Return a model from the cross-validation procedure or the final 'global' model. Caution: This function is still under development.
Usage
# S3 method for metapred
subset(
x,
select = "cv",
step = NULL,
model = NULL,
stratum = NULL,
add = TRUE,
...
)Arguments
- x
metapred object
- select
Which type of model to select: "cv" (default), "global", or (experimental) "stratified", or "stratum".
- step
Which step should be selected? Defaults to the best step. numeric is converted to name of the step: 0 for an unchanged model, 1 for the first change...
- model
Which model change should be selected? NULL (default, best change) or character name of variable or (integer) index of model change.
- stratum
Experimental. Stratum to return if select = "stratum".
- add
Logical. Add data, options and functions to the resulting object? Defaults to
TRUE. Experimental.- ...
For compatibility only.
Value
An object of class mp.cv for select = "cv" and an object of class mp.global for select = "global".
In both cases, additional data is added to the resulting object, thereby making it suitable for further methods.
Examples
data(DVTipd)
DVTipd$cluster <- letters[1:4] # Add a fictional clustering to the data.
mp <- metapred(DVTipd, strata = "cluster", formula = dvt ~ histdvt + ddimdich, family = binomial)
subset(mp) # best cross-validated model
#> Prediction models estimated in 4 strata. Coefficients:
#> (Intercept) ddimdich
#> a -3.198673 2.135779
#> b -3.555348 2.251292
#> c -19.566069 18.540216
#> d -3.891820 2.947359
#>
#> Meta-analytic models, estimated in 4 fold combinations. Coefficients:
#> (Intercept) ddimdich
#> b, c, d -3.724268 2.600774
#> a, c, d -3.432711 2.421694
#> a, b, d -3.463478 2.377572
#> a, b, c -3.318460 2.176257
#>
#> Cross-validation at stratum level yields the following performance:
#> val.strata estimate se var ci.lb ci.ub
#> b, c, d a 0.1285223 0.01986149 0.0003944788 0.08921088 0.1678338
#> a, c, d b 0.1293544 0.01729276 0.0002990394 0.09512723 0.1635816
#> a, b, d c 0.1123561 0.01717570 0.0002950048 0.07836058 0.1463516
#> a, b, c d 0.1297595 0.01948263 0.0003795729 0.09119796 0.1683211
#> measure
#> b, c, d mse
#> a, c, d mse
#> a, b, d mse
#> a, b, c mse
#>
#> Generalizability:
#> 1
#> 0.1244262
#>
subset(mp, select = "global") # Final model fitted on all strata.
#> Meta-analytic model of prediction models estimated in 4 strata. Coefficients:
#> (Intercept) ddimdich
#> -3.463480 2.377574
subset(mp, step = 1) # The best model of step 1
#> Prediction models estimated in 4 strata. Coefficients:
#> (Intercept) ddimdich
#> a -3.198673 2.135779
#> b -3.555348 2.251292
#> c -19.566069 18.540216
#> d -3.891820 2.947359
#>
#> Meta-analytic models, estimated in 4 fold combinations. Coefficients:
#> (Intercept) ddimdich
#> b, c, d -3.724268 2.600774
#> a, c, d -3.432711 2.421694
#> a, b, d -3.463478 2.377572
#> a, b, c -3.318460 2.176257
#>
#> Cross-validation at stratum level yields the following performance:
#> val.strata estimate se var ci.lb ci.ub
#> b, c, d a 0.1285223 0.01986149 0.0003944788 0.08921088 0.1678338
#> a, c, d b 0.1293544 0.01729276 0.0002990394 0.09512723 0.1635816
#> a, b, d c 0.1123561 0.01717570 0.0002950048 0.07836058 0.1463516
#> a, b, c d 0.1297595 0.01948263 0.0003795729 0.09119796 0.1683211
#> measure
#> b, c, d mse
#> a, c, d mse
#> a, b, d mse
#> a, b, c mse
#>
#> Generalizability:
#> 1
#> 0.1244262
#>
subset(mp, step = 1, model = "histdvt") # The model in which histdvt was removed, in step 1.
#> Prediction models estimated in 4 strata. Coefficients:
#> (Intercept) ddimdich
#> a -3.198673 2.135779
#> b -3.555348 2.251292
#> c -19.566069 18.540216
#> d -3.891820 2.947359
#>
#> Meta-analytic models, estimated in 4 fold combinations. Coefficients:
#> (Intercept) ddimdich
#> b, c, d -3.724268 2.600774
#> a, c, d -3.432711 2.421694
#> a, b, d -3.463478 2.377572
#> a, b, c -3.318460 2.176257
#>
#> Cross-validation at stratum level yields the following performance:
#> val.strata estimate se var ci.lb ci.ub
#> b, c, d a 0.1285223 0.01986149 0.0003944788 0.08921088 0.1678338
#> a, c, d b 0.1293544 0.01729276 0.0002990394 0.09512723 0.1635816
#> a, b, d c 0.1123561 0.01717570 0.0002950048 0.07836058 0.1463516
#> a, b, c d 0.1297595 0.01948263 0.0003795729 0.09119796 0.1683211
#> measure
#> b, c, d mse
#> a, c, d mse
#> a, b, d mse
#> a, b, c mse
#>
#> Generalizability:
#> 1
#> 0.1244262
#>