Movatterモバイル変換

Type:	Package
Title:	Fast Fixed-Effects Estimations
Version:	0.13.2
Imports:	stats, graphics, grDevices, tools, utils, methods, numDeriv,nlme, sandwich, Rcpp(≥ 1.0.5), dreamerr(≥ 1.4.0),stringmagic(≥ 1.2.0)
Suggests:	knitr, rmarkdown, data.table, plm, MASS, pander, ggplot2,lfe, tinytex, pdftools, emmeans, estimability, AER, Matrix
LinkingTo:	Rcpp
Depends:	R(≥ 3.5.0)
Description:	Fast and user-friendly estimation of econometric models with multiple fixed-effects. Includes ordinary least squares (OLS), generalized linear models (GLM) and the negative binomial. The core of the package is based on optimized parallel C++ code, scaling especially well for large data sets. The method to obtain the fixed-effects coefficients is based on Berge (2018)https://github.com/lrberge/fixest/blob/master/_DOCS/FENmlm_paper.pdf. Further provides tools to export and view the results of several estimations with intuitive design to cluster the standard-errors.
License:	GPL-3
BugReports:	https://github.com/lrberge/fixest/issues
URL:	https://lrberge.github.io/fixest/,https://github.com/lrberge/fixest
VignetteBuilder:	knitr
LazyData:	true
RoxygenNote:	7.3.2.9000
Encoding:	UTF-8
NeedsCompilation:	yes
Packaged:	2025-09-05 09:44:37 UTC; berge028
Author:	Laurent Berge [aut, cre], Sebastian Krantz [ctb], Grant McDermott [ctb], Russell Lenth [ctb], Kyle Butts [ctb]
Maintainer:	Laurent Berge <laurent.berge@u-bordeaux.fr>
Repository:	CRAN
Date/Publication:	2025-09-08 07:30:02 UTC

Fast and User-Friendly Fixed-Effects Estimations

Description

The packagefixest provides a family of functions to perform estimationswith multiple fixed-effects. Standard-errors can be easily and intuitively clustered.It also includes tools to seamlessly export the results of various estimations.

• To get started, look at theintroduction.

Details

The main features are:

• Estimation. The core functions are:feols,feglm andfemlm toestimate, respectively, linear models, generalized linear models and maximum likelihoodmodels with multiple fixed-effects. The functionfeNmlm allows the inclusion ofnon-linear in parameters right hand sides. Finallyfepoisandfenegbin are shorthands to estimate Poisson and NegativeBinomial models.

• Multiple estimations: You can perform multiple estimations at once withthestepwise functions. It's then very easy to manipulate multiple resultswith the associated methods. See an introduction in the dedicated vignette:Multiple estimations

• Easy and flexible clustering of standard-errors. By using the argumentsvcovandssc (seesummary.fixest). To have a sense of how the standard errors are computed,see the vignetteOn standard-errors.

• Visualization and exportation of results. You can visualize the results ofmultiple estimations in R, or export them in Latex using the functionetable.This vignette details how to customize the Latex tables:Exporting estimation tables.

• Plot multiple results. You can plot the coefficients and confidence intervals ofestimations easily with the functioncoefplot. This function also offers a specificlayout for interactions.

Author(s)

Maintainer: Laurent Bergelaurent.berge@u-bordeaux.fr

Other contributors:

• Sebastian Krantz [contributor]

• Grant McDermottgrantmcd@uoregon.edu (ORCID) [contributor]

• Russell Lenthrussell-lenth@uiowa.edu [contributor]

• Kyle Buttsbuttskyle96@gmail.com [contributor]

References

Berge, Laurent, 2018, "Efficient estimation of maximum likelihood modelswith multiple fixed-effects: the R package FENmlm." CREA Discussion Papers,13 ().

See Also

Useful links:

• https://lrberge.github.io/fixest/

• https://github.com/lrberge/fixest

• Report bugs athttps://github.com/lrberge/fixest/issues

Aikake's an information criterion

Description

This function computes the AIC (Aikake's, an information criterion) from afixest estimation.

Usage

## S3 method for class 'fixest'AIC(object, ..., k = 2)

Arguments

Details

The AIC is computed as:

AIC = -2\times LogLikelihood + k\times nbParams

with k the penalty parameter.

You can have more information on this criterion onAIC.

Value

It return a numeric vector, with length the same as the number of objects taken as arguments.

Author(s)

Laurent Berge

See Also

See also the main estimation functionsfemlm,feols orfeglm.Other statictics methods:BIC.fixest,logLik.fixest,nobs.fixest.

Examples

# two fitted models with different expl. variables:res1 = femlm(Sepal.Length ~ Sepal.Width + Petal.Length +             Petal.Width | Species, iris)res2 = femlm(Sepal.Length ~ Petal.Width | Species, iris)AIC(res1, res2)BIC(res1, res2)

Bayesian information criterion

Description

This function computes the BIC (Bayesian information criterion) from afixest estimation.

Usage

## S3 method for class 'fixest'BIC(object, ...)

Arguments

Details

The BIC is computed as follows:

BIC = -2\times LogLikelihood + \log(nobs)\times nbParams

with k the penalty parameter.

You can have more information on this criterion onAIC.

Value

It return a numeric vector, with length the same as the number of objects taken as arguments.

Author(s)

Laurent Berge

See Also

See also the main estimation functionsfemlm,feols orfeglm. Other statistics functions:AIC.fixest,logLik.fixest.

Examples

# two fitted models with different expl. variables:res1 = femlm(Sepal.Length ~ Sepal.Width + Petal.Length +            Petal.Width | Species, iris)res2 = femlm(Sepal.Length ~ Petal.Width | Species, iris)AIC(res1, res2)BIC(res1, res2)

Subsets a fixest_multi object

Description

Subsets a fixest_multi object using different keys.

Usage

## S3 method for class 'fixest_multi'x[i, sample, lhs, rhs, fixef, iv, I, reorder = TRUE, drop = FALSE]

Arguments

Details

The order with we we use the keys matter. Every time a keysample,lhs,rhs,fixef oriv is used, a reordering is performed to consider the leftmost-side keyto be the new root.

Use logical keys to easily reorder. For example, say the objectres contains amultiple estimation with multiple left-hand-sides, right-hand-sides and fixed-effects.By default the results are ordered as follows:lhs,fixef,rhs.If you useres[lhs = FALSE], then the new order is:fixef,rhs,lhs.Withres[rhs = TRUE, lhs = FALSE] it becomes:rhs,fixef,lhs. In both casesyou keep all estimations.

Value

It returns afixest_multi object. If there is only one estimation left in the object, thenthe result is simplified into afixest object only withdrop = TRUE.

See Also

The main fixest estimation functions:feols,fepois,fenegbin,feglm,feNmlm. Tools for mutliple fixestestimations:summary.fixest_multi,print.fixest_multi,as.list.fixest_multi,sub-sub-.fixest_multi,sub-.fixest_multi.

Examples

# Estimation with multiple samples/LHS/RHSaq = airquality[airquality$Month %in% 5:6, ]est_split = feols(c(Ozone, Solar.R) ~ sw(poly(Wind, 2), poly(Temp, 2)),                  aq, split = ~ Month)# By default: sample is the rootetable(est_split)# Let's reorder, by considering lhs the rootetable(est_split[lhs = 1:.N])# Selecting only one LHS and RHSetable(est_split[lhs = "Ozone", rhs = 1])# Taking the first root (here sample = 5)etable(est_split[I = 1])# The first and last estimationsetable(est_split[i = c(1, .N)])

Method to subselect from afixest_panel

Description

Subselection from afixest_panel which has been created with the functionpanel.Also allows to create lag/lead variables with functionsl/f ifthefixest_panel is also adata.table::data.table.

Usage

## S3 method for class 'fixest_panel'x[i, j, ...]

Arguments

Details

If the original data was also a data.table, some calls to⁠[.fixest_panel⁠ may dissolvethefixest_panel object and return a regular data.table. This is the case forsubselections with additional arguments. If so, a note is displayed on the console.

Value

It returns afixest_panel data base, with the attributes allowing to createlags/leads properly bookkeeped.

Author(s)

Laurent Berge

See Also

Alternatively, the functionpanel changes adata.frame into a panel from which thefunctionsl andf (creating leads and lags) can be called. Otherwise you can set thepanel 'live' during the estimation using the argumentpanel.id (see for example inthe functionfeols).

Examples

data(base_did)# Creating a fixest_panel objectpdat = panel(base_did, ~id+period)# Subselections of fixest_panel objects bookkeeps the leads/lags enginepdat_small = pdat[!pdat$period %in% c(2, 4), ]a = feols(y~l(x1, 0:1), pdat_small)# we obtain the same results, had we created the lags "on the fly"base_small = base_did[!base_did$period %in% c(2, 4), ]b = feols(y~l(x1, 0:1), base_small, panel.id = ~id+period)etable(a, b)# Using data.table to create new lead/lag variablesif(require("data.table")){  pdat_dt = panel(as.data.table(base_did), ~id+period)  # Variable creation  pdat_dt[, x_l1 := l(x1)]  pdat_dt[, c("x_l1", "x_f1_2") := .(l(x1), f(x1)**2)]  # Estimation on a subset of the data  #  (the lead/lags work appropriately)  feols(y~l(x1, 0:1), pdat_dt[!period %in% c(2, 4)])}

Extracts one element from afixest_multi object

Description

Extracts single elements from multiplefixest estimations.

Usage

## S3 method for class 'fixest_multi'x[[i]]

Arguments

Value

Afixest object is returned.

See Also

The main fixest estimation functions:feols,fepois,fenegbin,feglm,feNmlm. Tools for mutliple fixestestimations:summary.fixest_multi,print.fixest_multi,as.list.fixest_multi,sub-sub-.fixest_multi,sub-.fixest_multi.

Examples

base = irisnames(base) = c("y", "x1", "x2", "x3", "species")# Multiple estimationres = feols(y ~ csw(x1, x2, x3), base, split = ~species)# The first estimationres[[1]]# The second one, etcres[[2]]

Aggregates the values of DiD coefficients a la Sun and Abraham

Description

Simple tool that aggregates the value of CATT coefficients in staggereddifference-in-difference setups (see details).

Usage

## S3 method for class 'fixest'aggregate(x, agg, full = FALSE, use_weights = TRUE, ...)

Arguments

Details

This is a function helping to replicate the estimator from Sun and Abraham (2021).You first need to perform an estimation with cohort and relative periods dummies(typically using the functioni), this leads to estimators of the cohortaverage treatment effect on the treated (CATT). Then you can use this function toretrieve the average treatment effect on each relative period, or for any other wayyou wish to aggregate the CATT.

Note that contrary to the SA article, here the cohort share in the sample isconsidered to be a perfect measure for the cohort share in the population.

Value

It returns a matrix representing a table of coefficients.

Author(s)

Laurent Berge

References

Liyang Sun and Sarah Abraham, 2021, "Estimating Dynamic Treatment Effects inEvent Studies with Heterogeneous Treatment Effects". Journal of Econometrics.

Examples

## DiD example#data(base_stagg)# 2 kind of estimations:# - regular TWFE model# - estimation with cohort x time_to_treatment interactions, later aggregated# Note: the never treated have a time_to_treatment equal to -1000# Now we perform the estimationres_twfe = feols(y ~ x1 + i(time_to_treatment, treated,                            ref = c(-1, -1000)) | id + year, base_stagg)# we use the "i." prefix to force year_treated to be considered as a factorres_cohort = feols(y ~ x1 + i(time_to_treatment, i.year_treated,                              ref = c(-1, -1000)) | id + year, base_stagg)# Displaying the resultsiplot(res_twfe, ylim = c(-6, 8))att_true = tapply(base_stagg$treatment_effect_true,                  base_stagg$time_to_treatment, mean)[-1]points(-9:8 + 0.15, att_true, pch = 15, col = 2)# The aggregate effect for each periodagg_coef = aggregate(res_cohort, "(ti.*nt)::(-?[[:digit:]]+)")x = c(-9:-2, 0:8) + .35points(x, agg_coef[, 1], pch = 17, col = 4)ci_low = agg_coef[, 1] - 1.96 * agg_coef[, 2]ci_up = agg_coef[, 1] + 1.96 * agg_coef[, 2]segments(x0 = x, y0 = ci_low, x1 = x, y1 = ci_up, col = 4)legend("topleft", col = c(1, 2, 4), pch = c(20, 15, 17),       legend = c("TWFE", "True", "Sun & Abraham"))# The ATTaggregate(res_cohort, c("ATT" = "treatment::[^-]"))with(base_stagg, mean(treatment_effect_true[time_to_treatment >= 0]))# The total effect for each cohortaggregate(res_cohort, c("cohort" = "::[^-].*year_treated::([[:digit:]]+)"))

Transforms a character string into a dictionary

Description

Transforms a single character string containing a dictionary in a textual format into a proper dictionary, that is a named character vector

Usage

as.dict(x)

Arguments

Details

This function is mostly used in combination withsetFixest_dict to set the dictionary to beused in the functionetable.

Value

It returns a named character vector.

Author(s)

Laurent Berge

See Also

etable,setFixest_dict

Examples

x = "# Main vars     mpg: Miles per gallon     hp: Horsepower     # Categorical variables     cyl: Number of cylinders; vs: Engine"as.dict(x)

Transforms a fixest_multi object into a list

Description

Extracts the results from afixest_multi object and place them into a list.

Usage

## S3 method for class 'fixest_multi'as.list(x, ...)

Arguments

Value

Returns a list containing all the results of the multiple estimations.

See Also

The main fixest estimation functions:feols,fepois,fenegbin,feglm,feNmlm. Tools for mutliple fixestestimations:summary.fixest_multi,print.fixest_multi,as.list.fixest_multi,sub-sub-.fixest_multi,sub-.fixest_multi.

Examples

base = irisnames(base) = c("y", "x1", "x2", "x3", "species")# Multiple estimationres = feols(y ~ csw(x1, x2, x3), base, split = ~species)# All the results at onceas.list(res)

Sample data for difference in difference

Description

This data has been generated to illustrate the use of difference in difference functions inpackagefixest. This is a balanced panel of 104 individuals and 10 periods.About half the individuals are treated, the treatment having a positive effect onthe dependent variabley after the 5th period. The effect of the treatment ony is gradual.

Usage

data(base_did, package = "fixest")

Format

base_did is a data frame with 1,040 observations and 6 variables namedy,x1,id,period,post andtreat.

y

The dependent variable affected by the treatment.

x1

An explanatory variable.

id

Identifier of the individual.

period

From 1 to 10

post

Indicator taking value 1 if the period is strictly greater than 5, 0 otherwise.

treat

Indicator taking value 1 if the individual is treated, 0 otherwise.

Source

This data has been generated fromR.

Publication data sample

Description

This data reports the publication output (number of articles and number of citations received)for a few scientists from the start of their career to 2000.Most of the variables are processed from the Microsoft Academic Graph (MAG) data set. A few variables are randomly generated.

Usage

data(base_pub, package = "fixest")

Format

base_pub is a data frame with 4,024 observations and 10 variables. There are 200 different scientists and 51 different years (ends in 2000).

• author_id: scientist identifier

• year: current year

• affil_id: affiliation ID of the scientist's current affiliation

• affil_name: affiliation name of the scientist's current affiliation (character)

• field: field name of the scientist (character), time invariant

• nb_pub: number of publications of the scientist for the current year

• nb_cites: number of citations received by the publications of the scientist in the current year. Accounts for the citations received from articles published up to 2020.

• birth_year: birth year of the scientist (this is randomly generated)

• is_woman: 1 if the scientist is a woman, 0 otherwise (this is randomly generated)

• age: current age of the scientist (formallyyear - birth_year)

Source

The source of this data set is the Microsoft Academic Graph data set, extracted in 2020. Now a defunct project, you can find similar data onOpenAlex.

The variablesbirth_year,is_woman andage were randomly generated. All other variables have created from the raw MAG files.

Sample data for staggered difference in difference

Description

This data has been generated to illustrate the Sun and Abraham (Journal of Econometrics, 2021) method for staggered difference-in-difference. This is a balanced panel of 95 individuals and 10 periods. Half the individuals are treated. For those treated, the treatment date can vary from the second to the last period. The effect of the treatment depends on the time since the treatment: it is first negative and then increasing.

Usage

data(base_stagg, package = "fixest")

Format

base_stagg is a data frame with 950 observations and 7 variables:

• id: panel identifier.

• year: from 1 to 10.

• year_treated: the period at which the individual is treated.

• time_to_treatment: different between the year and the treatment year.

• treated: indicator taking value 1 if the individual is treated, 0 otherwise.

• treatment_effect_true: true effect of the treatment.

• x1: explanatory variable, correlated with the period.

• y: the dependent variable affected by the treatment.

Source

This data has been generated fromR.

Bins the values of a variable (typically a factor)

Description

Tool to easily group the values of a given variable.

Usage

bin(x, bin)

Arguments

Value

It returns a vector of the same length asx.

"Cutting" a numeric vector

Numeric vectors can be cut easily into: a) equal parts, b) user-specified bins.

Use"cut::n" to cut the vector inton (roughly) equal parts. Percentiles areused to partition the data, hence some data distributions can lead to create lessthann parts (for example if P0 is the same as P50).

The user can specify custom bins with the following syntax:"cut::a]b]c]". Herethe numbersa,b,c, etc, are a sequence of increasing numbers, each followedby an open or closed square bracket. The numbers can be specified as eitherplain numbers (e.g."cut::5]12[32["), quartiles (e.g."cut::q1]q3["),or percentiles (e.g."cut::p10]p15]p90]"). Values of different types can be mixed:"cut::5]q2[p80[" is valid provided the median (q2) is indeed greaterthan5, otherwise an error is thrown.

The square bracket right of each number tells whether the numbers should be includedor excluded from the current bin. For example, sayx ranges from 0 to 100,then"cut::5]" will create two bins: one from 0 to 5 and a second from 6 to 100.With"cut::5[" the bins would have been 0-4 and 5-100.

A factor is always returned. The labels always report the min and max values in each bin.

To have user-specified bin labels, just add them in the character vectorfollowing'cut::values'. You don't need to provide all of them, andNA valuesfall back to the default label. For example,bin = c("cut::4", "Q1", NA, "Q3")will modify only the first and third label that will be displayed as"Q1" and"Q3".

bin vsref

The functionsbin andref are able to do the same thing, then why use oneinstead of the other? Here are the differences:

• ref always returns a factor. This is in contrast withbin which returns,when possible, a vector of the same type as the vector in input.

• ref always places the values modified in the first place of the factor levels.On the other hand,bin tries to not modify the ordering of the levels. It is possibleto makebin mimic the behavior ofref by adding an"@" as the first element ofthe list in the argumentbin.

• when a vector (and not a list) is given in input,ref will place each element ofthe vector in the first place of the factor levels. The behavior ofbin istotally different,bin will transform all the values in the vector into a singlevalue inx (i.e. it's binning).

Author(s)

Laurent Berge

See Also

To re-factor variables:ref.

Examples

data(airquality)month_num = airquality$Monthtable(month_num)# Grouping the first two valuestable(bin(month_num, 5:6))# ... plus changing the name to '10'table(bin(month_num, list("10" = 5:6)))# ... and grouping 7 to 9table(bin(month_num, list("g1" = 5:6, "g2" = 7:9)))# Grouping every two monthstable(bin(month_num, "bin::2"))# ... every 2 consecutive elementstable(bin(month_num, "!bin::2"))# ... idem starting from the last onetable(bin(month_num, "!!bin::2"))# Using .() for list():table(bin(month_num, .("g1" = 5:6)))## with non numeric data#month_lab = c("may", "june", "july", "august", "september")month_fact = factor(month_num, labels = month_lab)# Grouping the first two elementstable(bin(month_fact, c("may", "jun")))# ... using regextable(bin(month_fact, "@may|jun"))# ...changing the nametable(bin(month_fact, list("spring" = "@may|jun")))# Grouping every 2 consecutive monthstable(bin(month_fact, "!bin::2"))# ...idem but starting from the lasttable(bin(month_fact, "!!bin::2"))# Relocating the months using "@d" in the nametable(bin(month_fact, .("@5" = "may", "@1 summer" = "@aug|jul")))# Putting "@" as first item means subsequent items will be placed firsttable(bin(month_fact, .("@", "aug", "july")))## "Cutting" numeric data#data(iris)plen = iris$Petal.Length# 3 parts of (roughly) equal sizetable(bin(plen, "cut::3"))# Three custom binstable(bin(plen, "cut::2]5]"))# .. same, excluding 5 in the 2nd bintable(bin(plen, "cut::2]5["))# Using quartilestable(bin(plen, "cut::q1]q2]q3]"))# Using percentilestable(bin(plen, "cut::p20]p50]p70]p90]"))# Mixing alltable(bin(plen, "cut::2[q2]p90]"))# NOTA:# -> the labels always contain the min/max values in each bin# Custom labels can be provided, just give them in the char. vector# NA values lead to the default labeltable(bin(plen, c("cut::2[q2]p90]", "<2", "]2; Q2]", NA, ">90%")))## With a formula#data(iris)plen = iris$Petal.Length# We need to use "x"table(bin(plen, list("< 2" = ~x < 2, ">= 2" = ~x >= 2)))

Extracts the bread matrix from fixest objects

Description

Extracts the bread matrix from fixest objects to be used to compute sandwich variance-covariance matrices.

Usage

## S3 method for class 'fixest'bread(x, ...)

Arguments

Value

Returns a matrix of the same dimension as the number of variables used in the estimation.

Examples

est = feols(Petal.Length ~ Petal.Width + Sepal.Width, iris)bread(est)

Check the fixed-effects convergence of afeols estimation

Description

Checks the convergence of afeols estimation by computing the first-order conditions of all fixed-effects (all should be close to 0)

Usage

check_conv_feols(x)## S3 method for class 'fixest_check_conv'summary(object, type = "short", ...)

Arguments

Value

It returns a list ofN elements,N being the number of variables in the estimation(dependent variable + explanatory variables +, if IV, endogenous variables and instruments). Foreach variable, all the first-order conditions for each fixed-effect are returned.

Examples

base = setNames(iris, c("y", "x1", "x2", "x3", "species"))base$FE = rep(1:30, 5)# one estimation with fixed-effects + varying slopesest = feols(y ~ x1 | species[x2] + FE[x3], base)# Checking the convergenceconv = check_conv_feols(est)# We can check that al values are close to 0summary(conv)summary(conv, "detail")

Extracts the coefficients from afixest estimation

Description

This function extracts the coefficients obtained from a model estimated withfemlm,feols orfeglm.

Usage

## S3 method for class 'fixest'coef(object, keep, drop, order, collin = FALSE, agg = TRUE, ...)## S3 method for class 'fixest'coefficients(object, keep, drop, order, collin = FALSE, agg = TRUE, ...)

Arguments

Details

The coefficients are the ones that have been found to maximize the log-likelihood of the specified model. More information can be found on the models from the estimations help pages:femlm,feols orfeglm.

Note that if the model has been estimated with fixed-effects, to obtain the fixed-effect coefficients, you need to use the functionfixef.fixest.

Value

This function returns a named numeric vector.

Author(s)

Laurent Berge

See Also

See also the main estimation functionsfemlm,feols orfeglm.summary.fixest,confint.fixest,vcov.fixest,etable,fixef.fixest.

Examples

# simple estimation on iris data, using "Species" fixed-effectsres = femlm(Sepal.Length ~ Sepal.Width + Petal.Length +            Petal.Width | Species, iris)# the coefficients of the variables:coef(res)# the fixed-effects coefficients:fixef(res)

Extracts the coefficients of fixest_multi objects

Description

Utility to extract the coefficients of multiple estimations and rearrange them into a matrix.

Usage

## S3 method for class 'fixest_multi'coef(  object,  keep,  drop,  order,  collin = FALSE,  long = FALSE,  na.rm = TRUE,  ...)## S3 method for class 'fixest_multi'coefficients(  object,  keep,  drop,  order,  collin = FALSE,  long = FALSE,  na.rm = TRUE,  ...)

Arguments

Examples

base = irisnames(base) = c("y", "x1", "x2", "x3", "species")# A multiple estimationest = feols(y ~ x1 + csw0(x2, x3), base)# Getting all the coefficients at once,# each row is a modelcoef(est)# Example of keep/drop/ordercoef(est, keep = "Int|x1", order = "x1")# To change the order of the model, use fixest_multi# extraction tools:coef(est[rhs = .N:1])# collin + long + na.rmbase$x1_bis = base$x1 # => collinearest = feols(y ~ x1_bis + csw0(x1, x2, x3), base, split = ~species)# does not display x1 since it is always collinearcoef(est)# now it doescoef(est, collin = TRUE)# longcoef(est, long = TRUE)# long but balanced (with NAs then)coef(est, long = TRUE, na.rm = FALSE)

Plots confidence intervals and point estimates

Description

This function plots the results of estimations (coefficients and confidence intervals).The functioniplot restricts the output to variables created withi, eitherinteractions with factors or raw factors.

Usage

coefplot(  ...,  objects = NULL,  style = NULL,  se,  ci_low,  ci_high,  df.t = NULL,  vcov = NULL,  cluster = NULL,  x,  x.shift = 0,  horiz = FALSE,  dict = NULL,  keep,  drop,  order,  ci.width = "1%",  ci_level = 0.95,  add = FALSE,  plot_prms = list(),  pch = c(20, 17, 15, 21, 24, 22),  col = 1:8,  cex = 1,  lty = 1,  lwd = 1,  ylim = NULL,  xlim = NULL,  pt.pch = pch,  pt.bg = NULL,  pt.cex = cex,  pt.col = col,  ci.col = col,  pt.lwd = lwd,  ci.lwd = lwd,  ci.lty = lty,  grid = TRUE,  grid.par = list(lty = 3, col = "gray"),  zero = TRUE,  zero.par = list(col = "black", lwd = 1),  pt.join = FALSE,  pt.join.par = list(col = pt.col, lwd = lwd),  ci.join = FALSE,  ci.join.par = list(lwd = lwd, col = col, lty = 2),  ci.fill = FALSE,  ci.fill.par = list(col = "lightgray", alpha = 0.5),  ref = "auto",  ref.line = "auto",  ref.line.par = list(col = "black", lty = 2),  lab.cex,  lab.min.cex = 0.85,  lab.max.mar = 0.25,  lab.fit = "auto",  xlim.add,  ylim.add,  only.params = FALSE,  sep,  as.multiple = FALSE,  bg,  group = "auto",  group.par = list(lwd = 2, line = 3, tcl = 0.75),  main = "Effect on __depvar__",  value.lab = "Estimate and __ci__ Conf. Int.",  ylab = NULL,  xlab = NULL,  sub = NULL,  i.select = NULL,  do_iplot = NULL)iplot(  ...,  i.select = 1,  objects = NULL,  style = NULL,  se,  ci_low,  ci_high,  df.t = NULL,  vcov = NULL,  cluster = NULL,  x,  x.shift = 0,  horiz = FALSE,  dict = NULL,  keep,  drop,  order,  ci.width = "1%",  ci_level = 0.95,  add = FALSE,  plot_prms = list(),  pch = c(20, 17, 15, 21, 24, 22),  col = 1:8,  cex = 1,  lty = 1,  lwd = 1,  ylim = NULL,  xlim = NULL,  pt.pch = pch,  pt.bg = NULL,  pt.cex = cex,  pt.col = col,  ci.col = col,  pt.lwd = lwd,  ci.lwd = lwd,  ci.lty = lty,  grid = TRUE,  grid.par = list(lty = 3, col = "gray"),  zero = TRUE,  zero.par = list(col = "black", lwd = 1),  pt.join = FALSE,  pt.join.par = list(col = pt.col, lwd = lwd),  ci.join = FALSE,  ci.join.par = list(lwd = lwd, col = col, lty = 2),  ci.fill = FALSE,  ci.fill.par = list(col = "lightgray", alpha = 0.5),  ref = "auto",  ref.line = "auto",  ref.line.par = list(col = "black", lty = 2),  lab.cex,  lab.min.cex = 0.85,  lab.max.mar = 0.25,  lab.fit = "auto",  xlim.add,  ylim.add,  only.params = FALSE,  sep,  as.multiple = FALSE,  bg,  group = "auto",  group.par = list(lwd = 2, line = 3, tcl = 0.75),  main = "Effect on __depvar__",  value.lab = "Estimate and __ci__ Conf. Int.",  ylab = NULL,  xlab = NULL,  sub = NULL)

Arguments

Functions

• iplot(): Plots the coefficients generated with i()

Setting custom default values

The functioncoefplot dispose of many arguments to parametrize the plots. Mostof these arguments can be set once an for all using the functionsetFixest_coefplot.See Example 3 below for a demonstration.

iplot

The functioniplot restrictscoefplot to interactions or factors createdwith the functioni. Onlyone of the i-variables will be plotted at a time.If you have several i-variables, you can navigate through them with thei.select argument.

The argumenti.select is an index that will go through all the i-variables.It will work well if the variables are pure, meaning not interacted with othervariables. If the i-variables are interacted, the index may have an odd behaviorbut will (in most cases) work all the same, just try some numbers up until you(hopefully) obtain the graph you want.

Note, importantly, that interactions of two factor variables are (in general)disregarded since they would require a 3-D plot to be properly represented.

Arguments keep, drop and order

The argumentskeep,drop andorder use regular expressions. If you are not awareof regular expressions, I urge you to learn it, since it is an extremely powerful wayto manipulate character strings (and it exists across most programming languages).

For example drop = "Wind" would drop any variable whose name contains "Wind". Note thatvariables such as "Temp:Wind" or "StrongWind" do contain "Wind", so would be dropped.To drop only the variable named "Wind", you need to usedrop = "^Wind$" (with "^" meaning beginning, resp. "$" meaning end,of the string => this is the language of regular expressions).

Although you can combine several regular expressions in a single characterstring using pipes,drop also accepts a vector of regular expressions.

You can use the special character "!" (exclamation mark) to reverse the effectof the regular expression (this feature is specific to this function).For exampledrop = "!Wind" would drop any variable that does not contain "Wind".

You can use the special character "%" (percentage) to make reference to theoriginal variable name instead of the aliased name. For example, you have avariable named"Month6", and use a dictionarydict = c(Month6="June").Thus the variable will be displayed as"June".If you want to delete that variable, you can use eitherdrop="June",ordrop="%Month6" (which makes reference to its original name).

The argumentorder takes in a vector of regular expressions, the order will follow theelements of this vector. The vector gives a list of priorities,on the left the elements with highest priority.For example, order = c("Wind", "!Inter", "!Temp") would give highest priorities tothe variables containing "Wind" (which would then appear first),second highest priority is the variables not containing "Inter", last,with lowest priority, the variables not containing "Temp".If you had the following variables: (Intercept), Temp:Wind, Wind, Temp youwould end up with the following order: Wind, Temp:Wind, Temp, (Intercept).

Author(s)

Laurent Berge

See Also

SeesetFixest_coefplot to set the default values ofcoefplot, and the estimationfunctions: e.g.feols,fepois,feglm,fenegbin.

Examples

## Example 1: Stacking two sets of results on the same graph## Estimation on Iris data with one fixed-effect (Species)# + we cluster the standard-errorsest = feols(Petal.Length ~ Petal.Width + Sepal.Width | Species,             iris, vcov = "cluster")# Now with "regular" standard-errorsest_std = summary(est, vcov = "iid")# You can plot the two results at oncecoefplot(est, est_std)# You could also use the argument vcovcoefplot(est, vcov = list("cluster", "iid"))# Alternatively, you can use the argument x.shift# to do it sequentially:# First graph with clustered standard-errorscoefplot(est, x.shift = -.2)# 'x.shift' was used to shift the coefficients to the left.# Second set of results: this time with#  standard-errors that are not clustered.coefplot(est, vcov = "iid", x.shift = .2,         add = TRUE, col = 2, ci.lty = 2, pch = 15)legend("topright", col = 1:2, pch = 20, lwd = 1, lty = 1:2,       legend = c("Clustered", "IID"), title = "Standard-Errors")## Example 2: Interactions## Now we estimate and plot the "yearly" treatment effectsdata(base_did)base_inter = base_did# We interact the variable 'period' with the variable 'treat'est_did = feols(y ~ x1 + i(period, treat, 5) | id + period, base_inter)# In the estimation, the variable treat is interacted#  with each value of period but 5, set as a reference# coefplot will show all the coefficients:coefplot(est_did)# Note that the grouping of the coefficients is due to 'group = "auto"'# If you want to keep only the coefficients# created with i() (ie the interactions), use iplotiplot(est_did)# We can see that the graph is different from before:#  - only interactions are shown,#  - the reference is present,# => this is fully flexibleiplot(est_did, ref.line = FALSE, pt.join = TRUE)## What if the interacted variable is not numeric?# Let's create a "month" variableall_months = c("aug", "sept", "oct", "nov", "dec", "jan",               "feb", "mar", "apr", "may", "jun", "jul")base_inter$period_month = all_months[base_inter$period]# The new estimationest = feols(y ~ x1 + i(period_month, treat, "oct") | id+period, base_inter)# Since 'period_month' of type character, coefplot sorts itiplot(est)# To respect a plotting order, use a factorbase_inter$month_factor = factor(base_inter$period_month, levels = all_months)est = feols(y ~ x1 + i(month_factor, treat, "oct") | id + period, base_inter)iplot(est)## Example 3: Setting defaults## coefplot has many arguments, which makes it highly flexible.# If you don't like the default style of coefplot. No worries,# you can set *your* default by using the function# setFixest_coefplot()dict = c("Petal.Length"="Length (Petal)", "Petal.Width"="Width (Petal)",         "Sepal.Length"="Length (Sepal)", "Sepal.Width"="Width (Sepal)")setFixest_coefplot(ci.col = 2, pt.col = "darkblue", ci.lwd = 3,                   pt.cex = 2, pt.pch = 15, ci.width = 0, dict = dict)est = feols(Petal.Length ~ Petal.Width + Sepal.Length +                Sepal.Width + i(Species), iris)# And that's itcoefplot(est)# You can set separate default values for iplotsetFixest_coefplot("iplot", pt.join = TRUE, pt.join.par = list(lwd = 2, lty = 2))iplot(est)# To reset to the default settings:setFixest_coefplot("all", reset = TRUE)coefplot(est)## Example 4: group + cleaning## You can use the argument group to group variables# You can further use the special character "^^" to clean#  the beginning of the coef. name: particularly useful for factorsest = feols(Petal.Length ~ Petal.Width + Sepal.Length +                Sepal.Width + Species, iris)# No grouping:coefplot(est)# now we group by Sepal and Speciescoefplot(est, group = list(Sepal = "Sepal", Species = "Species"))# now we group + clean the beginning of the names using the special character ^^coefplot(est, group = list(Sepal = "^^Sepal.", Species = "^^Species"))

Extracts the coefficients table from an estimation

Description

Methods to extracts the coefficients table and its sub-components from an estimation.

Usage

coeftable(object, ...)se(object, ...)pvalue(object, ...)tstat(object, ...)

Arguments

Value

Returns a matrix (coeftable) or vectors.

See Also

Please look at thecoeftable.fixest page for more detailed information.

Examples

est = lm(mpg ~ cyl, mtcars)coeftable(est)

Extracts the coefficients table from an estimation

Description

Default method to extracts the coefficients table and its sub-components from an estimation.

Usage

## Default S3 method:coeftable(object, keep, drop, order, ...)## Default S3 method:se(object, keep, drop, order, ...)## Default S3 method:tstat(object, keep, drop, order, ...)## Default S3 method:pvalue(object, keep, drop, order, ...)## S3 method for class 'matrix'se(object, keep, drop, order, ...)

Arguments

Value

Returns a matrix (coeftable) or vectors.

Functions

• se(default): Extracts the standard-errors from an estimation

• tstat(default): Extracts the standard-errors from an estimation

• pvalue(default): Extracts the p-values from an estimation

• se(matrix): Extracts the standard-errors from a VCOV matrix

Examples

# NOTA: This function is really made to handle fixest objects# The default methods works for simple structures, but you'd be# likely better off with broom::tidy for other modelsest = lm(mpg ~ cyl, mtcars)coeftable(est)se(est)

Obtain various statistics from an estimation

Description

Set of functions to directly extract some commonly used statistics, like the p-value orthe table of coefficients, from estimations. This was first implemented forfixest estimations, but has some support for other models.

Usage

## S3 method for class 'fixest'coeftable(  object,  vcov = NULL,  ssc = NULL,  cluster = NULL,  keep = NULL,  drop = NULL,  order = NULL,  list = FALSE,  ...)## S3 method for class 'fixest'se(  object,  vcov = NULL,  ssc = NULL,  cluster = NULL,  keep = NULL,  drop = NULL,  order = NULL,  ...)## S3 method for class 'fixest'tstat(  object,  vcov = NULL,  ssc = NULL,  cluster = NULL,  keep = NULL,  drop = NULL,  order = NULL,  ...)## S3 method for class 'fixest'pvalue(  object,  vcov = NULL,  ssc = NULL,  cluster = NULL,  keep = NULL,  drop = NULL,  order = NULL,  ...)

Arguments

Details

This set of tiny functions is primarily constructed forfixest estimations.

Value

Returns a table of coefficients, with in rows the variables and four columns: the estimate,the standard-error, the t-statistic and the p-value.

Iflist = TRUE then a nested list is returned, the first layer is accessed withthe coefficients names; the second layer with the following values:coef,se,tstat,pvalue. For example, withres = coeftable(est, list = TRUE)you can access the SE of the coefficientx1 withres$x1$se; and itscoefficient withres$x1$coef, etc.

Functions

• se(fixest): Extracts the standard-error of an estimation

• tstat(fixest): Extracts the t-statistics of an estimation

• pvalue(fixest): Extracts the p-value of an estimation

Examples

# Some data and estimationdata(trade)est = fepois(Euros ~ log(dist_km) | Origin^Product + Year, trade)## Coeftable/se/tstat/pvalue#coeftable(est)se(est)tstat(est)pvalue(est)# Now with two-way clustered standard-errors#  and using coeftable()coeftable(est, cluster = ~Origin + Product)se(est, cluster = ~Origin + Product)pvalue(est, cluster = ~Origin + Product)tstat(est, cluster = ~Origin + Product)# Or you can cluster only once using summary:est_sum = summary(est, cluster = ~Origin + Product)coeftable(est_sum)se(est_sum)tstat(est_sum)pvalue(est_sum)# You can use the arguments keep, drop, order# to rearrange the resultsbase = irisnames(base) = c("y", "x1", "x2", "x3", "species")est_iv = feols(y ~ x1 | x2 ~ x3, base)tstat(est_iv, keep = "x1")coeftable(est_iv, keep = "x1|Int")coeftable(est_iv, order = "!Int")## Using lists## Returning the coefficients table as a list can be useful for quick# reference in markdown documents.# Note that the "(Intercept)" is renamed into "constant"res = coeftable(est_iv, list = TRUE)# coefficient of the constant:res$constant$coef# pvalue of x1res$x1$pvalue

Extracts the coefficients tables fromfixest_multi estimations

Description

Series of methods to extract the coefficients table or its sub-components from afixest_multi objects (i.e. the outcome of multiple estimations).

Usage

## S3 method for class 'fixest_multi'coeftable(  object,  vcov = NULL,  keep = NULL,  drop = NULL,  order = NULL,  long = FALSE,  wide = FALSE,  ...)## S3 method for class 'fixest_multi'se(  object,  vcov = NULL,  keep = NULL,  drop = NULL,  order = NULL,  long = FALSE,  ...)## S3 method for class 'fixest_multi'tstat(  object,  vcov = NULL,  keep = NULL,  drop = NULL,  order = NULL,  long = FALSE,  ...)## S3 method for class 'fixest_multi'pvalue(  object,  vcov = NULL,  keep = NULL,  drop = NULL,  order = NULL,  long = FALSE,  ...)

Arguments

Value

It returns adata.frame containing the coefficients tables (or just the se/pvalue/tstat)along with the information on which model was estimated.

Ifwide = TRUE, then a list is returned. The elements of the list arecoef/se/tstat/pvalue. Each element of the list is a wide table with a column per coefficient.

Iflong = TRUE, then all the information is stacked. This removes the 4 columnscontaining the coefficient estimates to the p-values, and replace them with twonew columns:"param" and"value". The columnparam contains thevaluescoef/se/tstat/pvalue, and the columnvalues theassociated numerical information.

Functions

• se(fixest_multi): Extracts the standard-errors fromfixest_multi estimations

• tstat(fixest_multi): Extracts the t-stats fromfixest_multi estimations

• pvalue(fixest_multi): Extracts the p-values fromfixest_multi estimations

Examples

base = setNames(iris, c("y", "x1", "x2", "x3", "species"))est_multi = feols(y ~ csw(x.[,1:3]), base, split = ~species)# we get all the coefficient tables at oncecoeftable(est_multi)# Now just the standard-errorsse(est_multi)# wide = TRUE => leads toa  list of wide tablescoeftable(est_multi, wide = TRUE)# long = TRUE, all the information is stackedcoeftable(est_multi, long = TRUE)

Collinearity diagnostics forfixest objects

Description

In some occasions, the optimization algorithm offemlm may fail to converge, orthe variance-covariance matrix may not be available. The most common reason of whythis happens is collinearity among variables. This function helps to find out whichset of variables is problematic.

Usage

collinearity(x, verbose)

Arguments

Details

This function tests: 1) collinearity with the fixed-effect variables,2) perfect multi-collinearity between the variables, 3) perfect multi-collinearitybetween several variables and the fixed-effects, and 4) identification issueswhen there are non-linear in parameters parts.

Value

It returns a text message with the identified diagnostics.

Author(s)

Laurent Berge

Examples

# Creating an example data base:set.seed(1)fe_1 = sample(3, 100, TRUE)fe_2 = sample(20, 100, TRUE)x = rnorm(100, fe_1)**2y = rnorm(100, fe_2)**2z = rnorm(100, 3)**2dep = rpois(100, x*y*z)base = data.frame(fe_1, fe_2, x, y, z, dep)# creating collinearity problems:base$v1 = base$v2 = base$v3 = base$v4 = 0base$v1[base$fe_1 == 1] = 1base$v2[base$fe_1 == 2] = 1base$v3[base$fe_1 == 3] = 1base$v4[base$fe_2 == 1] = 1# Estimations:# Collinearity with the fixed-effects:res_1 = femlm(dep ~ log(x) + v1 + v2 + v4 | fe_1 + fe_2, base)collinearity(res_1)# => collinearity with the first fixed-effect identified, we drop v1 and v2res_1bis = femlm(dep ~ log(x) + v4 | fe_1 + fe_2, base)collinearity(res_1bis)# Multi-Collinearity:res_2 =  femlm(dep ~ log(x) + v1 + v2 + v3 + v4, base)collinearity(res_2)

Confidence interval for parameters estimated withfixest

Description

This function computes the confidence interval of parameter estimates obtained from amodel estimated withfemlm,feols orfeglm.

Usage

## S3 method for class 'fixest'confint(  object,  parm,  level = 0.95,  vcov,  se,  cluster,  ssc = NULL,  coef.col = FALSE,  ...)

Arguments

Value

Returns a data.frame with two columns giving respectively the lower and upper boundof the confidence interval. There is as many rows as parameters.

Author(s)

Laurent Berge

Examples

# Load trade datadata(trade)# We estimate the effect of distance on trade (with 3 fixed-effects)est_pois = femlm(Euros ~ log(dist_km) + log(Year) | Origin + Destination +                 Product, trade)# confidence interval with "normal" VCOVconfint(est_pois)# confidence interval with "clustered" VCOV (w.r.t. the Origin factor)confint(est_pois, se = "cluster")

Confidence intervals forfixest_multi objects

Description

Computes the confidence intervals of parameter estimates forfixest's multipleestimation objects (akafixest_multi).

Usage

## S3 method for class 'fixest_multi'confint(  object,  parm,  level = 0.95,  vcov = NULL,  se = NULL,  cluster = NULL,  ssc = NULL,  ...)

Arguments

Value

It returns a data frame whose first columns indicate which model has been estimated.The last three columns indicate the coefficient name, and the lower and upperconfidence intervals.

Examples

base = setNames(iris, c("y", "x1", "x2", "x3", "species"))est = feols(y ~ csw(x.[,1:3]) | sw0(species), base, vcov = "iid")confint(est)# focusing only on the coefficient 'x3'confint(est, "x3")# the 'id' provides the index of the estimationest[c(3, 6)]

Gets the degrees of freedom of afixest estimation

Description

Simple utility to extract the degrees of freedom from afixest estimation.

Usage

degrees_freedom(  x,  type,  vars = NULL,  vcov = NULL,  se = NULL,  cluster = NULL,  ssc = NULL,  stage = 2)degrees_freedom_iid(x, type)

Arguments

Functions

• degrees_freedom_iid(): Gets the degrees of freedom of afixest estimation

Examples

# First: an estimationbase = irisnames(base) = c("y", "x1", "x2", "x3", "species")est = feols(y ~ x1 + x2 | species, base)# "Normal" standard-errors (SE)est_standard = summary(est, se = "st")# Clustered SEsest_clustered = summary(est, se = "clu")# The different degrees of freedom# => different type 1 DoF (because of the clustering)degrees_freedom(est_standard, type = "k")degrees_freedom(est_clustered, type = "k") # fixed-effects are excluded# => different type 2 DoF (because of the clustering)degrees_freedom(est_standard, type = "resid") # => equivalent to the df.residual from lmdegrees_freedom(est_clustered, type = "resid")

Centers a set of variables around a set of factors

Description

User-level access to internal demeaning algorithm offixest.

Usage

demean(  X,  f,  slope.vars,  slope.flag,  data,  weights,  sample = "estimation",  nthreads = getFixest_nthreads(),  notes = getFixest_notes(),  iter = 2000,  tol = 1e-06,  fixef.reorder = TRUE,  fixef.algo = NULL,  na.rm = TRUE,  as.matrix = is.atomic(X),  im_confident = FALSE,  ...)

Arguments

Value

It returns a data.frame of the same number of columns as the number of variables to be centered.

Ifna.rm = TRUE, then the number of rows is equal to the number of rows in input minus thenumber of NA values (contained inX,f,slope.vars orweights). The default is to havean output of the same number of observations as the input (filled with NAs where appropriate).

A matrix can be returned ifas.matrix = TRUE.

Varying slopes

You can add variables with varying slopes in the fixed-effect part of the formula.The syntax is as follows:fixef_var[var1, var2]. Here the variables var1 and var2 willbe with varying slopes (one slope per value in fixef_var) and the fixed-effectfixef_var will also be added.

To add only the variables with varying slopes and not the fixed-effect,use double square brackets:fixef_var[[var1, var2]].

In other words:

• fixef_var[var1, var2] is equivalent tofixef_var + fixef_var[[var1]] + fixef_var[[var2]]

• fixef_var[[var1, var2]] is equivalent tofixef_var[[var1]] + fixef_var[[var2]]

In general, for convergence reasons, it is recommended to always add the fixed-effect andavoid using only the variable with varying slope (i.e. use single square brackets).

Examples

# Illustration of the FWL theoremdata(trade)base = tradebase$ln_dist = log(base$dist_km)base$ln_euros = log(base$Euros)# We center the two variables ln_dist and ln_euros#  on the factors Origin and DestinationX_demean = demean(X = base[, c("ln_dist", "ln_euros")],                  f = base[, c("Origin", "Destination")])base[, c("ln_dist_dm", "ln_euros_dm")] = X_demeanest = feols(ln_euros_dm ~ ln_dist_dm, base)est_fe = feols(ln_euros ~ ln_dist | Origin + Destination, base)# The results are the same as if we used the two factors# as fixed-effectsetable(est, est_fe, se = "st")## Variables with varying slopes## You can center on factors but also on variables with varying slopes# Let's have an illustrationbase = irisnames(base) = c("y", "x1", "x2", "x3", "species")## We center y and x1 on species and x2 * species# using a formulabase_dm = demean(y + x1 ~ species[x2], data = base)# using vectorsbase_dm_bis = demean(X = base[, c("y", "x1")], f = base$species,                     slope.vars = base$x2, slope.flag = 1)# Let's look at the equivalencesres_vs_1 = feols(y ~ x1 + species + x2:species, base)res_vs_2 = feols(y ~ x1, base_dm)res_vs_3 = feols(y ~ x1, base_dm_bis)# only the small sample adj. differ in the SEsetable(res_vs_1, res_vs_2, res_vs_3, keep = "x1")## center on x2 * species and on another FEbase$fe = rep(1:5, 10)# using a formula => double square brackets!base_dm = demean(y + x1 ~ fe + species[[x2]], data = base)# using vectors => note slope.flag!base_dm_bis = demean(X = base[, c("y", "x1")], f = base[, c("fe", "species")],                     slope.vars = base$x2, slope.flag = c(0, -1))# Explanations slope.flag = c(0, -1):# - the first 0: the first factor (fe) is associated to no variable# - the "-1":#    * |-1| = 1: the second factor (species) is associated to ONE variable#    *   -1 < 0: the second factor should not be included as such# Let's look at the equivalencesres_vs_1 = feols(y ~ x1 + i(fe) + x2:species, base)res_vs_2 = feols(y ~ x1, base_dm)res_vs_3 = feols(y ~ x1, base_dm_bis)# only the small sample adj. differ in the SEsetable(res_vs_1, res_vs_2, res_vs_3, keep = "x1")

Controls the parameters of the demeaning procedure

Description

Fine control of the demeaning procedure. Since the defaults are sensible,only use this function in case of difficult convergence (e.g. infeols ordemean).That is, look at the slot⁠$iterations⁠ of the returned object, if it's high (over 50),then it might be worth playing around with these settings.

Usage

demeaning_algo(  extraProj = 0,  iter_warmup = 15,  iter_projAfterAcc = 40,  iter_grandAcc = 4,  internal = FALSE)

Arguments

Details

The demeaning algorithm is a fixed-point algorithm. Basically a functionf is applieduntil⁠|f(X) - X| = 0⁠, i.e. there is no difference betweenX and its image.For terminology, let's call the application off a "projection".

For well behaved problems, the algorithm in its simplest form, i.e. just applyingf untilconvergence, works fine and you only need a few iterations to reach convergence.

The problems arise for non well behaved problems. In these cases, simply applying thefunctionf can lead to extremely slow convergence. To handle these cases, this algorithmapplies a fixed-point acceleration algorithm, namely the "Irons and Tuck" acceleration.

The main algorithm combines regular projections with accelerations. Unfortunatelysometimes this is not enough, so we also resort on internal cuisine, detailed below.

Sometimes the acceleration in its simplest form does not work well, and garbles theconvergence properties. In those cases:

• the argumentextraProj adds several standard projections in between two accelerations,which can improve the performance of the algorithm. By default there are no extraprojections. Note that while it can reduce the total number of iterations until convergence,each iterations is almost twice expensive in terms of computing time.

• the argumentiter_projAfterAcc controls whether, and when, to apply a simple projectionright after the acceleration step. This projection adds roughly a 33% increase incomputing time per iteration but can improve the convergence properties and speed. By defaultthis step starts at iteration 40 (when the convergence rate is already not great).

On top of this, in case of very difficult convergence, a "grand" acceleration is added tothe algorithm. The regular acceleration is overf. Sayg is the function equivalent tothe application of one regular iteration (which is a combination of one acceleration withseveral projections).By default the grand acceleration is over⁠h = g o g o g o g⁠, otherwiseg applied four times.The grand acceleration is controled with the argumentiter_grandAcc which correspondsto the number of iterations of the regular algorithm definingh.

Finally in case of 3+ fixed-effects (FE), the convergence in general takes more iterations.In cases of the absence of quick convergence, applying a first demeaning over the firsttwo largest FEs before applying the demeaning over all FEs can improve convergence speed.This is controlled with the argumentiter_warmup which gives the number of iterationsover all the FEs to run before going to the 2 FEs demeaning. By default, the deameaningover all FEs is run for 15 iterations before switching to the 2 FEs case.

The above defaults are the outcome of extended empirical applications, and try to strike abalance across a majority of cases. Of course you can always get better results by tailoringthe settings to your problem at hand.

Value

This function returns a list of 4 integers, equal to the arguments passed by the user.That list is of classdemeaning_algo.

References

B. M. Irons, R. Tuck, "A version of the Aitken accelerator for computer iteration",International journal of numerical methods in engineering 1 (1969) 670 275–277.

Extracts the deviance of a fixest estimation

Description

Returns the deviance from afixest estimation.

Usage

## S3 method for class 'fixest'deviance(object, ...)

Arguments

Value

Returns a numeric scalar equal to the deviance.

See Also

feols,fepois,feglm,fenegbin,feNmlm.

Examples

est = feols(Petal.Length ~ Petal.Width, iris)deviance(est)est_pois = fepois(Petal.Length ~ Petal.Width, iris)deviance(est_pois)

Residual degrees-of-freedom forfixest objects

Description

Returns the residual degrees of freedom for a fittedfixest object

Usage

## S3 method for class 'fixest'df.residual(object, ...)

Arguments

Value

It returns an integer scalar giving the residuals degrees of freedom of the estimation.

See Also

The functiondegrees_freedom infixest.

Examples

est = feols(mpg ~ hp, mtcars)df.residual(est)

Treated and control sample descriptives

Description

This function shows the means and standard-deviations of several variables conditional on whether they are from the treated or the control group. The groups can further be split according to a pre/post variable. Results can be seamlessly be exported to Latex.

Usage

did_means(  fml,  base,  treat_var,  post_var,  tex = FALSE,  treat_dict,  dict = getFixest_dict(),  file,  replace = FALSE,  title,  label,  raw = FALSE,  indiv,  treat_first,  prepostnames = c("Before", "After"),  diff.inv = FALSE)

Arguments

Details

By default, when the user tries to apply this function to nun-numeric variables, an error is raised. The exception is when the all variables are selected with the dot (like in. ~ treat. In this case, non-numeric variables are automatically omitted (with a message).

NAs are removed automatically: if the data contains NAs an information message will be prompted. First all observations containing NAs relating to the treatment or post variables are removed. Then if there are still NAs for the variables, they are excluded separately for each variable, and a new message detailing the NA breakup is prompted.

Value

It returns a data.frame or a Latex table with the conditional means and statistical differences between the groups.

Examples

# Playing around with the DiD datadata(base_did)# means of treat/controldid_means(y+x1+period~treat, base_did)# same but inverting the differencedid_means(y+x1+period~treat, base_did, diff.inv = TRUE)# now treat/control, before/afterdid_means(y+x1+period~treat|post, base_did)# same but with a new line giving the number of unique "indiv" for each casedid_means(y+x1+period~treat|post, base_did, indiv = "id")# same but with the treat case "0" coming firstdid_means(y+x1+period~treat|post, base_did, indiv = ~id, treat_first = 0)# Selecting all the variables with "."did_means(.~treat|post, base_did, indiv = "id")

Simple and powerful string manipulation with the dot square bracket operator

Description

Compactly performs many low level string operations. Advanced support for pluralization.

Usage

dsb(  ...,  frame = parent.frame(),  sep = "",  vectorize = FALSE,  nest = TRUE,  collapse = NULL)

Arguments

Value

It returns a character vector whose length depends on the elements and operations in".[]".

Examples

## BASIC USAGE #####x = c("Romeo", "Juliet")# .[x] inserts xdsb("Hello .[x]!")# elements in ... are collapsed with "" (default)dsb("Hello .[x[1]], ",    "how is .[x[2]] doing?")# Splitting a comma separated string# The mechanism is explained laterdsb("/J. Mills, David, Agnes, Dr Strong")# Nota: this is equivalent to (explained later)dsb("', *'S !J. Mills, David, Agnes, Dr Strong")## Applying low level operations to strings## Two main syntax:# A) expression evaluation# .[operation ? x]#             | |#             |  \-> the expression to be evaluated#              \-> ? means that the expression will be evaluated# B) verbatim# .[operation ! x]#             | |#             |  \-> the expression taken as verbatim (here ' x')#              \-> ! means that the expression is taken as verbatim# operation: usually 'arg'op with op an operation code.# Example: splittingx = "hello dear"dsb(".[' 's ? x]")# x is split by ' 'dsb(".[' 's !hello dear]")# 'hello dear' is split by ' '# had we used ?, there would have been an error# By default, the string is nested in .[], so in that case no need to use .[]:dsb("' 's ? x")dsb("' 's !hello dear")# There are 35 string operators# Operators usually have a default value# Operations can be chained by separating them with a comma# Example: default of 's' is ' ' + chaining with collapsedsb("s, ' my 'c!hello dear")## Nesting## .[operations ! s1.[expr]s2]#              |    |#              |     \-> expr will be evaluated then added to the string#               \-> nesting requires verbatim evaluation: '!'dsb("The variables are: .[C!x.[1:4]].")# This one is a bit ugly but it shows triple nestingdsb("The variables are: .[w, C!.[2* ! x.[1:4]].[S, 4** ! , _sq]].")## Splitting## s: split with fixed pattern, default is ' 'dsb("s !a b c")dsb("' b 's !a b c")# S: split with regex pattern, default is ', *'dsb("S !a, b, c")dsb("'[[:punct:] ]'S !a! b; c")## Collapsing## c and C do the same, their default is different# syntax: 's1||s2' with# - s1 the string used for collapsing# - s2 (optional) the string used for the last collapse# c: default is ' 'dsb("c?1:3")# C: default is ', || and 'dsb("C?1:3")dsb("', || or 'c?1:4")## Extraction## x: extracts the first pattern# X: extracts all patterns# syntax: 'pattern'x# Default is '[[:alnum:]]+'x = "This years is... 2020"dsb("x ? x")dsb("X ? x")dsb("'\\d+'x ? x")## STRING FORMATTING ####### u, U: uppercase first/all letters# first letterdsb("u!julia mills")# title case: split -> upper first letter -> collapsedsb("s, u, c!julia mills")# upper all lettersdsb("U!julia mills")## L: lowercasedsb("L!JULIA MILLS")## q, Q: single or double quotedsb("S, q, C!Julia, David, Wilkins")dsb("S, Q, C!Julia, David, Wilkins")## f, F: formats the string to fit the same lengthscore = c(-10, 2050)nm = c("Wilkins", "David")dsb("Monopoly scores:\n.['\n'c ! - .[f ? nm]: .[F ? score] US$]")# OK that example may have been a bit too complex,# let's make it simple:dsb("Scores: .[f ? score]")dsb("Names: .[F ? nm]")## w, W: reformat the white spaces# w: suppresses trimming white spaces + normalizes successive white spaces# W: same but also includes punctuationdsb("w ! The   white  spaces are now clean.  ")dsb("W ! I, really -- truly; love punctuation!!!")## %: applies sprintf formattingdsb("pi = .['.2f'% ? pi]")## a: appends text on each item# syntax: 's1|s2'a, adds s1 at the beginning and s2 at the end of the string# It accepts the special values :1:, :i:, :I:, :a:, :A:# These values create enumerations (only one such value is accepted)# appending square bracketsdsb("'[|]'a, ' + 'c!x.[1:4]")# Enumerationsacad = dsb("/you like admin, you enjoy working on weekends, you really love emails")dsb("Main reasons to pursue an academic career:\n .[':i:) 'a, C ? acad].")## A: same as 'a' but adds at the begging/end of the full string (not on the elements)# special values: :n:, :N:, give the number of elementscharacters = dsb("/David, Wilkins, Dora, Agnes")dsb("There are .[':N: characters: 'A, C ? characters].")## stop: removes basic English stopwords# the list is from the Snowball project: http://snowball.tartarus.org/algorithms/english/stop.txtdsb("stop, w!It is a tale told by an idiot, full of sound and fury, signifying nothing.")## k: keeps the first n characters# syntax: nk: keeps the first n characters#         'n|s'k: same + adds 's' at the end of shortened strings#         'n||s'k: same but 's' counts in the n characters keptwords = dsb("/short, constitutional")dsb("5k ? words")dsb("'5|..'k ? words")dsb("'5||..'k ? words")## K: keeps the first n elements# syntax: nK: keeps the first n elements#         'n|s'K: same + adds the element 's' at the end#         'n||s'K: same but 's' counts in the n elements kept## Special values :rest: and :REST:, give the number of items droppedbx = dsb("/Pessac Leognan, Saint Emilion, Marguaux, Saint Julien, Pauillac")dsb("Bordeaux wines I like: .[3K, ', 'C ? bx].")dsb("Bordeaux wines I like: .['3|etc..'K, ', 'C ? bx].")dsb("Bordeaux wines I like: .['3||etc..'K, ', 'C ? bx].")dsb("Bordeaux wines I like: .['3|and at least :REST: others'K, ', 'C ? bx].")## Ko, KO: special operator which keeps the first n elements and adds "others"# syntax: nKo# KO gives the rest in lettersdsb("Bordeaux wines I like: .[4KO, C ? bx].")## r, R: string replacement# syntax: 's'R: deletes the content in 's' (replaces with the empty string)#         's1 => s2'R replaces s1 into s2# r: fixed / R: perl = TRUEdsb("'e'r !The letter e is deleted")# adding a perl look-behinddsb("'(?<! )e'R !The letter e is deleted")dsb("'e => a'r !The letter e becomes a")dsb("'([[:alpha:]]{3})[[:alpha:]]+ => \\1.'R !Trimming the words")## *, *c, **, **c: replication, replication + collapse# syntax: n* or n*c# ** is the same as * but uses "each" in the replicationdsb("N.[10*c!o]!")dsb("3*c ? 1:3")dsb("3**c ? 1:3")## d: replaces the items by the empty string# -> useful in conditionsdsb("d!I am going to be annihilated")## ELEMENT MANIPULATION ####### D: deletes all elements# -> useful in conditionsx = dsb("/I'll, be, deleted")dsb("D ? x")## i, I: inserts an item# syntax: 's1|s2'i: inserts s1 first and s2 last# I: is the same as i but is 'invisibly' includedcharacters = dsb("/David, Wilkins, Dora, Agnes, Trotwood")dsb("'Heep|Spenlow'i, C ? characters")dsb("'Heep|Spenlow'I, C ? characters")## PLURALIZATION ###### There is support for pluralization## *s, *s_: adds 's' or 's ' depending on the number of elementsnb = 1:5dsb("Number.[*s, D ? nb]: .[C ? nb]")dsb("Number.[*s, D ? 2 ]: .[C ? 2 ]")# ordsb("Number.[*s, ': 'A, C ? nb]")## v, V: adds a verb at the beginning/end of the string# syntax: 'verb'v# Unpopular opinion?brand = c("Apple", "Samsung")dsb(".[V, C ? brand] overrated.")dsb(".[V, C ? brand[1]] overrated.")win = dsb("/Peggoty, Agnes, Emily")dsb("The winner.[*s_, v, C ? win].")dsb("The winner.[*s_, v, C ? win[1]].")# Other verbsdsb(".[' have'V, C ? win] won a prize.")dsb(".[' have'V, C ? win[1]] won a prize.")dsb(".[' was'V, C ? win] unable to come.")dsb(".[' was'V, C ? win[1]] unable to come.")## *A: appends text depending on the length of the vector# syntax: 's1|s2 / s3|s4'#         if length == 1: applies 's1|s2'A#         if length >  1: applies 's3|s4'Awin = dsb("/Barkis, Micawber, Murdstone")dsb("The winner.[' is /s are '*A, C ? win].")dsb("The winner.[' is /s are '*A, C ? win[1]].")## CONDITIONS ###### Conditions can be applied with 'if' statements.",# The syntax is 'type comp value'if(true : false), with# - type: either 'len', 'char', 'fixed' or 'regex'#   + len: number of elements in the vector#   + char: number of characters#   + fixed: fixed pattern#   + regex: regular expression pattern# - comp: a comparator:#   + valid for len/char: >, <, >=, <=, !=, ==#   + valid for fixed/regex: !=, ==# - value: a value for which the comparison is applied.# - true: operations to be applied if true (can be void)# - false: operations to be applied if false (can be void)dsb("'char <= 2'if('(|)'a : '[|]'a), ' + 'c ? c(1, 12, 123)")sentence = "This is a sentence with some longish words."dsb("s, 'char<=4'if(D), c ? sentence")dsb("s, 'fixed == e'if(:D), c ! Only words with an e are selected.")## ARGUMENTS FROM THE FRAME ###### Arguments can be evaluated from the calling frame.# Simply use backticks instead of quotes.dollar = 6reason = "glory"dsb("Why do you develop packages? For .[`dollar`*c!$]?",    "For money? No... for .[U,''s, c?reason]!", sep = "\n")

Support for emmeans package

Description

Ifemmeans is installed, its functionality is supported forfixestorfixest_multi objects. Its reference grid is based on the main partof the model, and does not include fixed effects or instrumental variables.Note that any desired arguments tovcov() may be passed as optionalarguments inemmeans::emmeans() oremmeans::ref_grid().

Note

When fixed effects are present, estimated marginal means (EMMs) are estimatedcorrectly, provided equal weighting is used. However, the SEs of these EMMswill be incorrect - often dramatically - because the estimated variance ofthe intercept is not available. However,contrasts among EMMs can beestimated and tested with no issues, because these do not involve theintercept.

Author(s)

Russell V. Lenth

Examples

if(requireNamespace("emmeans") && requireNamespace("AER")) {    data(Fatalities, package = "AER")    Fatalities$frate = with(Fatalities, fatal/pop * 10000)    fat.mod = feols(frate ~ breath * jail * beertax | state + year, data = Fatalities)    emm = emmeans::emmeans(fat.mod, ~ breath*jail, cluster = ~ state + year)    emm   ### SEs and CIs are incorrect    emmeans::contrast(emm, "consec", by = "breath")   ### results are reliable}

Estimates afixest estimation from afixest environment

Description

This is a function advanced users which allows to estimate anyfixest estimation from afixest environment obtained withonly.env = TRUE in afixest estimation.

Usage

est_env(env, y, X, weights, endo, inst)

Arguments

Details

This function has been created for advanced users, mostly to avoid overheadswhen making simulations withfixest.

How can it help you make simulations? First make a core estimation withonly.env = TRUE,and usually withonly.coef = TRUE (to avoid having extra things that take time to compute).Then loop while modifying the appropriate things directly in the environment. Beware thatif you make a mistake here (typically giving stuff of the wrong length),then you can make the R session crash because there is no more error-handling!Finally estimate withest_env(env = core_env) and store the results.

Instead ofest_env, you could use directlyfixest estimations too, likefeols,since they accept theenv argument. The functionest_env is only here to add abit of generality to avoid the trouble to the user to write conditions(look at the source, it's just a one liner).

Objects of main interest in the environment are:

lhs

The left hand side, or dependent variable.

linear.mat

The matrix of the right-hand-side, or explanatory variables.

iv_lhs

The matrix of the endogenous variables in IV regressions.

iv.mat

The matrix of the instruments in IV regressions.

weights.value

The vector of weights.

I strongly discourage changing the dimension of any of these elements, or else crash can occur.However, you can change their values at will (given the dimension stay the same).The only exception is the weights, which tolerates changing its dimension: it canbe identical to the scalar1 (meaning no weights), or to something of the length thenumber of observations.

I also discourage changing anything in the fixed-effects (even their value)since this will almost surely lead to a crash.

Note that this function is mostly useful when the overheads/estimation ratio is high.This means that OLS will benefit the most from this function. For GLM/Max.Lik. estimations,the ratio is small since the overheads is only a tiny portion of the total estimation time.Hence this function will be less useful for these models.

Value

It returns the results of afixest estimation: the one that was summoned whenobtaining the environment.

Author(s)

Laurent Berge

Examples

# Let's make a short simulation# Inspired from Grant McDermott bboot function# See https://twitter.com/grant_mcdermott/status/1487528757418102787# Simple function that computes a Bayesian bootstrapbboot = function(x, n_sim = 100){  # We bootstrap on the weights  # Works with fixed-effects/IVs  #  and with any fixest function that accepts weights  core_env = update(x, only.coef = TRUE, only.env = TRUE)  n_obs = x$nobs  res_all = vector("list", n_sim)  for(i in 1:n_sim){    ## begin: NOT RUN    ## We could directly assign in the environment:    # assign("weights.value", rexp(n_obs, rate = 1), core_env)    # res_all[[i]] = est_env(env = core_env)    ##   end: NOT RUN    ## Instead we can use the argument weights, which does the same    res_all[[i]] = est_env(env = core_env, weights = rexp(n_obs, rate = 1))  }  do.call(rbind, res_all)}est = feols(mpg ~ wt + hp, mtcars)boot_res = bboot(est)coef = colMeans(boot_res)std_err = apply(boot_res, 2, sd)# Comparing the results with the main estimationcoeftable(est)cbind(coef, std_err)

Extracts the scores from a fixest estimation

Description

Extracts the scores from a fixest estimation.

Usage

## S3 method for class 'fixest'estfun(x, ...)

Arguments

Value

Returns a matrix of the same number of rows as the number of observations used forthe estimation, and the same number of columns as there were variables.

Examples

data(iris)est = feols(Petal.Length ~ Petal.Width + Sepal.Width, iris)head(estfun(est))

Estimations table (export the results of multiples estimations to a DF or to Latex)

Description

Aggregates the results of multiple estimations and displays them in the form of either a Latextable or adata.frame. Note that you will need thebooktabs package for the Latex table torender properly. SeesetFixest_etable to set the default values, andstyle.tex to customize Latex output.

Usage

esttable(  ...,  vcov = NULL,  stage = 2,  agg = NULL,  se = NULL,  ssc = NULL,  cluster = NULL,  .vcov_args = NULL,  digits = 4,  digits.stats = 5,  fitstat = NULL,  coefstat = "se",  ci = 0.95,  se.row = NULL,  se.below = NULL,  keep = NULL,  drop = NULL,  order = NULL,  dict = TRUE,  file = NULL,  replace = TRUE,  create_dirs = FALSE,  convergence = NULL,  signif.code = NULL,  headers = list("auto"),  fixef_sizes = FALSE,  fixef_sizes.simplify = TRUE,  keepFactors = TRUE,  family = NULL,  powerBelow = -5,  interaction.combine = NULL,  interaction.order = NULL,  i.equal = NULL,  depvar = TRUE,  style.df = NULL,  group = NULL,  extralines = NULL,  fixef.group = NULL,  drop.section = NULL,  poly_dict = c("", " square", " cube"),  postprocess.df = NULL,  fit_format = "__var__",  coef.just = NULL,  highlight = NULL,  coef.style = NULL,  export = NULL,  page.width = "fit",  div.class = "etable")esttex(  ...,  vcov = NULL,  stage = 2,  agg = NULL,  se = NULL,  ssc = NULL,  cluster = NULL,  .vcov_args = NULL,  digits = 4,  digits.stats = 5,  fitstat = NULL,  caption = NULL,  coefstat = "se",  ci = 0.95,  se.row = NULL,  se.below = NULL,  keep = NULL,  drop = NULL,  order = NULL,  dict = TRUE,  file = NULL,  replace = TRUE,  create_dirs = FALSE,  convergence = NULL,  signif.code = NULL,  label = NULL,  float = NULL,  headers = list("auto"),  fixef_sizes = FALSE,  fixef_sizes.simplify = TRUE,  keepFactors = TRUE,  family = NULL,  powerBelow = -5,  interaction.combine = NULL,  interaction.order = NULL,  i.equal = NULL,  depvar = TRUE,  style.tex = NULL,  notes = NULL,  group = NULL,  extralines = NULL,  fixef.group = NULL,  placement = "htbp",  drop.section = NULL,  poly_dict = c("", " square", " cube"),  postprocess.tex = NULL,  tpt = FALSE,  arraystretch = NULL,  adjustbox = NULL,  fontsize = NULL,  fit_format = "__var__",  tabular = "normal",  highlight = NULL,  coef.style = NULL,  meta = NULL,  meta.time = NULL,  meta.author = NULL,  meta.sys = NULL,  meta.call = NULL,  meta.comment = NULL,  view = FALSE,  export = NULL,  markdown = NULL,  page.width = "fit",  div.class = "etable")etable(  ...,  vcov = NULL,  stage = 2,  agg = NULL,  se = NULL,  ssc = NULL,  cluster = NULL,  .vcov_args = NULL,  digits = 4,  digits.stats = 5,  tex,  fitstat = NULL,  caption = NULL,  coefstat = "se",  ci = 0.95,  se.row = NULL,  se.below = NULL,  keep = NULL,  drop = NULL,  order = NULL,  dict = TRUE,  file = NULL,  replace = TRUE,  create_dirs = FALSE,  convergence = NULL,  signif.code = NULL,  label = NULL,  float = NULL,  headers = list("auto"),  fixef_sizes = FALSE,  fixef_sizes.simplify = TRUE,  keepFactors = TRUE,  family = NULL,  powerBelow = -5,  interaction.combine = NULL,  interaction.order = NULL,  i.equal = NULL,  depvar = TRUE,  style.tex = NULL,  style.df = NULL,  notes = NULL,  group = NULL,  extralines = NULL,  fixef.group = NULL,  placement = "htbp",  drop.section = NULL,  poly_dict = c("", " square", " cube"),  postprocess.tex = NULL,  postprocess.df = NULL,  tpt = FALSE,  arraystretch = NULL,  adjustbox = NULL,  fontsize = NULL,  fit_format = "__var__",  coef.just = NULL,  tabular = "normal",  highlight = NULL,  coef.style = NULL,  meta = NULL,  meta.time = NULL,  meta.author = NULL,  meta.sys = NULL,  meta.call = NULL,  meta.comment = NULL,  view = FALSE,  export = NULL,  markdown = NULL,  page.width = "fit",  div.class = "etable")setFixest_etable(  digits = 4,  digits.stats = 5,  fitstat,  coefstat = c("se", "tstat", "confint", "pvalue"),  ci = 0.95,  se.below = TRUE,  keep,  drop,  order,  dict,  float,  signif.code = NULL,  fixef_sizes = FALSE,  fixef_sizes.simplify = TRUE,  family,  powerBelow = -5,  interaction.order = NULL,  depvar,  style.tex = NULL,  style.df = NULL,  notes = NULL,  group = NULL,  extralines = NULL,  fixef.group = NULL,  placement = "htbp",  drop.section = NULL,  view = FALSE,  markdown = NULL,  view.cache = TRUE,  page.width = "fit",  div.class = "etable",  postprocess.tex = NULL,  postprocess.df = NULL,  fit_format = "__var__",  meta.time = NULL,  meta.author = NULL,  meta.sys = NULL,  meta.call = NULL,  meta.comment = NULL,  reset = FALSE,  save = FALSE)getFixest_etable()## S3 method for class 'etable_tex'print(x, ...)## S3 method for class 'etable_df'print(x, ...)log_etable(type = "pdflatex")

Arguments

Details

The functionesttex is equivalent to the functionetable with argumenttex = TRUE.

The functionesttable is equivalent to the functionetable with argumenttex = FALSE.

To display the table, you will need the Latex packagebooktabs which containsthe⁠\\toprule⁠,⁠\\midrule⁠ and⁠\\bottomrule⁠ commands.

You can permanently change the way your table looks in Latex by usingsetFixest_etable.The following vignette gives an example as well as illustrates how to use thestyle andpostprocessing functions:Exporting estimation tables.

When the argumentpostprocess.tex is not missing, two additional tags willbe included in the character vector returned byetable:"%start:tab\\n" and"%end:tab\\n". These can be usedto identify the start and end of the tabular and are useful to insert codewithin thetable environment.

Value

Iftex = TRUE, the lines composing the Latex table are returned invisibly whilethe table is directly prompted on the console.

Iftex = FALSE, the data.frame is directly returned. If the argumentfile isnot missing, thedata.frame is printed and returned invisibly.

Functions

• esttable(): Exports the results of multiplefixest estimations in a Latex table.

• esttex(): Exports the results of multiplefixest estimations in a Latex table.

Latex dependencies

Some features require specific Latex dependencies, these are:

• always needed:⁠\\usepackage{booktabs}⁠,⁠\\usepackage{array}⁠,⁠\\usepackage{multirow}⁠,⁠\\usepackage{amsmath}⁠,⁠\\usepackage{amssymb}⁠

• if there are line break within cells:⁠\\usepackage{makecell}⁠

• if the tabularx environment is used:⁠\\usepackage{tabularx}⁠

• if threeparttable notes are used:⁠\\usepackage[flushleft]{threeparttable}⁠

• if you use adjustbox:⁠\\usepackage{adjustbox}⁠

• if you use any kind of colors in the table:⁠\\usepackage[dvipsnames,table]{xcolor}⁠

• if you highlight cells with a box:⁠\\usepackage{tikz}⁠ and⁠\\usetikzlibrary{matrix, shapes, arrows, fit, tikzmark}⁠

• if you highlight rows using the background color:⁠\\usepackage{colortbl}⁠

Here is a summary:

% required\usepackage{booktabs}\usepackage{array}\usepackage{multirow}\usepackage{amsmath}\usepackage{amssymb}% optionnal, dependent on context\usepackage{makecell}\usepackage{tabularx}\usepackage[flushleft]{threeparttable}\usepackage{adjustbox}\usepackage[dvipsnames,table]{xcolor}\usepackage{tikz}\usetikzlibrary{matrix, shapes, arrows, fit, tikzmark}\usepackage{colortbl}

How doesdigits handle the number of decimals displayed?

The default display of decimals is the outcome of an algorithm. Let's take the exampleofdigits = 3 which "kind of" requires 3 significant digits to be displayed.

For numbers greater than 1 (in absolute terms), their integral part isalways displayed and the number of decimals shown is equal todigitsminus the number of digits in the integral part.This means that12.345 will be displayed as12.3.If the number of decimals should be 0, then a single decimal is displayedto suggest that the number is not whole. This means that1234.56 willbe displayed as1234.5. Note that if the number is whole, no decimals are shown.

For numbers lower than 1 (in absolute terms), the number of decimals displayed is equaltodigits except if there are only 0s in which case the first significantdigit is shown.This means that0.01234 will be displayed as0.012 (first rule),and that 0.000123 will be displayed as0.0001 (second rule).

Arguments keep, drop and order

The argumentskeep,drop andorder use regular expressions. If you are not awareof regular expressions, I urge you to learn it, since it is an extremely powerful wayto manipulate character strings (and it exists across most programming languages).

For example drop = "Wind" would drop any variable whose name contains "Wind". Note thatvariables such as "Temp:Wind" or "StrongWind" do contain "Wind", so would be dropped.To drop only the variable named "Wind", you need to usedrop = "^Wind$" (with "^" meaning beginning, resp. "$" meaning end,of the string => this is the language of regular expressions).

Although you can combine several regular expressions in a single characterstring using pipes,drop also accepts a vector of regular expressions.

You can use the special character "!" (exclamation mark) to reverse the effectof the regular expression (this feature is specific to this function).For exampledrop = "!Wind" would drop any variable that does not contain "Wind".

You can use the special character "%" (percentage) to make reference to theoriginal variable name instead of the aliased name. For example, you have avariable named"Month6", and use a dictionarydict = c(Month6="June").Thus the variable will be displayed as"June".If you want to delete that variable, you can use eitherdrop="June",ordrop="%Month6" (which makes reference to its original name).

The argumentorder takes in a vector of regular expressions, the order will follow theelements of this vector. The vector gives a list of priorities,on the left the elements with highest priority.For example, order = c("Wind", "!Inter", "!Temp") would give highest priorities tothe variables containing "Wind" (which would then appear first),second highest priority is the variables not containing "Inter", last,with lowest priority, the variables not containing "Temp".If you had the following variables: (Intercept), Temp:Wind, Wind, Temp youwould end up with the following order: Wind, Temp:Wind, Temp, (Intercept).

The argumentextralines

The argumentextralines adds well... extra lines to the table.It accepts either a list, or a one-sided formula.

For each line, you can define the values taken by each cell using 4 different ways:a) a vector, b) a list, c) a function, and d) a formula.

If a vector, it should represent the values taken by each cell. Note that if thelength of the vector is smaller than the number of models, its values arerecycled across models, but the length of the vector is required to be adivisor of the number of models.

If a list, it should be of the form⁠list("item1" = #item1, "item2" = #item2, etc)⁠.For examplelist("A"=2, "B"=3) leads toc("A", "A", "B", "B", "B").Note that if the number of items is 1, you don't need to add⁠= 1⁠.For examplelist("A"=2, "B") is valid and leads to⁠c("A", "A", "B"⁠. As for the vector the values are recycled if necessary.

If a function, it will be applied to each model and should return a scalar (NA valuesreturned are accepted).

If a formula, it must be one-sided and the elements in the formula must represent eitherextralines macros, either fit statistics (i.e. valid types ofthe functionfitstat).One new line will be added for each element of the formula.To registerextralines macros, you must first register them inextralines_register.

Finally, you can combine as many lines as wished by nesting them in a list.The names of the nesting list are the row titles (values in the leftmost cell).For exampleextralines = list(~r2, Controls = TRUE, Group = list("A"=2, "B")) willadd three lines, the titles of which are "R2", "Controls" and "Group".

Controlling the placement of extra lines

The argumentsgroup,extralines andfixef.group allow to add customized lines in thetable. They can be defined via a list where the list name will be the row name.By default, the placement of the extra line is right after the coefficients(except forfixef.group, covered in the last paragraph).For instance,group = list("Controls" = "x[[:digit:]]") will create aline right after the coefficients telling which models contain the control variables.

But the placement can be customized. The previous example (of the controls) willbe used for illustration (the mechanism forextralines andfixef.group is identical).

The row names accept 2 special characters at the very start.The first character tells in which section the line should appear:it can be equal to"^","-", or"_", meaning respectivelythe coefficients, the fixed-effects and the statistics section(which typically appear at the top, mid and bottom of the table).The second one governs the placement of the new line withinthe section: it can be equal to"^", meaning first line, or"_", meaning last line.

Let's have some examples. Using the previous example, writing"_^Controls"would place the new line at the top of the statistics section.Writing"-_Controls" places it as the last row ofthe fixed-effects section;"^^Controls" at the top row ofthe coefficients section; etc...

The second character is optional, the default placement being in the bottom.This means that"_Controls" would place it at the bottom of the statistics section.

The placement infixef.group is defined similarly, only the defaultplacement is different.Its default placement is at the top of the fixed-effects section.

Escaping special Latex characters

By default on all instances (with the notable exception of the elements ofstyle.tex)special Latex characters are escaped. This means thatcaption="Exports in million $." will be exported as"Exports in million \\$.": the dollar sign will be escaped.This is true for the following characters: &,$, %, _, ^ and #.

Note, importantly, that equations are NOT escaped. This means thatcaption="Functional form $a_i \\times x^b$, variation in %." will be displayed as:"Functional form $a_i \\times x^b$, variation in \\%.": only thelast percentage will be escaped.

If for some reason you don't want the escaping to take place, the argumentsheaders andextralines are the only ones allowing that. To disable escaping, add the special token":tex:" in the row names.Example: inheaders=list(":tex:Row title"="weird & & %\\n tex stuff\\\\"),the elements will be displayed verbatim. Of course, since it can easily ruin your table,it is only recommended to super users.

Markdown markup

Within anything that is Latex-escaped (see previous section), you can use a markdown-stylemarkup to put the text in italic and/or bold. Use⁠*text*⁠,⁠**text**⁠ or⁠***text***⁠ toput some text in, respectively, italic (with⁠\\textit⁠),bold (with⁠\\textbf⁠) and italic-bold.

The markup can be escaped by using an backslash first. For example"***This: \\***, are three stars***" will leave the three stars in the middle untouched.

Author(s)

Laurent Berge

See Also

For styling the table:setFixest_etable,style.tex,style.df.

See also the main estimation functionsfemlm,feols orfeglm.Usesummary.fixestto see the results with the appropriate standard-errors,fixef.fixest to extract thefixed-effects coefficients.

Examples

est1 = feols(Ozone ~ i(Month) / Wind + Temp, data = airquality)est2 = feols(Ozone ~ i(Month, Wind) + Temp | Month, data = airquality)# Displaying the two results in a single tableetable(est1, est2)# keep/drop: keeping only interactionsetable(est1, est2, keep = " x ")# or using drop  (see regexp help):etable(est1, est2, drop = "^(Month|Temp|\\()")# keep/drop: dropping interactionsetable(est1, est2, drop = " x ")# or using keep ("!" reverses the effect):etable(est1, est2, keep = "! x ")# order: Wind variable first, intercept last (note the "!" to reverse the effect)etable(est1, est2, order = c("Wind", "!Inter"))# Month, then interactions, then the restetable(est1, est2, order = c("^Month", " x "))## dict## You can rename variables with dict = c(var1 = alias1, var2 = alias2, etc)# You can also rename values taken by factors.# Here's a full example:dict = c(Temp = "Temperature", "Month::5"="May", "6"="Jun")etable(est1, est2, dict = dict)# Note the difference of treatment between Jun and May# Assume the following dictionary:dict = c("Month::5"="May", "Month::6"="Jun", "Month::7"="Jul",         "Month::8"="Aug", "Month::9"="Sep")# We would like to keep only the Months, but now the names are all changed...# How to do?# We can use the special character '%' to make reference to the original names.etable(est1, est2, dict = dict, keep = "%Month")## signif.code#etable(est1, est2, signif.code = c(" A"=0.01, " B"=0.05, " C"=0.1, " D"=0.15, " F"=1))## Using the argument style to customize Latex exports## If you don't like the default layout of the table, no worries!# You can modify many parameters with the argument style# To drop the headers before each section, use:# Note that a space adds an extra linestyle_noHeaders = style.tex(var.title = "", fixef.title = "", stats.title = " ")etable(est1, est2, dict = dict, tex = TRUE, style.tex = style_noHeaders)# To change the lines of the table + dropping the table footerstyle_lines = style.tex(line.top = "\\toprule", line.bottom = "\\bottomrule",                    tablefoot = FALSE)etable(est1, est2, dict = dict, tex = TRUE, style.tex = style_lines)# Or you have the predefined type "aer"etable(est1, est2, dict = dict, tex = TRUE, style.tex = style.tex("aer"))## Group and extralines## Sometimes it's useful to group control variables into a single line# You can achieve that with the group argumentsetFixest_fml(..ctrl = ~ poly(Wind, 2) + poly(Temp, 2))est_c0 = feols(Ozone ~ Solar.R, data = airquality)est_c1 = feols(Ozone ~ Solar.R + ..ctrl, data = airquality)est_c2 = feols(Ozone ~ Solar.R + Solar.R^2 + ..ctrl, data = airquality)etable(est_c0, est_c1, est_c2, group = list(Controls = "poly"))# 'group' here does the same as drop = "poly", but adds an extra line# with TRUE/FALSE where the variables were found# 'extralines' adds an extra line, where you can add the value for each modelest_all  = feols(Ozone ~ Solar.R + Temp + Wind, data = airquality)est_sub1 = feols(Ozone ~ Solar.R + Temp + Wind, data = airquality,                 subset = ~ Month %in% 5:6)est_sub2 = feols(Ozone ~ Solar.R + Temp + Wind, data = airquality,                 subset = ~ Month %in% 7:8)est_sub3 = feols(Ozone ~ Solar.R + Temp + Wind, data = airquality,                 subset = ~ Month == 9)etable(est_all, est_sub1, est_sub2, est_sub3,       extralines = list("Sub-sample" = c("All", "May-June", "Jul.-Aug.", "Sept.")))# You can monitor the placement of the new lines with two special characters# at the beginning of the row name.# 1) "^", "-" or "_" which mean the coefficients, the fixed-effects or the# statistics section.# 2) "^" or "_" which mean first or last line of the section## Ex: starting with "_^" will place the line at the top of the stat. section#     starting with "-_" will place the line at the bottom of the FEs section#     etc.## You can use a single character which will represent the section,# the line would then appear at the bottom of the section.# Examplesetable(est_c0, est_c1, est_c2, group = list("_Controls" = "poly"))etable(est_all, est_sub1, est_sub2, est_sub3,       extralines = list("^^Sub-sample" = c("All", "May-June", "Jul.-Aug.", "Sept.")))## headers## You can add header lines with 'headers'# These lines will appear at the top of the table# first, 3 estimationsest_header = feols(c(Ozone, Solar.R, Wind) ~  poly(Temp, 2), airquality)# header => vector: adds a line w/t titleetable(est_header, headers = c("A", "A", "B"))# header => list: identical way to do the previous header# The form is: list(item1 = #item1, item2 = #item2,  etc)etable(est_header, headers = list("A" = 2, "B" = 1))# Adding a title +# when an element is to be repeated only once, you can avoid the "= 1":etable(est_header, headers = list(Group = list("A" = 2, "B")))# To change the placement, add as first character:# - "^" => top# - "-" => mid (default)# - "_" => bottom# Note that "mid" and "top" are only distinguished when tex = TRUE# Placing the new header line at the bottometable(est_header, headers = list("_Group" = c("A", "A", "B"),                                  "^Currency" = list("US $" = 2, "CA $" = 1)))# In Latex, you can add "grouped underlines" (cmidrule from the booktabs package)# by adding ":_:" in the title:etable(est_header, tex = TRUE,       headers = list("^:_:Group" = c("A", "A", "B")))## extralines and headers: .() for list()## In the two arguments extralines and headers, .() can be used for list()# For example:etable(est_header, headers = .("^Currency" = .("US $" = 2, "CA $" = 1)))## fixef.group## You can group the fixed-effects line with fixef.groupest_0fe = feols(Ozone ~ Solar.R + Temp + Wind, airquality)est_1fe = feols(Ozone ~ Solar.R + Temp + Wind | Month, airquality)est_2fe = feols(Ozone ~ Solar.R + Temp + Wind | Month + Day, airquality)# A) automatic way => simply use fixef.group = TRUEetable(est_0fe, est_2fe, fixef.group = TRUE)# Note that when grouping would lead to inconsistencies across models,# it is avoidedetable(est_0fe, est_1fe, est_2fe, fixef.group = TRUE)# B) customized way => use a listetable(est_0fe, est_2fe, fixef.group = list("Dates" = "Month|Day"))# Note that when a user grouping would lead to inconsistencies,# the term partial replaces yes/no and the fixed-effects are not removed.etable(est_0fe, est_1fe, est_2fe, fixef.group = list("Dates" = "Month|Day"))# Using customized placement => as with 'group' and 'extralines',# the user can control the placement of the new line.# See the previous 'group' examples and the dedicated section in the help.# On top of the coefficients:etable(est_0fe, est_2fe, fixef.group = list("^^Dates" = "Month|Day"))# Last line of the statisticsetable(est_0fe, est_2fe, fixef.group = list("_Dates" = "Month|Day"))## Using custom functions to compute the standard errors## You can use external functions to compute the VCOVs# by feeding functions in the 'vcov' argument.# Let's use some covariances from the sandwich packageetable(est_c0, est_c1, est_c2, vcov = sandwich::vcovHC)# To add extra arguments to vcovHC, you need to write your wrapper:etable(est_c0, est_c1, est_c2, vcov = function(x) sandwich::vcovHC(x, type = "HC0"))## Customize which fit statistic to display## You can change the fit statistics with the argument fitstat# and you can rename them with the dictionaryetable(est1, est2, fitstat = ~ r2 + n + G)# If you use a formula, '.' means the default:etable(est1, est2, fitstat = ~ ll + .)## Computing a different SE for each model#est = feols(Ozone ~ Solar.R + Wind + Temp, data = airquality)## Method 1: use summarys1 = summary(est, "iid")s2 = summary(est, cluster = ~ Month)s3 = summary(est, cluster = ~ Day)s4 = summary(est, cluster = ~ Day + Month)etable(list(s1, s2, s3, s4))## Method 2: using a list in the argument 'vcov'est_bis = feols(Ozone ~ Solar.R + Wind + Temp | Month, data = airquality)etable(est, est_bis, vcov = list("hetero", ~ Month))# When you have only one model, this model is replicated# along the elements of the vcov list.etable(est, vcov = list("hetero", ~ Month))## Method 3: Using "each" or "times" in vcov# If the first element of the list in 'vcov' is "each" or "times",# then all models will be replicated and all the VCOVs will be# applied to each model. The order in which they are replicated# are governed by the each/times keywords.# eachetable(est, est_bis, vcov = list("each", "iid", ~ Month, ~ Day))# timesetable(est, est_bis, vcov = list("times", "iid", ~ Month, ~ Day))## Notes and markup## Notes can be also be set in a dictionary# You can use markdown markup to put text into italic/bolddict = c("note 1" = "*Notes:* This data is not really random.",         "source 1" = "**Source:** the internet?")est = feols(Ozone ~ csw(Solar.R, Wind, Temp), data = airquality)etable(est, dict = dict, tex = TRUE, notes = c("note 1", "source 1"))

Registerextralines macros to be used inetable

Description

This function is used to createextralines (which is an argument ofetable) macrosthat can be easily summoned inetable.

Usage

extralines_register(type, fun, alias)

Arguments

Details

You can register as many macros as you wish, the only constraint is that the type name should not conflict with afitstat type name.

Examples

# We register a function computing the standard-deviation of the dependent variablemy_fun = function(x) sd(model.matrix(x, type = "lhs"))extralines_register("sdy", my_fun, "SD(y)")# An estimationdata(iris)est = feols(Petal.Length ~ Sepal.Length | Species, iris)# Now we can easily create a row with the SD of y.# We just "summon" it in a one-sided formulaetable(est, extralines = ~ sdy)# We can change the alias on the fly:etable(est, extralines = list("_Standard deviation of the dep. var." = ~ sdy))

Lags a variable in afixest estimation

Description

Produce lags or leads in the formulas offixest estimations or when creating variables inadata.table::data.table. The data must be set as a panel beforehand (either withthe functionpanel or with the argumentpanel.id in the estimation).

Usage

f(x, k = 1, fill = NA)d(x, k = 1, fill = NA)l(x, k = 1, fill = NA)

Arguments

Value

These functions can only be used i) in a formula of afixest estimation, or ii) whencreating variables within afixest_panel object (obtained with functionpanel) whichis alaos adata.table::data.table.

Functions

• f(): Forwards a variable (inverse of lagging) in afixest estimation

• d(): Creates differences (i.e. x - lag(x)) in afixest estimation

See Also

The functionpanel changesdata.frames into a panel from which the functionslandf can be called. Otherwise you can set the panel 'live' during the estimation usingthe argumentpanel.id (see for example in the functionfeols).

Examples

data(base_did)# Setting a data set as a panel...pdat = panel(base_did, ~ id + period)# ...then using the functions l and fest1 = feols(y ~ l(x1, 0:1), pdat)est2 = feols(f(y) ~ l(x1, -1:1), pdat)est3 = feols(l(y) ~ l(x1, 0:3), pdat)etable(est1, est2, est3, order = c("f", "^x"), drop = "Int")# or using the argument panel.idfeols(f(y) ~ l(x1, -1:1), base_did, panel.id = ~id + period)feols(d(y) ~ d(x1), base_did, panel.id = ~id + period)# l() and f() can also be used within a data.table:if(require("data.table")){  pdat_dt = panel(as.data.table(base_did), ~id+period)  # Now since pdat_dt is also a data.table  #   you can create lags/leads directly  pdat_dt[, x1_l1 := l(x1)]  pdat_dt[, x1_d1 := d(x1)]  pdat_dt[, c("x1_l1_fill0", "y_f2") := .(l(x1, fill = 0), f(y, 2))]}

Formatted dimension

Description

Prints the dimension of a data set, in an user-readable way

Usage

fdim(x)

Arguments

Value

It does not return anything, the output is directly printed on the console.

Author(s)

Laurent Berge

Examples

fdim(iris)fdim(iris$Species)

Fixed effects nonlinear maximum likelihood models

Description

This function estimates maximum likelihood models (e.g., Poisson or Logit) with non-linearin parameters right-hand-sides and is efficient to handle any number of fixed effects.If you do not use non-linear in parameters right-hand-side, usefemlm orfeglminstead (their design is simpler).

Usage

feNmlm(  fml,  data,  family = c("poisson", "negbin", "logit", "gaussian"),  NL.fml,  vcov,  fixef,  fixef.rm = "perfect_fit",  NL.start,  lower,  upper,  NL.start.init,  offset,  subset,  split,  fsplit,  split.keep,  split.drop,  cluster,  se,  ssc,  panel.id,  panel.time.step = NULL,  panel.duplicate.method = "none",  start = 0,  jacobian.method = "simple",  useHessian = TRUE,  hessian.args = NULL,  opt.control = list(),  nthreads = getFixest_nthreads(),  lean = FALSE,  verbose = 0,  theta.init,  fixef.tol = 1e-05,  fixef.iter = 10000,  deriv.tol = 1e-04,  deriv.iter = 1000,  warn = TRUE,  notes = getFixest_notes(),  fixef.keep_names = NULL,  mem.clean = FALSE,  only.env = FALSE,  only.coef = FALSE,  data.save = FALSE,  env,  ...)

Arguments

Details

This function estimates maximum likelihood models where the conditional expectationsare as follows:

Gaussian likelihood:

E(Y|X)=X\beta

Poisson and Negative Binomial likelihoods:

E(Y|X)=\exp(X\beta)

where in the Negative Binomial there is the parameter\theta used tomodel the variance as\mu+\mu^2/\theta, with\mu theconditional expectation.Logit likelihood:

E(Y|X)=\frac{\exp(X\beta)}{1+\exp(X\beta)}

When there are one or more fixed-effects, the conditional expectation can be written as:

E(Y|X) = h(X\beta+\sum_{k}\sum_{m}\gamma_{m}^{k}\times C_{im}^{k}),

whereh(.) is the function corresponding to the likelihood function as shown before.C^k is the matrix associated to fixed-effect dimensionk such thatC^k_{im}is equal to 1 if observationi is of categorym in thefixed-effect dimensionk and 0 otherwise.

When there are non linear in parameters functions, we can schematically splitthe set of regressors in two:

f(X,\beta)=X^1\beta^1 + g(X^2,\beta^2)

with first a linear term and then a non linear part expressed by the function g. That is,we add a non-linear term to the linear terms (which areX*beta andthe fixed-effects coefficients). It is always better (more efficient) to putinto the argumentNL.fml only the non-linear in parameter terms, andadd all linear terms in thefml argument.

To estimate only a non-linear formula without even the intercept, you mustexclude the intercept from the linear formula by using, e.g.,fml = z~0.

The over-dispersion parameter of the Negative Binomial family, theta,is capped at 10,000. If theta reaches this high value, it means that there is no overdispersion.

Value

Afixest object. Note thatfixest objects contain many elements and most of themare for internal use, they are presented here only for information. To access them,it is safer to use the user-level methods (e.g.vcov.fixest,resid.fixest,etc) or functions (like for instancefitstat to access any fit statistic).

@seealsoSee alsosummary.fixest to see the results with the appropriate standard-errors,fixef.fixest to extract the fixed-effects coefficients, and the functionetableto visualize the results of multiple estimations.

And other estimation methods:feols,femlm,feglm,fepois,fenegbin.

Lagging variables

To use leads/lags of variables in the estimation, you can: i) either provide the argumentpanel.id, ii) either set your data set as a panel with the functionpanel,f andd.

You can provide several leads/lags/differences at once: e.g. if your formula is equal tof(y) ~ l(x, -1:1), it means that the dependent variable is equal to the lead ofy,and you will have as explanatory variables the lead ofx1,x1 and the lag ofx1.See the examples in functionl for more details.

Interactions

You can interact a numeric variable with a "factor-like" variable by usingi(factor_var, continuous_var, ref), wherecontinuous_var will be interacted witheach value offactor_var and the argumentref is a value offactor_vartaken as a reference (optional).

Using this specific way to create interactions leads to a different display of theinteracted values inetable. See examples.

It is important to note thatif you do not care about the standard-errors ofthe interactions, then you can add interactions in the fixed-effects part of the formula,it will be incomparably faster (using the syntaxfactor_var[continuous_var], as explainedin the section “Varying slopes”).

The functioni has in fact more arguments, please see details in its associated help page.

On standard-errors

Standard-errors can be computed in different ways, you can use the argumentsse andsscinsummary.fixest to define how to compute them. By default, the VCOV is the "standard" one.

The following vignette:On standard-errors describes in details how the standard-errors are computed infixest and how you can replicate standard-errors from other software.

You can use the functionssetFixest_vcov andsetFixest_ssc topermanently set the way the standard-errors are computed.

Multiple estimations

Multiple estimations can be performed at once, they just have to be specified in the formula.Multiple estimations yield afixest_multi object which is ‘kind of’ a list ofall the results but includes specific methods to access the results in a handy way.Please have a look at the dedicated vignette:Multiple estimations.

To include multiple dependent variables, wrap them inc() (list() also works).For instancefml = c(y1, y2) ~ x1 would estimate the modelfml = y1 ~ x1 andthen the modelfml = y2 ~ x1.

To include multiple independent variables, you need to use the stepwise functions.There are 4 stepwise functions:sw,sw0,csw,csw0, andmvsw. Of courseswstands for stepwise, andcsw for cumulative stepwise. Finallymvsw is a bit special,it stands for multiverse stepwise. Let's explain that.Assume you have the following formula:fml = y ~ x1 + sw(x2, x3).The stepwise functionsw will estimate the following two models:y ~ x1 + x2 andy ~ x1 + x3. That is, each element insw() is sequentially, and separately,added to the formula. Would have you usedsw0 in lieu ofsw, then the modely ~ x1 would also have been estimated. The0 in the name means that the modelwithout any stepwise element also needs to be estimated.The prefixc means cumulative: each stepwise element is added to the next. That is,fml = y ~ x1 + csw(x2, x3) would lead to the following modelsy ~ x1 + x2 andy ~ x1 + x2 + x3. The0 has the same meaning and would also lead to the model withoutthe stepwise elements to be estimated: in other words,fml = y ~ x1 + csw0(x2, x3)leads to the following three models:y ~ x1,y ~ x1 + x2 andy ~ x1 + x2 + x3.Finallymvsw will add, in a stepwise fashion all possible combinations of the variablesin its arguments. For examplemvsw(x1, x2, x3) is equivalent tosw0(x1, x2, x3, x1 + x2, x1 + x3, x2 + x3, x1 + x2 + x3). The number of modelsto estimate grows at a factorial rate: so be cautious!

Multiple independent variables can be combined with multiple dependent variables, as infml = c(y1, y2) ~ cw(x1, x2, x3) which would lead to 6 estimations. Multipleestimations can also be combined to split samples (with the argumentssplit,fsplit).

You can also add fixed-effects in a stepwise fashion. Note that you cannot performstepwise estimations on the IV part of the formula (feols only).

If NAs are present in the sample, to avoid too many messages, only NA removalconcerning the variables common to all estimations is reported.

A note on performance. The feature of multiple estimations has been highly optimized forfeols, in particular in the presence of fixed-effects. It is faster to estimatemultiple models using the formula rather than with a loop. For non-feols models usingthe formula is roughly similar to using a loop performance-wise.

Argument sliding

When the data set has been set up globally usingsetFixest_estimation(data = data_set), the argumentvcov can be used implicitly.This means that calls such asfeols(y ~ x, "HC1"), orfeols(y ~ x, ~id), are valid:i) the data is automatically deduced from the global settings, and ii) thevcovis deduced to be the second argument.

Piping

Although the argument 'data' is placed in second position, the data can be piped to theestimation functions. For example, with R >= 4.1,mtcars |> feols(mpg ~ cyl) works asfeols(mpg ~ cyl, mtcars).

Tricks to estimate multiple LHS

To use multiple dependent variables infixest estimations, you need to include themin a vector: like inc(y1, y2, y3).

First, if names are stored in a vector, they can readily be inserted in a formula toperform multiple estimations using the dot square bracket operator. For instance ifmy_lhs = c("y1", "y2"), callingfixest with, sayfeols(.[my_lhs] ~ x1, etc) isequivalent to usingfeols(c(y1, y2) ~ x1, etc). Beware that this is a special featureunique to theleft-hand-side offixest estimations (the default behavior of the DSBoperator is to aggregate with sums, seexpd).

Second, you can use a regular expression to grep the left-hand-sides on the fly. When the..("regex") (reregex("regex")) feature is used naked on the LHS,the variables grepped are inserted intoc(). For example⁠..("Pe") ~ Sepal.Length, iris⁠ is equivalent to⁠c(Petal.Length, Petal.Width) ~ Sepal.Length, iris⁠. Beware that this is aspecial feature unique to theleft-hand-side offixest estimations(the default behavior of..("regex") is to aggregate with sums, seexpd).

Note that if the dependent variable is also on the right-hand-side, it is automaticallyremoved from the set of explanatory variable.For example, feols(y ~ y + x, base) works as feols(y ~ x, base).This is particulary useful to batch multiple estimations with multiple left hand sides.

Dot square bracket operator in formulas

In a formula, the dot square bracket (DSB) operator can: i) create manifold variablesat once, or ii) capture values from the current environment and put themverbatim in the formula.

Say you want to include the variablesx1 tox3 in your formula. You can usexpd(y ~ x.[1:3]) and you'll gety ~ x1 + x2 + x3.

To summon values from the environment, simply put the variable in square brackets.For example:for(i in 1:3) xpd(y.[i] ~ x) will create the formulasy1 ~ x toy3 ~ xdepending on the value ofi.

You can include a full variable from the environment in the same way:for(y in c("a", "b")) xpd(.[y] ~ x) will create the two formulasa ~ x andb ~ x.

The DSB can even be used within variable names, but then the variable must be nested incharacter form. For exampley ~ .["x.[1:2]_sq"] will createy ~ x1_sq + x2_sq. Using thecharacter form is important to avoid a formula parsing error.Double quotes must be used. Note that the character string that is nested willbe parsed with the functiondsb, and thus it will return a vector.

By default, the DSB operator expands vectors into sums. You can add a comma,like in.[, x],to expand with commas–the content can then be used within functions. For instance:c(x.[, 1:2]) will createc(x1, x2) (andnotc(x1 + x2)).

In allfixest estimations, this special parsing is enabled, so you don't need to usexpd.

One-sided formulas can be expanded with the DSB operator: letx = ~sepal + petal, thenxpd(y ~ .[x]) leads tocolor ~ sepal + petal.

You can even use multiple square brackets within a single variable,but then the use of nesting is required.For example, the followingxpd(y ~ .[".[letters[1:2]]_.[1:2]"]) will createy ~ a_1 + b_2. Remember that the nested character string is parsed withdsb,which explains this behavior.

When the element to be expanded i) is equal to the empty string or,ii) is of length 0, it is replaced with a neutral element, namely1.For example,⁠x = "" ; xpd(y ~ .[x])⁠ leads toy ~ 1.

Author(s)

Laurent Berge

References

Berge, Laurent, 2018, "Efficient estimation of maximum likelihood models withmultiple fixed-effects: the R package FENmlm." CREA Discussion Papers,13 ().

For models with multiple fixed-effects:

Gaure, Simen, 2013, "OLS with multiple high dimensional category variables",Computational Statistics & Data Analysis 66 pp. 8–18

On the unconditionnal Negative Binomial model:

Allison, Paul D and Waterman, Richard P, 2002, "Fixed-Effects NegativeBinomial Regression Models", Sociological Methodology 32(1) pp. 247–265

Examples

# This section covers only non-linear in parameters examples# For linear relationships: use femlm or feglm instead# Generating data for a simple exampleset.seed(1)n = 100x = rnorm(n, 1, 5)**2y = rnorm(n, -1, 5)**2z1 = rpois(n, x*y) + rpois(n, 2)base = data.frame(x, y, z1)# Estimating a 'linear' relation:est1_L = femlm(z1 ~ log(x) + log(y), base)# Estimating the same 'linear' relation using a 'non-linear' callest1_NL = feNmlm(z1 ~ 1, base, NL.fml = ~a*log(x)+b*log(y), NL.start = list(a=0, b=0))# we compare the estimates with the function esttable (they are identical)etable(est1_L, est1_NL)# Now generating a non-linear relation (E(z2) = x + y + 1):z2 = rpois(n, x + y) + rpois(n, 1)base$z2 = z2# Estimation using this non-linear formest2_NL = feNmlm(z2 ~ 0, base, NL.fml = ~log(a*x + b*y),               NL.start = 2, lower = list(a=0, b=0))# we can't estimate this relation linearily# => closest we can do:est2_L = femlm(z2 ~ log(x) + log(y), base)# Difference between the two models:etable(est2_L, est2_NL)# Plotting the fits:plot(x, z2, pch = 18)points(x, fitted(est2_L), col = 2, pch = 1)points(x, fitted(est2_NL), col = 4, pch = 2)

Fixed-effects GLM estimations

Description

Estimates GLM models with any number of fixed-effects.

Usage

feglm(  fml,  data,  family = "gaussian",  vcov,  offset,  weights,  subset,  split,  fsplit,  split.keep,  split.drop,  cluster,  se,  ssc,  panel.id,  panel.time.step = NULL,  panel.duplicate.method = "none",  start = NULL,  etastart = NULL,  mustart = NULL,  fixef,  fixef.rm = "perfect_fit",  fixef.tol = 1e-06,  fixef.iter = 10000,  fixef.algo = NULL,  collin.tol = 1e-09,  glm.iter = 25,  glm.tol = 1e-08,  nthreads = getFixest_nthreads(),  lean = FALSE,  warn = TRUE,  notes = getFixest_notes(),  verbose = 0,  only.coef = FALSE,  data.save = FALSE,  fixef.keep_names = NULL,  mem.clean = FALSE,  only.env = FALSE,  env,  ...)feglm.fit(  y,  X,  fixef_df,  family = "gaussian",  vcov,  offset,  split,  fsplit,  split.keep,  split.drop,  cluster,  se,  ssc,  weights,  subset,  start = NULL,  etastart = NULL,  mustart = NULL,  fixef.rm = "perfect_fit",  fixef.tol = 1e-06,  fixef.iter = 10000,  fixef.algo = NULL,  collin.tol = 1e-09,  glm.iter = 25,  glm.tol = 1e-08,  nthreads = getFixest_nthreads(),  lean = FALSE,  warn = TRUE,  notes = getFixest_notes(),  mem.clean = FALSE,  verbose = 0,  only.env = FALSE,  only.coef = FALSE,  env,  ...)fepois(  fml,  data,  vcov,  offset,  weights,  subset,  split,  fsplit,  split.keep,  split.drop,  cluster,  se,  ssc,  panel.id,  panel.time.step = NULL,  panel.duplicate.method = "none",  start = NULL,  etastart = NULL,  mustart = NULL,  fixef,  fixef.rm = "perfect_fit",  fixef.tol = 1e-06,  fixef.iter = 10000,  fixef.algo = NULL,  collin.tol = 1e-09,  glm.iter = 25,  glm.tol = 1e-08,  nthreads = getFixest_nthreads(),  lean = FALSE,  warn = TRUE,  notes = getFixest_notes(),  verbose = 0,  fixef.keep_names = NULL,  mem.clean = FALSE,  only.env = FALSE,  only.coef = FALSE,  data.save = FALSE,  env,  ...)

Arguments

Details

The core of the GLM are the weighted OLS estimations. These estimations are performedwithfeols. The method used to demean each variable along the fixed-effectsis based on Berge (2018), since this is the same problem to solve as for the Gaussiancase in a ML setup.

Value

Afixest object. Note thatfixest objects contain many elements and most of themare for internal use, they are presented here only for information. To access them,it is safer to use the user-level methods (e.g.vcov.fixest,resid.fixest,etc) or functions (like for instancefitstat to access any fit statistic).

Combining the fixed-effects

You can combine two variables to make it a new fixed-effect using^.The syntax is as follows:fe_1^fe_2. Here you created a new variable which is the combinationof the two variables fe_1 and fe_2. This is identical to doingpaste0(fe_1, "_", fe_2)but more convenient.

Note that pasting is a costly operation, especially for large data sets.Hence, by default this paste is done only when the number of unique valuesis lower than 50,000 observations.

In case you are using a large data set and want to keep the identity of the fixed-effects,you need to use the argumentfixef.keep_names = TRUE.

Note that these “identities” are useful only if you're interested inthe value of the fixed-effects (that you can extract withfixef.fixest).

Varying slopes

You can add variables with varying slopes in the fixed-effect part of the formula.The syntax is as follows:fixef_var[var1, var2]. Here the variables var1 and var2 willbe with varying slopes (one slope per value in fixef_var) and the fixed-effectfixef_var will also be added.

To add only the variables with varying slopes and not the fixed-effect,use double square brackets:fixef_var[[var1, var2]].

In other words:

• fixef_var[var1, var2] is equivalent tofixef_var + fixef_var[[var1]] + fixef_var[[var2]]

• fixef_var[[var1, var2]] is equivalent tofixef_var[[var1]] + fixef_var[[var2]]

In general, for convergence reasons, it is recommended to always add the fixed-effect andavoid using only the variable with varying slope (i.e. use single square brackets).

Lagging variables

To use leads/lags of variables in the estimation, you can: i) either provide the argumentpanel.id, ii) either set your data set as a panel with the functionpanel,f andd.

You can provide several leads/lags/differences at once: e.g. if your formula is equal tof(y) ~ l(x, -1:1), it means that the dependent variable is equal to the lead ofy,and you will have as explanatory variables the lead ofx1,x1 and the lag ofx1.See the examples in functionl for more details.

Interactions

You can interact a numeric variable with a "factor-like" variable by usingi(factor_var, continuous_var, ref), wherecontinuous_var will be interacted witheach value offactor_var and the argumentref is a value offactor_vartaken as a reference (optional).

Using this specific way to create interactions leads to a different display of theinteracted values inetable. See examples.

It is important to note thatif you do not care about the standard-errors ofthe interactions, then you can add interactions in the fixed-effects part of the formula,it will be incomparably faster (using the syntaxfactor_var[continuous_var], as explainedin the section “Varying slopes”).

The functioni has in fact more arguments, please see details in its associated help page.

On standard-errors

Standard-errors can be computed in different ways, you can use the argumentsse andsscinsummary.fixest to define how to compute them. By default, the VCOV is the "standard" one.

The following vignette:On standard-errors describes in details how the standard-errors are computed infixest and how you can replicate standard-errors from other software.

You can use the functionssetFixest_vcov andsetFixest_ssc topermanently set the way the standard-errors are computed.

Multiple estimations

Multiple estimations can be performed at once, they just have to be specified in the formula.Multiple estimations yield afixest_multi object which is ‘kind of’ a list ofall the results but includes specific methods to access the results in a handy way.Please have a look at the dedicated vignette:Multiple estimations.

To include multiple dependent variables, wrap them inc() (list() also works).For instancefml = c(y1, y2) ~ x1 would estimate the modelfml = y1 ~ x1 andthen the modelfml = y2 ~ x1.

To include multiple independent variables, you need to use the stepwise functions.There are 4 stepwise functions:sw,sw0,csw,csw0, andmvsw. Of courseswstands for stepwise, andcsw for cumulative stepwise. Finallymvsw is a bit special,it stands for multiverse stepwise. Let's explain that.Assume you have the following formula:fml = y ~ x1 + sw(x2, x3).The stepwise functionsw will estimate the following two models:y ~ x1 + x2 andy ~ x1 + x3. That is, each element insw() is sequentially, and separately,added to the formula. Would have you usedsw0 in lieu ofsw, then the modely ~ x1 would also have been estimated. The0 in the name means that the modelwithout any stepwise element also needs to be estimated.The prefixc means cumulative: each stepwise element is added to the next. That is,fml = y ~ x1 + csw(x2, x3) would lead to the following modelsy ~ x1 + x2 andy ~ x1 + x2 + x3. The0 has the same meaning and would also lead to the model withoutthe stepwise elements to be estimated: in other words,fml = y ~ x1 + csw0(x2, x3)leads to the following three models:y ~ x1,y ~ x1 + x2 andy ~ x1 + x2 + x3.Finallymvsw will add, in a stepwise fashion all possible combinations of the variablesin its arguments. For examplemvsw(x1, x2, x3) is equivalent tosw0(x1, x2, x3, x1 + x2, x1 + x3, x2 + x3, x1 + x2 + x3). The number of modelsto estimate grows at a factorial rate: so be cautious!

Multiple independent variables can be combined with multiple dependent variables, as infml = c(y1, y2) ~ cw(x1, x2, x3) which would lead to 6 estimations. Multipleestimations can also be combined to split samples (with the argumentssplit,fsplit).

You can also add fixed-effects in a stepwise fashion. Note that you cannot performstepwise estimations on the IV part of the formula (feols only).

If NAs are present in the sample, to avoid too many messages, only NA removalconcerning the variables common to all estimations is reported.

A note on performance. The feature of multiple estimations has been highly optimized forfeols, in particular in the presence of fixed-effects. It is faster to estimatemultiple models using the formula rather than with a loop. For non-feols models usingthe formula is roughly similar to using a loop performance-wise.

Argument sliding

When the data set has been set up globally usingsetFixest_estimation(data = data_set), the argumentvcov can be used implicitly.This means that calls such asfeols(y ~ x, "HC1"), orfeols(y ~ x, ~id), are valid:i) the data is automatically deduced from the global settings, and ii) thevcovis deduced to be the second argument.

Piping

Although the argument 'data' is placed in second position, the data can be piped to theestimation functions. For example, with R >= 4.1,mtcars |> feols(mpg ~ cyl) works asfeols(mpg ~ cyl, mtcars).

Tricks to estimate multiple LHS

To use multiple dependent variables infixest estimations, you need to include themin a vector: like inc(y1, y2, y3).

First, if names are stored in a vector, they can readily be inserted in a formula toperform multiple estimations using the dot square bracket operator. For instance ifmy_lhs = c("y1", "y2"), callingfixest with, sayfeols(.[my_lhs] ~ x1, etc) isequivalent to usingfeols(c(y1, y2) ~ x1, etc). Beware that this is a special featureunique to theleft-hand-side offixest estimations (the default behavior of the DSBoperator is to aggregate with sums, seexpd).

Second, you can use a regular expression to grep the left-hand-sides on the fly. When the..("regex") (reregex("regex")) feature is used naked on the LHS,the variables grepped are inserted intoc(). For example⁠..("Pe") ~ Sepal.Length, iris⁠ is equivalent to⁠c(Petal.Length, Petal.Width) ~ Sepal.Length, iris⁠. Beware that this is aspecial feature unique to theleft-hand-side offixest estimations(the default behavior of..("regex") is to aggregate with sums, seexpd).

Note that if the dependent variable is also on the right-hand-side, it is automaticallyremoved from the set of explanatory variable.For example, feols(y ~ y + x, base) works as feols(y ~ x, base).This is particulary useful to batch multiple estimations with multiple left hand sides.

Dot square bracket operator in formulas

In a formula, the dot square bracket (DSB) operator can: i) create manifold variablesat once, or ii) capture values from the current environment and put themverbatim in the formula.

Say you want to include the variablesx1 tox3 in your formula. You can usexpd(y ~ x.[1:3]) and you'll gety ~ x1 + x2 + x3.

To summon values from the environment, simply put the variable in square brackets.For example:for(i in 1:3) xpd(y.[i] ~ x) will create the formulasy1 ~ x toy3 ~ xdepending on the value ofi.

You can include a full variable from the environment in the same way:for(y in c("a", "b")) xpd(.[y] ~ x) will create the two formulasa ~ x andb ~ x.

The DSB can even be used within variable names, but then the variable must be nested incharacter form. For exampley ~ .["x.[1:2]_sq"] will createy ~ x1_sq + x2_sq. Using thecharacter form is important to avoid a formula parsing error.Double quotes must be used. Note that the character string that is nested willbe parsed with the functiondsb, and thus it will return a vector.

By default, the DSB operator expands vectors into sums. You can add a comma,like in.[, x],to expand with commas–the content can then be used within functions. For instance:c(x.[, 1:2]) will createc(x1, x2) (andnotc(x1 + x2)).

In allfixest estimations, this special parsing is enabled, so you don't need to usexpd.

One-sided formulas can be expanded with the DSB operator: letx = ~sepal + petal, thenxpd(y ~ .[x]) leads tocolor ~ sepal + petal.

You can even use multiple square brackets within a single variable,but then the use of nesting is required.For example, the followingxpd(y ~ .[".[letters[1:2]]_.[1:2]"]) will createy ~ a_1 + b_2. Remember that the nested character string is parsed withdsb,which explains this behavior.

When the element to be expanded i) is equal to the empty string or,ii) is of length 0, it is replaced with a neutral element, namely1.For example,⁠x = "" ; xpd(y ~ .[x])⁠ leads toy ~ 1.

Author(s)

Laurent Berge

References

Berge, Laurent, 2018, "Efficient estimation of maximum likelihood models withmultiple fixed-effects: the R package FENmlm." CREA Discussion Papers,13 ().

For models with multiple fixed-effects:

Gaure, Simen, 2013, "OLS with multiple high dimensional category variables",Computational Statistics & Data Analysis 66 pp. 8–18

See Also

See alsosummary.fixest to see the results with the appropriate standard-errors,fixef.fixest to extract the fixed-effects coefficients, and the functionetableto visualize the results of multiple estimations.And other estimation methods:feols,femlm,fenegbin,feNmlm.

Examples

# Poisson estimationres = feglm(Sepal.Length ~ Sepal.Width + Petal.Length | Species, iris, "poisson")# You could also use fepoisres_pois = fepois(Sepal.Length ~ Sepal.Width + Petal.Length | Species, iris)# With the fit method:res_fit = feglm.fit(iris$Sepal.Length, iris[, 2:3], iris$Species, "poisson")# All results are identical:etable(res, res_pois, res_fit)# Note that you have many more examples in feols## Multiple estimations:## 6 estimationsest_mult = fepois(c(Ozone, Solar.R) ~ Wind + Temp + csw0(Wind:Temp, Day), airquality)# We can display the results for the first lhs:etable(est_mult[lhs = 1])# And now the second (access can be made by name)etable(est_mult[lhs = "Solar.R"])# Now we focus on the two last right hand sides# (note that .N can be used to specify the last item)etable(est_mult[rhs = 2:.N])# Combining with splitest_split = fepois(c(Ozone, Solar.R) ~ sw(poly(Wind, 2), poly(Temp, 2)),                  airquality, split = ~ Month)# You can display everything at once with the print methodest_split# Different way of displaying the results with "compact"summary(est_split, "compact")# You can still select which sample/LHS/RHS to displayest_split[sample = 1:2, lhs = 1, rhs = 1]

Fixed-effects maximum likelihood models

Description

This function estimates maximum likelihood models with any number of fixed-effects.

Usage

femlm(  fml,  data,  family = c("poisson", "negbin", "logit", "gaussian"),  vcov,  start = 0,  fixef,  fixef.rm = "perfect_fit",  offset,  subset,  split,  fsplit,  split.keep,  split.drop,  cluster,  se,  ssc,  panel.id,  panel.time.step = NULL,  panel.duplicate.method = "none",  fixef.tol = 1e-05,  fixef.iter = 10000,  nthreads = getFixest_nthreads(),  lean = FALSE,  verbose = 0,  warn = TRUE,  notes = getFixest_notes(),  theta.init,  fixef.keep_names = NULL,  mem.clean = FALSE,  only.env = FALSE,  only.coef = FALSE,  data.save = FALSE,  env,  ...)fenegbin(  fml,  data,  vcov,  theta.init,  start = 0,  fixef,  fixef.rm = "perfect_fit",  offset,  subset,  split,  fsplit,  split.keep,  split.drop,  cluster,  se,  ssc,  panel.id,  panel.time.step = NULL,  panel.duplicate.method = "none",  fixef.tol = 1e-05,  fixef.iter = 10000,  nthreads = getFixest_nthreads(),  lean = FALSE,  verbose = 0,  warn = TRUE,  notes = getFixest_notes(),  fixef.keep_names = NULL,  mem.clean = FALSE,  only.env = FALSE,  only.coef = FALSE,  data.save = FALSE,  env,  ...)

Arguments

Details

Note that the functionsfeglm andfemlm provide the same results when usingthe same families but differ in that the latter is a direct maximum likelihoodoptimization (so the two can really have different convergence rates).

Value

Afixest object. Note thatfixest objects contain many elements and most ofthem are for internal use, they are presented here only for information.To access them, it is safer to use the user-level methods(e.g.vcov.fixest,resid.fixest, etc) or functions (like for instancefitstat to access any fit statistic).

Combining the fixed-effects

You can combine two variables to make it a new fixed-effect using^.The syntax is as follows:fe_1^fe_2. Here you created a new variable which is the combinationof the two variables fe_1 and fe_2. This is identical to doingpaste0(fe_1, "_", fe_2)but more convenient.

Note that pasting is a costly operation, especially for large data sets.Hence, by default this paste is done only when the number of unique valuesis lower than 50,000 observations.

In case you are using a large data set and want to keep the identity of the fixed-effects,you need to use the argumentfixef.keep_names = TRUE.

Note that these “identities” are useful only if you're interested inthe value of the fixed-effects (that you can extract withfixef.fixest).

Lagging variables

To use leads/lags of variables in the estimation, you can: i) either provide the argumentpanel.id, ii) either set your data set as a panel with the functionpanel,f andd.

You can provide several leads/lags/differences at once: e.g. if your formula is equal tof(y) ~ l(x, -1:1), it means that the dependent variable is equal to the lead ofy,and you will have as explanatory variables the lead ofx1,x1 and the lag ofx1.See the examples in functionl for more details.

Interactions

You can interact a numeric variable with a "factor-like" variable by usingi(factor_var, continuous_var, ref), wherecontinuous_var will be interacted witheach value offactor_var and the argumentref is a value offactor_vartaken as a reference (optional).

Using this specific way to create interactions leads to a different display of theinteracted values inetable. See examples.

It is important to note thatif you do not care about the standard-errors ofthe interactions, then you can add interactions in the fixed-effects part of the formula,it will be incomparably faster (using the syntaxfactor_var[continuous_var], as explainedin the section “Varying slopes”).

The functioni has in fact more arguments, please see details in its associated help page.

On standard-errors

Standard-errors can be computed in different ways, you can use the argumentsse andsscinsummary.fixest to define how to compute them. By default, the VCOV is the "standard" one.

The following vignette:On standard-errors describes in details how the standard-errors are computed infixest and how you can replicate standard-errors from other software.

You can use the functionssetFixest_vcov andsetFixest_ssc topermanently set the way the standard-errors are computed.

Multiple estimations

Multiple estimations can be performed at once, they just have to be specified in the formula.Multiple estimations yield afixest_multi object which is ‘kind of’ a list ofall the results but includes specific methods to access the results in a handy way.Please have a look at the dedicated vignette:Multiple estimations.

To include multiple dependent variables, wrap them inc() (list() also works).For instancefml = c(y1, y2) ~ x1 would estimate the modelfml = y1 ~ x1 andthen the modelfml = y2 ~ x1.

To include multiple independent variables, you need to use the stepwise functions.There are 4 stepwise functions:sw,sw0,csw,csw0, andmvsw. Of courseswstands for stepwise, andcsw for cumulative stepwise. Finallymvsw is a bit special,it stands for multiverse stepwise. Let's explain that.Assume you have the following formula:fml = y ~ x1 + sw(x2, x3).The stepwise functionsw will estimate the following two models:y ~ x1 + x2 andy ~ x1 + x3. That is, each element insw() is sequentially, and separately,added to the formula. Would have you usedsw0 in lieu ofsw, then the modely ~ x1 would also have been estimated. The0 in the name means that the modelwithout any stepwise element also needs to be estimated.The prefixc means cumulative: each stepwise element is added to the next. That is,fml = y ~ x1 + csw(x2, x3) would lead to the following modelsy ~ x1 + x2 andy ~ x1 + x2 + x3. The0 has the same meaning and would also lead to the model withoutthe stepwise elements to be estimated: in other words,fml = y ~ x1 + csw0(x2, x3)leads to the following three models:y ~ x1,y ~ x1 + x2 andy ~ x1 + x2 + x3.Finallymvsw will add, in a stepwise fashion all possible combinations of the variablesin its arguments. For examplemvsw(x1, x2, x3) is equivalent tosw0(x1, x2, x3, x1 + x2, x1 + x3, x2 + x3, x1 + x2 + x3). The number of modelsto estimate grows at a factorial rate: so be cautious!

Multiple independent variables can be combined with multiple dependent variables, as infml = c(y1, y2) ~ cw(x1, x2, x3) which would lead to 6 estimations. Multipleestimations can also be combined to split samples (with the argumentssplit,fsplit).

You can also add fixed-effects in a stepwise fashion. Note that you cannot performstepwise estimations on the IV part of the formula (feols only).

If NAs are present in the sample, to avoid too many messages, only NA removalconcerning the variables common to all estimations is reported.

A note on performance. The feature of multiple estimations has been highly optimized forfeols, in particular in the presence of fixed-effects. It is faster to estimatemultiple models using the formula rather than with a loop. For non-feols models usingthe formula is roughly similar to using a loop performance-wise.

Argument sliding

When the data set has been set up globally usingsetFixest_estimation(data = data_set), the argumentvcov can be used implicitly.This means that calls such asfeols(y ~ x, "HC1"), orfeols(y ~ x, ~id), are valid:i) the data is automatically deduced from the global settings, and ii) thevcovis deduced to be the second argument.

Piping

Although the argument 'data' is placed in second position, the data can be piped to theestimation functions. For example, with R >= 4.1,mtcars |> feols(mpg ~ cyl) works asfeols(mpg ~ cyl, mtcars).

Tricks to estimate multiple LHS

To use multiple dependent variables infixest estimations, you need to include themin a vector: like inc(y1, y2, y3).

First, if names are stored in a vector, they can readily be inserted in a formula toperform multiple estimations using the dot square bracket operator. For instance ifmy_lhs = c("y1", "y2"), callingfixest with, sayfeols(.[my_lhs] ~ x1, etc) isequivalent to usingfeols(c(y1, y2) ~ x1, etc). Beware that this is a special featureunique to theleft-hand-side offixest estimations (the default behavior of the DSBoperator is to aggregate with sums, seexpd).

Second, you can use a regular expression to grep the left-hand-sides on the fly. When the..("regex") (reregex("regex")) feature is used naked on the LHS,the variables grepped are inserted intoc(). For example⁠..("Pe") ~ Sepal.Length, iris⁠ is equivalent to⁠c(Petal.Length, Petal.Width) ~ Sepal.Length, iris⁠. Beware that this is aspecial feature unique to theleft-hand-side offixest estimations(the default behavior of..("regex") is to aggregate with sums, seexpd).

Note that if the dependent variable is also on the right-hand-side, it is automaticallyremoved from the set of explanatory variable.For example, feols(y ~ y + x, base) works as feols(y ~ x, base).This is particulary useful to batch multiple estimations with multiple left hand sides.

Dot square bracket operator in formulas

In a formula, the dot square bracket (DSB) operator can: i) create manifold variablesat once, or ii) capture values from the current environment and put themverbatim in the formula.

Say you want to include the variablesx1 tox3 in your formula. You can usexpd(y ~ x.[1:3]) and you'll gety ~ x1 + x2 + x3.

To summon values from the environment, simply put the variable in square brackets.For example:for(i in 1:3) xpd(y.[i] ~ x) will create the formulasy1 ~ x toy3 ~ xdepending on the value ofi.

You can include a full variable from the environment in the same way:for(y in c("a", "b")) xpd(.[y] ~ x) will create the two formulasa ~ x andb ~ x.

The DSB can even be used within variable names, but then the variable must be nested incharacter form. For exampley ~ .["x.[1:2]_sq"] will createy ~ x1_sq + x2_sq. Using thecharacter form is important to avoid a formula parsing error.Double quotes must be used. Note that the character string that is nested willbe parsed with the functiondsb, and thus it will return a vector.

By default, the DSB operator expands vectors into sums. You can add a comma,like in.[, x],to expand with commas–the content can then be used within functions. For instance:c(x.[, 1:2]) will createc(x1, x2) (andnotc(x1 + x2)).

In allfixest estimations, this special parsing is enabled, so you don't need to usexpd.

One-sided formulas can be expanded with the DSB operator: letx = ~sepal + petal, thenxpd(y ~ .[x]) leads tocolor ~ sepal + petal.

You can even use multiple square brackets within a single variable,but then the use of nesting is required.For example, the followingxpd(y ~ .[".[letters[1:2]]_.[1:2]"]) will createy ~ a_1 + b_2. Remember that the nested character string is parsed withdsb,which explains this behavior.

When the element to be expanded i) is equal to the empty string or,ii) is of length 0, it is replaced with a neutral element, namely1.For example,⁠x = "" ; xpd(y ~ .[x])⁠ leads toy ~ 1.

Author(s)

Laurent Berge

References

Berge, Laurent, 2018, "Efficient estimation of maximum likelihood models withmultiple fixed-effects: the R package FENmlm." CREA Discussion Papers,13 ().

For models with multiple fixed-effects:

Gaure, Simen, 2013, "OLS with multiple high dimensional category variables",Computational Statistics & Data Analysis 66 pp. 8–18

On the unconditionnal Negative Binomial model:

Allison, Paul D and Waterman, Richard P, 2002, "Fixed-Effects NegativeBinomial Regression Models", Sociological Methodology 32(1) pp. 247–265

See Also

See alsosummary.fixest to see the results with the appropriate standard-errors,fixef.fixest to extract the fixed-effects coefficients, and the functionetable to visualize the results of multiple estimations.And other estimation methods:feols,feglm,fepois,feNmlm.

Examples

# Load trade datadata(trade)# We estimate the effect of distance on trade => we account for 3 fixed-effects# 1) Poisson estimationest_pois = femlm(Euros ~ log(dist_km) | Origin + Destination + Product, trade)# 2) Log-Log Gaussian estimation (with same FEs)est_gaus = update(est_pois, log(Euros+1) ~ ., family = "gaussian")# Comparison of the results using the function etableetable(est_pois, est_gaus)# Now using two way clustered standard-errorsetable(est_pois, est_gaus, se = "twoway")# Comparing different types of standard errorssum_hetero   = summary(est_pois, se = "hetero")sum_oneway   = summary(est_pois, se = "cluster")sum_twoway   = summary(est_pois, se = "twoway")sum_threeway = summary(est_pois, se = "threeway")etable(sum_hetero, sum_oneway, sum_twoway, sum_threeway)## Multiple estimations:## 6 estimationsest_mult = femlm(c(Ozone, Solar.R) ~ Wind + Temp + csw0(Wind:Temp, Day), airquality)# We can display the results for the first lhs:etable(est_mult[lhs = 1])# And now the second (access can be made by name)etable(est_mult[lhs = "Solar.R"])# Now we focus on the two last right hand sides# (note that .N can be used to specify the last item)etable(est_mult[rhs = 2:.N])# Combining with splitest_split = fepois(c(Ozone, Solar.R) ~ sw(poly(Wind, 2), poly(Temp, 2)),                  airquality, split = ~ Month)# You can display everything at once with the print methodest_split# Different way of displaying the results with "compact"summary(est_split, "compact")# You can still select which sample/LHS/RHS to displayest_split[sample = 1:2, lhs = 1, rhs = 1]

Fixed-effects OLS estimation

Description

Estimates OLS with any number of fixed-effects.

Usage

feols(  fml,  data,  vcov,  weights,  offset,  subset,  split,  fsplit,  split.keep,  split.drop,  cluster,  se,  ssc,  panel.id,  panel.time.step = NULL,  panel.duplicate.method = "none",  fixef,  fixef.rm = "perfect_fit",  fixef.tol = 1e-06,  fixef.iter = 10000,  fixef.algo = NULL,  collin.tol = 1e-09,  nthreads = getFixest_nthreads(),  lean = FALSE,  verbose = 0,  warn = TRUE,  notes = getFixest_notes(),  only.coef = FALSE,  data.save = FALSE,  fixef.keep_names = NULL,  demeaned = FALSE,  mem.clean = FALSE,  only.env = FALSE,  env,  ...)feols.fit(  y,  X,  fixef_df,  vcov,  offset,  split,  fsplit,  split.keep,  split.drop,  cluster,  se,  ssc,  weights,  subset,  fixef.rm = "perfect_fit",  fixef.tol = 1e-06,  fixef.iter = 10000,  fixef.algo = NULL,  collin.tol = 1e-09,  nthreads = getFixest_nthreads(),  lean = FALSE,  warn = TRUE,  notes = getFixest_notes(),  mem.clean = FALSE,  verbose = 0,  only.env = FALSE,  only.coef = FALSE,  env,  ...)

Arguments

Details

The method used to demean each variable along the fixed-effects is based on Berge (2018), sincethis is the same problem to solve as for the Gaussian case in a ML setup.

Value

Afixest object. Note thatfixest objects contain many elements and most of them arefor internal use, they are presented here only for information. To access them, it is saferto use the user-level methods (e.g.vcov.fixest,resid.fixest, etc) or functions(like for instancefitstat to access any fit statistic).

Combining the fixed-effects

You can combine two variables to make it a new fixed-effect using^.The syntax is as follows:fe_1^fe_2. Here you created a new variable which is the combinationof the two variables fe_1 and fe_2. This is identical to doingpaste0(fe_1, "_", fe_2)but more convenient.

Note that pasting is a costly operation, especially for large data sets.Hence, by default this paste is done only when the number of unique valuesis lower than 50,000 observations.

In case you are using a large data set and want to keep the identity of the fixed-effects,you need to use the argumentfixef.keep_names = TRUE.

Note that these “identities” are useful only if you're interested inthe value of the fixed-effects (that you can extract withfixef.fixest).

Varying slopes

You can add variables with varying slopes in the fixed-effect part of the formula.The syntax is as follows:fixef_var[var1, var2]. Here the variables var1 and var2 willbe with varying slopes (one slope per value in fixef_var) and the fixed-effectfixef_var will also be added.

To add only the variables with varying slopes and not the fixed-effect,use double square brackets:fixef_var[[var1, var2]].

In other words:

• fixef_var[var1, var2] is equivalent tofixef_var + fixef_var[[var1]] + fixef_var[[var2]]

• fixef_var[[var1, var2]] is equivalent tofixef_var[[var1]] + fixef_var[[var2]]

In general, for convergence reasons, it is recommended to always add the fixed-effect andavoid using only the variable with varying slope (i.e. use single square brackets).

Lagging variables

To use leads/lags of variables in the estimation, you can: i) either provide the argumentpanel.id, ii) either set your data set as a panel with the functionpanel,f andd.

You can provide several leads/lags/differences at once: e.g. if your formula is equal tof(y) ~ l(x, -1:1), it means that the dependent variable is equal to the lead ofy,and you will have as explanatory variables the lead ofx1,x1 and the lag ofx1.See the examples in functionl for more details.

Interactions

You can interact a numeric variable with a "factor-like" variable by usingi(factor_var, continuous_var, ref), wherecontinuous_var will be interacted witheach value offactor_var and the argumentref is a value offactor_vartaken as a reference (optional).

Using this specific way to create interactions leads to a different display of theinteracted values inetable. See examples.

It is important to note thatif you do not care about the standard-errors ofthe interactions, then you can add interactions in the fixed-effects part of the formula,it will be incomparably faster (using the syntaxfactor_var[continuous_var], as explainedin the section “Varying slopes”).

The functioni has in fact more arguments, please see details in its associated help page.

On standard-errors

Standard-errors can be computed in different ways, you can use the argumentsse andsscinsummary.fixest to define how to compute them. By default, the VCOV is the "standard" one.

The following vignette:On standard-errors describes in details how the standard-errors are computed infixest and how you can replicate standard-errors from other software.

You can use the functionssetFixest_vcov andsetFixest_ssc topermanently set the way the standard-errors are computed.

Instrumental variables

To estimate two stage least square regressions, insert the relationship betweenthe endogenous regressor(s) and the instruments in a formula, after a pipe.

For example,fml = y ~ x1 | x_endo ~ x_inst will use the variablesx1 andx_inst inthe first stage to explainx_endo. Then will use the fitted value ofx_endo(which will be namedfit_x_endo) andx1 to explainy.To include several endogenous regressors, just use "+",like in:fml = y ~ x1 | x_endo1 + x_end2 ~ x_inst1 + x_inst2.

Of course you can still add the fixed-effects, but the IV formula must always come last,like infml = y ~ x1 | fe1 + fe2 | x_endo ~ x_inst.

If you want to estimate a model without exogenous variables, use"1" as aplaceholder: e.g.fml = y ~ 1 | x_endo ~ x_inst.

By default, the second stage regression is returned. You can access the first stage(s)regressions either directly in the slotiv_first_stage (not recommended),or using the argumentstage = 1 from the functionsummary.fixest.For examplesummary(iv_est, stage = 1) will give the first stage(s).Note that using summary you can display both the second and first stages atthe same time using, e.g.,stage = 1:2 (using2:1 would reverse the order).

Multiple estimations

Multiple estimations can be performed at once, they just have to be specified in the formula.Multiple estimations yield afixest_multi object which is ‘kind of’ a list ofall the results but includes specific methods to access the results in a handy way.Please have a look at the dedicated vignette:Multiple estimations.

To include multiple dependent variables, wrap them inc() (list() also works).For instancefml = c(y1, y2) ~ x1 would estimate the modelfml = y1 ~ x1 andthen the modelfml = y2 ~ x1.

To include multiple independent variables, you need to use the stepwise functions.There are 4 stepwise functions:sw,sw0,csw,csw0, andmvsw. Of courseswstands for stepwise, andcsw for cumulative stepwise. Finallymvsw is a bit special,it stands for multiverse stepwise. Let's explain that.Assume you have the following formula:fml = y ~ x1 + sw(x2, x3).The stepwise functionsw will estimate the following two models:y ~ x1 + x2 andy ~ x1 + x3. That is, each element insw() is sequentially, and separately,added to the formula. Would have you usedsw0 in lieu ofsw, then the modely ~ x1 would also have been estimated. The0 in the name means that the modelwithout any stepwise element also needs to be estimated.The prefixc means cumulative: each stepwise element is added to the next. That is,fml = y ~ x1 + csw(x2, x3) would lead to the following modelsy ~ x1 + x2 andy ~ x1 + x2 + x3. The0 has the same meaning and would also lead to the model withoutthe stepwise elements to be estimated: in other words,fml = y ~ x1 + csw0(x2, x3)leads to the following three models:y ~ x1,y ~ x1 + x2 andy ~ x1 + x2 + x3.Finallymvsw will add, in a stepwise fashion all possible combinations of the variablesin its arguments. For examplemvsw(x1, x2, x3) is equivalent tosw0(x1, x2, x3, x1 + x2, x1 + x3, x2 + x3, x1 + x2 + x3). The number of modelsto estimate grows at a factorial rate: so be cautious!

Multiple independent variables can be combined with multiple dependent variables, as infml = c(y1, y2) ~ cw(x1, x2, x3) which would lead to 6 estimations. Multipleestimations can also be combined to split samples (with the argumentssplit,fsplit).

You can also add fixed-effects in a stepwise fashion. Note that you cannot performstepwise estimations on the IV part of the formula (feols only).

If NAs are present in the sample, to avoid too many messages, only NA removalconcerning the variables common to all estimations is reported.

A note on performance. The feature of multiple estimations has been highly optimized forfeols, in particular in the presence of fixed-effects. It is faster to estimatemultiple models using the formula rather than with a loop. For non-feols models usingthe formula is roughly similar to using a loop performance-wise.

Tricks to estimate multiple LHS

To use multiple dependent variables infixest estimations, you need to include themin a vector: like inc(y1, y2, y3).

First, if names are stored in a vector, they can readily be inserted in a formula toperform multiple estimations using the dot square bracket operator. For instance ifmy_lhs = c("y1", "y2"), callingfixest with, sayfeols(.[my_lhs] ~ x1, etc) isequivalent to usingfeols(c(y1, y2) ~ x1, etc). Beware that this is a special featureunique to theleft-hand-side offixest estimations (the default behavior of the DSBoperator is to aggregate with sums, seexpd).

Second, you can use a regular expression to grep the left-hand-sides on the fly. When the..("regex") (reregex("regex")) feature is used naked on the LHS,the variables grepped are inserted intoc(). For example⁠..("Pe") ~ Sepal.Length, iris⁠ is equivalent to⁠c(Petal.Length, Petal.Width) ~ Sepal.Length, iris⁠. Beware that this is aspecial feature unique to theleft-hand-side offixest estimations(the default behavior of..("regex") is to aggregate with sums, seexpd).

Note that if the dependent variable is also on the right-hand-side, it is automaticallyremoved from the set of explanatory variable.For example, feols(y ~ y + x, base) works as feols(y ~ x, base).This is particulary useful to batch multiple estimations with multiple left hand sides.

Argument sliding

When the data set has been set up globally usingsetFixest_estimation(data = data_set), the argumentvcov can be used implicitly.This means that calls such asfeols(y ~ x, "HC1"), orfeols(y ~ x, ~id), are valid:i) the data is automatically deduced from the global settings, and ii) thevcovis deduced to be the second argument.

Piping

Although the argument 'data' is placed in second position, the data can be piped to theestimation functions. For example, with R >= 4.1,mtcars |> feols(mpg ~ cyl) works asfeols(mpg ~ cyl, mtcars).

Dot square bracket operator in formulas

In a formula, the dot square bracket (DSB) operator can: i) create manifold variablesat once, or ii) capture values from the current environment and put themverbatim in the formula.

Say you want to include the variablesx1 tox3 in your formula. You can usexpd(y ~ x.[1:3]) and you'll gety ~ x1 + x2 + x3.

To summon values from the environment, simply put the variable in square brackets.For example:for(i in 1:3) xpd(y.[i] ~ x) will create the formulasy1 ~ x toy3 ~ xdepending on the value ofi.

You can include a full variable from the environment in the same way:for(y in c("a", "b")) xpd(.[y] ~ x) will create the two formulasa ~ x andb ~ x.

The DSB can even be used within variable names, but then the variable must be nested incharacter form. For exampley ~ .["x.[1:2]_sq"] will createy ~ x1_sq + x2_sq. Using thecharacter form is important to avoid a formula parsing error.Double quotes must be used. Note that the character string that is nested willbe parsed with the functiondsb, and thus it will return a vector.

By default, the DSB operator expands vectors into sums. You can add a comma,like in.[, x],to expand with commas–the content can then be used within functions. For instance:c(x.[, 1:2]) will createc(x1, x2) (andnotc(x1 + x2)).

In allfixest estimations, this special parsing is enabled, so you don't need to usexpd.

One-sided formulas can be expanded with the DSB operator: letx = ~sepal + petal, thenxpd(y ~ .[x]) leads tocolor ~ sepal + petal.

You can even use multiple square brackets within a single variable,but then the use of nesting is required.For example, the followingxpd(y ~ .[".[letters[1:2]]_.[1:2]"]) will createy ~ a_1 + b_2. Remember that the nested character string is parsed withdsb,which explains this behavior.

When the element to be expanded i) is equal to the empty string or,ii) is of length 0, it is replaced with a neutral element, namely1.For example,⁠x = "" ; xpd(y ~ .[x])⁠ leads toy ~ 1.

Author(s)

Laurent Berge

References

Berge, Laurent, 2018, "Efficient estimation of maximum likelihood models withmultiple fixed-effects: the R package FENmlm." CREA Discussion Papers, 13 ().

For models with multiple fixed-effects:

Gaure, Simen, 2013, "OLS with multiple high dimensional category variables",Computational Statistics & Data Analysis 66 pp. 8–18

See Also

See alsosummary.fixest to see the results with the appropriate standard-errors,fixef.fixest to extract the fixed-effects coefficients, and the functionetableto visualize the results of multiple estimations. For plotting coefficients: seecoefplot.

And other estimation methods:femlm,feglm,fepois,fenegbin,feNmlm.

Examples

## Basic estimation#res = feols(Sepal.Length ~ Sepal.Width + Petal.Length, iris)# You can specify clustered standard-errors in summary:summary(res, cluster = ~Species)## Just one set of fixed-effects:#res = feols(Sepal.Length ~ Sepal.Width + Petal.Length | Species, iris)# Here we have "default" SEssummary(res)## Varying slopes:#res = feols(Sepal.Length ~ Petal.Length | Species[Sepal.Width], iris)summary(res)## Combining the FEs:#base = irisbase$fe_2 = rep(1:10, 15)res_comb = feols(Sepal.Length ~ Petal.Length | Species^fe_2, base)summary(res_comb)fixef(res_comb)[[1]]## Using leads/lags:#data(base_did)# We need to set up the panel with the arg. panel.idest1 = feols(y ~ l(x1, 0:1), base_did, panel.id = ~id+period)est2 = feols(f(y) ~ l(x1, -1:1), base_did, panel.id = ~id+period)etable(est1, est2, order = "f", drop = "Int")## Using interactions:#data(base_did)# We interact the variable 'period' with the variable 'treat'est_did = feols(y ~ x1 + i(period, treat, 5) | id + period, base_did)# Now we can plot the result of the interaction with coefplotcoefplot(est_did)# You have many more example in coefplot help## Instrumental variables## To estimate Two stage least squares,# insert a formula describing the endo. vars./instr. relation after a pipe:data(fulton)# Using exogenous control, 1 endogenous var. and 1 instrumentres_iv = feols(qty ~ t | price ~ speed2, fulton)# The second stage is the defaultsummary(res_iv)# To show the first stage:summary(res_iv, stage = 1)# To show both the first and second stages:summary(res_iv, stage = 1:2)# Adding a fixed-effect => IV formula always last!res_iv_fe = feols(qty ~ t | day | price ~ speed2, fulton)# With two instrumentsres_iv2 = feols(qty ~ t | day | price ~ speed2 + wave2, fulton)# Now there's two first stages => a fixest_multi object is returnedsum_res_iv2 = summary(res_iv2, stage = 1)# You can navigate through it by subsetting:sum_res_iv2[iv = 1]# The stage argument also works in etable:etable(res_iv, res_iv_fe, res_iv2, order = "endo")etable(res_iv, res_iv_fe, res_iv2, stage = 1:2, order = c("endo", "inst"),       group = list(control = "!endo|inst"))## Multiple estimations:## 6 estimationsest_mult = feols(c(Ozone, Solar.R) ~ Wind + Temp + csw0(Wind:Temp, Day), airquality)# We can display the results for the first lhs:etable(est_mult[lhs = 1])# And now the second (access can be made by name)etable(est_mult[lhs = "Solar.R"])# Now we focus on the two last right hand sides# (note that .N can be used to specify the last item)etable(est_mult[rhs = 2:.N])# Combining with splitest_split = feols(c(Ozone, Solar.R) ~ sw(poly(Wind, 2), poly(Temp, 2)),                  airquality, split = ~ Month)# You can display everything at once with the print methodest_split# Different way of displaying the results with "compact"summary(est_split, "compact")# You can still select which sample/LHS/RHS to displayest_split[sample = 1:2, lhs = 1, rhs = 1]## Split sample estimations#base = setNames(iris, c("y", "x1", "x2", "x3", "species"))est  = feols(y ~ x.[1:3], base, split = ~species)etable(est)# You can select specific values with the %keep% and %drop% operators# By default, partial matching is enabled. It should refer to a single variable.est  = feols(y ~ x.[1:3], base, split = ~species %keep% c("set", "vers"))etable(est)# You can supply regular expression by using an @ first.# regex can match several values.est  = feols(y ~ x.[1:3], base, split = ~species %keep% c("@set|vers"))etable(est)## Argument sliding## When the data set is set up globally, you can use the vcov argument implicitlybase = setNames(iris, c("y", "x1", "x2", "x3", "species"))no_sliding = feols(y ~ x1 + x2, base, ~species)# With slidingsetFixest_estimation(data = base)# ~species is implicitly deduced to be equal to 'vcov'sliding = feols(y ~ x1 + x2, ~species)etable(no_sliding, sliding)# Resetting the global optionssetFixest_estimation(data = NULL)## Formula expansions## By default, the features of the xpd function are enabled in# all fixest estimations# Here's a few examplesbase = setNames(iris, c("y", "x1", "x2", "x3", "species"))# dot square bracket operatorfeols(y ~ x.[1:3], base)# fetching variables via regular expressions: ..("regex")feols(y ~ ..("1|2"), base)# NOTA: it also works for multiple LHSmult1 = feols(x.[1:2] ~ y + species, base)mult2 = feols(..("y|3") ~ x.[1:2] + species, base)etable(mult1, mult2)# Use .[, stuff] to include variables in functions:feols(y ~ csw(x.[, 1:3]), base)# Same for ..(, "regex")feols(y ~ csw(..(,"x")), base)

Computes fit statistics of fixest objects

Description

Computes various fit statistics forfixest estimations.

Usage

fitstat(  x,  type,  vcov = NULL,  cluster = NULL,  ssc = NULL,  simplify = FALSE,  verbose = TRUE,  show_types = FALSE,  frame = parent.frame(),  ...)

Arguments

Value

By default an object of classfixest_fitstat is returned. Usingverbose = FALSEreturns a simple a list. Finally, if only one type is selected,simplify = TRUEleads to the selected type to be returned.

Registering your own types

You can register custom fit statistics with the functionfitstat_register.

Available types

The types are case sensitive, please use lower case only. The types available are:

n,ll,aic,bic,rmse:

The number of observations, the log-likelihood,the AIC, the BIC and the root mean squared error, respectively.

my:

Mean of the dependent variable.

g:

The degrees of freedom used to compute the t-test (it influences the p-valuesof the coefficients). When the VCOV is clustered, this value is equal to the minimumcluster size, otherwise, it is equal to the sample size minus the number of variables.

r2,ar2,wr2,awr2,pr2,apr2,wpr2,awpr2:

All r2 that can beobtained with the functionr2. Thea stands for 'adjusted', thew for 'within' andthep for 'pseudo'. Note that the order of the lettersa,w andp does not matter.The pseudo R2s are McFadden's R2s (ratios of log-likelihoods).

theta:

The over-dispersion parameter in Negative Binomial models. Low values meanhigh overdispersion.

f,wf:

The F-tests of nullity of the coefficients. Thew stands for'within'. These types return the following values:stat,p,df1 anddf2.If you want to display only one of these, use their name after a dot: e.g.f.statwill give the statistic of the F-test, orwf.p will give the p-values of the F-teston the projected model (i.e. projected onto the fixed-effects).

wald:

Wald test of joint nullity of the coefficients. This test always excludesthe intercept and the fixed-effects. These type returns the following values:stat,p,df1,df2 andvcov. The elementvcov reports the way the VCOVmatrix was computed since it directly influences this statistic.

ivf,ivf1,ivf2,ivfall:

These statistics are specific to IV estimations.They report either the IV F-test (namely the Cragg-Donald F statistic in the presenceof only one endogenous regressor) of the first stage (ivf orivf1), of thesecond stage (ivf2) or of both (ivfall). The F-test of the first stage iscommonly named weak instrument test. The value ofivfall is only useful inetablewhen both the 1st and 2nd stages are displayed (it leads to the 1st stage F-test(s)to be displayed on the 1st stage estimation(s), and the 2nd stage one on the2nd stage estimation – otherwise,ivf1 would also be displayed on the 2nd stageestimation). These types return the following values:stat,p,df1 anddf2.

ivwald,ivwald1,ivwald2,ivwaldall:

These statistics are specific to IVestimations. They report either the IV Wald-test of the first stage (ivwald orivwald1),of the second stage (ivwald2) or of both (ivwaldall). The Wald-test of the first stageis commonly named weak instrument test. Note that if the estimation was done with a robustVCOV and there is only one endogenous regressor, this is equivalent to theKleibergen-Paap statistic. The value ofivwaldall is only useful inetable when boththe 1st and 2nd stages are displayed (it leads to the 1st stage Wald-test(s) to be displayedon the 1st stage estimation(s), and the 2nd stage one on the 2nd stage estimation –otherwise,ivwald1 would also be displayed on the 2nd stage estimation). These typesreturn the following values:stat,p,df1,df2, andvcov.

cd:

The Cragg-Donald test for weak instruments.

kpr:

The Kleibergen-Paap test for weak instruments.

wh:

This statistic is specific to IV estimations. Wu-Hausman endogeneity test.H0 is the absence of endogeneity of the instrumented variables. It returns the followingvalues:stat,p,df1,df2.

sargan:

Sargan test of overidentifying restrictions. H0: the instruments arenot correlated with the second stage residuals. It returns thefollowing values:stat,p,df.

lr,wlr:

Likelihood ratio and within likelihood ratio tests. It returnsthe following elements:stat,p,df. Concerning the within-LR test, note that,contrary to estimations withfemlm orfeNmlm, estimations withfeglm/fepoisneed to estimate the model with fixed-effects only which may prove time-consuming(depending on your model). Bottom line, if you really need the within-LR and estimate aPoisson model, usefemlm instead offepois (the former uses direct ML maximization forwhich the only FEs model is a by product).