make_many_predictions: Generate a collection of raw counterfactual trajectories
In simITS: Analysis via Simulation of Interrupted Time Series (ITS) Data

Description Usage Arguments Value Functions References Examples

Given a fit linear model 'fit0', generate R prediction series starting at t0. This takes model uncertainty into account by pulling from the pseudo-posterior of the model parameters (from Gelman and Hill arm package).

1
2
3

make_many_predictions(fit0, dat, R, outcomename, t0)

make_many_predictions_plug(fit0, dat, R, outcomename, t0)

`fit0`	The fit linear model to simulate from.
`dat`	A dataframe with the covariates needed by the model fit0 for both pre and post-policy months.
`R`	Number of series to generate.
`outcomename`	The name of the column in dat which is our outcome.
`t0`	Last month of pre-policy. Will start predicting at t0+1.

A data.frame with the collection of predicted series, one row per month per replicate (so will have R*nrow(dat) rows).

make_many_predictions_plug: This version makes multiple predictions using estimated parameters without additional uncertainty. This takes point estimates from the fit model as fixed parameters. WARNING: This method will not capture true uncertainty as it is not taking parameter uncertainty into account.

The 'arm' package, see https://cran.r-project.org/package=arm

Also see Gelman, A., & Hill, J. (2007). Data analysis using regression and multilevelhierarchical models (Vol. 1). New York, NY, USA: Cambridge University Press.

data("mecklenberg" )
mecklenberg = add_lagged_covariates( mecklenberg, "pbail"  )
mecklenberg.pre = dplyr::filter( mecklenberg, month <= 0 )
M0 = fit_model_default( mecklenberg.pre, "pbail" )
res = make_many_predictions( M0, dat=mecklenberg, outcome="pbail", t0=0, R=2 )
tail( res )