Please enable JavaScript.
Coggle requires JavaScript to display documents.
1.Principle Component VS 2.Exploratory factor analysis (2.2 EFA:假设common…
1.Principle Component VS 2.Exploratory factor analysis
2.2 EFA:假设
common factors are independent variables; specific factors(error) 均值=0。
假设
所有itemsX和factors的variance=1。
•The
communality of X ---amount of variation of the variable explained by the common factors
(the common or shared variances).
•
1-communality=V(error)=1—入的平方之和(
此X对所有common factors的 入)
•
入The factor loading element is the correlation between X and 此factor
表格“commonality”
看extraction,表示一个X被选出common factors的所解释的X的variance
表格“
Correlation Matrix” VS “Reproduced correlations”
前者是X-items之间的observed correlations
后者有(1)X-items在extraction后,各自的commonality
(2)X-items之间的reproduced correlations, 与observed比较的residual
表格“Total Variance Explained”:
每个factor解释的variance。 左边Extraction Sums of Squared Loadings针对principal methods 右边Rotation Sums of Squared Loadings针对factor analysis。总variance不变,但是每个factor解释的variance会在rotate后改变
表格 Rotated factor matrix--对应factors之间不相关的。表格 Pattern factor matrix--对应factors之间有相关性。以上两个表格确定哪些X-items去load哪个factor
Variance(X)=1=common factors的系数(入)的平方之和+错误的平方(specific factor的variance)
确定common factors 1) K方法-- Factor在rotation后解释的variance大于1,就可以成为。2)scree plot
Principle component VS Factor analysis
principal axis factoring
PC:增加variance---得到equation组合
,但不去区分是common factors还是specific factors
rotation of PC ----求得 error variance---直至items的 error or specific factor之间的correlation=0
假设:implied correlation约等observed correlation
看 rotated factors
旧的factors---尽量和更少的item相关---到新的factors
factors相关:观察oblique factor rotation 而不是orthonormal rotate
2.1Factor score coefficients:
每个X--item=系数✖️common factor 这里的系数是factor score系数,要从factor score coefficient table里找。它与PC分析中 u matrix 中的系数不同。 注意方法是:MLE(maximum likelihood estimation--针对CFA)or PAF(principal axis factoring--针对EFA)
已知X1...Xp,怎么估计factor1....factor c? factor score coefficient : b作为系数,形成 线形组合
1.1 Umatrix: u✖️入的开方=f (CM),方法principle component,求得Z与X的线性equation有两个条件:1)使得Z取得the largest possible variance 2)equation里每个X的系数u的平方的总和=1 重要:3)这些Z(X的线形组合)之间的correlation=0
注意虽然求得的所有Z之间的correlation不为0,X之间是的correlation coefficient是不都为0的,这也是进行principle component的必要性。
X的独立性检验
:
X的独立性检验
: all the items X1...Xp are independent ---卡方检验 CV−BM3
Zij与Xij的correlation(component matrix Zj/ matrix of factor loadings) fij
= uij ✖️(square root o f 入)
入是这个factorZ最大分到的variance
Commenalities
选择 入 大与1的factors Z留下。equation:Xi=u1Z+u2Z....这些选出的Z 与Xi的每个fij的平方,再求和。叫做communality o f Xi,或sum of squares of factor loadings
pick pup factors1scree plot. 2.kaiser's rule--->1
贯穿1,2,3章的---X是标准化后的数据: X=【Q-E(Q)】➗SD(Q)所以两个X1,2之间的correlation coefficient:r12=每组(X1✖️X2)的总和,再除以(n组-1);如果不是标准化的数据,要看covariance表。------发现X--items share common things, so try to extract factors by increase variance
Z---对应principle方法;K si表示factor analysis