#------------#
# INPUT FILE 
#------------#

# 1. INTRODUCTION: 	
# The function "continuous_time" provides continuous time parameter estimates for panel data. It allows 
#		(a) the estimation of nonoscillating and oscillating processes (with and without dampening), 
#		(b) the estimation of uni- and multivariate processes (with and without coupling), 
#		(c) individually varying time intervals (different across individuals and/or across time)
# A description of the approach (including all equations implemented in the "continuous_time" function) is provided in Voelkle & Oud (2012)

# 2. PREPARATION:
# The program consists of two files, the present input file and the CT_SEM.R file. Both files must be located in the same folder. All specifications are 
# made in the present file. The CT_SEM.R file must not be opened or changed. In order to run the syntax it is necessary:
#		(a) to have R installed (http://www.r-project.org/)
# 		(b) to have OpenMx installed (type "source('http://openmx.psyc.virginia.edu/getOpenMx.R'" into the R Editor) 
# 		(c) to set the path of the working directory to the folder where the input file, CT_SEM.R, and the data are saved (all in one folder)

# 3. DATA:
# The data should be saved in an ASCII .dat format and should be prepared so that
#		(a) all indicators (of possibly different factors) at one time point are in a consecutive order,
#    		followed by all indicators of the second time point and so on: (e.g., x1t1, x2t1, ..., x1t2, x2t2, ..., x1T, ..., xqT)
#		(b) The last T-1 variables contain the length of the time interval for each person and time interval

# 4. MODEL SPECIFICATION:
# The model is specified by providing starting values for all parameters to be estimated. All parameters that are specified with starting values other 
# than 0 or 1 are being freely estimated. Only values of exactly 1 or 0 are treated as fixed. As a consequence it is not possible to specify a 
# starting value of 1 or 0 (but 1.001 or 0.001 is possible). All matrices below must be specified. If some are not relevant (e.g. if there are no latent
# factors but just manifest variables) simply set them to zero (i.e., LAMBDA to one and THETA to zero)

# 5. EXAMPLES
# In line with Voelkle & Oud (2012) we provide two input files and example datasets. One example is a bivariate coupled model
# without oscillation and individually varying time intervals. The other is an example of a univariate damped linear oscillator with individually varying
# time intervals. [depending on the number of different time intervals, oversamples, and computing power, the computation may take up to 20min] 

#setwd("/ophth/Continuous_Time_Lagged_Modeling/Web")		# set working directory
#setwd("U:\\supplemental_material")				# set working directory

# name of input file [rows = subjects; columns = variables in the order described above]
#input_file 	<- "/ophth/Continuous_Time_Lagged_Modeling/Web/1_data_nonoscillating_xy.dat"	
input_file 	<- "BMSP_2043_sm_data1.dat"

n.manifest	<- 2						# number of indicators for all factors 
n.latent	<- 2						# number of latent variables
Tpoints		<- 11						# number of time points

PHI1		<- matrix(c(1.01, 0.10,				# var/cov matrix of latent variables at first time point
	                    0.10, 1.01		
			),nrow=n.latent, ncol=n.latent, byrow=TRUE)			

latentM1	<- matrix(c(1.55, 1.01				# means of latent variables at first time point 
			),nrow=n.latent, ncol=1, byrow=TRUE)		

manifestM	<- matrix(c(0,0					# intercepts of manifest variables [usually fixed to zero]
			),nrow=n.manifest, ncol=1, byrow=TRUE)				

LAMBDA		<- matrix(c(1,0,				# factor loading matrix
		            0,1		
			),nrow=n.manifest, ncol=n.latent,byrow=TRUE)	

THETA		<- matrix(c(0,0,
			    0,0					# var/cov matrix of measurement error
			),nrow=n.manifest, ncol=n.manifest)
	
DRIFT     	<- matrix(c(-0.66,  0.25,			# drift matrix
			     0.50, -0.33 		
			),n.latent,n.latent, byrow=TRUE)

CINT		<- matrix(c(0.15, 0.25				# continuous time intercepts
			),ncol=n.latent, nrow=1, byrow=TRUE)

G<- matrix(c(1.1, 0.1,0,1.1),n.latent, n.latent)				# G matrix [lower left triangle must be zero] 			

D<- 30 					# number of oversamples [for lower values computation 
								# time decreases but parameter estimates become less precise]			

#do not change#
source("BMSP_2043_sm_input3.R")			
#continuous_time(input_file, Tpoints, n.latent, DRIFT, CINT, G, D, n.manifest, LAMBDA, PHI1, THETA, latentM1, manifestM)
continuous_time(input_file, Tpoints, n.latent, DRIFT, CINT, G, D, n.manifest, LAMBDA, PHI1, THETA, latentM1, manifestM)