process_outcome_model {simITS}

Generate an ITS extrapolation simulation.

Description

This is the primary function to use to use this approach on a given dataset.

Usage

process_outcome_model(
  outcomename,
  dat,
  t0,
  R = 400,
  summarize = FALSE,
  smooth = FALSE,
  smoother = NULL,
  fit_model = fit_model_default,
  covariates = NULL,
  plug_in = FALSE,
  ...
)

Arguments

outcomename

Name of column in dat containing the time series.

dat

Dataframe with a 'month' column for time. 'month' is assumed to be a sequence of integer values.

t0

Last pre-policy timepoint

R

Number of simulated pre-policy extrapolations to generate.

summarize

Summarise the series? (TRUE/FALSE)

smooth

Smooth the series? (TRUE/FALSE)

smoother

Function to smooth residuals, if smoothing set to TRUE. If NULL, will dynamically make a model smoother based on the fit_model method if covariates are passed. Otherwise it will use the simple smoother on the outcomes.

fit_model

The function used to fit the model to simulate from. (This model could be a seasonality model. Default is simple linear model with no covariates.)

covariates

Vector of covariate names of all covariates used in the passed model function fit_model. If null, will attempt to get list of covariates form the "lags" attribute of the passed 'fit_model'.

plug_in

Use the estimated parameters as fixed and do not include extra uncertainty of parameter estimation in the simulation. (Not recommended as it destroys inference.)

...

Extra arguments to be passed to 'extrapolate_model()'

Details

Take a given outcome variable, fit an ITS model, use it to extrapolate R plusible trajectories, and then using these trajectories, generate final impact results by averaging (if summarize is set to TRUE).

This function is basically a wrapper for 'extrapolate_model()' with some extra calls to 'make_model_smoother()' to prepare, in the case of smoothing, and adding on a summary trend via 'generate_Ybars()' in the case of summarizing.

Value

If summarize=TRUE, A dataframe with several columns of interest and one row per month of data. The columns are Ymin and Ymax, the limits of the envelope, 'range', the range of the envelope, 'SE', the standard deviation of the trajectories at that time point, 'Ysmooth' the median smoothed value at that time point (if smoothing), 'Ystar' the median unsmoothed value at that time point (regardless of smooth flag), 'Y', the observed outcome, 'Ysmooth1', the smoothed observed outcomes, and 'Ybar' the predicted outcome given the model with no autoregressive aspect.

If summarize=FALSE, a dataframe of all the raw series generated.

See Also

The core internal function that this method is a wrapper for is extrapolate_model.

Examples

data( "mecklenberg" )
t0 = 0
envelope = process_outcome_model( "pbail", mecklenberg, 
                                  t0=t0, R = 10, 
                                  summarize = TRUE, smooth=FALSE )
make_envelope_graph(envelope, t0=t0, ylab = "Proportion given bail") +
  ggplot2::labs( title="Sample ITS plot")