computeModelUncertainty: Approximate uncertainty contributed by models (using the...

In saharaja/imputeORS: Impute Missing Values in the BLS Occupational Requirements Survey

Description

One of the major sources of uncertainty in this analysis is from the models themselves. We attempt to quantify this using the predictions generated for the known values.

Usage

computeModelUncertainty(
  model.results.soc2,
  model.results.soc3,
  conv.iter.soc2,
  conv.iter.soc3,
  soc2.prop,
  soc3.prop
)

Arguments

model.results.soc2	Results of iterative modeling, usually from SOC2 smart guessed data (output of iterateModel())
model.results.soc3	Results of iterative modeling, usually from SOC3 smart guessed data (output of iterateModel())
conv.iter.soc2	Convergence iteration of model.results.soc2 (calculated by computeConvergence())
conv.iter.soc3	Convergence iteration of model.results.soc3 (calculated by computeConvergence())
soc2.prop	Contribution of model.results.soc2 to blending (calculated by computeBlendingRatio())
soc3.prop	Contribution of model.results.soc3 to blending (calculated by computeBlendingRatio())

Details

At each iteration, predictions are generated for the known values. However, during the adjustment phase, these predictions are reset to the actual value associated with these known observations in preparation for the next iteration. Prior to this reset however, the predictions are corrected to adhere to boundary constraints (all estimates must be on the interval [0,1]). These boundary-corrected predictions are saved for each iteration, and then used to calculate the MAE and ME of known value predictions. This is done for each simulation at the convergence iteration, and then the average MAE and ME across simulations is also calculated.

Note that the predictions used to complete this calculation are the result of blending the boundary-corrected predictions according to the proportions calculated in the k-folds cross validation portion of the analysis.

Value

A list of length four, containing the MAE by simulation, the ME by simulation, the average MAE across simulations, and the average ME across simulations

See Also

iterateModel()

computeConvergence()

computeBlendingRatio()

ModelMetrics::mae()

saharaja/imputeORS documentation built on Feb. 4, 2022, 12:27 a.m.