求助潜变量统计值

基本懂了，谢谢各位！

在Google学术上找到一个英文的问答题：http://www-01.ibm.com/support/docview.wss?uid=swg21600240

Obtaining Factor Score Weights in AMOS and applying them
Technote (FAQ)

Question
Can I save estimates of scores on the latent variables, i.e. factors, back to the data file? Can I display factor score weights in my output? If so, how can I apply these weights to estimate factor scores in a raw data set?

Does the formula for combining factor weights to form factor scores remain the same even when factor loadings are estimated 'with intercepts'? Are the weights affected by means and intercepts constraints in a multiple group model?
Answer
AMOS can not save factor score estimates to the original data file that was analyzed by AMOS. However, AMOS 6 and later versions can use imputation to save the estimated factor scores, along with the observed variables, to a new data file. Also, factor score weights can be printed in the AMOS text output. These weights can be applied in another program, such as SPSS, to estimate factor scores.
To impute factor scores to a new file, Choose 'Data Imputation' from the bottom of the Analyze menu in AMOS. In the 'AMOS Data Imputation' dialog that opens, there are 3 imputation methods available. The default method is regression imputation, which will return the same factor scores that would be computed by applying the factor score weights in SPSS compute commands, as described below. Values will also be imputed for missing data on the observed variables. The alternate imputation methods are stochastic regression and Bayesian imputation.. Each of these alternate methods will allow you to make a single imputation or multiple imputations. If you choose multiple imputations, you can save the imputed values to multiple files or a single file.

The data imputation dialog will list the 'Incomplete data files' and 'Complete Data Files' in a box at the bottom of the dialog. The 'incomplete' files are the input data files for the AMOS model. The 'Complete' files are the files that will be created by the imputation process and will contain the latent variable and observed variable scores. Missing values in the observed variables in the input data will be replaced by imputed values, while nonmissing values among the observed variables are simply copied from the original data files to the 'complete' files. The latent variable scores will
be imputed according to the model.

AMOS assigns a default name for the complete file and saves it in the same directory as the original data. You can change the directory and file name by clicking the File Names button, choosing a directory from the 'Save in' scroll bar, changing the file name and/or type in the 'File name' box and clicking the Save button to return to the AMOS Data Imputation dialog. Click the Impute button to perform the imputation. AMOS will alert you that the new files have been created. You can then click the OK button for that alert and then close the AMOS Data Imputation window.

To request the printing of factor score weights from AMOS Graphics, open the View menu (View/Set in early versions of AMOS) and choose Analysis Properties. Click the Output tab and check the box beside 'Factor score weights' on the right side of the Output dialog. See Example 6 of the AMOS User's Guide for an example with related screen shots. The VB.NET command to request the weights is:

Sem.FactorScoreWeights()

See Example 6C in the AMOS User's Guide for an example of an AMOS syntax program with the request included.

Note that the factor score weights should be applied to the unstandardized observed variables in the raw data file. Based on their experience with factor analysis procedures in general statistical packages (such as the FACTOR procedure in SPSS), analysts often expect that they will need to transform the observed variables to standardized scores before applying the factor score weights. This is not the case with the weights generated by AMOS.

Amos converts all scores (on both measured and unmeasured variables) to deviation scores before estimating factor score weights. That is, it gives each variable a mean of zero. After that pre-treatment of the scores, the intercept is zero. You are not told what the intercept would be if you used raw scores.

Any change that you make to a model that affects the implied covariance matrix can affect the factor score weights. In a multiple-group factor analysis model, constraining the factor means to be equal across groups would affect the implied covariances in both groups, and so it would affect the factor score weights.

Here is an example of the application of factor score weights with a set of SPSS commands. It is based on Example 8 in the AMOS User's Guide. The example file, Ex08.amw, was opened in AMOS and a request for factor score weights was added. The weights for this example are reprinted below.

Factor Score Weights

wordmean sentence paragrap lozenges cubes visperc
-------- -------- -------- -------- -------- --------
verbal 0.118 0.176 0.371 0.010 0.013 0.010
spatial 0.028 0.041 0.087 0.237 0.293 0.234


To compute the factor scores in the raw data file for Example 8, Grnt_fem.sav, that file was opened in the SPSS Data Editor and the following SPSS commands were run: The observed variables were centered (i.e. each variable's mean was subtracted from the raw measurements on that variable) so that the factor scores had means of 0.


* Use Aggregate to save observed variable means to the active data file.
AGGREGATE
/OUTFILE=* MODE=ADDVARIABLES
/BREAK=
/visperc_mean=MEAN(visperc)
/cubes_mean=MEAN(cubes)
/lozenges_mean=MEAN(lozenges)
/paragrap_mean=MEAN(paragrap)
/sentence_mean=MEAN(sentence)
/wordmean_mean=MEAN(wordmean).

* Compute centered versions of the observed variables as visperc_c to wordmean_c .
do repeat raw = visperc cubes lozenges paragrap sentence wordmean
/avg = visperc_mean cubes_mean lozenges_mean paragrap_mean sentence_mean wordmean_mean
/inp_c = visperc_c cubes_c lozenges_c paragrap_c sentence_c wordmean_c .
compute inp_c = raw - avg .
end repeat.
execute.

* compute factor scores with centered input variables .
COMPUTE verbal = 0.118*wordmean_c + 0.176*sentence_c + 0.371*paragrap_c
+ 0.010*lozenges_c + 0.013*cubes_c + 0.010*visperc_c .
COMPUTE spatial = 0.028*wordmean_c + 0.041*sentence_c + 0.087*paragrap_c
+ 0.237*lozenges_c + 0.293*cubes_c + 0.234*visperc_c .

EXECUTE .

The estimated factor scores, verbal and spatial, have now been added to the active data file. That file will need to be resaved (or saved to a new name) to retain the scores for future analyses.

使用AMOS里的 data imputation 可以直接得到 潜变量的估算值  
此外，在OUTPUT中，也可以列出“factor score weights”

法鼓山 发表于 2016-4-18 16:19
使用AMOS里的 data imputation 可以直接得到 潜变量的估算值  
此外，在OUTPUT中，也可以列出 ...

data imputation 请问这个在AMOS哪里可以找到呢？

法鼓山 发表于 2016-4-18 16:19
使用AMOS里的 data imputation 可以直接得到 潜变量的估算值  
此外，在OUTPUT中，也可以列出 ...

请问请问，output列出factor score weights 之后，在哪里可以看到结果呢？

法鼓山 发表于 2016-4-18 16:19
使用AMOS里的 data imputation 可以直接得到 潜变量的估算值  
此外，在OUTPUT中，也可以列出 ...

很棒！！

这个帖子很棒，从理论到实践都说明白了，我补充一下实操部分，data imputation在AMOS顶部菜单栏的Analyze下，在跳出的对话框里直接点impute，AMOS会直接输出一个sav文件，最后的几个变量即潜变量得分。output里查看因子得分权重在Esitimates下的Matrices里，每一个潜变量的得分等于纵列的每一个观测变量乘以对应的系数之和，比较了一下手动计算结果和data imputation的结果，有些许差异，原因在于手动计算采用的是保留小数，data imputation用的是内部原始数据。