Message/Author |
|
|
Dear all, I have pooled data stemming from two cross-sectional surveys (with independent respondents, Firebaugh 1997)and want to estimate a path model for four variables using Mplus. However, yet I haven't found any article applying SEM to repeated CS surveys - thus it would be extremely helpful if you could tell me 1. if such an analysis could, in principle, be conducted at all 2. Mplus has specific defaults to be considered 3. point to related literature applying SEM for repeated cross-sectional data Many thanks in advance |
|
|
Can you give me the Firebaugh reference? |
|
|
Firebaugh, Glenn (1997). Analyzing repeated surveys. Thousand Oaks, Calif.: Sage Publ. See also: Firebaugh, G. (1992). "Where Does Social Change Come From? Estimating the Relative Contributions of Individual Change and Population Turnover." Population Research and Policy Review 11:1-20. Please note that these publications do NOT cover the question of SEM for repeated CS. |
|
|
Dear all, My situation is similar to Elmar, only a little bit more complicated. I am testing a path analysis using a HUGE repeated complex sample dataset (Cross-Sectional). The problem is that the individual respondents are different from year to year, and only the group/organisational level can be possibly matched up. The fit indices are very good for 3 repeated models (TYPE=TWOLEVEL MISSING). However, the parameters vary to some extent, especially on BETWEEN level. How could we possibly take advantages of repeated data (on between level only) to explain the changes? Is that possible to compare means and parameters on between level from three years model to explain the change curve over time? How robust it would be? Also, if I draw path diagram, which coefficient should I put in, std or stdYx? Many thanks in advance. (Survey was conducted in more than 500 healthcare organisations in UK, with more than 200,000 respondents annually.) http://www.healthcarecommission.org.uk/NationalFindings/Surveys/StaffSurveys/fs/en?CONTENT_ID=4000125&chk=3p67/v |
|
|
It sounds like you have the same organizations over time but different individuals from the organization and you want to create a data set where organization is the analysis unit using the average over the individuals at each year as the organization's data points. I think that is a little shaky given that the data points from each year comes from different individuals who are very likely not randomly selected. You may be criticized for this. STDYX is usually used. |
|
|
I was really amazed by your prompt reply and very many thanks for this. If I only use one year data to establish a multilevel model with similar constructs across level(homologous model), it maybe good enough for a paper. Given your comments, how should i capture the advantages of repeated survey? Could you please give me some suggestions? |
|
|
I'm sorry but I think you may have misunderstood me. My point is that using data where different subjects are observed at each time point to estimate a repeated measures model is not definsible in my opinion. |
|
|
Yes. I did understand your points. Strictly speaking, it is not a repeated survey, isn't it? However, our concerns are mainly on the organisational level rather than individual level. Is there any way to work it around? Or alternatively, you would suggest to take annually survey separately, rather than messing them up. |
|
|
I can't make any recommendations. |
|
|
Very many thanks. |
|
Anwar Hasan posted on Tuesday, January 14, 2014 - 7:55 pm
|
|
|
Dear all I have secondary data (panel data)and i= company t= time N= 30 companies(i) T= 84 months Number of observation 2520 I have 16 variables One observe variable as dependent (na1) variables Tow latent variables as independents variable (f1 f2) and one observe control(c1) variables as independent also, So my model structure is like that: Usevariables are na1 a1 a2 a3 a4 a5 a6 b1 b2 b3 b4 b5 b6 b7 b8 c1; Model f1 BY a1 a2 a3 a4 a5 a6 f2 By b1 b2 b3 b4 b5 b6 b7 b8 na1 on f1 f2 na1 on c1 So, please I need your advise how to run the Mplus with this type of data. Kind regards |
|
|
Sounds like you have the same companies measured 84 times, in which case you could do a Type=twolevel analysis with time as level 1 and company as level 2. There are many UG examples of 2-level SEM. |
|
Anwar Hasan posted on Wednesday, January 15, 2014 - 6:03 pm
|
|
|
thank you Prof Bengt O. Muthen. |
|
Anwar Hasan posted on Thursday, January 16, 2014 - 6:22 am
|
|
|
Please if you do not mind to write the model for me as it suppose to be in twolevel model.because I do not have idea how to write the model in my case as I have (i) and (time). I would like to thank in in advance for your help. Kind regards |
|
Anwar Hasan posted on Thursday, January 16, 2014 - 6:28 am
|
|
|
Hi Dear prof Muthen I could not start the analysis because still have problem with "w" ? I am asking also whether to create new variable in my data under the name of "w" and also I am asking whether I have to create one column in my data for the time variable. I have secondary data (panel data) and i= company t= time N= 30 companies(i) T= 84 months Number of observation 2520 I have 16 variables One observe variable as dependent (na1) variable Tow latent variables as independents variable (f1, f2) and one observe control(c1) variables as independent variable also, So, my model structure is like that: Usevariables are na1 a1 a2 a3 a4 a5 a6 b1 b2 b3 b4 b5 b6 b7 b8 c1; my proposed Model f1 BY a1 a2 a3 a4 a5 a6 f2 By b1 b2 b3 b4 b5 b6 b7 b8 na1 on f1 f2 na1 on c1 Please if you do not mind to write the model for me as it suppose to be in twolevel model.because I do not have idea how to write the model in my case as I have (i) and (time). I would like to thank in in advance for your help. Kind regards |
|
|
I don't know what role your w has in your model. As for 2-level modeling of your data, here's a brief outline. It's related to UG ex 9.16 for growth modeling done in a two-level, "long format". You have as variables (columns) in your data id (varying across your companies), time (varying from 1 to 84), and your 16 variables. You specify Cluster = id; so that is your level-2, whereas time is your level-1. Then you have your Model statements: %Within% f1 BY a1 a2 a3 a4 a5 a6 f2 By b1 b2 b3 b4 b5 b6 b7 b8 na1 on f1 f2 na1 on c1 %Between% fb1 BY a1 a2 a3 a4 a5 a6 fb2 By b1 b2 b3 b4 b5 b6 b7 b8 na1 on fb1 fb2 na1 on c1 That gives you a start, where you don't explore trends in your variables but let them correlate over time using the random means/intercepts of the variables, that is, the subject-level part of the variables. |
|
Anwar Hasan posted on Saturday, January 18, 2014 - 5:01 am
|
|
|
Dear Prof Bengt O. Muthen I would like to thank you for your attention and respond, I found the following report from the Mplus software: Mplus VERSION 7.11 MUTHEN & MUTHEN 01/18/2014 8:30 PM INPUT INSTRUCTIONS TITLE: MICRO and MACRO DATA: FILE IS "C:\Users\Anwar_1\Desktop\M-Plus\penal-data Ln- 2012.dat"; VARIABLE: NAMES ARE id time nav dv hp s lfz lmex lhf lcpi lipi tbr lm3 fer lop lci npe fc; USEVARIABLES ARE id time nav dv hp s lfz lmex lhf lcpi lipi tbr lm3 fer lop lci npe fc; Cluster = id; Model: %Within% f1 BY dv hp s lfz lmex; f2 By lhf lcpi lipi tbr lm3 fer lop lci npe; nav on f1 f2; nav on fc; f1 on f2; f2 on f1; %Between% fb1 BY dv hp s lfz lmex; fb2 By lhf lcpi lipi tbr lm3 fer lop lci npe; nav on fb1 fb2; nav on fc; fb1 on fb2; fb2 on fb1; ANALYSIS: TYPE IS TWOLEVEL; ESTIMATOR IS MLM; ITERATIONS = 1000; CONVERGENCE = 0.000001; OUTPUT: SAMPSTAT RESIDUAL STANDARDIZED CINTERVAL TECH1 TECH2 TECH3 TECH4 TECH5; SAVEDATA: FILE IS Anwar; FORMAT IS FREE; RECORDLENGTH = 1000; RESULTS IS anwar1; *** ERROR This analysis is only available with the Multilevel or Combination Add-On. |
|
Anwar Hasan posted on Saturday, January 18, 2014 - 5:16 am
|
|
|
Dear Prof, when I run the software, I follow the folwing steps: 1-Implus- Language Generator. 2- SEM with clustered data- next. 3- I wrote the title- data fine-next. 4- Data format (free)- Missing data(no-next. 5- variable names-(entr all the variables)-next. 6- variable List-Uservariabl list(enter all the variables)-next 7- dos not select anything- next 8-analysis Type- twoleve-estmiator-MLM-next 9-output-I selected all the options.next 10-seved option- I selected the raw data and result file- finish. then I found the above result. Kind regards Anwar |
|
|
You must have the Base Program or the Base Program with the Mixture Add-On. These programs do not support multilevel modeling. |
|
Anwar Hasan posted on Saturday, January 18, 2014 - 6:53 am
|
|
|
Please where I can get this program |
|
|
See Order on the website. |
|
Anwar Hasan posted on Tuesday, January 21, 2014 - 11:35 pm
|
|
|
Dear Prof Muthen could you please explaine to me why goodness fit of the model not apear in the result |
|
|
If means, variances, and covariances are not sufficient statistics for model estimation, chi-square and related fit statistics are not available. Differences in nested models can be tested using -2 times the loglikelihood difference which is distributed as chi-square. |
|
Back to top |