psatGLM: Inference for Generalized Linear Models Selected via an...
In ammeir2/PSAT: Inference after Selection with Aggregate Testing

Description Usage Arguments Details Value See Also

psatGLM is used to estimate regression models that were selected based on a single aggregate test of the of form

β' K β > c > 0

a'β < l or a'β > u

where β is the MLE estimate for the regression coefficients.

psatGLM(X, y, contrasts = NULL, test = "wald",
  test_direction = c("two-sided", "lower", "upper"), family = "gaussian",
  resid_sd = c("null", "naive"), threshold = NULL, pval_threshold = 0.05,
  estimate_type = c("mle", "naive"), pvalue_type = c("hybrid", "polyhedral",
  "naive"), ci_type = c("switch", "polyhedral", "naive"),
  confidence_level = 0.95, verbose = TRUE, control = psatControl())

`X`	the design matrix of the regression model without an intercept.
`y`	the observed dependent variable.
`contrasts`	an optional matrix of contrasts to be tested: must have number of columns identical to `ncol(X)`. If left as `NULL`, the regression coefficients will be tested by default.
`test`	For a quadratic test, the string 'wald' or a semi-positive matrix of dimension `ncol(X) X ncol{X}`. For a linear test a vector of length `ncol(X)`.
`test_direction`	see `mvnLinear` for details.
`family`	see `glm` for details.
`resid_sd`	method for estimating the covariance matrix or, optionaly a standard deviation estimate for linear model. Built in estimation methods are Re-fitted least squares if 'naive' or intercept model if 'null'.
`threshold`	a threshold that the aggregate test must cross in order for the model to be estimated. Should be a positive scalar for quadratic tests and a vector of size 2 for linear aggregate tests.
`pval_threshold`	if threshold is not specified, then a threshold will be set such that the selection occures iff `aggregate_pval < pval_threshold`.
`estimate_type`	see `mvnQuadratic` for details.
`pvalue_type`	see `mvnQuadratic` for details.
`ci_type`	see `mvnQuadratic` for details.
`confidence_level`	see `mvnQuadratic` for details.
`verbose`	whether the progress of the inference function should be reported.
`control`	an object of type `psatControl`.

psatGLM is used to perform inference in generalized linear models that were selected via a single aggregate test. This function essentially works as a wrapper for mvnQuadratic and mvnLinear. See the documentation of mvnQuadratic for details regrading the available inference methods.

The function works by fitting a regression model using the glm function. If resid_sd is set to "naive" then the covariance matrix is estimated based on the initial model fit. If resid_sd is set to "null" then the covariance matrix will be estimated based on an intercept model. Once a model has been estimated, an aggregate test is performed based on the estimated covariance with the intercept removed and the vector of regression coefficients is passed to mvnQuadratic or mvnLinear without the intercept.

An object of type psatGLM

mvnQuadratic, mvnLinear, getCI, getPval, plot.psatGLM, summary.psatGLM, coef.pastGLM, predict.psatGLM