multisynth: Fit staggered synth

In ebenmichael/augsynth: The Augmented Synthetic Control Method

Fit staggered synth

Description

Fit staggered synth

Usage

multisynth(
  form,
  unit,
  time,
  data,
  n_leads = NULL,
  n_lags = NULL,
  nu = NULL,
  lambda = 0,
  V = NULL,
  fixedeff = TRUE,
  n_factors = 0,
  scm = T,
  time_cohort = F,
  how_match = "knn",
  cov_agg = NULL,
  eps_abs = 1e-04,
  eps_rel = 1e-04,
  verbose = FALSE,
  ...
)

Arguments

form	outcome ~ treatment | weighting covariates | approximate matching covaraites | exact matching covariates outcomeName of the outcome of interest treatmentName of the treatment assignment variable weighting covariatesAuxiliary covariates to weight on approximate matching covariatesAuxiliary covariates to approximately match one before weighting exact matching covariatesAuxiliary covariates to exactly match on before weighting If covariates are time-varying, their average value before the first unit is treated will be used. This can be changed by supplying a custom aggregation function to cov_agg.
unit	Name of unit column
time	Name of time column
data	Panel data as dataframe
n_leads	How long past treatment effects should be estimated for, default is number of post treatment periods for last treated unit
n_lags	Number of pre-treatment periods to balance, default is to balance all periods
nu	Fraction of balance for individual balance
lambda	Regularization hyperparameter, default = 0
V	Scaling matrix for synth optimization, default NULL is identity
fixedeff	Whether to include a unit fixed effect, default TRUE
n_factors	Number of factors for interactive fixed effects, setting to NULL fits with CV, default is 0
scm	Whether to fit scm weights
time_cohort	Whether to average synthetic controls into time cohorts, default FALSE
cov_agg	Covariate aggregation function
eps_abs	Absolute error tolerance for osqp
eps_rel	Relative error tolerance for osqp
verbose	Whether to print logs for osqp
...	Extra arguments

Value

multisynth object that contains:

• "weights"weights matrix where each column is a set of weights for a treated unit

• "data"Panel data as matrices

• "imbalance"Matrix of treatment minus synthetic control for pre-treatment time periods, each column corresponds to a treated unit

• "global_l2"L2 imbalance for the pooled synthetic control

• "scaled_global_l2"L2 imbalance for the pooled synthetic control, scaled by the imbalance for unitform weights

• "ind_l2"Average L2 imbalance for the individual synthetic controls

• "scaled_ind_l2"Average L2 imbalance for the individual synthetic controls, scaled by the imbalance for unitform weights

• "n_leads", "n_lags"Number of post treatment outcomes (leads) and pre-treatment outcomes (lags) to include in the analysis

• "nu"Fraction of balance for individual balance

• "lambda"Regularization hyperparameter

• "scm"Whether to fit scm weights

• "grps"Time periods for treated units

• "y0hat"Pilot estimates of control outcomes

• "residuals"Difference between the observed outcomes and the pilot estimates

• "n_factors"Number of factors for interactive fixed effects

ebenmichael/augsynth documentation built on Sept. 9, 2024, 3:29 p.m.