I have a model of 4 categorical independent variables and 1 dependent categorical variable that are in a class of 1...40 possible values. I have a set of over 1000 previous "draws" (I am using a term from the mc1.pdf document posted on this site that gave me a track of using Mplus for my problem). I would like to use some version (not sure which) of montecarlo model that could allow me to estimate the N+1 dependent variable based on the previous 1...N independent variable sets. My question is how many of 1...N "draws" are required counting that this variables come in a chaotic, random (Markov Chain) fashion. Could I define a model that could use (based on a parameter) a lag model for estimating the N+1 dependent variable? By lag I mean spliting the 1..N data in subsets of 10/9/7...starting backwards form N and using this subsets to predict the next_in_series dependent variable? I hope I am not too confusing, if someone has found among the examples posted on this site something similar I would be grateful to receive a hint.
Witk thanks in advance
bmuthen posted on Saturday, February 26, 2005 - 6:02 pm
Let me ask you some questions to give me a clearer picture of what you have in mind.
1. why do you regard the dependent variable as categorical if it has 40 possible values? Non-equidistance?
2. Are the 4 categorical independent variables and the 1 dependent variable all observed at several time points? N time points?
3. Is the Markov chain with respect to the dependent variable or also wrt the independents?