Anonymous posted on Friday, September 09, 2005 - 8:53 am
My question is one that is more methodological. The issue is that I have a dataset that asks juveniles about writing habits. The data are responses that individuals gave about writing while they are in 1998. The subsequent waves ask individuals in 1999-2004 the same set of questions about writing, but these individuals share demographics like age, race, and sex. I'm wondering if my data are suited for latent growth curve modeling, and where there may be citations that I may read to figure out how to argue that this analysis plan?
I am not sure that I understand your question. Do you have a group of the same individuals who were asked the same questions in 1998, 1999, 2000, etc. up to 2004? What do you mean that they share demographics?
Anonymous posted on Friday, September 09, 2005 - 4:25 pm
The individuals are different, but the questions are the same. Some of my colleagues have suggested a trend analysis. With this in mind, I'm wondering if latent growth curve analysis is possible?
That is, between 1998 through 2004, the students are between 10-12 years old during each year. The other demographics: race and sex are similar as well. With these issues, I'm really wondering if the data is in a proper format for latent growth curve analysis?
Anonymous posted on Friday, September 09, 2005 - 4:36 pm
The individuals are not the same. The questions are the same. The demographics are the same (e.g., ages ranged from 10-12 for each year). I certainly hope that this is proper data for latent growth curve analysis. What do you think?
I have used latent growth curve analysis for a data of 40 intervention group individuals with 3 assessment time points. despite restrictions of a small sample size, the CFI and chi-square values indicate that the model is a good fit for most varibles. However, for one variable, the CFI is 0.650 and chi-square is 11.50 (significance 0.003. I am unclear as to why the values for this variable are low...also, unclear as to whether i can term this as an acceptable or poor fit.. I would be grateful if you could help me clear my doubt
It is easier to get a good fit with a small sample, not harder. This is due to lower power to reject. A CFI of 0.65 is very poor and there should be ways to improve the model - there must be something different about this variable.
RuoShui posted on Wednesday, November 20, 2013 - 9:31 pm
Hello Dr. Muthen,
I am quite new to LGCM. I have a question regarding running LGCM. When I only modeled the growth of factors across time points, there was a significant slope. However, after I brought in predictors of the slope and intercept, the slope became non significant. I am not sure what this means. Actually, if the predictors are observed score, the slope became non significant; however, if I put the predictor as a factor with indicators, the slope remained significant and had a similar size. What does this mean?
I think you are talking about the mean of the slope growth factor. In a model with covariates, it is the intercept that is estimated not the mean. It is often seen that latent variables explain less variance than observed variables.
RuoShui posted on Thursday, November 21, 2013 - 1:43 pm
Thank you very much Dr. Muthen. Yes, I did look at the intercept of the slope growth factor. Thank you for your explanation. It did seem like latent variables explained less variance.
But there is one thing I still don't understand. The slope growth factor was significant; so I used predictors to predict the slope. But what does it mean when the slope growth factor became non significant? Is there still a growth? Thank you very much.
The slope growth factor is a variable. When you regress it on a covariate, the object is to explain the variance in the slope growth factor. You now have a conditional model where the intercept and residual variance of the slope growth factor are estimated. The mean and variance of the slope growth factor may have been significant but that does not mean that the intercept and residual variance are. These are two different models with different parameters. See the Topic 3 video and course handout on the website where growth modeling is discussed.
RuoShui posted on Thursday, November 21, 2013 - 7:29 pm
Dear Dr. Muthen, Can MPLUS handle large data for latent growth curve modeling? I have 575,3068 observations and 3 variables. When I ran the latent growth curve modeling, it looks like MPLUS is always reading the data, but no further progress. Thank you!
I am new to Latent Growth Modelling using MPlus. While studying a text book on it I decided to try out some analyses with a simplified version of my data set. I tried conducting a first order LGCM but kept getting an error message. Below are my input and the error message. The number of observations is not zero and I could not find any invalid symbol in the data. I do not understand this message. Thank you for your help.
I did as advised and found that there was this symbol "ï»¿" in the first line of my data set. It was not the case in SPSS or in Notepad. I tried several times saving the data set again but each time the problem remained the same in Mplus. Where is this symbol coming from and how can I get rid of it in order to proceed with my analyses? thank you so much for your help.
Hi, We want to model the relation between prenatal head growth and a behavioral trait. Fetal head size was measured at 3 timepoints and coded as gestational-age adjusted z-scores. We want to adjust the regression of the intercept and the slope of head growth on the behavioral trait for confounders. We used this script: analysis: estimator=mlr; model: i s | sdhcg1@email@example.com@17.5; Z_srs_sqrt on i s gender ETHN_d1 ETHN_d2 mdrink_1 mdrink_2 mdrink_3 msmoke_1 msmoke_2 EDUCM5_1 EDUCM5_2 AGE_M; gender ETHN_d1 ETHN_d2 mdrink_1 mdrink_2 mdrink_3 msmoke_1 msmoke_2 EDUCM5_1 EDUCM5_2 age_m;
This warning appeared in the output: 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 0.235D-19. PROBLEM INVOLVING THE FOLLOWING PARAMETER: Parameter 122, AGE_M
This warning isn't related to a specific covariate (it refers to other covariates when ‘problematic’ covariates are removed) and occurs as soon as any covariate is entered, but not in the basic model.
Please could you say if the model is still trustworthy or if there are ways to remove the warning? Thanks!