Modelling the data
In this example, the correlation matrix of the data is a linear combination of the following covariates:
- The intercept
2. X1
2. X2
- The interaction between X1 and X2
Let’s combine all these covariates into a list.
- There is also a spatial effect, which has the following adjacency matrix:
Preparing the data
We simulate data from the standard normal distribution:
The correlation matrix of this distribution is a weighted sum of the covariates:
Computing the WSCE, SCE and IVE
The SCE estimates the linear coefficients of the covariates to estimate the correlation matrix. Notice how the squares representing different covariates have different sizes and colors.
Suppose that one suspects that the data does not follow a normal distribution. In that case, one should use our semiparamteric estimator, the IVE.
Let’s initialize the data from a different distribution,
and compute the IVE:
 One might also be worried about the model not being specified correctly. For example, one of the covariates could be unknown. In this case, one should use the WSCE.
Let us, e.g., suppose that we do not know that interactions are present. Let us simulate the data from the same normal distribution,
but compute the WSCE without X2 (NOTE: a lot faster if parallelize=TRUE and ncores>1):
 Notice that the parameter lambda is far away from 1, indicating that the model is misspecified.
## [1] "lambda: 0.0726239379078317"