ACE Model

1. The c² is 2.332136e^-14 instead of 4.377829 e^-14. I installed OpenMx now from our website and add the recommended optimizer code after the require(OpenMx) command. Funny enough I get know the value 7.337437e^12 for c². And also the numbers for the estimated a11, c11 and e11 have the wrong sign ;( But I don't think it is to important because for my results I only have to make a table with -2LL, df, AIC, A, C and E values (and therefore I read it is common to use the standardized ones). They say nothing about the platform they used.

That discrepancy in c² values is too small to be of concern. All three of those values are basically zero [edit: I think? Should 7.337437e^12 actually be 7.337437e^-12?] Also, the path coefficients labeled "a11", "c11", and "e11" are square roots of the corresponding variance components, so their sign is immaterial.

2. I somehow don't understand how it doesn't really matter which starting values I choose (they just need to be reasonable).

The optimizer is trying to find values of the free parameters that minimize the fitfunction, -2logL (equivalent to maximizing the likelihood). It has to start its search someplace, namely, at the start values we give it. Giving it reasonable start values helps it find a minimum (i.e, a solution) more quickly., and make it more likely that the minimum it finds is the global minimum, out of the whole parameter space. It's possible (depending on the difficulty of the optimization problem) for the optimizer to reach the same minimum from different sets of reasonable start values. By "reasonable" start values, we mean start values that are somewhat close to the free parameter values at the solution, that satisfy all explicit and implicit constraints on the optimization problem, and which do not create numerical difficulties for the optimizer. You might be interested in some relevant presentations from an OpenMx workshop, here and here.

Is it okey if I choose for svPa <- 0.422 (Falconer-style start)? What about svMe? Could you make an example with my data how you would make it? (Sorry it is hard to understand for me, I am quite new (2 month) in this field)

I can't even get as far as the descriptive statistics in your script without making some changes. See below:

# zyg as factor
Data$zyg <- factor(Data$zyg, levels= c(1:2),labels= c("MZ", "DZ")) #<--Probably unnecessary
#OpenMx does not allow periods in data column names:
colnames(Data) <- c("Subjects","zyg","duration_1","duration_2","quality_1","quality_2")

# ACE Model
# Matrices declared to store a, c, and e Path Coefficients
pathA <- mxMatrix( type="Full", nrow=nv, ncol=nv,
free=TRUE, values=svPa, label="a11", name="a" ) # matrix names with 11 referring to the first row and column of the matrix
pathC <- mxMatrix( type="Full", nrow=nv, ncol=nv,
free=TRUE, values=svPc, label="c11", name="c" )
pathE <- mxMatrix( type="Full", nrow=nv, ncol=nv,
free=TRUE, values=svPe, label="e11", name="e" )


I am working on a data set with 10 DZ and 15 MZ twin pairs. I measured 8 different parameters for sleep. Is it right if I do for each parameter a different ACE,CE,AE model (choosing the best according to AIC value) or am I barking up the wrong tree?
(Sorry may a stupid question, but my social enviroment only gave me the data and are not able to help me with the analysis)

Ordinarily I'd recommend doing a multitrait analysis of all 8 phenotypes, but your sample size is so small that the approach you suggest here--doing 8 separate monophenotype analyses--is probably best in your case.

Also, unless your twins are all the same sex and have minimal variation in age, you're going to want to adjust for age and sex in your model's means.