@@xuyang2776 If you define independent variable as a "natural number sequence", then my explanation is not correct. Using that definition, there are no independent variables in the model. However, the literature that I draw from does not define independent variables that way: the independent variables are assumed to have a distribution and be from (possibly a subset) of real numbers. They can be either observed or latent. With this definition, I do not see any errors in my anser.
Is it possible to estimate the intercept and slope of latent variables across time instead of using observed variables? I ask this because one may have a construct that is measured using multiple items at each time point.
Yes, that is possible. You just need to establish measurement invariance to at least strong factorial invariance level. I have an example here ua-cam.com/video/XGg5Lz_MBis/v-deo.html
Thank you very much! I finally understand this kind of model by your video. I have a question. If I use AMOS' imputation function to calculate the latent factors' score(here they are intercept and slope) based on the first model, can I get two series which are all constant?
I would caution against using factor scores calculated the way you describe. See my video on factor score estimation. Factor scores have limited usefulness for research. I do not use AMOS myself, but if you get a series of constants as factor scores, it seems that either you are not doing something right or the software does not calculate factor scores correctly for this kind of model.
You misunderstand my question. In the first model in your video, intercept and slope are all constants, not random variables, right? In this logic, the corresponding factor scores are deserved to be constant. My question is if the estimated factor scores have such a trait@@mronkko
@@xuyang2776 OK, I see. The intercept and slope are not constant but are latent variables. They receive different (unobserved) values for each row in the data.
Thanks for your answer. I can understand when the intercept and slope are random variables. But in Multi-level models, they could be modeled as constant, at least one of them could be set as a constant, such as slope. In such a case, what will happen ? @@mronkko
@@xuyang2776@xuyang2776 A variable, by definition, cannot be constant because a variable is something that varies (i.e., has variance). You model constants by leaving out variables. Consider multilevel models: The simplest multilevel model has two levels where the only level-two random effect is the random intercept. If you set the random intercept to have no variance by leaving it out, you will just have a normal regression, and the predicted intercept for each cluster is just the intercept of the regression model.
Very nice, concise, and short overview, also appreciated the comparison to linear mixed models. Even if your accent is sometimes a bit difficult to understand, the videos are really helpful.
Hi i have 2 moderator and 2 mediators variables and i will conduct experimental design within- between (factorial design ) . i will collect the data 3 times my question for testing moderators and mediators how many time do i need to collect the data for them ?
@@talzabidi1569 I cannot really provide research design consulting through UA-cam comments. I do my best to answer questions that are easily answerable, but this is not the kind of question.
Another question. In the model shown in the video(0.11~1.45m), ln(crime) is the dependent variable, but where are the independent variables?
The independent variables are the latent slope and the latent intercept. They can be explained by other variables, like X in my example.
@@mronkko Thanks. I got it. But I think your explanation is not correct, the independent variable is a natural number sequence.
@@xuyang2776 If you define independent variable as a "natural number sequence", then my explanation is not correct. Using that definition, there are no independent variables in the model.
However, the literature that I draw from does not define independent variables that way: the independent variables are assumed to have a distribution and be from (possibly a subset) of real numbers. They can be either observed or latent. With this definition, I do not see any errors in my anser.
Is it possible to estimate the intercept and slope of latent variables across time instead of using observed variables? I ask this because one may have a construct that is measured using multiple items at each time point.
Yes, that is possible. You just need to establish measurement invariance to at least strong factorial invariance level. I have an example here ua-cam.com/video/XGg5Lz_MBis/v-deo.html
Thank you very much! I finally understand this kind of model by your video. I have a question. If I use AMOS' imputation function to calculate the latent factors' score(here they are intercept and slope) based on the first model, can I get two series which are all constant?
I would caution against using factor scores calculated the way you describe. See my video on factor score estimation. Factor scores have limited usefulness for research.
I do not use AMOS myself, but if you get a series of constants as factor scores, it seems that either you are not doing something right or the software does not calculate factor scores correctly for this kind of model.
You misunderstand my question. In the first model in your video, intercept and slope are all constants, not random variables, right? In this logic, the corresponding factor scores are deserved to be constant. My question is if the estimated factor scores have such a trait@@mronkko
@@xuyang2776 OK, I see. The intercept and slope are not constant but are latent variables. They receive different (unobserved) values for each row in the data.
Thanks for your answer. I can understand when the intercept and slope are random variables. But in Multi-level models, they could be modeled as constant, at least one of them could be set as a constant, such as slope. In such a case, what will happen ? @@mronkko
@@xuyang2776@xuyang2776 A variable, by definition, cannot be constant because a variable is something that varies (i.e., has variance). You model constants by leaving out variables. Consider multilevel models: The simplest multilevel model has two levels where the only level-two random effect is the random intercept. If you set the random intercept to have no variance by leaving it out, you will just have a normal regression, and the predicted intercept for each cluster is just the intercept of the regression model.
Very nice, concise, and short overview, also appreciated the comparison to linear mixed models. Even if your accent is sometimes a bit difficult to understand, the videos are really helpful.
Cool, thanks!
Hi
i have 2 moderator and 2 mediators variables and i will conduct experimental design within- between (factorial design ) . i will collect the data 3 times my question for testing moderators and mediators how many time do i need to collect the data for them ?
That really depends on the broader research question and is impossible to answer from the information that you provided.
@@mronkko
What additional information do you need?
@@talzabidi1569 I cannot really provide research design consulting through UA-cam comments. I do my best to answer questions that are easily answerable, but this is not the kind of question.