SEM in AMOS when you have incomplete data (new, 2018)
Вставка
- Опубліковано 2 тра 2018
- This video provides an overview of SEM (using path analysis) in AMOS when you have missing data. It demonstrates how to estimate the basic model using FIML estimation (enacted by clicking on Estimate Means and Intercepts under Analysis Properties). It also demonstrates the use of the Regression Imputation approach to generate a complete dataset (which will allow you to use other options in AMOS such as Modification indices and Bootstrapping). The video does not cover the theory behind missing data mechanisms or the approaches recommended with particular patterns of missingness. The viewer is encouraged to read more on those topics.
A copy of the SPSS data file used in the video can be downloaded here: drive.google.com/file/d/18azD...
A copy of the .amw file containing the path model generated for the video can be downloaded here: drive.google.com/file/d/10ryZ...
A pdf copy of the page referenced in the video on interpreting fit statistics can be obtained here: drive.google.com/file/d/14zM-...
You can also access another video on dealing with missing data by going here: • Using multiple imputat...
For more instructional videos and other materials on various statistics topics, be sure to my webpages at the links below:
Introductory statistics:
sites.google.com/view/statist...
Multivariate statistics:
sites.google.com/view/statist...
Dr. Crowson, you are providing a much needed and appreciated service of visually walking us through the details of various SEMs. As a PHD student in Ed research and measurement, I visit your UA-cam videos more than I’d like, but doing so alleviates the anxiety of trial and error. Thank you. You rock !!!!
You may have saved my life with this video. Thank you very much.
This is a very useful video, saved my morning. Thank you so much!
I really like your videos.
when you find let say invalid characters on the data, you go to the file and fix the error, save and upload a new file and run it again, but it is still showing error signs, then the AMOS is still using the previous file, although you think that have uploaded the new file. So every time I found error on the data and edit it, I close the entire program and re do it, otherwise SPSS might give you incorrect results, so just to be safe, close the entire program and redo it all over again. No need to make new construct, use the construct you saved and upload the new data once its been reopened. Crying over the weekend.
Great video but would it be possible to ask a question? In one area of my data there is a group that wasn't asked specific questions regarding post separation manipulation of children as they didn't have children. If I still want to run the analysis including this particular variable (it's important) and the full data set (those with children and those without) do I just replace the blank data as zeros in order that they are not treaded as missing- although this may affect the results? Or do I have to do separate tests (with/without children). Sorry to ask, I'm new to AMOS & Path analysis, but if you could point me in the right direction I would be most grateful.
How can I turn my data into a complete data if I get this issue?what is missing?
Dear sir, in my data set 200 rows and 15 columns, but when I run the CFA the red result tab is not glowing and showing xx default model. Please help me to know the reason. Jena
Hi, great video! I had a few questions: 1) When using ML estimation, how does AMOS estimate the means and intercepts? How is it account for the missing data? Is it using Maximum likihood (a model based method) to estimate the missing data and the model fit at the same time? 2) Also, when you did the data imputation, were the missing data estimated using single imputation methods (i.e., regression) or model based methods (i.e., multiple imputation)? It is not clear to me how regression is being used to predict missing data. Thank you so much in advance for your help!
Hi. Thanks for the feedback. When you run your analysis on complete data, the estimation is standard maximum likelihood. When there is missing data and you select "Estimate means and intercepts" AMOS invokes Full information maximum likelihood (FIML). There's are nice discussions of FIML as well as the regression approach here (www.theanalysisfactor.com/missing-data-two-recommended-solutions/), here (www.spss.com.hk/amos/missing_values_application.htm) and here (stat.utexas.edu/software-faqs/general#missingincompletedata). The AMOS manual (go to: ftp://public.dhe.ibm.com/software/analytics/spss/documentation/statistics/24.0/en/amos/Manuals/IBM_SPSS_Amos_User_Guide.pdf) also presents comprehensive discussion of different imputation methods in examples 30 and 31. I'll be honest, I'm not a huge fan of the regression approach I used in the example, but you have to have a complete dataset when using AMOS to generate boostrap standard errors and CI's for testing indirect effects using standard ML estimation (and I was trying to keep things fairly simple in the example). AMOS does provide an avenue for using multiple datasets (see AMOS manual examples referenced above) using model-based procedures. But unfortunately there is no automated way in AMOS to combine results across multiply imputed datasets. .Hope this helps!
Hi, I wanted to follow up and let you know that I just added another video on using multiple imputation: ua-cam.com/video/eWHqxrsI0NI/v-deo.html
Dear Dr. Crowson, thank you very much for all of the help! I look forward to watching your new video.