Package 'ggfixest'

Aggregates event-study treatment effects.

Description

Aggregates post- (and/or pre-) treatment effects of an "event-study" estimation, also known as a dynamic difference-in-differences (DDiD) model. The event-study should have been estimated using the fixest package, which provides a specialised i() operator for this class of models. By default, the function will return the average post-treatment effect (i.e. across multiple periods). However, it can also return the cumulative post-treatment effect and can be used to aggregate pre-treatment effects too. At its heart, aggr_es() is a convenience wrapper around marginaleffects::hypotheses(), which is used to perform the underlying joint hypothesis test.

Usage

aggr_es(
  object,
  rhs = 0,
  period = "post",
  aggregation = c("mean", "cumulative"),
  abbr_term = TRUE,
  ...
)

Arguments

object	A model object of class fixest, where the i() operator has been used to facilitate an "event-study" DiD design. See Examples.
rhs	Numeric. The null hypothesis value. Defaults to 0.
period	Keyword string or numeric sequence. Which group of periods are we aggregating? Can either be one of three convenience strings—i.e., "post" (the default), "prep", or "both"—or a numeric sequence that matches a subset of periods in the data (e.g. 6:8).
aggregation	Character string. The aggregation type. Either "mean" (the default) or "cumulative".
abbr_term	Logical. Should the leading "term" column of the return data frame be abbreviated? The default is TRUE. If FALSE, then the term column will retain the full hypothesis test string as per usual with marginaleffects(). Note that this information is retained as an attribute of the return object, regardless.
...	Additional arguments passed to marginaleffects::hypotheses().

Value

A "tidy" data frame of aggregated (pre and/or post) treatment effects, plus inferential information about standard errors, confidence intervals, etc. Potentially useful information about the underlying hypothesis test is also provided as an attribute. See Examples.

See Also

marginaleffects::hypotheses()

Examples

library(ggfixest) ## Will load fixest too

est = feols(y ~ x1 + i(period, treat, 5) | id + period, base_did)

# Default hypothesis test is a null mean post-treatment effect
(post_mean = aggr_es(est))
# The underlying hypothesis is saved as an attribute
attributes(post_mean)["hypothesis"]

# Other hypothesis and aggregation options
aggr_es(est, aggregation = "cumulative") # cumulative instead of mean effects
aggr_es(est, period = "pre")             # pre period instead of post
aggr_es(est, period = "both")            # pre & post periods separately
aggr_es(est, period = 6:8)               # specific subset of periods
aggr_es(est, rhs = -1, period = "pre")   # pre period with H0 value of 1
# Etc.

---

Draw coefficient plots and interaction plots from fixest regression objects.

Description

Draws the ggplot2 equivalents of fixest::coefplot and fixest::iplot. These "gg" versions do their best to recycle the same arguments and plotting logic as their original base counterparts. But they also support additional features via the ggplot2 API and infrastructure. The overall goal remains the same as the original functions. To wit: ggcoefplot plots the results of estimations (coefficients and confidence intervals). The function ggiplot restricts the output to variables created with i, either interactions with factors or raw factors.

Usage

ggcoefplot(
  object,
  geom_style = c("pointrange", "errorbar"),
  multi_style = c("dodge", "facet"),
  facet_args = NULL,
  theme = NULL,
  ...
)

ggiplot(
  object,
  geom_style = c("pointrange", "errorbar", "ribbon"),
  multi_style = c("dodge", "facet"),
  aggr_eff = NULL,
  aggr_eff.par = list(col = "grey50", lwd = 1, lty = 1),
  facet_args = NULL,
  theme = NULL,
  ...
)

Arguments

object	A model object of class fixest or fixest_multi, or a list thereof.
geom_style	Character string. One of c('pointrange', 'errorbar', 'ribbon') describing the preferred geometric representation of the coefficients. Note that ribbon plots not supported for ggcoefplot, since we cannot guarantee a continuous relationship among the coefficients.
multi_style	Character string. One of c('dodge', 'facet'), defining how multi-model objects should be presented.
facet_args	A list of arguments passed down to ggplot::fact_wrap(). E.g. facet_args = list(ncol = 2, scales = 'free_y'). Only used if multi_style = 'facet'.
theme	ggplot2 theme. Defaults to theme_linedraw() with some minor adjustments, such as centered plot title. Can also be defined on an existing ggiplot object to redefine theme elements. See examples.
...	Arguments passed down to, or equivalent to, the corresponding fixest::coefplot/fixest::iplot arguments. Note that some of these require list objects. Currently used are: keep and drop for subsetting variables using regular expressions. The fixest::iplot help page includes more detailed examples, but these should generally work as you expect. One useful regexp trick worth mentioning briefly for event studies with many pre-/post-periods is drop = "[[:digit:]]{2}". This will cause the plot to zoom in around single digit pre-/post-periods. group a list indicating variables to group over. Each element of the list reports the coefficients to be grouped while the name of the element is the group name. Each element of the list can be either: i) a character vector of length 1, ii) of length 2, or iii) a numeric vector. Special patterns such as "^^var_start" can be used to more appealing plotting, where group labels are separated from their subsidiary labels. This can be especially useful for plotting interaction terms. See the Details section of fixest::coefplot for more information. i.select Integer scalar, default is 1. In ggiplot, used to select which variable created with i() to select. Only used when there are several variables created with i. See the Details section of fixest::iplot for more information. main, xlab, and ylab for setting the plot title, x- and y-axis labels, respectively. zero and zero.par for defining or adjusting the zero line. For example, zero.par = list(col = 'orange'). ref.line and ref.line.par for defining or adjusting the vertical reference line. For example, ref.line.par = list(col = 'red', lty = 4). pt.pch and pt.join for overriding the default point estimate shapes and joining them, respectively. col for manually defining line, point, and ribbon colours. ci_level for changing the desired confidence level (default = 0.95). Note that multiple levels are allowed, e.g. ci_level = c(0.8, 0.95). ci.width for changing the width of the extremities of the confidence intervals. Only used if geom_style = "errorbar" (or if multiple CI levels are requested for the default pointrange style). The default value is 0.2. ci.fill.par for changing the confidence interval fill. Only used when geom_style = "ribbon" and currently only affects the alpha (transparency) channel. For example, we can make the CI band lighter with ci.fill.par = list(alpha = 0.2) (the default alpha is 0.3). dict a dictionary for overriding coefficient names. vcov, cluster or se as alternative options for adjusting the standard errors of the model object(s) on the fly. See summary.fixest for details. Written here in superseding order; cluster will only be considered if vcov is not null, etc.
aggr_eff	A keyword string or numeric sequence, indicating whether mean treatment effects for some subset of the model should be displayed as part of the plot. For example, the "post" keyword means that the mean post-treatment effect will be plotted alongside the individual period effects. Passed to aggr_es; see that function's documentation for other valid options.
aggr_eff.par	List. Parameters of the aggregated treatment effect line, if plotted. The default values are col = 'gray50', lwd = 1, lty = 1.

Details

These functions generally try to mimic the functionality and (where appropriate) arguments of fixest::coefplot and fixest::iplot as closely as possible. However, by leveraging the ggplot2 API and infrastructure, they are able to support some more complex plot arrangements out-of-the-box that would be more difficult to achieve using the base coefplot/iplot alternatives.

Value

A ggplot2 object.

Functions

• ggiplot(): This function plots the results of estimations (coefficients and confidence intervals). The function ggiplot restricts the output to variables created with i, either interactions with factors or raw factors.

See Also

fixest::coefplot(), fixest::iplot().

Examples

library(ggfixest)

##
# Author note: The examples that follow deliberately follow the original
#   examples from the coefplot/iplot help pages. A few "gg-" specific
#   features are sprinkled within, with the final set of examples in
#   particular highlighting unique features of this package.


#
# Example 1: Basic use and stacking two sets of results on the same graph
#

# Estimation on Iris data with one fixed-effect (Species)
est = feols(Petal.Length ~ Petal.Width + Sepal.Length + Sepal.Width | Species, iris)

ggcoefplot(est)

# Show multiple CIs
ggcoefplot(est, ci_level = c(0.8, 0.95))

# By default, fixest model standard errors are clustered by the first fixed
# effect (here: Species).
# But we can easily switch to "regular" standard-errors
est_std = summary(est, se = "iid")

# You can plot both results at once in the same plot frame...
ggcoefplot(list("Clustered" = est, "IID" = est_std))
# ... or as separate facets
ggcoefplot(list("Clustered" = est, "IID" = est_std), multi_style = "facet") +
	theme(legend.position = "none")


#
# Example 2: Interactions
#


# Now we estimate and plot the "yearly" treatment effects

data(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

# ggcoefplot will show all the coefficients:
ggcoefplot(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 ggiplot
ggiplot(est_did)

# We can see that the graph is different from before:
#  - only interactions are shown,
#  - the reference is present,
# => this is fully flexible

ggiplot(est_did, ci_level = c(0.8, 0.95))
ggiplot(est_did, ref.line = FALSE, pt.join = TRUE, geom_style = "errorbar")
ggiplot(est_did, geom_style = "ribbon", col = "orange")
# etc

# We can also use a dictionary to replace label values. The dicionary should
# take the form of a named vector or list, e.g. c("old_lab1" = "new_lab1", ...)

# Let's create a "month" variable
all_months = c("aug", "sept", "oct", "nov", "dec", "jan",
							 "feb", "mar", "apr", "may", "jun", "jul")
# Turn into a dictionary by providing the old names
# Note the implication that treatment occured here in December (5 month in our series)
dict = all_months; names(dict) = 1:12
# Pass our new dictionary to our ggiplot call
ggiplot(est_did, pt.join = TRUE, geom_style = "errorbar", dict = dict)

#
# What if the interacted variable is not numeric?

# let's re-use our all_months vector from the previous example, but add it
# directly to the dataset
base_inter$period_month = all_months[base_inter$period]

# The new estimation
est = feols(y ~ x1 + i(period_month, treat, "oct") | id+period, base_inter)
# Since 'period_month' of type character, iplot/coefplot both sort it
ggiplot(est)

# To respect a plotting order, use a factor
base_inter$month_factor = factor(base_inter$period_month, levels = all_months)
est     = feols(y ~ x1 + i(month_factor, treat, "oct") | id + period, base_inter)
ggiplot(est)

# dict -> c("old_name" = "new_name")
dict = all_months; names(dict) = 1:12; dict
ggiplot(est_did, dict = dict)

#
# Example 3: Setting defaults
#

# The customization logic of ggcoefplot/ggiplot works differently than the
# original base fixest counterparts, so we don't have "gg" equivalents of
# setFixest_coefplot and setFixest_iplot. However, you can still invoke some
# global fixest settings like setFixest_dict(). SImple example:

base_inter$letter = letters[base_inter$period]
est_letters = feols(y ~ x1 + i(letter, treat, 'e') | id+letter, base_inter)

# Set global dictionary for capitalising the letters
dict = LETTERS[1:10]; names(dict) = letters[1:10]
setFixest_dict(dict)

ggiplot(est_letters)

setFixest_dict() # reset

#
# 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 factors

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

# No grouping:
ggcoefplot(est)

# now we group by Sepal and Species
ggcoefplot(est, group = list(Sepal = "Sepal", Species = "Species"))

# now we group + clean the beginning of the names using the special character ^^
ggcoefplot(est, group = list(Sepal = "^^Sepal.", Species = "^^Species"))


#
# Example 5: Some more ggcoefplot/ggiplot extras
#

# We'll demonstrate using the staggered treatment example from the
# introductory fixest vignette.

data(base_stagg)
est_twfe = feols(
  y ~ x1 + i(time_to_treatment, treated, ref = c(-1, -1000)) | id + year,
  base_stagg
)
est_sa20 = feols(
  y ~ x1 + sunab(year_treated, year) | id + year,
  data = base_stagg
)

# Plot both regressions in a faceted plot
ggiplot(
  list('TWFE' = est_twfe, 'Sun & Abraham (2020)' = est_sa20),
  main = 'Staggered treatment', ref.line = -1, pt.join = TRUE
)

# So far that's no different than base iplot (automatic legend  aside). But an
# area where ggiplot shines is in complex multiple estimation cases, such as
# lists of fixest_multi objects. To illustrate, let's add a split variable
# (group) to our staggered dataset.
base_stagg_grp = base_stagg
base_stagg_grp$grp = ifelse(base_stagg_grp$id %% 2 == 0, 'Evens', 'Odds')

# Now re-run our two regressions from earlier, but splitting the sample to
# generate fixest_multi objects.
est_twfe_grp = feols(
  y ~ x1 + i(time_to_treatment, treated, ref = c(-1, -1000)) | id + year,
  data = base_stagg_grp, split = ~ grp
)
est_sa20_grp = feols(
  y ~ x1 + sunab(year_treated, year) | id + year,
  data = base_stagg_grp, split = ~ grp
)

# ggiplot combines the list of multi-estimation objects without a problem...
ggiplot(list('TWFE' = est_twfe_grp, 'Sun & Abraham (2020)' = est_sa20_grp),
				ref.line = -1, main = 'Staggered treatment: Split multi-sample')

# ... but is even better when we use facets instead of dodged errorbars.
# Let's use this an opportunity to construct a fancy plot that invokes some
# additional arguments and ggplot theming.
ggiplot(
  list('TWFE' = est_twfe_grp, 'Sun & Abraham (2020)' = est_sa20_grp),
  ref.line = -1,
  main = 'Staggered treatment: Split multi-sample',
  xlab = 'Time to treatment',
  multi_style = 'facet',
  geom_style = 'ribbon',
  facet_args = list(labeller = labeller(id = \(x) gsub(".*: ", "", x))),
  theme = theme_minimal() +
    theme(
      text = element_text(family = 'HersheySans'),
      plot.title = element_text(hjust = 0.5),
      legend.position = 'none'
    )
)

#
# Aside on theming and scale adjustments
#

# Setting the theme inside the `ggiplot()` call is optional and not strictly
# necessary, since the ggplot2 API allows programmatic updating of existing
# plots. E.g.
last_plot() +
	labs(caption = 'Note: Super fancy plot brought to you by ggiplot')
last_plot() +
	theme_grey() +
	theme(legend.position = 'none') +
	scale_fill_brewer(palette = 'Set1', aesthetics = c("colour", "fill"))
# etc.

---

Internal function for grabbing and preparing iplot data.

Description

Grabs the underlying data used to construct fixest::iplot, with some added functionality and tweaks for the ggiplot equivalents.

Usage

iplot_data(
  object,
  .ci_level = 0.95,
  .keep = NULL,
  .drop = NULL,
  .dict = fixest::getFixest_dict(),
  .internal.only.i = TRUE,
  .i.select = 1,
  .aggr_es = NULL,
  .group = "auto",
  .vcov = NULL,
  .cluster = NULL,
  .se = NULL
)

coefplot_data(
  object,
  .ci_level = 0.95,
  .keep = NULL,
  .drop = NULL,
  .group = "auto",
  .dict = fixest::getFixest_dict(),
  .internal.only.i = FALSE,
  .i.select = 1,
  .aggr_es = "none",
  .vcov = NULL,
  .cluster = NULL,
  .se = NULL
)

Arguments

object	A model object of class fixest or fixest_multi, where the i() operator has been used to construct an interaction, or set of interactions.
.ci_level	A number between 0 and 1 indicating the desired confidence level, Defaults to 0.95.
.keep	Character vector used to subset the coefficients of interest. Passed down to fixest::iplot(..., keep = .keep) and should take the form of an acceptable regular expression.
.drop	Character vector used to subset the coefficients of interest (complement of .keep). Passed down to fixest::iplot(..., drop = .drop) and should take the form of an acceptable regular expression.
.dict	A dictionary (i.e. named character vector or a logical scalar). Used for changing coefficient names. Defaults to the values in getFixest_dict(). See the ?fixest::coefplot documentation for more information. Note: This argument applies dictionary changes directly to the return object for coefplot_data. However, it is ignored for iplot_data, since we want to preserve the numeric ordering for potential event study plots. (And imposing an ordered factor would create its own downstream problems in the case of continuous plot features like ribbons.) Instead, any dictionary replacement for ggiplot is deferred to the actual plot call and applied directly to the labels.
.internal.only.i	Logical variable used for some internal function handling when passing on to coefplot/iplot.
.i.select	Integer scalar, default is 1. In (gg)iplot, used to select which variable created with i() to select. Only used when there are several variables created with i. This is an index, just try increasing numbers to hopefully obtain what you want. Passed down to fixest::iplot(..., i.select = .i.select)
.aggr_es	A keyword string or numeric sequence indicating whether the aggregated mean treatment effects for some subset of the model should be added as a column to the returned data frame. Passed to aggr_es(..., aggregation = "mean").
.group	A list, default is missing. Each element of the list reports the coefficients to be grouped while the name of the element is the group name. Passed down to fixest::coefplot(..., group = .group). Example of valid uses: group=list(group_name="pattern") group=list(group_name=c("var_start", "var_end")) group=list(group_name=1:2) See the Details section of ?fixest::coefplot for more.
.vcov, .cluster, .se	Alternative options for adjusting the standard errors of the model object on the fly. See summary.fixest for details (although note that the "." period prefix should be ignored in the latter's argument documentation). Written here in superseding order; .cluster will only be considered if .vcov is not null, etc.

Details

This function is a wrapper around fixest::iplot(..., only.params = TRUE), but with various checks and tweaks to better facilitate plotting with ggplot2 and handling of complex object types (e.g. lists of fixest_multi models)

Value

A data frame consisting of estimate values, confidence intervals, relative x-axis positions, and other aesthetic information needed to draw a ggplot2 object.

Functions

• coefplot_data(): Internal function for grabbing and preparing coefplot data

See Also

fixest::iplot(), aggr_es().

Examples

library(fixest)

est_did = feols(y ~ x1 + i(period, treat, 5) | id+period,
                data = base_did)
iplot(est_did, only.params = TRUE) # The "base" version
iplot_data(est_did)                # The wrapper provided by this package

# Illustrative fixest_multi case, where the sample has been split by odd and
# even ID numbers.
est_split = feols(y ~ x1 + i(period, treat, 5) | id+period,
                  data = base_did, split = ~id%%2)
iplot(est_split, only.params = TRUE) # The "base" version
iplot_data(est_split)                # The wrapper provided by this package

---