Results Module

Container class for difference-in-differences estimation outputs.

This module provides the LWDIDResults class for encapsulating estimation outputs from both common timing and staggered adoption DiD designs. The container stores point estimates, standard errors, confidence intervals, period-specific effects, and diagnostic information. Multiple export formats are supported for reproducible research and integration with statistical reporting workflows.

The LWDIDResults object is returned by the main lwdid.lwdid() function and provides a unified interface for accessing estimation outputs regardless of whether the design uses common timing or staggered adoption. For staggered designs, additional attributes provide cohort-specific and cohort-time specific effect estimates.

Result container for difference-in-differences estimation outputs.

This module provides the LWDIDResults class for encapsulating estimation outputs from rolling transformation DiD methodology, supporting three scenarios:

  1. Small-sample common timing: Results include exact t-based inference statistics under classical linear model assumptions.

  2. Large-sample common timing: Results include asymptotic inference with heteroskedasticity-robust standard errors.

  3. Staggered adoption: Results include cohort-time specific effects, cohort-level aggregations, and overall weighted effects with flexible control group strategies.

The class implements immutable core attributes via properties to ensure result integrity, provides multiple summary formats (text, LaTeX, Excel, CSV), supports event study visualization for staggered designs, and includes period-specific effects for common timing designs.

Attributes Reference

The LWDIDResults object provides read-only access to estimation outputs through properties. Attributes are organized into categories based on their purpose.

Core Estimates

These attributes provide the primary treatment effect estimates and inference statistics.

  • att (float): Average treatment effect on the treated (ATT).

  • se_att (float): Standard error of ATT.

  • t_stat (float): t-statistic for H0: ATT = 0.

  • pvalue (float): Two-sided p-value for the t-test.

  • ci_lower (float): Lower bound of 95% confidence interval.

  • ci_upper (float): Upper bound of 95% confidence interval.

Sample Information

These attributes describe the estimation sample.

  • nobs (int): Number of observations in the regression.

  • n_treated (int): Number of treated units in the sample.

  • n_control (int): Number of control units in the sample.

  • K (int): Last pre-treatment period index.

  • tpost1 (int): First post-treatment period index.

Inference Configuration

These attributes describe the variance estimation and inference settings.

  • df_resid (int): Residual degrees of freedom from the regression.

  • df_inference (int): Degrees of freedom used for inference. For cluster-robust SE, this is G - 1; otherwise equals df_resid.

  • vce_type (str or None): Variance estimator: ‘ols’, ‘hc0’, ‘robust’/’hc1’, ‘hc2’, ‘hc3’, ‘hc4’, or ‘cluster’.

  • cluster_var (str or None): Clustering variable name (if vce='cluster').

  • n_clusters (int or None): Number of clusters (if vce='cluster').

Transformation and Controls

These attributes describe the transformation method and control variables used.

  • rolling (str): Transformation method: ‘demean’, ‘detrend’, ‘demeanq’, or ‘detrendq’.

  • controls_used (bool): Whether control variables were included.

  • controls (list): List of control variable names used.

Period-Specific Effects

  • att_by_period (pd.DataFrame or None): Period-specific ATT estimates with columns: period, tindex, beta, se, ci_lower, ci_upper, tstat, pval, N. First row contains the overall average effect.

Randomization Inference

These attributes are available only when ri=True is specified in the lwdid() call.

  • ri_pvalue (float or None): Randomization inference p-value.

  • ri_seed (int or None): Random seed used for RI.

  • rireps (int or None): Number of RI replications.

  • ri_method (str or None): RI method: ‘bootstrap’ or ‘permutation’.

  • ri_valid (int or None): Number of valid (successful) RI replications.

  • ri_failed (int or None): Number of failed RI replications.

Regression Coefficients

These attributes provide access to the full regression output.

  • params (array-like): Full vector of regression coefficients.

  • bse (array-like): Standard errors of all coefficients.

  • vcov (array-like): Variance-covariance matrix.

Staggered-Specific Attributes

These attributes are available only for staggered adoption designs (when gvar is specified instead of post).

Design Information:

  • is_staggered (bool): Whether this is a staggered DiD design.

  • cohorts (list): Sorted list of treatment cohorts (first treatment periods).

  • cohort_sizes (dict): Number of units in each cohort.

  • n_never_treated (int or None): Number of never-treated units.

  • control_group (str or None): User-specified control group strategy.

  • control_group_used (str or None): Actual control group strategy used (may differ due to auto-switching).

  • aggregate (str or None): Aggregation level: ‘none’, ‘cohort’, or ‘overall’.

  • estimator (str or None): Estimation method: ‘ra’, ‘ipw’, ‘ipwra’, or ‘psm’.

Effect Estimates:

  • att_by_cohort_time (pd.DataFrame or None): (g,r)-specific ATT estimates with columns: cohort, period, event_time, att, se, ci_lower, ci_upper, t_stat, pvalue, n_treated, n_control.

  • att_by_cohort (pd.DataFrame or None): Cohort-specific ATT estimates (if aggregate='cohort' or 'overall').

  • att_overall (float or None): Overall weighted ATT (if aggregate='overall').

  • se_overall (float or None): Standard error of overall ATT.

  • ci_overall_lower (float or None): 95% CI lower bound for overall ATT.

  • ci_overall_upper (float or None): 95% CI upper bound for overall ATT.

  • t_stat_overall (float or None): t-statistic for overall ATT.

  • pvalue_overall (float or None): p-value for overall ATT.

  • cohort_weights (dict): Cohort weights for overall effect (\(\omega_g = N_g / N_{treat}\)).

Methods Reference

Display Methods

summary()

Generate a formatted summary of estimation results. For staggered designs, automatically dispatches to summary_staggered(). Returns a string suitable for console output.

print(results.summary())
summary_staggered()

Generate a formatted summary for staggered DiD results. Displays treatment cohorts, sample sizes, control group strategy, overall weighted effect (if aggregate='overall'), and cohort-specific effects. Raises ValueError if called on non-staggered results.

Visualization Methods

plot(gid=None, graph_options=None)

Generate a time series plot of residualized outcomes for treated and control groups. A vertical line indicates the treatment start period.

Parameters:

  • gid (str or int, optional): Specific unit ID to highlight.

  • graph_options (dict, optional): Matplotlib customization options.

Returns: matplotlib.figure.Figure

fig = results.plot()
fig.savefig('trends.png')
plot_event_study(…)

Generate an event study diagram for staggered DiD results. Aggregates cohort-time specific effects by event time (e = r - g) and visualizes dynamic treatment effects relative to a reference period.

Key parameters:

  • ref_period (int, optional): Reference period for normalization. Default 0.

  • show_ci (bool): Display 95% confidence interval shading. Default True.

  • aggregation ({‘mean’, ‘weighted’}): Cross-cohort aggregation method.

  • include_pre_treatment (bool): Include pre-treatment periods. Default True.

  • return_data (bool): Also return the aggregated event study DataFrame.

Returns: (fig, ax) or (fig, ax, event_df) if return_data=True.

fig, ax = results.plot_event_study(
    title='Policy Effect',
    ylabel='Treatment Effect'
)
fig.savefig('event_study.png', dpi=300)

Export Methods

to_excel(path)

Export results to an Excel file with multiple sheets. For common timing: Summary sheet with core statistics, ByPeriod sheet with period-specific effects, and RI sheet if randomization inference was performed. For staggered designs: Summary, Overall, Cohort, CohortTime, Weights, and Metadata sheets.

results.to_excel('results.xlsx')
to_csv(path)

Export period-specific effects (att_by_period) to CSV format.

results.to_csv('period_effects.csv')
to_latex(path)

Export results to a LaTeX table file. Generates publication-ready tables containing summary statistics and period-specific effects.

results.to_latex('table.tex')

Usage by Design Type

Common Timing Design

For common timing designs (specified with the post parameter), the following attributes are available:

from lwdid import lwdid

results = lwdid(data, y='outcome', d='treated', ivar='unit',
                tvar='year', post='post', rolling='detrend')

# Core estimates
print(f"ATT: {results.att:.4f}")
print(f"SE: {results.se_att:.4f}")
print(f"t-stat: {results.t_stat:.3f}")
print(f"p-value: {results.pvalue:.4f}")
print(f"95% CI: [{results.ci_lower:.4f}, {results.ci_upper:.4f}]")

# Sample information
print(f"N = {results.nobs}, df = {results.df_inference}")
print(f"Treated: {results.n_treated}, Control: {results.n_control}")

# Period-specific effects
print(results.att_by_period)

# Staggered-specific attributes are None
assert results.is_staggered == False
assert results.att_overall is None

Staggered Adoption Design

For staggered adoption designs (specified with the gvar parameter), additional attributes are available:

from lwdid import lwdid

results = lwdid(data, y='outcome', ivar='unit', tvar='year',
                gvar='first_treat_year', rolling='demean',
                aggregate='overall', control_group='not_yet_treated')

# Staggered design indicator
assert results.is_staggered == True

# Overall weighted effect (when aggregate='overall')
print(f"Overall ATT: {results.att_overall:.4f}")
print(f"SE: {results.se_overall:.4f}")
print(f"95% CI: [{results.ci_overall_lower:.4f}, {results.ci_overall_upper:.4f}]")

# Cohort information
print(f"Cohorts: {results.cohorts}")
print(f"Cohort sizes: {results.cohort_sizes}")
print(f"Cohort weights: {results.cohort_weights}")

# Cohort-specific effects (when aggregate='cohort' or 'overall')
print(results.att_by_cohort)

# (g,r)-specific effects (always available)
print(results.att_by_cohort_time)

# Control group information
print(f"Control group requested: {results.control_group}")
print(f"Control group used: {results.control_group_used}")

# Event study visualization
fig, ax = results.plot_event_study()

Attribute Availability Summary

The following table summarizes attribute availability by design type:

  • att: Common timing = Yes, Staggered = Yes

  • se_att: Common timing = Yes, Staggered = Yes

  • att_by_period: Common timing = Yes, Staggered = No

  • is_staggered: Common timing = False, Staggered = True

  • att_overall: Common timing = No, Staggered = when aggregate='overall'

  • att_by_cohort: Common timing = No, Staggered = when aggregate='cohort' or 'overall'

  • att_by_cohort_time: Common timing = No, Staggered = Yes (always)

  • cohorts: Common timing = Empty list, Staggered = Yes

  • cohort_weights: Common timing = Empty dict, Staggered = Yes

LWDIDResults Class

class lwdid.results.LWDIDResults(results_dict, metadata, att_by_period=None, cohort_time_effects=None)[source]

Bases: object

Container for difference-in-differences estimation results.

Stores all estimation outputs from the lwdid() function implementing the rolling transformation methodology. Supports three scenarios:

  1. Small-sample common timing: Exact t-based inference under classical linear model assumptions.

  2. Large-sample common timing: Asymptotic inference with heteroskedasticity-robust standard errors.

  3. Staggered adoption: Cohort-time specific effects with flexible control group strategies.

All core attributes are read-only properties to ensure result integrity. Provides methods for displaying, visualizing, and exporting results.

Variables:

  • att (float) – Average treatment effect on the treated (ATT) point estimate.

  • se_att (float) – Standard error of ATT.

  • t_stat (float) – t-statistic for H0: ATT = 0.

  • pvalue (float) – Two-sided p-value for t-test.

  • ci_lower (float) – Lower bound of 95% confidence interval.

  • ci_upper (float) – Upper bound of 95% confidence interval.

  • nobs (int) – Number of observations in the regression.

  • n_treated (int) – Number of treated units.

  • n_control (int) – Number of control units.

  • K (int) – Last pre-treatment period index.

  • tpost1 (int) – First post-treatment period index.

  • df_resid (int) – Residual degrees of freedom from the main regression.

  • df_inference (int) – Degrees of freedom used for inference. For cluster-robust standard errors, this is G - 1 (number of clusters minus 1). For other variance estimators, this equals df_resid.

  • rolling (str) – Transformation method used (‘demean’, ‘detrend’, ‘demeanq’, or ‘detrendq’).

  • vce_type (str) – Variance estimator type (‘ols’, ‘robust’, ‘hc1’, ‘hc3’, or ‘cluster’).

  • cluster_var (str or None) – Clustering variable name (if vce=’cluster’).

  • n_clusters (int or None) – Number of clusters (if vce=’cluster’).

  • controls_used (bool) – Whether control variables were included.

  • controls (list) – List of control variable names used.

  • params (array-like) – Full vector of regression coefficients.

  • bse (array-like) – Standard errors of all coefficients.

  • vcov (array-like) – Variance-covariance matrix.

  • att_by_period (pd.DataFrame) – Period-specific ATT estimates with columns: period, tindex, beta, se, ci_lower, ci_upper, tstat, pval, N.

  • ri_pvalue (float or None) – Randomization inference p-value (if ri=True was specified).

  • ri_seed (int or None) – Random seed used for RI.

  • rireps (int or None) – Number of RI permutations.

  • ri_method (str or None) – Randomization inference method used (‘bootstrap’ or ‘permutation’). Only available if ri=True was specified.

  • ri_valid (int or None) – Number of valid (successful) RI replications. Only available if ri=True.

  • ri_failed (int or None) – Number of failed RI replications. Only available if ri=True.

  • data (pd.DataFrame) – Transformed data used for regression. Contains the rolling-transformed outcome variable and other regression variables. If the original ivar was string type, this DataFrame contains numeric IDs (1, 2, 3, …). The mapping between original and numeric IDs is stored in data.attrs[‘id_mapping’].

  • is_staggered (bool) – Whether this is a staggered DiD estimation.

  • cohorts (list of int) – Sorted list of treatment cohorts (first treatment periods). Only available when is_staggered=True.

  • cohort_sizes (dict) – Number of units in each cohort. Only available when is_staggered=True.

  • att_by_cohort_time (pd.DataFrame or None) – (g,r)-specific ATT estimates with columns: cohort, period, event_time, att, se, ci_lower, ci_upper, t_stat, pvalue, n_treated, n_control. Only available when is_staggered=True.

  • att_by_cohort (pd.DataFrame or None) – Cohort-specific ATT estimates (if aggregate=’cohort’ or ‘overall’). Only available when is_staggered=True.

  • att_overall (float or None) – Overall weighted ATT (if aggregate=’overall’). Only available when is_staggered=True.

  • se_overall (float or None) – Standard error of overall ATT. Only available when is_staggered=True.

  • cohort_weights (dict) – Cohort weights used for overall effect (omega_g = N_g / N_treat). Only available when is_staggered=True.

  • control_group (str or None) – User-specified control group strategy. Only available when is_staggered=True.

  • control_group_used (str or None) – Actual control group strategy used (may differ due to auto-switching). Only available when is_staggered=True.

  • aggregate (str or None) – Aggregation level (‘none’, ‘cohort’, ‘overall’). Only available when is_staggered=True.

  • estimator (str or None) – Estimation method (‘ra’, ‘ipw’, ‘ipwra’, ‘psm’). Only available when is_staggered=True.

summary()[source]

Returns formatted results summary string.

plot(gid=None, graph_options=None)[source]

Generates plot of residualized outcomes (control vs. treated).

plot_event_study(**kwargs)[source]

Generates event study diagram for staggered designs.

to_excel(path)[source]

Exports results to Excel file with multiple sheets.

to_csv(path)[source]

Exports period-specific effects to CSV.

to_latex(path)[source]

Exports results to LaTeX table format.

See also

lwdid

Main estimation function that produces LWDIDResults objects.

__init__(results_dict, metadata, att_by_period=None, cohort_time_effects=None)[source]

Initialize LWDIDResults container with estimation outputs.

Parameters:

  • results_dict (dict) – Estimation results containing keys: ‘att’, ‘se_att’, ‘t_stat’, ‘pvalue’, ‘ci_lower’, ‘ci_upper’, ‘nobs’, ‘df_resid’, ‘params’, ‘bse’, ‘vcov’, ‘resid’, ‘vce_type’.

  • metadata (dict) – Metadata containing keys: ‘K’, ‘tpost1’, ‘depvar’, ‘N_treated’, ‘N_control’.

  • att_by_period (pd.DataFrame, optional) – Period-specific effect estimates for common timing design.

  • cohort_time_effects (list, optional) – List of CohortTimeEffect objects for staggered designs.

property att: float

ATT point estimate.

property se_att: float

Standard error of ATT.

property t_stat: float

t-statistic for ATT.

property pvalue: float

Two-sided p-value.

property ci_lower: float

95% CI lower bound.

property ci_upper: float

95% CI upper bound.

property nobs: int

Number of observations.

property n_treated: int

Number of treated units.

property n_control: int

Number of control units.

property df_resid: int

Residual degrees of freedom.

property df_inference: int

Degrees of freedom for inference.

property K: int

Last pre-treatment period index.

property tpost1: int

First post-treatment period index.

property cmd: str

Command name.

property depvar: str

Dependent variable name.

property rolling: str

Transformation method used.

property vce_type: str | None

Variance estimator type.

property cluster_var: str | None

Clustering variable name.

property n_clusters: int | None

Number of clusters.

property controls_used: bool

Whether control variables were included.

property controls: list

List of control variable names.

property params

Full coefficient vector.

property bse

Standard errors of coefficients.

property vcov

Variance-covariance matrix.

property att_by_period: DataFrame | None

Period-specific ATT estimates (returns copy).

property ri_pvalue: float | None

Randomization inference p-value.

property ri_seed: int | None

Random seed used for RI.

property rireps: int | None

Number of RI replications.

property ri_method: str | None

Randomization inference method.

property ri_valid: int | None

Number of valid RI replications.

property ri_failed: int | None

Number of failed RI replications.

property data: DataFrame | None

Transformed data used for regression.

property is_staggered: bool

Whether this is a staggered DiD estimation.

property cohorts: list

List of treatment cohorts.

property cohort_sizes: dict

Number of units in each cohort.

property att_by_cohort_time: DataFrame | None

Cohort-time specific ATT estimates (returns copy).

property att_by_cohort: DataFrame | None

Cohort-specific ATT estimates (returns copy).

property att_overall: float | None

Overall weighted ATT estimate.

property se_overall: float | None

Standard error of overall ATT.

property ci_overall_lower: float | None

95% CI lower bound for overall ATT.

property ci_overall_upper: float | None

95% CI upper bound for overall ATT.

property t_stat_overall: float | None

t-statistic for overall ATT.

property pvalue_overall: float | None

p-value for overall ATT.

property cohort_weights: dict

Cohort weights for overall effect.

property control_group: str | None

User-specified control group strategy.

property control_group_used: str | None

Actual control group strategy used.

property aggregate: str | None

Aggregation level.

property estimator: str | None

Estimation method.

property n_never_treated: int | None

Number of never-treated units.

property att_pre_treatment: DataFrame | None

Pre-treatment ATT estimates (returns copy).

DataFrame with columns: cohort, period, event_time, att, se, ci_lower, ci_upper, t_stat, pvalue, n_treated, n_control, is_anchor, rolling_window_size.

Only available when include_pretreatment=True was specified during estimation.

property parallel_trends_test: ParallelTrendsTestResult | None

Parallel trends test results.

Contains individual t-tests for each pre-treatment period and joint F-test for H0: all pre-treatment ATT = 0.

Only available when include_pretreatment=True and pretreatment_test=True were specified during estimation.

property include_pretreatment: bool

Whether pre-treatment dynamics were computed.

summary()[source]

Generate a formatted summary of estimation results.

For staggered designs, dispatches to summary_staggered(). For common timing designs, displays ATT estimate, standard error, t-statistic, p-value, confidence interval, and period-specific effects if available.

Returns:

Formatted results summary string suitable for console output.

Return type:

str

summary_staggered()[source]

Generate a formatted summary for staggered DiD estimation results.

Displays treatment cohorts, sample sizes, control group strategy, overall weighted effect (if aggregate=’overall’), and cohort-specific effects (if aggregate=’cohort’ or ‘overall’).

Returns:

Formatted results summary string suitable for console output.

Return type:

str

Raises:

ValueError – If called on non-staggered estimation results.

Notes

The summary output varies by aggregation level. When aggregate=’none’, only (g,r)-specific effects are available. When aggregate=’cohort’, cohort-specific effects are shown. When aggregate=’overall’, both cohort-specific and overall weighted effects are displayed. If the control group strategy was automatically switched from the user-specified value, a notification is included.

__repr__()[source]

Return a concise string representation of the results object.

Return type:

str

__str__()[source]

Return the formatted summary as the string representation.

Return type:

str

property metadata: dict[str, Any]

Internal metadata dictionary (returns copy).

plot(gid=None, graph_options=None)[source]

Generate a plot of residualized outcomes for treated and control groups.

Creates a time series plot comparing the average residualized outcomes between treated and control units across all time periods. A vertical line indicates the treatment start period.

Parameters:

  • gid (str or int, optional) – Specific unit ID to highlight. If provided, plots the individual unit trajectory along with group averages.

  • graph_options (dict, optional) – Matplotlib customization options including ‘title’, ‘xlabel’, ‘ylabel’, ‘figsize’, ‘colors’, and other styling parameters.

Returns:

The generated matplotlib figure object.

Return type:

matplotlib.figure.Figure

Raises:

ValueError – If results.data is not set (plotting requires transformed data).

See also

plot_event_study

Event study visualization for staggered designs.

plot_event_study(ref_period=0, show_ci=True, aggregation='mean', include_pre_treatment=True, alpha=0.05, df_strategy='conservative', title=None, xlabel=None, ylabel=None, figsize=(10, 6), savefig=None, dpi=150, ax=None, return_data=False, **kwargs)[source]

Generate an event study diagram for staggered DiD results.

Aggregates cohort-time specific effects by event time (e = r - g) and visualizes dynamic treatment effects relative to a reference period.

Parameters:

  • ref_period (int or None, optional) – Reference period for normalization (event time). Default is 0 (first treatment period). If None, no normalization is performed.

  • show_ci (bool, optional) – Whether to display confidence interval shading. Default True.

  • aggregation ({'mean', 'weighted'}, optional) – Cross-cohort aggregation method. ‘mean’ computes simple average with SE = sqrt(sum(se^2))/n. ‘weighted’ uses cohort weights with SE = sqrt(sum(w^2 * se^2)). Default ‘mean’.

  • include_pre_treatment (bool, optional) – Whether to include pre-treatment periods (e < 0). Default True.

  • alpha (float, optional) – Significance level for confidence intervals. Default 0.05 (95% CI).

  • df_strategy ({'conservative', 'weighted', 'fallback'}, optional) – Strategy for selecting degrees of freedom for t-distribution: - ‘conservative’: min(df_g) across cohorts (default) - ‘weighted’: weighted average of df_g - ‘fallback’: n_cohorts - 1

  • title (str, optional) – Plot title. Default ‘Event Study: Dynamic Treatment Effects’.

  • xlabel (str, optional) – X-axis label. Default ‘Event Time (Periods Since Treatment)’.

  • ylabel (str, optional) – Y-axis label. Default ‘Treatment Effect’.

  • figsize (tuple of int, optional) – Figure size in inches (width, height). Default (10, 6).

  • savefig (str, optional) – File path to save the figure. If provided, saves automatically.

  • dpi (int, optional) – Resolution for saved figure. Default 150.

  • ax (matplotlib.axes.Axes, optional) – Existing axes object to plot on. If None, creates new figure.

  • return_data (bool, optional) – If True, also returns the aggregated event study DataFrame. Default False.

  • **kwargs – Additional keyword arguments passed to matplotlib plotting functions.

Returns:

  • fig (matplotlib.figure.Figure) – The matplotlib figure object.

  • ax (matplotlib.axes.Axes) – The matplotlib axes object.

  • event_df (pd.DataFrame) – Aggregated event study data. Only returned if return_data=True.

Raises:

ValueError – If called on non-staggered results or if att_by_cohort_time is empty or None.

Notes

Confidence intervals use t-distribution rather than normal distribution for proper small-sample inference. The degrees of freedom are selected based on the df_strategy parameter.

See also

plot

Residualized outcomes plot for common timing designs.

to_excel(path)[source]

Export estimation results to an Excel file.

For common timing designs, creates a workbook with Summary sheet containing ATT, SE, t-statistic, p-value, CI bounds, and sample sizes. If period-specific effects are available, includes a ByPeriod sheet. For staggered designs, dispatches to to_excel_staggered().

Parameters:

path (str) – File path for the Excel output (.xlsx extension recommended).

See also

to_csv

Export period-specific effects to CSV format.

to_latex

Export results to LaTeX table format.

to_excel_staggered(path)[source]

Export staggered DiD results to a multi-sheet Excel file.

Creates an Excel workbook with sheets tailored to the aggregation level. The Summary sheet is always included. Additional sheets depend on the aggregate parameter used during estimation.

Parameters:

path (str) – File path for the Excel output (.xlsx extension required).

Raises:

  • ValueError – If called on non-staggered estimation results.

  • ImportError – If openpyxl package is not installed.

Notes

Sheet structure varies by aggregation level:

  • aggregate=’overall’: Summary, Overall, Cohort, CohortTime, Weights, Metadata

  • aggregate=’cohort’: Summary, Cohort, CohortTime, Weights, Metadata

  • aggregate=’none’: Summary, CohortTime, Metadata

to_csv(path)[source]

Export period-specific treatment effects to a CSV file.

Parameters:

path (str) – File path for the CSV output.

Raises:

ValueError – If att_by_period is not available (None or empty DataFrame).

See also

to_excel

Export comprehensive results to Excel format.

to_latex(path)[source]

Export estimation results to a LaTeX table file.

Generates a LaTeX document containing summary statistics (ATT, SE, t-statistic, p-value, CI bounds, sample sizes) and period-specific effects if available.

Parameters:

path (str) – File path for the LaTeX output (.tex extension recommended).

See also

to_excel

Export comprehensive results to Excel format.

to_csv

Export period-specific effects to CSV format.

Example Usage

Basic Example

from lwdid import lwdid

# Run estimation
results = lwdid(data, y='outcome', d='treated', ivar='unit',
                tvar='year', post='post', rolling='detrend')

# Display summary
print(results.summary())

# Access attributes
print(f"ATT: {results.att:.4f} (SE: {results.se_att:.4f})")
print(f"95% CI: [{results.ci_lower:.4f}, {results.ci_upper:.4f}]")

# Export results
results.to_excel('results.xlsx')

Staggered Design Example

from lwdid import lwdid

# Estimate with staggered design
results = lwdid(data, y='lhomicide', ivar='state', tvar='year',
                gvar='effyear', rolling='demean',
                aggregate='overall', control_group='not_yet_treated')

# Display staggered-specific summary
print(results.summary())

# Access overall effect
print(f"Overall ATT: {results.att_overall:.4f}")

# Access cohort-level effects
print(results.att_by_cohort)

# Generate event study plot
fig, ax = results.plot_event_study()
fig.savefig('event_study.png')

# Export with all sheets
results.to_excel('staggered_results.xlsx')

See Also