selSLSE {causalSLSE}

Knots Selection Method

Description

This is the main function to select the knots in cslseModel or slseModel objects. It returns a model with an optimal set of knots.

Usage

## S3 method for class 'cslseModel'
selSLSE(model, selType=c("BLSE", "FLSE"),
        selCrit = c("AIC", "BIC", "PVT"), 
        pvalT = function(p) 1/log(p),
        vcovType = c("HC0", "Classical", "HC1", "HC2", "HC3"),
        reSelect=FALSE, ...)
## S3 method for class 'slseModel'
selSLSE(model, selType=c("BLSE", "FLSE"),
        selCrit = c("AIC", "BIC", "PVT"), 
        pvalT = function(p) 1/log(p),
        vcovType = c("HC0", "Classical", "HC1", "HC2", "HC3"),
        reSelect=FALSE, ...)

Arguments

model

A model of class cslseModel or slseModel.

selType

The selection method: backward ("BLSE") or forward ("FLSE").

selCrit

The criterion to select the piecewise polynomial knots.

pvalT

A function to determine the p-value threshold for the significance of the coefficients. It has to be a function of one parameter, which is the average number of knots in the model.

vcovType

The type of least squares covariance matrix used to compute the p-values needed for the selection.

reSelect

By default, the stored selections are used. If reSelect is set to TRUE, the selection is re-computed.

...

Additional arguments to pass to other methods. Currently not used.

Details

It selects the knots using one of the two methods, FLSE or BLSE, with either the AIC, BIC or a p-value threshold (see the vignette for more details). Any of these selection methods requires several least squares estimations and it is performed only if the method has not been applied yet and reSelect is set to TRUE. This is possible because any new knots selection is saved into the returned model. A model may have more than one selection saved into it. The active knots (the ones used when we estimate the model) is stored into the element knots of the model and the saved selections are stored into the element selections. See below for what is included in this element.

Note that the selections for the three criteria AIC, BIC and PVT are computed and saved automatically in the returned model when selCrit is set to either "AIC" or "BIC", because it does not require many more operations to select them all once we do it for AIC or BIC. However, it is only computed for PVT when selCrit is set to "PVT".

The knots are selected jointly for all treatment groups in cslseModel objects. However, the active knots and all saved selections are stored separately for each treatment group. For example, the active knots for the treated in the cslseModel object mod are stored in mod$treated$knots. See the Vignette for more details.

Value

The method returns an object of class slseModel or cslseModel depending on which object it is applied to. When it does not already exist, the element selections is added to the slseModel object (or to each slseModel object in cslseModel objects). The element selections is a list with one or more of the following elements:

originalKnots

The original knots as selected initially by slseModel or cslseModel. This element is always included.

FLSE, BLSE

This is where selections based on the forward (FLSE) and the backward (BLSE) methods are stored.

Finally, BLSE and FLSE are lists that may contain the following elements:

AIC, BIC

A list of integer vectors, one for each covariate in the slseModel model. The vectors of integers indicate which sets the original set of knots minimizes the AIC or the BIC.

PVT

Same as the AIC and BIC, but the selection is based on a p-value threshold.

JAIC, JBIC

This criteria is for cslseModel models. It is like the AIC and BIC, but it indicates that the AIC or BIC was computes by estimating the models of all treatment groups jointly. This selection is not the same as a selection group by group.

Threshold

The p-value threshold used for the PVT criterion.

pval

A list of p-values, one for each original knots. See vignette for a definition of the p-values.

See Also

slseModel and cslseModel for model objects description and update for ways of selecting stored selections

Examples

data(simDat3)
mod1 <- cslseModel(Y~Z|~X1*X2, data=simDat3)
mod1 <- selSLSE(mod1, selType="FLSE", selCrit="AIC")

## The following does not require additional computation
## because the selection is stored in mod1

mod1 <- selSLSE(mod1, selType="FLSE", selCrit="BIC")


## But the following does

mod1 <- selSLSE(mod1, selType="BLSE", selCrit="BIC")

## See one selection:

mod1$treated$selections$BLSE$JBIC