| Message/Author |
|
|
|
|
| I am planning to run a latent growth curve model involving U.S. States. I have 17 years of repeated measures for all 50 states plus DC. So, essentially my study involves the entire population of U.S. states. So, I could easily just report descriptive stats, but I want to analyze growth overtime in the outcome variable as well as examine the factors that predict the growth rate. Since I am analyzing the population, should I worry about the fit indices? My thinking is that the implied covariance matrix will be the same as the population covariance matrix, which should make the chi-square test statistic and others unnecessary. Please weigh in on this and let me know what you think. |
|
|
|
|
| Even if you don't do inference, you still want to know if the means, variance and covariances of your estimated model comes close to those in your data/population. |
|
|
| Lazarus Adua posted on Wednesday, October 18, 2017 - 10:41 am
|
|
|
| Thank you This is very helpful. With your suggestion I can say something about the model-data fit in my reporting. |
|
|
| Lazarus Adua posted on Wednesday, October 25, 2017 - 10:59 am
|
|
|
| I hope this is not a silly question. I am running a latent growth curve model for a sample of 50 and 24 time points (1990-2014). I also have 3 time-invariant and 4 time-varying predictors. Is it technically feasible to model all 24 time points? Or will you rather recommend I do every other year, which will give rise to 12 time points? My preference will be to use all 24 time points, but I've in past experienced some difficulty estimating a model with so many time points. Thank you. |
|
|
|
|
| 24 time points should work for N=50. If not with ML in wide format, then using long format and perhaps Bayes. |
|
|
| Lazarus Adua posted on Wednesday, October 25, 2017 - 11:08 am
|
|
|
| Thank you. |
|
|
| Back to top |
