Amazing video! I just had a few questions: 1) what made you decided to use varimax rotation first 2) after varimax why did you change rotations and use promax (I know it's oblique, but why did you want to use oblique instead of orthogonal) 3) I have heard that you should use oblique rotations if your factors are correlated, but how correlated do they need to be to make you decide to use an oblique rotation (e.g., even one correlation above .4) 4) the two items that didn't load on the factors (for the promax rotation) did you just delete those from the final solution then? Thanks!!
Hi there. I just saw your question. It is ok if the pattern matrix loading (assuming correlated factors following oblique rotation) is greater than 1. I believe it is possible. However, if you used orthogonal rotation, then the pattern matrix=structure matrix, which contains correlations between the latent variables and indicator variables. In that case, there should be no loadings > 1. Cheers!
Hi Mike, your video really was of great help to me. Thanks a lot. Just one question: How important is the determinant and can I possibly get more than the three decimals?
Thanks a million!!! Dr. Crowson. Your video is really helpful! I am wondering if there is any difference between Exploratory factor analysis and Principal component analysis. They look very similar to me!
Once the factors are identified and qualitatively named is there a way to compare the difference between the reduced factors in terms of importance (basically a significance test looking at factor differences). In other words is the reduced factor 1 different from the reduced factor 2 in importance. Sorry if this is a bit fuzzy of an idea.
That is a very interesting question. I've not seen this approach taken in the literature using EFA. However, what you are asking is not really different from what often done in the context of CFA (confirmatory factor analysis; using SEM). If you were doing CFA, you might pose both one- and two-factor models as candidate models, and then compare the fit of the two models using a chi-square difference test (to determine whether there is a significant difference in model fit between the two). In general, the chi-square value for the more restricted model (e.g., one factor) indicate worse fit than the less restricted one (e.g., two-factor). [chi-square value is higher in the worse fitting model] Since the former is nested within the latter, a chi-square difference test could be applied to determine if the one-factor model would yield a significant reduction in fit relative to the two-factor model. If not, then you could go with the more parsimonious (one-factor) model. To obtain chi-square results for each model requires maximum likelihood estimation, which is the default estimation approach in SEM programs. There is a maximum likelihood factor analysis option in SPSS, so in principle the steps I just discussed might be applied (but again, I haven't really seen this done using EFA). However, you might just as easily re-run your data using an SEM program and compare the two candidate models that way. I hope this helps!
Hi Mike, thank you for your video. you explained it really well. I was wondering if you could also make a video about Multivariate Multiple Regression using SPSS? Specifically, I have 2 DV and 5 IV ( plus 3 control variables). Looking forward to hearing from you!
Hi there. Thanks for your comment and request. I'll certainly put it on my to-do list of topics to cover. But really, if you are running multivariate regression with two dv's in SPSS, there is a good demo on page 124-128 of the multivariate text by Pituch and Stevens (2016). You are using syntax though. In their demo, they were regressing two dv's (PEVOCAB & RAVEN) onto three predictors (NS, NA, SS) using the MANOVA command (see below). MANOVA PEVOCAB RAVEN WITH NS NA SS/ PRINT = CELLINFO(MEANS, COR). Essentially, you follow the MANOVA command with the names of your DV's, then type WITH and then follow with the names of your predictors. Be sure to add the '/' to the line. The multivariate effect of the predictors on the outcome variables (as a set) is found in the portion of the output called "Effect...Within cells regression". In the following sections you essentially get the results of standard multiple regressions, where each DV from the multivariate regression is regressed onto the predictors in the model. Hope this helps.
Thank you for your detailed reply ! It is helpful however, I did not mention that my DV are categorical variables ( 5 categories per variable ) , when I do the MANOVA, I don’t get the results per category but instead SPSS assumes it’s a binary variable ( when I clearly defined it as categorical)? I’m a bit confused how to go about that.
@@abdc8240 Ah, I didn't know your DV's are categorical. As far as I know, you really can't perform a multivariate regression when the dv's are categorical (the approach assumes they are continuous) in order to obtain a multivariate effect of the IV's on the DV's. I don't know if the overall multivariate effect is what you are mainly interested in or if you are concerned about correlations among the dv's. If the issue is the latter, you might try separate multinomial (if your dv's are nominal) or ordinal (if your dv's are ordered categorical) logistic regression, where you regress one of your dv's on the IV's AND the other dv (the latter included as a control variable). Then do the same the other way around. I have videos on these topics here (ruclips.net/video/1BL5cL8_Cyc/видео.html) and here (ruclips.net/video/rSCdwZD1DuM/видео.html)
How to write hypothesis for above analysis... Any way nice explanation sir... Would please explain the result relating to hypothesis.. My question is.. When we choose null or alternative hypothesis on the basis of result
hello professor i have already distributed each questionnaire with different variables so when i do the analysis can i do it altogether or i need to do the analysis for each variables separately
Thanks,Sir Crowson. This was more than informative and thanks so much for adding the PPt which equally simplified and well explained. Kindly share your email address for further questions i might have for you.
In line with everyone else-- thank you for such a clear video and providing the citations for different decisions. This is excellent.
Fantastic video. I really need a refresher and this was perfect. Thank you!
HUGE HELP. Thanks so much. The parallel analysis link is particularly helpful.
Thank you for giving an access to data from example!
Thanks for sharing your expertise on EFA.
I wish you were my prof :( this saved my life
Amazing video! I just had a few questions:
1) what made you decided to use varimax rotation first
2) after varimax why did you change rotations and use promax (I know it's oblique, but why did you want to use oblique instead of orthogonal)
3) I have heard that you should use oblique rotations if your factors are correlated, but how correlated do they need to be to make you decide to use an oblique rotation (e.g., even one correlation above .4)
4) the two items that didn't load on the factors (for the promax rotation) did you just delete those from the final solution then?
Thanks!!
hi Mike, thanks for the video. May I ask what I should do/ conclude if 1 loading is greater than 1 in the pattern matrix?
Hi there. I just saw your question. It is ok if the pattern matrix loading (assuming correlated factors following oblique rotation) is greater than 1. I believe it is possible. However, if you used orthogonal rotation, then the pattern matrix=structure matrix, which contains correlations between the latent variables and indicator variables. In that case, there should be no loadings > 1. Cheers!
Very informative. Thanks a lot
Hi Mike, your video really was of great help to me. Thanks a lot. Just one question: How important is the determinant and can I possibly get more than the three decimals?
Very informative! Thank you!
Thanks for visiting, Javid! best wishes!
Dear Sir,
Thank you for this video. For Parallel analysis, what value should be filled in the tab "Number of random correlation matrices to generate"?
Thanks a million!!! Dr. Crowson. Your video is really helpful! I am wondering if there is any difference between Exploratory factor analysis and Principal component analysis. They look very similar to me!
Hi Bobbie, I just saw your posting. Here is a video link on PCA: ruclips.net/video/6Ycw3_iQHD0/видео.html . Cheers!
Once the factors are identified and qualitatively named is there a way to compare the difference between the reduced factors in terms of importance (basically a significance test looking at factor differences). In other words is the reduced factor 1 different from the reduced factor 2 in importance. Sorry if this is a bit fuzzy of an idea.
That is a very interesting question. I've not seen this approach taken in the literature using EFA. However, what you are asking is not really different from what often done in the context of CFA (confirmatory factor analysis; using SEM). If you were doing CFA, you might pose both one- and two-factor models as candidate models, and then compare the fit of the two models using a chi-square difference test (to determine whether there is a significant difference in model fit between the two). In general, the chi-square value for the more restricted model (e.g., one factor) indicate worse fit than the less restricted one (e.g., two-factor). [chi-square value is higher in the worse fitting model] Since the former is nested within the latter, a chi-square difference test could be applied to determine if the one-factor model would yield a significant reduction in fit relative to the two-factor model. If not, then you could go with the more parsimonious (one-factor) model. To obtain chi-square results for each model requires maximum likelihood estimation, which is the default estimation approach in SEM programs. There is a maximum likelihood factor analysis option in SPSS, so in principle the steps I just discussed might be applied (but again, I haven't really seen this done using EFA). However, you might just as easily re-run your data using an SEM program and compare the two candidate models that way. I hope this helps!
Hi Mike, thank you for your video. you explained it really well.
I was wondering if you could also make a video about Multivariate Multiple Regression using SPSS?
Specifically, I have 2 DV and 5 IV ( plus 3 control variables).
Looking forward to hearing from you!
Hi there. Thanks for your comment and request. I'll certainly put it on my to-do list of topics to cover. But really, if you are running multivariate regression with two dv's in SPSS, there is a good demo on page 124-128 of the multivariate text by Pituch and Stevens (2016). You are using syntax though.
In their demo, they were regressing two dv's (PEVOCAB & RAVEN) onto three predictors (NS, NA, SS) using the MANOVA command (see below).
MANOVA PEVOCAB RAVEN WITH NS NA SS/
PRINT = CELLINFO(MEANS, COR).
Essentially, you follow the MANOVA command with the names of your DV's, then type WITH and then follow with the names of your predictors. Be sure to add the '/' to the line.
The multivariate effect of the predictors on the outcome variables (as a set) is found in the portion of the output called "Effect...Within cells regression".
In the following sections you essentially get the results of standard multiple regressions, where each DV from the multivariate regression is regressed onto the predictors in the model.
Hope this helps.
Thank you for your detailed reply ! It is helpful however, I did not mention that my DV are categorical variables ( 5 categories per variable ) , when I do the MANOVA, I don’t get the results per category but instead SPSS assumes it’s a binary variable ( when I clearly defined it as categorical)? I’m a bit confused how to go about that.
@@abdc8240 Ah, I didn't know your DV's are categorical. As far as I know, you really can't perform a multivariate regression when the dv's are categorical (the approach assumes they are continuous) in order to obtain a multivariate effect of the IV's on the DV's. I don't know if the overall multivariate effect is what you are mainly interested in or if you are concerned about correlations among the dv's. If the issue is the latter, you might try separate multinomial (if your dv's are nominal) or ordinal (if your dv's are ordered categorical) logistic regression, where you regress one of your dv's on the IV's AND the other dv (the latter included as a control variable). Then do the same the other way around. I have videos on these topics here (ruclips.net/video/1BL5cL8_Cyc/видео.html) and here (ruclips.net/video/rSCdwZD1DuM/видео.html)
How to write hypothesis for above analysis... Any way nice explanation sir... Would please explain the result relating to hypothesis.. My question is.. When we choose null or alternative hypothesis on the basis of result
Excellent
superb cool
Do we need to interpret the component covariance matrix also? Thanks.
hello professor
i have already distributed each questionnaire with different variables so when i do the analysis can i do it altogether or i need to do the analysis for each variables separately
HI Dr,I have great thanks for the video.Could you make some video showing us about using DHS data for analysis?
Thanks,Sir Crowson. This was more than informative and thanks so much for adding the PPt which equally simplified and well explained. Kindly share your email address for further questions i might have for you.
Thanks. Informative.. Tried to download the data and PPT, but not successful
How to get the diagram like SEM in exploratory factor analysis in SPSS
Mike, how did you get the figures in the Data View(as in 6:07)?
Hi there, Azam. I'm not sure what you are referring to at 6:07 in the video.
@@mikecrowson2462 the figures in the Variable View-NOT Data View(sorry for the typo error).
God, just no one can simplify the differences between am oblique and orthogonal rotations...maybe it is just a bit too rooted in Statistics
these videos assume that the viewer knows statistics. I dont. It did not hep me.