smooth_residuals: Smooth residuals after model fit
In simITS: Analysis via Simulation of Interrupted Time Series (ITS) Data

Description Usage Arguments Details Value See Also Examples

Smooth a series by fitting the model to the data, smoothing the residuals, and then putting the model predictions back.

smooth_residuals(
  res,
  t0,
  outcomename,
  post.only = TRUE,
  smooth_k = SMOOTH_K,
  fit_model = fit_model_default,
  covariates = res,
  full_output = FALSE
)

`res`	A dataframe with a month column and an 'outcomename' column (which is the column that will be smoothed).
`t0`	last pre-policy timepoint
`outcomename`	String name of the outcome variable in dat.
`post.only`	If TRUE fit model and smooth post-policy only. WHY fit model on post-policy data only? Because this will make sure the fixed pre-policy does not dominate too much? We are focusing on post-policy so we want a good fitting model for that so we can get our residuals as "white noise" as possible before smoothing.
`smooth_k`	A rough proxy for the number of observations to primarily consider to kernal weight in the neighborhood of each timepoint (this is a bandwidth, and the loess smoother gets smooth_k / n as a span value). We want to smooth with an absolute bandwidth, not one as function of how long the time series is.
`fit_model`	A function that takes data, fits a linear model, and returns the fit model. This function needs an option to include (or not) lagged covariates.
`covariates`	A dataframe with all covariates needed in the model fitting defined by fit_model.
`full_output`	If TRUE give back pieces for diagnostics of smoothing process.

Use loess smoother on complete series of residuals including original data pre-policy and synthetic data post policy (i.e., smooth the entire plausible series).

A numeric vector of the smoothed residuals. If full_output=TRUE return a dataframe with several other columns: 'resid', the residuals based on Ystar and the model, ‘residStar' the smoothed residuals, ’Ybar.sm' the structural predictions of the model used for smoothing. Here the smoothed values will be 'Ysmooth'.

See smooth_series for a more vanilla version that smooths without the model fitting step.

data( "newjersey" )
smooth = smooth_series( newjersey, outcomename = "n.warrant", t0= -8,
                        smooth_k = 30,
                        post.only = FALSE)
plot( newjersey$month, newjersey$n.warrant )
lines( newjersey$month, smooth, col="red" )  

mod =  make_fit_season_model( ~ temperature )
newjersey = add_lagged_covariates( newjersey, outcomename = "n.warrant", 
                                   covariates = c("temperature") )

smooth = smooth_residuals( newjersey, outcomename = "n.warrant", t0=-8,
                           smooth_k = 30,
                           post.only = FALSE,
                           fit_model = mod )
plot( newjersey$month, newjersey$n.warrant )
lines( newjersey$month, smooth, col="red" )