I'm so impressed. I've been looking at sooo many videos about cfa in r, and this is the best by a long shot. Thank you for explaining the output so well!!!!
Hi ! This was exactly what I was looking for. And I've searched for videos like this throughout UA-cam, in Spanish and English. Still the clearest and best explained video that I found. Thanks thanks a lot !
You showed the calculation for a normed chi-square, then you show another number that you indicated there was debate over using it (chi-square/ 1df). How did you calculate that number (around 25:06 minuted).
Hi Sara, very helpful video. Quick question, do you abide by any cut-off factor loading guidelines for deciding whether an indicator can be regarding as loading onto a factor. I've heard that any indicator with a standardised loading of >.4 can be taken as indicative of the associated latent concept. Do you just go by the z-test significance?
Great video!!!! Quick question, do you think we can use the standardized factor loadings to get the weight of each variable and then use that for further graphing purposes with each person's response? Kind of multiplying each person's response with the weight? Thanks in advance!
But your assumptions is based on a normal dist but your data is skewed and categorical ? Do you have any advise/vid to handle the poly matrix to the CFA ?
Thanks Sara! This is so clear. One question: when reporting the factor loadings from a CFA, do you report the estimates, or the standardized estimates?
Glad to hear it! I usually try to report both in a table, but if I had to choose one over the other, I'd probably go with standardized estimates. I find them easier to interpret, especially if different variables in the model use different metrics.
Thanks Sara, I am a noob with all this. My test statistics are 755.768, df: 424. If I divide this my normed chi square is 1.78 - not sure by what to judge now - how well/bad does my model fit? Chi Square should be around one? So, 1.78 isn't bad, but 755.768 is, or is it not?
Great question! The problem is chi-square is that it's biased to be large when samples are large. Given that SEM is a large N statistic, this means that your chi-square is coming to be significant (i.e., indicate poor model fit) the majority of the time. This is why we apply that correction using degrees of freedom, since it's supposed to counteract that large N. The problem, though, is that no one seems to agree on what your normed chi-square should be less than, with some saying 1, some saying larger values like 5. I usually use 3.841, which is the chi-square critical value for df = 1. Depending on who you talk to, that will either be too conservative, too lenient, or just right. But what people DO seem to agree on is the chi-square shouldn't be the deciding factor on whether your model fits well. Your other fit statistics, like RMSEA, CFI, TLI, etc., give you much better info. So if those indicators show good model fit, you're probably okay.
@@DeeplyTrivial Thanks! I have N = 220 and about 30 variables. So my sample is not large. I was looking at different models, with 5 to 8 factors and normed chi-square ranges from1.5 to 1.8. Differences seem to be minor between all different models. Therefore, I am thinking I should look at the questions and what factors make the most sense. One more issue, I have very weak correlations among the variables, and the sample comes from a very diverse population (Likert Data). I wonder if CFA makes even sense for that matter. Thank you so much Sara.
I'm so impressed. I've been looking at sooo many videos about cfa in r, and this is the best by a long shot. Thank you for explaining the output so well!!!!
Hi !
This was exactly what I was looking for. And I've searched for videos like this throughout UA-cam, in Spanish and English. Still the clearest and best explained video that I found. Thanks thanks a lot !
Great video. Very nice way of explaining stuff. I had R and my CFA open and understand everything now :)
10/10 video. many thanks!
Great explanation, really help my analysis, thank you.
Thanks for the video! Really clear explanation!
Thanks for this. Very clear explanation.
You showed the calculation for a normed chi-square, then you show another number that you indicated there was debate over using it (chi-square/ 1df). How did you calculate that number (around 25:06 minuted).
Hi Sara, very helpful video. Quick question, do you abide by any cut-off factor loading guidelines for deciding whether an indicator can be regarding as loading onto a factor. I've heard that any indicator with a standardised loading of >.4 can be taken as indicative of the associated latent concept. Do you just go by the z-test significance?
Hi! thank you so much!! :D
Actually helped with my analysis. Thank you.
Great video!!!! Quick question, do you think we can use the standardized factor loadings to get the weight of each variable and then use that for further graphing purposes with each person's response? Kind of multiplying each person's response with the weight? Thanks in advance!
But your assumptions is based on a normal dist but your data is skewed and categorical ? Do you have any advise/vid to handle the poly matrix to the CFA ?
Such a great video! Thank you!
How do we improve model fit in CFA?
Thanks Sara! This is so clear. One question: when reporting the factor loadings from a CFA, do you report the estimates, or the standardized estimates?
Glad to hear it! I usually try to report both in a table, but if I had to choose one over the other, I'd probably go with standardized estimates. I find them easier to interpret, especially if different variables in the model use different metrics.
Thanks Sara, it helped a lot :)!
Thanks Sara, I am a noob with all this. My test statistics are 755.768, df: 424. If I divide this my normed chi square is 1.78 - not sure by what to judge now - how well/bad does my model fit? Chi Square should be around one? So, 1.78 isn't bad, but 755.768 is, or is it not?
Great question! The problem is chi-square is that it's biased to be large when samples are large. Given that SEM is a large N statistic, this means that your chi-square is coming to be significant (i.e., indicate poor model fit) the majority of the time. This is why we apply that correction using degrees of freedom, since it's supposed to counteract that large N. The problem, though, is that no one seems to agree on what your normed chi-square should be less than, with some saying 1, some saying larger values like 5. I usually use 3.841, which is the chi-square critical value for df = 1. Depending on who you talk to, that will either be too conservative, too lenient, or just right. But what people DO seem to agree on is the chi-square shouldn't be the deciding factor on whether your model fits well. Your other fit statistics, like RMSEA, CFI, TLI, etc., give you much better info. So if those indicators show good model fit, you're probably okay.
@@DeeplyTrivial Thanks! I have N = 220 and about 30 variables. So my sample is not large. I was looking at different models, with 5 to 8 factors and normed chi-square ranges from1.5 to 1.8. Differences seem to be minor between all different models. Therefore, I am thinking I should look at the questions and what factors make the most sense. One more issue, I have very weak correlations among the variables, and the sample comes from a very diverse population (Likert Data). I wonder if CFA makes even sense for that matter. Thank you so much Sara.
epic video thank you very based yes