R API Reference

Complete documentation of leanfe R functions

leanfe

The main function for fixed effects regression in R.

library(leanfe)

result <- leanfe(
    data,
    y_col = NULL,
    x_cols = NULL,
    fe_cols = NULL,
    formula = NULL,
    weights = NULL,
    demean_tol = 1e-3,
    n_iter = 100,
    vcov = "iid",
    cluster_cols = NULL,
    ssc = FALSE,
    sample_frac = NULL,
    backend = "polars"
)

Parameters

Parameter	Type	Default	Description
`data`	`character` or `data.frame`	required	Input data: data frame or path to parquet file
`y_col`	`character`	`NULL`	Dependent variable column (optional if formula provided)
`x_cols`	`character vector`	`NULL`	Independent variable columns (optional if formula provided)
`fe_cols`	`character vector`	`NULL`	Fixed effect columns (optional if formula provided)
`formula`	`character`	`NULL`	R-style formula: `"y ~ x1 + x2 \| fe1 + fe2"`
`weights`	`character`	`NULL`	Column name for regression weights (WLS)
`demean_tol`	`numeric`	`1e-3`	Convergence tolerance for iterative demeaning
`n_iter`	`integer`	`100`	Maximum iterations for demeaning
`vcov`	`character`	`"iid"`	Variance estimator: `"iid"`, `"HC1"`, or `"cluster"`
`cluster_cols`	`character vector`	`NULL`	Clustering variables (required if `vcov="cluster"`)
`ssc`	`logical`	`FALSE`	Small sample correction for clustered SEs
`sample_frac`	`numeric`	`NULL`	Fraction of data to sample (e.g., 0.1 for 10%)
`backend`	`character`	`"polars"`	Computation backend: `"polars"` or `"duckdb"`

Returns

A list containing:

Element	Type	Description
`coefficients`	`named numeric`	Variable names → coefficient estimates
`std_errors`	`named numeric`	Variable names → standard errors
`n_obs`	`integer`	Number of observations used
`iterations`	`integer`	Number of demeaning iterations
`vcov_type`	`character`	Type of variance estimator used
`is_iv`	`logical`	Whether IV/2SLS was used
`n_instruments`	`integer`	Number of instruments (if IV)
`n_clusters`	`integer`	Number of clusters (if clustered SEs)

Formula Syntax

The formula follows R-style syntax with extensions:

y ~ x1 + x2 | fe1 + fe2 | z1 + z2
│    │         │           │
│    │         │           └── Instruments (optional, for IV)
│    │         └── Fixed effects to absorb
│    └── Independent variables
└── Dependent variable

Special syntax:

Syntax	Description	Example
`i(var)`	Factor variable expansion	`i(region)`
`i(var, ref='val')`	Custom reference category	`i(region, ref='North')`
`x:i(var)`	Interaction with factor	`treatment:i(region)`

Examples

Basic Usage

library(leanfe)
library(dplyr)

# Generate sample data
set.seed(42)
n <- 10000

df <- tibble(
    y = rnorm(n),
    treatment = rbinom(n, 1, 0.3),
    x1 = rnorm(n),
    firm_id = sample(1:100, n, replace = TRUE),
    year = sample(2015:2024, n, replace = TRUE)
)

# Basic two-way fixed effects
result <- leanfe(
    formula = "y ~ treatment + x1 | firm_id + year",
    data = df,
    vcov = "iid"
)

print(result$coefficients)
print(result$std_errors)
print(result$n_obs)

Clustered Standard Errors

result_cluster <- leanfe(
    formula = "y ~ treatment + x1 | firm_id + year",
    data = df,
    vcov = "cluster",
    cluster_cols = c("firm_id")
)

print(result_cluster$std_errors)
print(result_cluster$n_clusters)

Using DuckDB Backend

result_duckdb <- leanfe(
    formula = "y ~ treatment + x1 | firm_id + year",
    data = df,
    vcov = "iid",
    backend = "duckdb"
)

print(result_duckdb$coefficients)

Reading from Parquet

# For large datasets, read directly from parquet
result <- leanfe(
    formula = "y ~ treatment | firm_id + year",
    data = "large_data.parquet",
    backend = "duckdb"
)

Instrumental Variables

# Add instrument
df$instrument <- rnorm(n)

result_iv <- leanfe(
    formula = "y ~ treatment | firm_id + year | instrument",
    data = df,
    vcov = "iid"
)

print(result_iv$coefficients)
print(result_iv$is_iv)
print(result_iv$n_instruments)

Weighted Regression

# Add weights
df$weight <- abs(rnorm(n, 1, 0.3))

result_wls <- leanfe(
    formula = "y ~ treatment + x1 | firm_id + year",
    data = df,
    weights = "weight",
    vcov = "iid"
)

print(result_wls$coefficients)

Package Structure

leanfe/
├── R/
│   ├── leanfe.R    # Main entry point
│   ├── polars.R        # Polars backend
│   ├── duckdb.R        # DuckDB backend
│   └── common.R        # Shared utilities
├── man/                # Documentation
└── tests/              # Unit tests

Comparison with fixest

leanfe is designed to be compatible with fixest syntax:

Feature	leanfe	fixest
Formula syntax	`y ~ x \\| fe`	`y ~ x \\| fe`
Clustered SE	`vcov="cluster"`	`vcov="cluster"`
Factor variables	`i(var)`	`i(var)`
IV regression	`y ~ x \\| fe \\| z`	`y ~ x \\| fe \\| z`

Key differences:

leanfe offers DuckDB backend for memory efficiency
leanfe uses Polars for faster in-memory computation
fixest has more advanced features (multi-way FE, etc.)