Learn R Programming

PanelMatch (version 3.0.0)

Matching Methods for Causal Inference with Time-Series Cross-Sectional Data

Description

Implements a set of methodological tools that enable researchers to apply matching methods to time-series cross-sectional data. Imai, Kim, and Wang (2023) proposes a nonparametric generalization of the difference-in-differences estimator, which does not rely on the linearity assumption as often done in practice. Researchers first select a method of matching each treated observation for a given unit in a particular time period with control observations from other units in the same time period that have a similar treatment and covariate history. These methods include standard matching methods based on propensity score and Mahalanobis distance, as well as weighting methods. Once matching and refinement is done, treatment effects can be estimated with standard errors. The package also offers diagnostics for researchers to assess the quality of their results.

Copy Link

Version

Install

install.packages('PanelMatch')

Monthly Downloads

680

Version

3.0.0

License

GPL (>= 3)

Maintainer

In Kim

Last Published

March 3rd, 2025

Functions in PanelMatch (3.0.0)

Get distances See distances.matched.set method

enforce_lead_restrictions

enforce_lead_restrictions check treatment and control units for treatment reversion in the lead window. Treated units must stay treated and control units must stay in control (according to the specified qoi)

extract.PanelMatch

Extract matched.set objects from PanelMatch results

distances.matched.set

Extract the distances of matched control units

Country-year level democratization data

check_time_data

check_time_data

clean_leads Function to check the lead windows in treated and control units for missing outcome data. If data is missing, remove those units from matched sets.

findBinaryTreated

findBinaryTreated

get_unrefined_balance.PanelBalance

Extract unrefined covariate balance results, if they exist

get_set_treatment_effects

Calculate matched set level treatment effects

get.matchedsets

get.matchedsets

getDits returns a vector of Dit values, as defined in the paper. They should be in the same order as the data frame containing the original problem data.

get_covariate_balance

Calculate covariate balance measures for refined and unrefined matched sets

getWits returns a vector of Wits, as defined in the paper (equation 25 or equation 23). They should be in the same order as the data frame containing the original problem data. The pts, pcs, and getWits functions act for a specific lead. So, for instance if our lead window is 0,1,2,3,4, these function must be called for each of those -- so for 0, then for 1, etc.

handle_bootstrap_placebo_parallel

handle_bootstrap_placebo_parallel

handle_conditional_se

handle_conditional_se Calculates conditional standard errors analytically, as defined in Imai et al. (2023). See PanelEstimate() for a more complete description of the standard error types.

handle_mahalanobis_calculations

handle_mahalanobis_calculations Returns a matched.set object with weights for control units, along with some other metadata

A constructor for the matched.set class.

handle_missing_data

handle_missing_data

Extract matched.set objects from PanelMatch results

handle_bootstrap_placebo

handle_bootstrap_placebo

Prepare Control Units pcs and pts create data frames with the time/id combinations--that need to be found so that they can be easily looked up in the data frame via a hash table. The data frame also contains information about the weight of that unit at particular times, so we use the hash table to look up where to put this data so that we can easily assign the appropriate weights in the original data frame containing the problem data. pcs does this for all control units in a matched set. pts does this for all treated units.

extract_differences

extract_differences This function calculates the differences from t-1 to 1 for treated and control units in the treatment variable. While functionality is somewhat trivial for current implementation of package, it will be needed for continuous treatment version of the package.

lwd_units helper function that actually subsets sets down to contain units with complete data

handle_ps_match

handle_ps_match Returns a matched.set object with weights for control units, along with some other metadata

handle_moderating_variable

handle_moderating_variable

handle_bootstrap_parallel

handle_bootstrap_parallel

handle_ps_weighted

handle_ps_weighted

get_unrefined_balance

Extract just the unrefined covariate balance results, if they exist

Create basic plots of PanelData objects

plot_matched_set

Helper function for plotting the distribution of matched set sizes

handle_bootstrap

handle_bootstrap

handle_unconditional_se

handle_conditional_se Calculates conditional standard errors analytically, as defined in Imai et al. (2023). See PanelEstimate() for a more complete description of the standard error types.

plot.PanelBalance

Plot covariate balance results Create figures displaying covariate balance results for one or more PanelMatch configurations. Users can customize these visualizations.

Conduct a placebo test

plot.PanelEstimate

Plot point estimates and standard errors from a PanelEstimate calculation.

summary.PanelData

Summarize the number of unique units and time periods in a PanelData object

perform_refinement

perform_refinement Performs refinement of matched sets, ultimately returning sets of treated observations and controls with weights. This function mostly acts as an intermediary between PanelMatch and lower level functions that do the dirty work of refinement. The function takes a lot of the same arguments as PanelMatch()

prepare_data The calculation of point estimates and standard errors first requires the calculation of a variety of different weights, parameters, and indicator variables. This function prepares the data within PanelEstimate() such that the estimates can be calculated easily. In practical terms, the function calls the lower level helpers to calculate W_its and D_its as described in Imai et al. (2023) and merges those results together with the original data to facilitate calculations.

summary.PanelBalance

Summarize covariate balance over time

plot.PanelMatch

Plot the distribution of the sizes of matched sets.

summary.PanelEstimate

Get summaries of PanelEstimate objects and calculations

print.PanelEstimate

Print point estimates and standard errors

print.PanelMatch

Print PanelMatch objects.

plot.matched.set

Plot the distribution of control unit weights

Get weights of matched control units See weights.matched.set method

print.matched.set

Print matched.set objects.

lwd_refinement master function that performs refinement with listwise deletion = TRUE

identifyDirectionalChanges

identifyDirectionalChanges Identifies changes in treatment variable for treated and control observations

weights.matched.set

Extract the weights of matched control units

summary.matched.set

Summarize information about a matched.set object and the matched sets contained within them.

summary.PanelMatch

Summarize information about a PanelMatch object and the matched sets contained within them.

set_lwd_refinement

set_lwd_refinement Performs the set-level operations for refinement with listwise deletion. See documentation for lwd_refinement for descriptions of most parameters.

print.PanelBalance

Print basic information about PanelBalance objects

Subset matched.set object

Subset PanelBalance objects

print.PanelData

Print PanelData objects and basic metadata

perunitSum This is a low level function that is used to calculate a value associated with each unit. This value is a weighted summation of the dependent variable, based on the Wit values discussed in Imai et al. (2023)

perunitSum_Dit Similar to perunitSum, this is a low level helper function for calculating specific values defined in Imai et al. (2023). This focuses on Dit rather than Wit

Estimate a causal quantity of interest

Pre-process and balance panel data

Create and refine sets of matched treated and control observations

DisplayTreatment

Visualize the treatment distribution across units and time in a panel data set

expand_treated_ts

expand_treated_ts Builds a list that contains all times in a lag window that correspond to a particular treated unit. This is structured as a list of vectors. Each vector is lag + 1 units long. The overall list will be the same length as the number of matched sets

calculate_point_estimates

calculate_point_estimates Helper function that calculates the point estimates for the specified QOI

calculate_placebo_estimates

calculate_placebo_estimates

PanelMatch-package

Matching Methods for Causal Inference with Time-Series Cross-Sectional Data

calculate_estimates

calculate_estimates

build_maha_mats

build_maha_mats Builds the matrices that we will then use to calculate the mahalanobis distances for each matched set

equality_four Small helper function implementing estimation function from Imai, Kim, and Wang (2023)

equality_four_placebo

equality_four_placebo