s3alfisc / summclust

R module for cluster specific information (as in the Stata summclust module)

summclust

{summclust} is an R module for cluster level measures of leverage and influence, and further implements CRV3 and CRV3J cluster robust variance estimators.

For an introduction to the package, take a look at its vignette.

For a quick overview of different CRV estimators, take a look at the cluster robust variance estimation vignette.

For a very detailed description of the implemented methods, in particular a discussion of the different leverage and influence metrics, see:

MacKinnon, J.G., Nielsen, M.Ø., Webb, M.D., 2022. Leverage, influence, and the jackknife in clustered regression models: Reliable inference using summclust. QED Working Paper 1483. Queen’s University.

For the Stata version of the package, see here.

Installation

You can install the development version of summclust from CRAN, GitHub and r-universe with:

# install from CRAN
install.packages('summclust')

# from r-universe (windows & mac, compiled R > 4.0 required)
install.packages('summclust', repos ='https://s3alfisc.r-universe.dev')

# install.packages("devtools")
devtools::install_github("s3alfisc/summclust")

summclust's Issues

fixest: sparse matrix support

For very high-dimensional fixed effects problems, using sparse matrices should make the MNW jackknife estimator computationally feasible (based on a suggestion by @kylebutts).

replace `coeftable()` method by `tidy()` method

For easier interplay with modelsummary().

summclust() overrides type = "CRV3J"

If type = "CRV3J" is specified via summclust(), the function still returns a CRV3 variance covariance matrix. No downstream errors, but misleading output.

Separate computation of cluster leverages and vcov matrix

Reason:

I will use summclust to compute CRV3 vcov's for use in fwildclusterboot, and computing leverage statistics for many clusters k can be (relatively) slow. See here.

add center option to autoplot

Add option to center beta_kj's around beta for all covariances k, to guarantee beta_kj's are all on the same scale.

Pinv() via RcppEigen

• as it is much faster

Support Multiway Clustering

Support for models with unit specific time trends?

Hi, thanks for a great package!

I am trying to calculate CR3 standard errors for a model that includes unit specific time trends. Do you have any advice on how / if this can be implemented in summclust?

Here's a minimal example, showing the error that I get when I try to use the summclust() function on a model that includes unit specific time trends.

library(summclust)
library(fixest)
df <- fixest::trade

fit  <- feols(Euros ~ dist_km | Year + Origin + Origin[Year], data = df)
summ <- summclust(fit, params = ~dist_km, cluster = ~Origin)

The error is pasted below:

Error in `[[.default`(Origin, Year) : 
  attempt to select more than one element in vectorIndex

Thanks so much for your time and your help

fix fixed effects

Only fixed effects that are nested within clusters are allowed to be projected out in the bootstrap. For numerical stability, it is further recommended to project these fixed effects out. Therefore

• add an option to summclust.lm and summclust.fixest, e.g. absorb_cluster_fe and set it to TRUE by default.
• for regression objects that project out multiple fixed effects that are non-nested, make sure to properly add them as dummies to the design matrix X (see fwildclusterboot code)
• potentially, experiment with out-projection of nested fe's in every jackknife iteration

move all loops over the number of clusters G and covariates k to c++

Purpose: decrease runtime

Add extractor methods

For

• - leverage
• - partial leverage
• - p-values
• - t-statistics
• - confidence intervals
• - jackknife beta's

add measures

Add measures of

• leverage
• partial leverage
• coefficient of variation
• $\hat{\beta}_{j}^G}$
• a_harm, a_geo, a_quad
• effective number of clusters as in Carter et al

As described in MacKinnon, Nielsen & Webb (2022).

Add warning for singular matrices as summclust.ado

See here.

Add Support for Effective Number of Clusters following CSS (2017)

• Carter, Schnepel and Steigerwald (2017)
• see the discussion on numerical challenges in MNW (page 17)

Adjust to CRAN submission comments

• do not start the description with "This package", package name,

• The Description field is intended to be a (one paragraph) description of what the package does and why it may be useful.
  Please add more details about the package functionality and implemented methods in your Description text.

• Please add \value to .Rd files regarding exported methods and explain the functions results in the documentation. Please write about the structure of the output (class) and also what the output means. (If a function does not return a value, please document that too, e.g. \value{No return value, called for side effects} or similar)

  • coeftable.Rd: \value
  • coeftable.summclust.Rd: \value
  • plot.summclust.Rd: \value
  • summary.summclust.Rd: \value
  • summclust.fixest.Rd: \value
  • summclust.lm.Rd: \value
  • summclust.Rd: \value
  • vcov_CR3J.fixest.Rd: \value
  • vcov_CR3J.lm.Rd: \value
  • vcov_CR3J.Rd: \value

• \dontrun{} should only be used if the example really cannot be executed (e.g. because of missing additional software, missing API keys, ...) by the user. That's why wrapping examples in \dontrun{} adds the comment ("# Not run:") as a warning for the user.
  Does not seem necessary. Please replace \dontrun with \donttest.

• Please unwrap the examples if they are executable in < 5 sec, or replace \dontrun{} with \donttest{}.

• Please wrap examples that need packages in ‘Suggests’ in if(requireNamespace("pkgname")){} instead.

• Finally, resubmit.

Additional comments from my side:

• improve docs
• add @nord tags to internal functions

pkgcheck results - main

Checks for summclust (v0.2)

git hash: f66d35ff

• ✔️ Package name is available
• ✖️ does not have a 'codemeta.json' file.
• ✔️ has a 'contributing' file.
• ✔️ uses 'roxygen2'.
• ✔️ 'DESCRIPTION' has a URL field.
• ✔️ 'DESCRIPTION' has a BugReports field.
• ✖️ Package has no HTML vignettes
• ✖️ These functions do not have examples: [coeftable, coeftable.summclust, model_matrix.fixest, model_matrix.lm, model_matrix, plot.summclust, summary.summclust, summclust].
• ✖️ Function names are duplicated in other packages
• ✔️ Package has continuous integration checks.
• ✖️ Package coverage is 67.8% (should be at least 75%).
• ✔️ R CMD check found no errors.
• ✔️ R CMD check found no warnings.

Important: All failing checks above must be addressed prior to proceeding

Package License: MIT + file LICENSE

Support WLS

Support the summclust / the cluster jackknife for WLS.

Bug in tidy.summclust

Change the following:

From

  se <- diag(sqrt(vcov[param_, param_, drop = FALSE]))

to

  se <- sqrt(diag(vcov[param_, param_, drop = FALSE]))

in order to avoid warning message `Warning message:

In sqrt(vcov[param_, param_, drop = FALSE]) : NaNs produced.

Note that this error has no consequences on the displayed results.

CRAN fix

Got an email from CRAN.

"
Please see the problems shown on
https://cran.r-project.org/web/checks/check_results_summclust.html.

Please correct before 2023-08-09 to safely retain your package on CRAN.
"

So I'll do that =)

Allow to absorb a fixed effect nested within a cluster