simITS (version 0.1.1)

smooth_series: Smooth a series using a static loess smoother

Description

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).

Usage

smooth_series(res, outcomename, t0, smooth_k = SMOOTH_K, post.only = TRUE, ...)

Arguments

res

A dataframe with a month column and an 'outcomename' column (which is the column that will be smoothed).

outcomename

String name of the outcome variable in dat.

t0

last pre-policy timepoint

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.

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.

...

Extra arguments (not used in this function).

Value

An updated version of the 'res' dataframe with `Ysmooth`, the smoothed predictions of the original Ystar outcome. Also includes 'Ystar' the original sequence to be smoothed.

Details

This method takes several parameters it does not use, to maintain compatability with smooth_residuals.

Examples

# NOT RUN {
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" )  
# }