tregress_em: EM algorithm to find regression coefficients using...
In dcgerard/vicar: Various Ideas for Confounder Adjustment in Regression

Description Usage Arguments Author(s) References See Also

When using a non-standard t-distribution for your regression likelihood, there is a simple latent variable representation that allows us to develop an EM algorithm. This is a very similar procedure to Lange et al (1989).

tregress_em(
  Y,
  alpha,
  sig_diag,
  nu,
  lambda_init = NULL,
  Z_init = NULL,
  control_args = list()
)

`Y`	A matrix of numerics with one column. The response variables.
`alpha`	A matrix of numerics, the covariates. It must be that `nrow(Y)` is equal to `nrow(alpha)`.
`sig_diag`	A vector of numerics. The variances of the elements in `Y`, but only assumed to be known up to a scaling factor.
`nu`	A positive numeric scalar. The known degrees of freedom of the t-distribution.
`lambda_init`	A positive numeric scalar. The initial value of the variance inflation parameter. Defaults to the estimate under Gaussian errors.
`Z_init`	A matrix of numerics with one column. The initial value of the coefficients of `alpha`. Defaults to the estimate under Gaussian errors.
`control_args`	A list of control arguments for the EM algorithm that is passed to SQUAREM.

David Gerard

Lange, K.L., Little, R.J. and Taylor, J.M., 1989. "Robust statistical modeling using the t distribution." Journal of the American Statistical Association, 84(408), pp.881-896. doi: 10.1080/01621459.1989.10478852

tregress_obj for the objective function that this function maximizes, tregress_fix for the fixed point iteration that this function calls.

dcgerard/vicar documentation built on July 7, 2021, 1:08 p.m.