Hi, Dr. Geise! Your videos are very helpful! I would like to ask, I see a lot of papers' plots with standardized z-score of means, should I standardize these means after I get the results, or should I standardize all the variables before I do the LPA?
I would not routinely standardize the variables prior to an LPA because (1) it is usually not necessary and (2) it forces all variable to have the same (unit) variance, which may be problematic in some cases. Best, Christian Geiser
Very helpful video, thank you! How do I interpret the p-values given in the model results for each variable mean? Does the p-value indicate whether or not the class mean differs significantly from the overall sample mean? I am currently working on an LPA, in whiche not all of these p-values are significant.
Hi Ariadne, Thank you for watching! The p value for the class-specific mean is for a test of the null hypothesis that the parameter (mean) is zero in the population in this particular latent class. This hypothesis is usually not of interest. Best, Christian Geiser
Hi Dr Geiser, thank you so much for your sharing! It is really informative and helpful! Meanwhile, I have a question about interpreting the relationship between each profile and variables. It seems reasonable to argue that every profile is constituted by multiple variables with different proportions of influence? For example, let's say, the respondents of profile 1 are partly constituted by variable ES (70%), variable E (20%) and O (10%). Is there any method we can do this? Thank you!
Hi CAI Lawrance, I'm not sure I understand what you are asking. I don't think that it would make sense to say that respondents are "constituted by variables." Typically, we just show/describe the profiles as is. If you could elaborate a bit on your question, perhaps I would be able to provide a better answer. Best, Christian Geiser
@@QuantFish Thank you for your reply, Dr Geiser! In my sense, each profile generation is highly related to (and indeed it is based on) the observed variables. For example, we try to find out the latent profiles based on cultural values, which contain five constructs: power distance (po), uncertainty avoidance (ua), collectivism (co), masculinity (ma), and long-term oriented (lo). Finally, we find a two-profile structure. And theoretically, each profile represents different extents to cultural value construct. Such as profile 1 is high in po, ua and co, but low in ma and lo; whereas profile 2 is low in po, ua and co but high in ma and lo. Therefore, the relationship between the profile that each person belongs to and the construct that we input as observe variables in mplus is not extremely exclusive, but varying in different to different extents (please correct me if I get it wrong). Given that, I was wondering how I can visualise such “extent" more clearly, like I want to know to what extent profile 1 is generated by each cultural value constructs (demonstrated by statistical values). Further, if my data is with multi-levels, let's say between countries, can I just aggregate them by country and take the mean value? It would be appreciated if you could answer my question. Thank you!
Hi Dr. Geiser, in addition to LPA, I used the automatic BCH procedure to predict outcome variables based on profile membership. Regarding the BCH procedure output, am I correct in thinking that the "overall test" must be significant in order to interpret the pairwise comparisons?
Hello Hannah, Thank you for watching! You are probably right, but I would check with the Mplus support team to be absolutely sure. Best, Christian Geiser
Guten Tag Herr Geiser und herzlichen Dank für die tollen Videos! Darf ich nachfragen, ob es möglich ist, die Klassenzugehörigkeit (z.B. "the resilients, class 3) der Fälle (Studierenden) herauszulesen? Also kann ich irgendwo sehen z.B. Student XY gehört zu Class 3 beispielsweise? So wie es in der explorativen Clusteranalyse möglich ist. Danke für die Beantwortung und beste Grüsse, Marina Bregy
Liebe Frau Bregy, Sie koennen im Rahmen von LCA, LCA etc. die individuelle Klassenzugehoerigkeitswahrscheinlichkeiten schaetzen lassen und Personen anhand ihrer hoechsten Klassenwahrscheinlichkeit manifest zuordnen. Siehe mein Video hier: ruclips.net/video/eK6LRqEcYoU/видео.html Best, Christian Geiser
Hello Dr. Geiser, I was wondering if you had a similar demonstration for Latent Transition Analysis after doing an LPA on Mplus? I am having trouble with coding and could have used your help, suggestion, advice.
Hello Mish 799, I will post some videos on how to run LTAs in Mplus this coming summer. In case you cannot wait that long, I also offer personal consulting services (christiangeiser.com). Best, Christian Geiser
Hi there, this output doesn't have a chi square value. How do we evaluate model fit without it? Is there something I need to put in the syntax to get the chi square model fit value? Thanks!
There is currently no chi-square test of absolute model fit available for LPA. But you can use the BIC or TECH14 output (bootstrap likelihood ratio test) to compare models with different numbers of classes. Best, Christian Geiser
Hi Dr. Geiser, I am using mplus to calculate the sum and mean of my current variables, but the output file says that "Left numeric operand cannot be found." I wonder what I could do to fix it? Thank you in advance!
It is likely that there is an error in your syntax. I would email the Mplus support team with your syntax, data file, and license number. They can help you figure it out. Best, Christian Geiser
@@QuantFish Hi Dr. Geiser, I checked my syntax again today and luckily, I found the mistake! Now my analyses are running, but I have some more specific questions about my analyses, I hope it is okay to aks here! I want to compare the (two) group differences with two control variables (gender:0,1; and a z-score ), my questions are : 1. I want to compare the sum score of scales between two groups, however, when I searched online, they all recommended to use the item-level analysis. I wonder whether I could still use the sum score calculated under "define-sum()- command" because that is how the score should be calculated for that scale. 2. for the comparison syntax, I used the "GROUPING is.." in the usevariables, and also listed the variables (sum scores) under "model group" syntax, but it did not work. May I get some suggestion on the syntax? 3. I also tried to use ON command under the model syntax to test the regression of sum score on group, but the model results only showed the p value, I could not find which group has higher group mean. I wonder whether this is normal. I am sorry that I have too many questions, thanks a lot!!! Best, Sunny
Hi! Can you help me figure out what to do with this LCA output I get? Thanks a lot! THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS 0.240D-20.
Hi Vero, It would be best for you to directly contact the Mplus support. They can look over your input and data file and tell you more about what's going on. Best, Christian Geiser
Dear Dr. Geiser, I have a question regarding latent profile analysis with distal outcomes and covariates. Is this the right syntax to control for both ACE and AAS DATA: FILE IS total.dat; VARIABLE: NAMES ARE ACE AAS WV PHQ GH MI; CLASSES = L (3); AUXILIARY = PHQ (bch) GH (bch) MI (bch); ANALYSIS: TYPE = MIXTURE; starts = 500 50; stiterations = 50; MODEL: %OVERALL% L on ACE AAS; OUTPUT: TECH11; TECH14;
This is so helpful! As all our of your videos. They've been essential in helping me to complete my masters thesis. Much appreciated Dr. Geiser!
Go Ikela! We're happy to be part of your journey.
Very helpful. Excellent!
Thank you for this really interesting and helpful video!
Hi, Dr. Geise! Your videos are very helpful! I would like to ask, I see a lot of papers' plots with standardized z-score of means, should I standardize these means after I get the results, or should I standardize all the variables before I do the LPA?
I would not routinely standardize the variables prior to an LPA because (1) it is usually not necessary and (2) it forces all variable to have the same (unit) variance, which may be problematic in some cases.
Best, Christian Geiser
Very helpful video, thank you! How do I interpret the p-values given in the model results for each variable mean? Does the p-value indicate whether or not the class mean differs significantly from the overall sample mean? I am currently working on an LPA, in whiche not all of these p-values are significant.
Hi Ariadne, Thank you for watching! The p value for the class-specific mean is for a test of the null hypothesis that the parameter (mean) is zero in the population in this particular latent class. This hypothesis is usually not of interest. Best, Christian Geiser
@@QuantFish I see - thank you very much!
Hi Dr Geiser, thank you so much for your sharing! It is really informative and helpful!
Meanwhile, I have a question about interpreting the relationship between each profile and variables. It seems reasonable to argue that every profile is constituted by multiple variables with different proportions of influence? For example, let's say, the respondents of profile 1 are partly constituted by variable ES (70%), variable E (20%) and O (10%). Is there any method we can do this?
Thank you!
Hi CAI Lawrance, I'm not sure I understand what you are asking. I don't think that it would make sense to say that respondents are "constituted by variables." Typically, we just show/describe the profiles as is. If you could elaborate a bit on your question, perhaps I would be able to provide a better answer. Best, Christian Geiser
@@QuantFish Thank you for your reply, Dr Geiser!
In my sense, each profile generation is highly related to (and indeed it is based on) the observed variables. For example, we try to find out the latent profiles based on cultural values, which contain five constructs: power distance (po), uncertainty avoidance (ua), collectivism (co), masculinity (ma), and long-term oriented (lo). Finally, we find a two-profile structure. And theoretically, each profile represents different extents to cultural value construct. Such as profile 1 is high in po, ua and co, but low in ma and lo; whereas profile 2 is low in po, ua and co but high in ma and lo.
Therefore, the relationship between the profile that each person belongs to and the construct that we input as observe variables in mplus is not extremely exclusive, but varying in different to different extents (please correct me if I get it wrong). Given that, I was wondering how I can visualise such “extent" more clearly, like I want to know to what extent profile 1 is generated by each cultural value constructs (demonstrated by statistical values).
Further, if my data is with multi-levels, let's say between countries, can I just aggregate them by country and take the mean value? It would be appreciated if you could answer my question.
Thank you!
Hi Dr. Geiser, in addition to LPA, I used the automatic BCH procedure to predict outcome variables based on profile membership. Regarding the BCH procedure output, am I correct in thinking that the "overall test" must be significant in order to interpret the pairwise comparisons?
Hello Hannah, Thank you for watching! You are probably right, but I would check with the Mplus support team to be absolutely sure. Best, Christian Geiser
Guten Tag Herr Geiser und herzlichen Dank für die tollen Videos! Darf ich nachfragen, ob es möglich ist, die Klassenzugehörigkeit (z.B. "the resilients, class 3) der Fälle (Studierenden) herauszulesen? Also kann ich irgendwo sehen z.B. Student XY gehört zu Class 3 beispielsweise? So wie es in der explorativen Clusteranalyse möglich ist. Danke für die Beantwortung und beste Grüsse, Marina Bregy
Liebe Frau Bregy,
Sie koennen im Rahmen von LCA, LCA etc. die individuelle Klassenzugehoerigkeitswahrscheinlichkeiten schaetzen lassen und Personen anhand ihrer hoechsten Klassenwahrscheinlichkeit manifest zuordnen. Siehe mein Video hier:
ruclips.net/video/eK6LRqEcYoU/видео.html
Best,
Christian Geiser
Hello Dr. Geiser, I was wondering if you had a similar demonstration for Latent Transition Analysis after doing an LPA on Mplus? I am having trouble with coding and could have used your help, suggestion, advice.
Hello Mish 799,
I will post some videos on how to run LTAs in Mplus this coming summer. In case you cannot wait that long, I also offer personal consulting services (christiangeiser.com).
Best, Christian Geiser
Also, you can sign up for alerts about my upcoming LTA on-demand workshop here:
www.goquantfish.com/courses/latent-transition-analysis-with-mplus
Hi there, this output doesn't have a chi square value. How do we evaluate model fit without it? Is there something I need to put in the syntax to get the chi square model fit value? Thanks!
There is currently no chi-square test of absolute model fit available for LPA. But you can use the BIC or TECH14 output (bootstrap likelihood ratio test) to compare models with different numbers of classes.
Best, Christian Geiser
Hi Dr. Geiser, I am using mplus to calculate the sum and mean of my current variables, but the output file says that "Left numeric operand cannot be found." I wonder what I could do to fix it? Thank you in advance!
It is likely that there is an error in your syntax. I would email the Mplus support team with your syntax, data file, and license number. They can help you figure it out.
Best, Christian Geiser
@@QuantFish Thank you so much Dr. Geiser, I will then ask them this question.
Best, Sunny
@@QuantFish Hi Dr. Geiser, I checked my syntax again today and luckily, I found the mistake! Now my analyses are running, but I have some more specific questions about my analyses, I hope it is okay to aks here!
I want to compare the (two) group differences with two control variables (gender:0,1; and a z-score ), my questions are :
1. I want to compare the sum score of scales between two groups, however, when I searched online, they all recommended to use the item-level analysis. I wonder whether I could still use the sum score calculated under "define-sum()- command" because that is how the score should be calculated for that scale.
2. for the comparison syntax, I used the "GROUPING is.." in the usevariables, and also listed the variables (sum scores) under "model group" syntax, but it did not work. May I get some suggestion on the syntax?
3. I also tried to use ON command under the model syntax to test the regression of sum score on group, but the model results only showed the p value, I could not find which group has higher group mean. I wonder whether this is normal.
I am sorry that I have too many questions, thanks a lot!!!
Best,
Sunny
Hi! Can you help me figure out what to do with this LCA output I get? Thanks a lot!
THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE
TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE
FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING
VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE
CONDITION NUMBER IS 0.240D-20.
Hi Vero, It would be best for you to directly contact the Mplus support. They can look over your input and data file and tell you more about what's going on. Best, Christian Geiser
@@QuantFish Ok! Thank you so much for the quick reply!
Dear Dr. Geiser, I have a question regarding latent profile analysis with distal outcomes and covariates.
Is this the right syntax to control for both ACE and AAS
DATA: FILE IS total.dat;
VARIABLE: NAMES ARE ACE AAS WV PHQ GH MI;
CLASSES = L (3);
AUXILIARY = PHQ (bch) GH (bch) MI (bch);
ANALYSIS: TYPE = MIXTURE;
starts = 500 50;
stiterations = 50;
MODEL: %OVERALL%
L on ACE AAS;
OUTPUT: TECH11;
TECH14;
Yes, in this model, ACE and AAS are used as predictors (covariates) of class membership using logistic regression analysis.
Best, Christian Geiser