femlm: Fixed-effects maximum likelihood models

In fixest: Fast Fixed-Effects Estimations

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",
  offset,
  subset,
  split,
  fsplit,
  split.keep,
  split.drop,
  cluster,
  se,
  ssc,
  panel.id,
  fixef.tol = 1e-05,
  fixef.iter = 10000,
  nthreads = getFixest_nthreads(),
  lean = FALSE,
  verbose = 0,
  warn = TRUE,
  notes = getFixest_notes(),
  theta.init,
  combine.quick,
  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",
  offset,
  subset,
  split,
  fsplit,
  split.keep,
  split.drop,
  cluster,
  se,
  ssc,
  panel.id,
  fixef.tol = 1e-05,
  fixef.iter = 10000,
  nthreads = getFixest_nthreads(),
  lean = FALSE,
  verbose = 0,
  warn = TRUE,
  notes = getFixest_notes(),
  combine.quick,
  mem.clean = FALSE,
  only.env = FALSE,
  only.coef = FALSE,
  data.save = FALSE,
  env,
  ...
)

Arguments

fml	A formula representing the relation to be estimated. For example: fml = z~x+y. To include fixed-effects, insert them in this formula using a pipe: e.g. fml = z~x+y|fixef_1+fixef_2. Multiple estimations can be performed at once: for multiple dep. vars, wrap them in c(): ex c(y1, y2). For multiple indep. vars, use the stepwise functions: ex x1 + csw(x2, x3). The formula fml = c(y1, y2) ~ x1 + cw0(x2, x3) leads to 6 estimation, see details. Square brackets starting with a dot can be used to call global variables: y.[i] ~ x.[1:2] will lead to y3 ~ x1 + x2 if i is equal to 3 in the current environment (see details in xpd).
data	A data.frame containing the necessary variables to run the model. The variables of the non-linear right hand side of the formula are identified with this data.frame names. Can also be a matrix.
family	Character scalar. It should provide the family. The possible values are "poisson" (Poisson model with log-link, the default), "negbin" (Negative Binomial model with log-link), "logit" (LOGIT model with log-link), "gaussian" (Gaussian model).
vcov	Versatile argument to specify the VCOV. In general, it is either a character scalar equal to a VCOV type, either a formula of the form: vcov_type ~ variables. The VCOV types implemented are: "iid", "hetero" (or "HC1"), "cluster", "twoway", "NW" (or "newey_west"), "DK" (or "driscoll_kraay"), and "conley". It also accepts object from vcov_cluster, vcov_NW, NW, vcov_DK, DK, vcov_conley and conley. It also accepts covariance matrices computed externally. Finally it accepts functions to compute the covariances. See the vcov documentation in the vignette.
start	Starting values for the coefficients. Can be: i) a numeric of length 1 (e.g. start = 0, the default), ii) a numeric vector of the exact same length as the number of variables, or iii) a named vector of any length (the names will be used to initialize the appropriate coefficients).
fixef	Character vector. The names of variables to be used as fixed-effects. These variables should contain the identifier of each observation (e.g., think of it as a panel identifier). Note that the recommended way to include fixed-effects is to insert them directly in the formula.
fixef.rm	Can be equal to "perfect" (default), "singleton", "both" or "none". Controls which observations are to be removed. If "perfect", then observations having a fixed-effect with perfect fit (e.g. only 0 outcomes in Poisson estimations) will be removed. If "singleton", all observations for which a fixed-effect appears only once will be removed. Note, importantly, that singletons are removed in just one pass, there is no recursivity implemented. The meaning of "both" and "none" is direct.
offset	A formula or a numeric vector. An offset can be added to the estimation. If equal to a formula, it should be of the form (for example) ~0.5*x**2. This offset is linearly added to the elements of the main formula 'fml'.
subset	A vector (logical or numeric) or a one-sided formula. If provided, then the estimation will be performed only on the observations defined by this argument.
split	A one sided formula representing a variable (eg split = ~var) or a vector. If provided, the sample is split according to the variable and one estimation is performed for each value of that variable. If you also want to include the estimation for the full sample, use the argument fsplit instead. You can use the special operators ⁠%keep%⁠ and ⁠%drop%⁠ to select only a subset of values for which to split the sample. E.g. split = ~var %keep% c("v1", "v2") will split the sample only according to the values v1 and v2 of the variable var; it is equivalent to supplying the argument split.keep = c("v1", "v2"). By default there is partial matching on each value, you can trigger a regular expression evaluation by adding a '@' first, as in: ~var %drop% "@^v[12]" which will drop values starting with "v1" or "v2" (of course you need to know regexes!).
fsplit	A one sided formula representing a variable (eg fsplit = ~var) or a vector. If provided, the sample is split according to the variable and one estimation is performed for each value of that variable. This argument is the same as split but also includes the full sample as the first estimation. You can use the special operators ⁠%keep%⁠ and ⁠%drop%⁠ to select only a subset of values for which to split the sample. E.g. fsplit = ~var %keep% c("v1", "v2") will split the sample only according to the values v1 and v2 of the variable var; it is equivalent to supplying the argument split.keep = c("v1", "v2"). By default there is partial matching on each value, you can trigger a regular expression evaluation by adding an '@' first, as in: ~var %drop% "@^v[12]" which will drop values starting with "v1" or "v2" (of course you need to know regexes!).
split.keep	A character vector. Only used when split, or fsplit, is supplied. If provided, then the sample will be split only on the values of split.keep. The values in split.keep will be partially matched to the values of split. To enable regular expressions, you need to add an '@' first. For example split.keep = c("v1", "@other|var") will keep only the value in split partially matched by "v1" or the values containing "other" or "var".
split.drop	A character vector. Only used when split, or fsplit, is supplied. If provided, then the sample will be split only on the values that are not in split.drop. The values in split.drop will be partially matched to the values of split. To enable regular expressions, you need to add an '@' first. For example split.drop = c("v1", "@other|var") will drop only the value in split partially matched by "v1" or the values containing "other" or "var".
cluster	Tells how to cluster the standard-errors (if clustering is requested). Can be either a list of vectors, a character vector of variable names, a formula or an integer vector. Assume we want to perform 2-way clustering over var1 and var2 contained in the data.frame base used for the estimation. All the following cluster arguments are valid and do the same thing: cluster = base[, c("var1", "var2")], cluster = c("var1", "var2"), cluster = ~var1+var2. If the two variables were used as fixed-effects in the estimation, you can leave it blank with vcov = "twoway" (assuming var1 [resp. var2] was the 1st [resp. 2nd] fixed-effect). You can interact two variables using ^ with the following syntax: cluster = ~var1^var2 or cluster = "var1^var2".
se	Character scalar. Which kind of standard error should be computed: “standard”, “hetero”, “cluster”, “twoway”, “threeway” or “fourway”? By default if there are clusters in the estimation: se = "cluster", otherwise se = "iid". Note that this argument is deprecated, you should use vcov instead.
ssc	An object of class ssc.type obtained with the function ssc. Represents how the degree of freedom correction should be done.You must use the function ssc for this argument. The arguments and defaults of the function ssc are: adj = TRUE, fixef.K="nested", cluster.adj = TRUE, cluster.df = "min", t.df = "min", ⁠fixef.force_exact=FALSE)⁠. See the help of the function ssc for details.
panel.id	The panel identifiers. Can either be: i) a one sided formula (e.g. panel.id = ~id+time), ii) a character vector of length 2 (e.g. panel.id=c('id', 'time'), or iii) a character scalar of two variables separated by a comma (e.g. panel.id='id,time'). Note that you can combine variables with ^ only inside formulas (see the dedicated section in feols).
fixef.tol	Precision used to obtain the fixed-effects. Defaults to 1e-5. It corresponds to the maximum absolute difference allowed between two coefficients of successive iterations. Argument fixef.tol cannot be lower than 10000*.Machine$double.eps. Note that this parameter is dynamically controlled by the algorithm.
fixef.iter	Maximum number of iterations in fixed-effects algorithm (only in use for 2+ fixed-effects). Default is 10000.
nthreads	The number of threads. Can be: a) an integer lower than, or equal to, the maximum number of threads; b) 0: meaning all available threads will be used; c) a number strictly between 0 and 1 which represents the fraction of all threads to use. The default is to use 50% of all threads. You can set permanently the number of threads used within this package using the function setFixest_nthreads.
lean	Logical, default is FALSE. If TRUE then all large objects are removed from the returned result: this will save memory but will block the possibility to use many methods. It is recommended to use the arguments se or cluster to obtain the appropriate standard-errors at estimation time, since obtaining different SEs won't be possible afterwards.
verbose	Integer, default is 0. It represents the level of information that should be reported during the optimisation process. If verbose=0: nothing is reported. If verbose=1: the value of the coefficients and the likelihood are reported. If verbose=2: 1 + information on the computing time of the null model, the fixed-effects coefficients and the hessian are reported.
warn	Logical, default is TRUE. Whether warnings should be displayed (concerns warnings relating to convergence state).
notes	Logical. By default, two notes are displayed: when NAs are removed (to show additional information) and when some observations are removed because of only 0 (or 0/1) outcomes in a fixed-effect setup (in Poisson/Neg. Bin./Logit models). To avoid displaying these messages, you can set notes = FALSE. You can remove these messages permanently by using setFixest_notes(FALSE).
theta.init	Positive numeric scalar. The starting value of the dispersion parameter if family="negbin". By default, the algorithm uses as a starting value the theta obtained from the model with only the intercept.
combine.quick	Logical. When you combine different variables to transform them into a single fixed-effects you can do e.g. y ~ x | paste(var1, var2). The algorithm provides a shorthand to do the same operation: y ~ x | var1^var2. Because pasting variables is a costly operation, the internal algorithm may use a numerical trick to hasten the process. The cost of doing so is that you lose the labels. If you are interested in getting the value of the fixed-effects coefficients after the estimation, you should use combine.quick = FALSE. By default it is equal to FALSE if the number of observations is lower than 50,000, and to TRUE otherwise.
mem.clean	Logical, default is FALSE. Only to be used if the data set is large compared to the available RAM. If TRUE then intermediary objects are removed as much as possible and gc is run before each substantial C++ section in the internal code to avoid memory issues.
only.env	(Advanced users.) Logical, default is FALSE. If TRUE, then only the environment used to make the estimation is returned.
only.coef	Logical, default is FALSE. If TRUE, then only the estimated coefficients are returned. Note that the length of the vector returned is always the length of the number of coefficients to be estimated: this means that the variables found to be collinear are returned with an NA value.
data.save	Logical scalar, default is FALSE. If TRUE, the data used for the estimation is saved within the returned object. Hence later calls to predict(), vcov(), etc..., will be consistent even if the original data has been modified in the meantime. This is especially useful for estimations within loops, where the data changes at each iteration, such that postprocessing can be done outside the loop without issue.
env	(Advanced users.) A fixest environment created by a fixest estimation with only.env = TRUE. Default is missing. If provided, the data from this environment will be used to perform the estimation.
...	Not currently used.

Details

Note that the functions feglm and femlm provide the same results when using the same families but differ in that the latter is a direct maximum likelihood optimization (so the two can really have different convergence rates).

Value

A fixest object. Note that fixest objects contain many elements and most of them 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 instance fitstat to access any fit statistic).

nobs	The number of observations.
fml	The linear formula of the call.
call	The call of the function.
method	The method used to estimate the model.
family	The family used to estimate the model.
data	The original data set used when calling the function. Only available when the estimation was called with data.save = TRUE
fml_all	A list containing different parts of the formula. Always contain the linear formula. Then, if relevant: fixef: the fixed-effects; NL: the non linear part of the formula.
nparams	The number of parameters of the model.
fixef_vars	The names of each fixed-effect dimension.
fixef_id	The list (of length the number of fixed-effects) of the fixed-effects identifiers for each observation.
fixef_sizes	The size of each fixed-effect (i.e. the number of unique identifier for each fixed-effect dimension).
convStatus	Logical, convergence status.
message	The convergence message from the optimization procedures.
obs_selection	(When relevant.) List containing vectors of integers. It represents the sequential selection of observation vis a vis the original data set.
fixef_removed	(When relevant.) In the case there were fixed-effects and some observations were removed because of only 0/1 outcome within a fixed-effect, it gives the list (for each fixed-effect dimension) of the fixed-effect identifiers that were removed.
coefficients	The named vector of estimated coefficients.
coeftable	The table of the coefficients with their standard errors, z-values and p-values.
loglik	The log-likelihood.
iterations	Number of iterations of the algorithm.
ll_null	Log-likelihood of the null model (i.e. with the intercept only).
ll_fe_only	Log-likelihood of the model with only the fixed-effects.
ssr_null	Sum of the squared residuals of the null model (containing only with the intercept).
pseudo_r2	The adjusted pseudo R2.
fitted.values	The fitted values are the expected value of the dependent variable for the fitted model: that is E(Y|X).
residuals	The residuals (y minus the fitted values).
sq.cor	Squared correlation between the dependent variable and the expected predictor (i.e. fitted.values) obtained by the estimation.
hessian	The Hessian of the parameters.
cov.iid	The variance-covariance matrix of the parameters.
se	The standard-error of the parameters.
scores	The matrix of the scores (first derivative for each observation).
residuals	The difference between the dependent variable and the expected predictor.
sumFE	The sum of the fixed-effects coefficients for each observation.
offset	(When relevant.) The offset formula.
weights	(When relevant.) The weights formula.

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 combination of the two variables fe_1 and fe_2. This is identical to doing paste0(fe_1, "_", fe_2) but more convenient.

Note that pasting is a costly operation, especially for large data sets. Thus, the internal algorithm uses a numerical trick which is fast, but the drawback is that the identity of each observation is lost (i.e. they are now equal to a meaningless number instead of being equal to paste0(fe_1, "_", fe_2)). These “identities” are useful only if you're interested in the value of the fixed-effects (that you can extract with fixef.fixest). If you're only interested in coefficients of the variables, it doesn't matter. Anyway, you can use combine.quick = FALSE to tell the internal algorithm to use paste instead of the numerical trick. By default, the numerical trick is performed only for large data sets.

Lagging variables

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

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

Interactions

You can interact a numeric variable with a "factor-like" variable by using i(factor_var, continuous_var, ref), where continuous_var will be interacted with each value of factor_var and the argument ref is a value of factor_var taken as a reference (optional).

Using this specific way to create interactions leads to a different display of the interacted values in etable. See examples.

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

The function i 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 arguments se and ssc in summary.fixest to define how to compute them. By default, in the presence of fixed-effects, standard-errors are automatically clustered.

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

You can use the functions setFixest_vcov and setFixest_ssc to permanently 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 a fixest_multi object which is ‘kind of’ a list of all 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 in c() (list() also works). For instance fml = c(y1, y2) ~ x1 would estimate the model fml = y1 ~ x1 and then the model fml = y2 ~ x1.

To include multiple independent variables, you need to use the stepwise functions. There are 4 stepwise functions: sw, sw0, csw, csw0, and mvsw. Of course sw stands for stepwise, and csw for cumulative stepwise. Finally mvsw 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 function sw will estimate the following two models: y ~ x1 + x2 and y ~ x1 + x3. That is, each element in sw() is sequentially, and separately, added to the formula. Would have you used sw0 in lieu of sw, then the model y ~ x1 would also have been estimated. The 0 in the name means that the model without any stepwise element also needs to be estimated. The prefix c means cumulative: each stepwise element is added to the next. That is, fml = y ~ x1 + csw(x2, x3) would lead to the following models y ~ x1 + x2 and y ~ x1 + x2 + x3. The 0 has the same meaning and would also lead to the model without the 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 and y ~ x1 + x2 + x3. Finally mvsw will add, in a stepwise fashion all possible combinations of the variables in its arguments. For example mvsw(x1, x2, x3) is equivalent to sw0(x1, x2, x3, x1 + x2, x1 + x3, x2 + x3, x1 + x2 + x3). The number of models to estimate grows at a factorial rate: so be cautious!

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

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

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

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

Argument sliding

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

Piping

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

Tricks to estimate multiple LHS

To use multiple dependent variables in fixest estimations, you need to include them in a vector: like in c(y1, y2, y3).

First, if names are stored in a vector, they can readily be inserted in a formula to perform multiple estimations using the dot square bracket operator. For instance if my_lhs = c("y1", "y2"), calling fixest with, say feols(.[my_lhs] ~ x1, etc) is equivalent to using feols(c(y1, y2) ~ x1, etc). Beware that this is a special feature unique to the left-hand-side of fixest estimations (the default behavior of the DSB operator is to aggregate with sums, see xpd).

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

Dot square bracket operator in formulas

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

Say you want to include the variables x1 to x3 in your formula. You can use xpd(y ~ x.[1:3]) and you'll get y ~ 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 formulas y1 ~ x to y3 ~ x depending on the value of i.

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 formulas a ~ x and b ~ x.

The DSB can even be used within variable names, but then the variable must be nested in character form. For example y ~ .["x.[1:2]_sq"] will create y ~ x1_sq + x2_sq. Using the character form is important to avoid a formula parsing error. Double quotes must be used. Note that the character string that is nested will be parsed with the function dsb, 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 create c(x1, x2) (and not c(x1 + x2)).

In all fixest estimations, this special parsing is enabled, so you don't need to use xpd.

One-sided formulas can be expanded with the DSB operator: let x = ~sepal + petal, then xpd(y ~ .[x]) leads to color ~ sepal + petal.

You can even use multiple square brackets within a single variable, but then the use of nesting is required. For example, the following xpd(y ~ .[".[letters[1:2]]_.[1:2]"]) will create y ~ a_1 + b_2. Remember that the nested character string is parsed with dsb, 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, namely 1. For example, ⁠x = "" ; xpd(y ~ .[x])⁠ leads to y ~ 1.

Author(s)

Laurent Berge

References

Berge, Laurent, 2018, "Efficient estimation of maximum likelihood models with multiple 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 Negative Binomial Regression Models", Sociological Methodology 32(1) pp. 247–265

See Also

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

Examples

# Load trade data
data(trade)

# We estimate the effect of distance on trade => we account for 3 fixed-effects
# 1) Poisson estimation
est_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 etable
etable(est_pois, est_gaus)
# Now using two way clustered standard-errors
etable(est_pois, est_gaus, se = "twoway")

# Comparing different types of standard errors
sum_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 estimations
est_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 split
est_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 method
est_split

# Different way of displaying the results with "compact"
summary(est_split, "compact")

# You can still select which sample/LHS/RHS to display
est_split[sample = 1:2, lhs = 1, rhs = 1]

fixest documentation built on June 22, 2024, 9:12 a.m.