I’m taking a multi-variate data analysis course this semester and this video was so very helpful. I could follow along in SPSS and the explanations along the way were great. Thank you.
Thanks for this video, Dr. Crowson. As I said before, I am a BIG fan of you. Can you do me a favor? I would like you to upload a tutorial example of EFA using MPLUS. This is because I am a Mplus user. Please give me your help!!
@Mike Crowson thank you for this insightful video. I really learnt a lot from your presentation. Please, I kindly want to know after identifying what items load unto which component or factor. what is the next level of analysis to bring all the associated items together into a variable.
Hi Vincent, thanks for your comment and question. From what you are asking about bringing "all the associated items together", it sounds like you are asking about creating full scale (or composite) scores used to measure individuals' responses to the latent factors (constructs). Typically, you create composite measures when you seek to study/test relationships between these and other variables. The factor analysis gives you an idea of which variables you might combine in future analyses and are generally the basis behind the scoring procedures authors provide with respect to their scales (in other words, the factor analysis facilitates the creation of a rule for which items to combine into a composite measure of a construct). Thus, in future studies (or perhaps the same study), you use that rule to create full or perhaps subscale scores that can be included in additional analyses, such as correlation, linear regression, ANOVA, etc. When you read methods sections for quantitative studies that include various scales designed to measure different constructs, they often (but not always) contain information about which items are being used to measure which factor. The decision about which items go with which factors is often (but not always, since sometimes they develop the rule based solely on logical instead of empirical grouping) based on earlier factor analytic work. If you are curious about the mechanics of creating composite scores, here is a video I put together awhile back: ruclips.net/video/9jj-oPcu23M/видео.html Sometimes, a researcher will perform the factor analysis and then save the factor scores from that analysis, with those scores serving as the input to other analyses (such as linear regression). An issue with this approach is that it does not really pivot off any rule generated (as described above) for combining the items; the factor scores are not only based on the salient items on a factor on ALL items (irrespective of whether the loadings are trivial or salient). In other words, the factor scores are still defined largely as a function of the most salient loading items; but also the other items are still contributing to the computation of those scores (even if their loadings are small). This might be viewed as unpalatable for those who seek to use these scores in subsequent analyses. Additionally, this approach can be problematic in that it would necessitate a new factor analysis and saving factor scores in every study. This can be problematic in particular if you have smaller sample data that does not lend itself to factor analysis; or the possibility of factor scores reflecting sampling nuances where the factor definition might change from study to study. In general, I most researchers (at least in my areas of psychology and education) use the factor analysis to develop a rule for combining items, and then (assuming they feel they can trust the factor analytic results enough) use that rule to compute composite scores that are then used to carry out other statistical analyses to test pertinent hypotheses, etc. If your exploratory factor analysis is an initial step during scale validation, a 'next phase' may be to collect new sample data and to perform a confirmatory factor analysis (CFA). With the CFA, you are submitting the factor model you develop for EFA to further scrutiny. Unlike EFA where all items are allowed to load on all factors, CFA generally proceeds by allowing factor loadings to be estimated for those items that are 'supposed' to be associated with particular factors (based on the rule you develop during EFA) and fixing the remaining loadings to 0. This is a more restrictive model that typically should fit less well to the data than the more unrestrictive EFA model. However, it is a stronger test of your factor model. If this holds up well in a new sample and under the more restrictive conditions, then you have greater support for your claims regarding the factorial structure underlying the items in your measure. Some folks try to use this process as a basis for claiming they have proven the validity of the scale. However, validity is not something you can prove with a few studies. Building a strong case for validity typically involves other forms of evidence (with factor analytic evidence providing some information bearing on your claims). As you can see, there are many roads you can take following an EFA, and those roads generally depend on the questions you are seeking to answer in your study. I hope this answers your question, Vincent. Cheers!
Thanks so much for this! Just had one question- when I try to run the syntax on my SPSS (changing only the variable names) it comes up with the following error message: 183 GET The GET command does not start with the FILE subcommand. The FILE subcommand is required and must be the first subcommand. I was wondering if you might be able to advise on this?
Thank you Prof for this tutorial. After doing factor analysis , I got the determinant value as 2.949E-20. How can this be interpreted? Is the data fit for Factor analysis?
Super great presentation and learning experience. First time doing EFA, so I am grateful you made this video available. Thank you!!! ❤
Thank you, Dr. Crowson, for your invaluable insights. Your guidance has significantly enhanced my comprehension of EFA.
I’m taking a multi-variate data analysis course this semester and this video was so very helpful. I could follow along in SPSS and the explanations along the way were great. Thank you.
Thank you Dr. Crowson. This really helped me understand EFA a lot better.
Thanks for the wonderful and exciting presentation @DR Crowson
You are very welcome!
Great presentation, Dr.
Thank you so much for this video! This was so well explained!
Marvelous. This video has helped me a lot. I subscribe to the channel. Explains very clearly!!!!!
It's a fantastic tutorial! Many of my questions about EFA have been cleared. May I ask how you often deal with the cross-loading indicators?
Thanks for this video, Dr. Crowson. As I said before, I am a BIG fan of you. Can you do me a favor? I would like you to upload a tutorial example of EFA using MPLUS. This is because I am a Mplus user. Please give me your help!!
@Mike Crowson thank you for this insightful video. I really learnt a lot from your presentation. Please, I kindly want to know after identifying what items load unto which component or factor. what is the next level of analysis to bring all the associated items together into a variable.
Have always asked this question. @Mike Crowson, Kindly offer assistance
Hi Vincent, thanks for your comment and question. From what you are asking about bringing "all the associated items together", it sounds like you are asking about creating full scale (or composite) scores used to measure individuals' responses to the latent factors (constructs). Typically, you create composite measures when you seek to study/test relationships between these and other variables. The factor analysis gives you an idea of which variables you might combine in future analyses and are generally the basis behind the scoring procedures authors provide with respect to their scales (in other words, the factor analysis facilitates the creation of a rule for which items to combine into a composite measure of a construct). Thus, in future studies (or perhaps the same study), you use that rule to create full or perhaps subscale scores that can be included in additional analyses, such as correlation, linear regression, ANOVA, etc. When you read methods sections for quantitative studies that include various scales designed to measure different constructs, they often (but not always) contain information about which items are being used to measure which factor. The decision about which items go with which factors is often (but not always, since sometimes they develop the rule based solely on logical instead of empirical grouping) based on earlier factor analytic work. If you are curious about the mechanics of creating composite scores, here is a video I put together awhile back: ruclips.net/video/9jj-oPcu23M/видео.html
Sometimes, a researcher will perform the factor analysis and then save the factor scores from that analysis, with those scores serving as the input to other analyses (such as linear regression). An issue with this approach is that it does not really pivot off any rule generated (as described above) for combining the items; the factor scores are not only based on the salient items on a factor on ALL items (irrespective of whether the loadings are trivial or salient). In other words, the factor scores are still defined largely as a function of the most salient loading items; but also the other items are still contributing to the computation of those scores (even if their loadings are small). This might be viewed as unpalatable for those who seek to use these scores in subsequent analyses. Additionally, this approach can be problematic in that it would necessitate a new factor analysis and saving factor scores in every study. This can be problematic in particular if you have smaller sample data that does not lend itself to factor analysis; or the possibility of factor scores reflecting sampling nuances where the factor definition might change from study to study.
In general, I most researchers (at least in my areas of psychology and education) use the factor analysis to develop a rule for combining items, and then (assuming they feel they can trust the factor analytic results enough) use that rule to compute composite scores that are then used to carry out other statistical analyses to test pertinent hypotheses, etc.
If your exploratory factor analysis is an initial step during scale validation, a 'next phase' may be to collect new sample data and to perform a confirmatory factor analysis (CFA). With the CFA, you are submitting the factor model you develop for EFA to further scrutiny. Unlike EFA where all items are allowed to load on all factors, CFA generally proceeds by allowing factor loadings to be estimated for those items that are 'supposed' to be associated with particular factors (based on the rule you develop during EFA) and fixing the remaining loadings to 0. This is a more restrictive model that typically should fit less well to the data than the more unrestrictive EFA model. However, it is a stronger test of your factor model. If this holds up well in a new sample and under the more restrictive conditions, then you have greater support for your claims regarding the factorial structure underlying the items in your measure. Some folks try to use this process as a basis for claiming they have proven the validity of the scale. However, validity is not something you can prove with a few studies. Building a strong case for validity typically involves other forms of evidence (with factor analytic evidence providing some information bearing on your claims).
As you can see, there are many roads you can take following an EFA, and those roads generally depend on the questions you are seeking to answer in your study. I hope this answers your question, Vincent. Cheers!
Hi Kwame, I just responded to Vincent's question. I hope you find it useful. Cheers!
@@mikecrowson2462 Thank you for the elaborative response. I hope to get the platform to interact with you more on statistical analysis.
@@mikecrowson2462 Thank you, I have been wondering about this for a while.
Thanks so much for this! Just had one question- when I try to run the syntax on my SPSS (changing only the variable names) it comes up with the following error message:
183 GET The GET command does not start with the FILE subcommand. The FILE subcommand is required and must be the first subcommand.
I was wondering if you might be able to advise on this?
Thank you Prof for this tutorial. After doing factor analysis , I got the determinant value as 2.949E-20. How can this be interpreted? Is the data fit for Factor analysis?