: Does the data factor well?
　　-Varies from 0 to 1
　　-Value at least .50.(or more than .60) larger values are better. Higher 　　　values indicate overlap.
＊Indicates amount of overlap or shared variance btw pairs of variables.
We want to see some overlap (but not too much that it would produce spurious results) so that the items overlapping enough so that they move together but unique.

6-1）Principle factor(default) (本番)
Statistics>Multivariate>Factor and Principal component analysis> Factor analysis—factor ae1 ae2 ae3 ae4 ae5 as1 as2 as3 as4 as5 tc1 tc2 tc3 tc4
（結果）Factor #  | Eigenvalue, Difference, Proportion, Cumulative

6-2)Principle component factor analysis
Statistics>Multivariate>Factor and Principal component analysis> Factor analysis>model2>principal-component factor

7)improve interpretability (読みやすくする)

rotate　
Statistics>Multivariate>Factor and Principal component analysis> postestimation＞ “orthogonal” (Varimax)　but not always 

Steven(2009) :retaining factor loadings of <0.4 

0.2 difference between factor loadings for it to be unique

8) もっと読みやすくする　

（初めての時は）findit sortl (sort by loadings)　sortlをゲットしたらsortl

9)check  internal validity (内的妥当性) 

Statistics>Multivariate>Cronbach’s alpha
Scale reliability coefficient: 0.8439 　　　( 0.8<a<0.9 (intercoliniarity?)