Placebo Tests

A placebo test is used to assess the significance of a synthetic control study by running the study once for each control unit set as treated unit and the remaining control units set as controls. See [ADH15] (section I.B) for a motivation. An example of usage is in the python notebook reproducing the weights from that paper in the package repository here

The PlaceboTest class

class pysyncon.utils.PlaceboTest

Class that carries out placebo tests by running a synthetic control study using each possible control unit as the treated unit and the remaining control units as controls. See [ADH15] for more details.

fit(dataprep: Dataprep, scm: BaseSynth, scm_options: dict = {}, max_workers: int | None = None, verbose: bool = True)

Run the placebo tests. This method is multi-process and by default will use all available processors. Use the max_workers option to change this behaviour.

Parameters:

  • dataprep (Dataprep) – Dataprep object containing data to model, by default None.

  • scm (Synth | AugSynth) – Synthetic control study to use

  • scm_options (dict, optional) – Options to provide to the fit method of the synthetic control study, valid options are any valid option that the scm_type takes, by default {}

  • max_workers (Optional[int], optional) – Maximum number of processes to use, if not provided then will use all available, by default None

  • verbose (bool, optional) – Whether or not to output progress, by default True

gaps_plot(time_period: Iterable | Series | dict | None = None, grid: bool = True, treatment_time: int | None = None, mspe_threshold: float | None = None, exclude_units: list | None = None)

Plot the gaps between the treated unit and the synthetic control for each placebo test.

Parameters:

  • time_period (Iterable | pandas.Series | dict, optional) – Time range to plot, if none is supplied then the time range used is the time period over which the optimisation happens, by default None

  • grid (bool, optional) – Whether or not to plot a grid, by default True

  • treatment_time (int, optional) – If supplied, plot a vertical line at the time period that the treatment time occurred, by default None

  • mspe_threshold (float, optional) – Remove any non-treated units whose MSPE pre-treatment is \(>\) mspe_threshold \(\times\) the MSPE of the treated unit pre-treatment. This serves to exclude any non-treated units whose synthetic control had a poor pre-treatment match to the actual relative to how the actual treated unit matched pre-treatment.

Raises:

  • ValueError – if no placebo test has been run yet

  • ValueError – if mspe_threshold is supplied but treatment_year is not.

pvalue(treatment_time: int) → float

Calculate p-value of Abadie et al’s version of Fisher’s exact hypothesis test for no effect of treatment null, see also section 2.2. of [FP18].

Parameters:

treatment_time (int) – The time period that the treatment time occurred

Returns:

p-value for null hypothesis of no effect of treatment

Return type:

float

Raises:

ValueError – if no placebo test has been run yet