No plans to stop. There are still things that I need to cover on my courses and the papers that I write normally benefit from video explanations. There might be breaks though depending on what I need to focus on at work (teaching vs. research)
This is the clearest explanation of factor scores I have ever heard, thank you Mikko! For categorical indicators, do you use the same three types of factor scores?
Which factors scores to for categorical indicators depends on what purpose you will be using the indicators and how you assume the indicators to be related to the latent variable. If you are using a linear factor model, then the factor scores discussed in the video might be appropriate. (Or you might just take a sum of the indicators.) But if a linear factor model is not appropriate and you use an IRT (item response theory) model instead, then IRT scores would be more approriate.
Excellent content! I came across your post linking This video in a lavaan Google group. However, I’m still a little confused. It sounds like what you are saying undermines the use of SEM, no? Presumably, factor scores are passed into the structural part of an SEM. Yet, you are saying that summing up the factor indicators is sufficient for representing a construct for a given observation. Any clarification would be helpful! Thanks
SEM does not use factor scores but estimates the relationships between latent variables without estimating latent variable values. " factor scores are passed into the structural part of an SEM" is thus not what SEM does.
hey mikko... i saw your video "seemingly unrelated regression " thank you so much but i want to ask you in seemingly what are the main indicators /results in this method??? do you have a video you talk about that ?
SURE is just like regression except that there are more than one equation. There is really not more to it than that, so I do not think that the video would be very long ;)
Thanks for the viedo. I have two questions. (1) How does AMOS or other software calculate the factor score weights and make them sum up to one? In the formula of the estimation of factor score weights, there seems no constrains which can let the weights sum up to one. (2)If a model just includs one factor, is the estimator from regression method a unbiased? Thanks again.
1) I am not sure if the factor score weights sum to one. But you can make the sum to one by taking a sum of the weights and then divide each weight with the sum. 2) Yes, though being "unbiaesed by other factors" is not really a meaningful concept of if there are no other factors in the model. More generally, if you have just one factor, I think all factor score techniques produce the same weights.
I'm sure those weights sum up to one. it is also true in your case. I wonder if those software did the procedure that "taking a sum of the weights and then divide each weight with the sum" ? Thanks for the reply. @@mronkko
@@xuyang2776 Factor score weights can be positive or negative. If the factor model explains the data well, the factor score weights tend to have the same sign as the factor lodings.
@@xuyang2776 The weights do not sum to one, at least not generally. For example the Bartlett and tenBerge weights clearly sum to a number greater than one. With regression weights, this is not as obvious. But the sums of the weights for the example are: 1.140571 1.141877 1.365617 1.371021 1.248114 1.251167
Saved me some days of reading! Thank you.
You are welcome!
I hope you dont stop because this is precious information for me to have an A level masther thesis...Thank you so much for all this effort
No plans to stop. There are still things that I need to cover on my courses and the papers that I write normally benefit from video explanations. There might be breaks though depending on what I need to focus on at work (teaching vs. research)
This is the clearest explanation of factor scores I have ever heard, thank you Mikko! For categorical indicators, do you use the same three types of factor scores?
Which factors scores to for categorical indicators depends on what purpose you will be using the indicators and how you assume the indicators to be related to the latent variable. If you are using a linear factor model, then the factor scores discussed in the video might be appropriate. (Or you might just take a sum of the indicators.) But if a linear factor model is not appropriate and you use an IRT (item response theory) model instead, then IRT scores would be more approriate.
Excellent content! I came across your post linking This video in a lavaan Google group. However, I’m still a little confused. It sounds like what you are saying undermines the use of SEM, no? Presumably, factor scores are passed into the structural part of an SEM. Yet, you are saying that summing up the factor indicators is sufficient for representing a construct for a given observation. Any clarification would be helpful! Thanks
SEM does not use factor scores but estimates the relationships between latent variables without estimating latent variable values. " factor scores are passed into the structural part of an SEM" is thus not what SEM does.
hey mikko... i saw your video "seemingly unrelated regression " thank you so much but i want to ask you in seemingly what are the main indicators /results in this method??? do you have a video you talk about that ?
SURE is just like regression except that there are more than one equation. There is really not more to it than that, so I do not think that the video would be very long ;)
Thank you so much for this, very helpful!
You are welcome
Thanks for the viedo. I have two questions. (1) How does AMOS or other software calculate the factor score weights and make them sum up to one? In the formula of the estimation of factor score weights, there seems no constrains which can let the weights sum up to one. (2)If a model just includs one factor, is the estimator from regression method a unbiased? Thanks again.
1) I am not sure if the factor score weights sum to one. But you can make the sum to one by taking a sum of the weights and then divide each weight with the sum. 2) Yes, though being "unbiaesed by other factors" is not really a meaningful concept of if there are no other factors in the model. More generally, if you have just one factor, I think all factor score techniques produce the same weights.
I'm sure those weights sum up to one. it is also true in your case. I wonder if those software did the procedure that "taking a sum of the weights and then divide each weight with the sum" ? Thanks for the reply. @@mronkko
also why the factor score weights are all possitive? is it possible to be negative?Thanks again@@mronkko
@@xuyang2776 Factor score weights can be positive or negative. If the factor model explains the data well, the factor score weights tend to have the same sign as the factor lodings.
@@xuyang2776 The weights do not sum to one, at least not generally. For example the Bartlett and tenBerge weights clearly sum to a number greater than one. With regression weights, this is not as obvious. But the sums of the weights for the example are:
1.140571 1.141877 1.365617 1.371021 1.248114 1.251167