Skip to content

Commit

Permalink
Remove table, add more written comparisons
Browse files Browse the repository at this point in the history
  • Loading branch information
@jacob-long
jacob-long committed Aug 9, 2024
1 parent 32eb6f6 commit 7b1e203
Show file tree
Hide file tree
Showing 2 changed files with 55 additions and 18 deletions.
22 changes: 22 additions & 0 deletions paper.bib
View file
Original file line number Diff line number Diff line change
Expand Up @@ -172,4 +172,26 @@ @article{kastellec2007
annotation = {00199}
}

@Article{ggeffects,
title = {ggeffects: Tidy Data Frames of Marginal Effects from Regression Models.},
volume = {3},
doi = {10.21105/joss.00772},
number = {26},
journal = {Journal of Open Source Software},
author = {Daniel Lüdecke},
year = {2018},
pages = {772},
}

@Article{gtsummary,
author = {Daniel D. Sjoberg and Karissa Whiting and Michael Curry and Jessica A. Lavery and Joseph Larmarange},
title = {Reproducible Summary Tables with the gtsummary Package},
journal = {{The R Journal}},
year = {2021},
url = {https://doi.org/10.32614/RJ-2021-053},
doi = {10.32614/RJ-2021-053},
volume = {13},
issue = {1},
pages = {570-580},
}

51 changes: 33 additions & 18 deletions paper.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -79,13 +79,41 @@ is using third-party packages for computation and simply repackaging the results
comparable functionality to `summ()` and has some advantages, such as a
greater range
of supported models and more support for exporting to external documents.
`marginaleffects` [@marginaleffects], `sjPlot` [@sjPlot], and `see` [@see]
offer support for plotting predicted
`gtsummary` [@gtsummary] also has extensive functionality for exporting to
external documents, although its ability to include and calculate customized
statistics is more limited. These may be preferred for those whose sole goal is
to quickly export straightforward regression summaries to LaTeX or other
documents. Neither `modelsummary` nor `gtsummary` are designed for interactive (console) use as `jtools` is.

`marginaleffects` [@marginaleffects], `sjPlot` [@sjPlot], `see` [@see],
and `ggeffects` [@ggeffects] offer support for plotting predicted
values from fitted regression models and again have some of their own
advantages, such as more supported model types. There is also overlap in
functionality with the `car` package [@car], like the computation of variance
advantages, such as more supported model types. Each of these offer overlapping
functionality with emphasis on `ggplot2` graphical output. Of these, only
`ggeffects` offers the same partial residual functionality that `jtools` does.
A design difference for `jtools` in comparison to others (to varying degrees)
is that `jtools` tends to perform many steps in a single function call, with the
user making choices via function arguments. Many of these other packages tend to
have multi-step usage patterns, in which the user creates an object with one
function which is then passed to another function, and so on. For
instance, `ggeffects` recommends that users fit a model, then pass the model to
generate marginal effect estimates to another function, pass the outputted
marginal effects to a hypothesis testing function, then passing the output of
that function to a `plot()` S3 method (although it is possible to skip one or
more steps). This type of interface can provide
for more flexibility for the user as well as R code that is more explicit
On the other hand, less-experienced programmers may find these multi-step
workflows more confusing or error-prone.

There is also overlap in
functionality with the `car` and `effects` packages [@car], like the computation of variance
inflation factors and partial residuals, although `jtools` aims to improve
the user interface and support more model types in some cases.
the user interface, such as by allowing non-standard evaluation, and support more
model types in some cases. The use of `ggplot2` for graphics is another important
distinction from `car`. `car` does not produce model summaries in general, so
the user would need to incorporate `car`'s variance inflation factors into a
table through some other means or otherwise just inspect them devoid of that
context.

## Real-world use

Expand All @@ -101,17 +129,4 @@ Finally, @spalti2023 use `weights_tests()` to
assess the sensitivity of their estimates to the influence of survey sampling
weights in their research on scientific misperceptions.

\footnotesize

| **Feature** | `jtools` | `marginaleffects` | `modelsummary` | `sjPlot` | `ggeffects` | `gtsummary` | `parameters` |
|--------------------------------|----------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------|--------------------------------------------------------------------|-------------------------------------------------------------------|--------------------------------------------------------------------------------|-----------------------------------------------------|--------------------------------------------------|
| **Model information** | Flexible summaries, robust SEs, VIFs, centered and scaled variables, model fit statistics | Marginal effects, contrasts, equivalence tests | All information available via `broom` and `parameters`, plus robust SEs | Flexible summaries, robust SEs, scaled variables, model fit statistics | Marginal effects, contrasts, equivalence tests (from `marginaleffects` or `emmeans`) | Information available via `broom`, pairwise comparisons | Coefficient estimates, intervals, test statistics |
| **Supported models** | lm, glm, mixed models, survey data for summaries, many more for visualization | Over 100 types | All models supported by `broom` and `parameters` | lm, glm, mixed models, Bayesian models | Large number of models | All that are supported by `broom` | Large number, including structural models |
| **Visualization** | Effect plots, coefficient plots (`ggplot2`) | Visualization of marginal effects, similar to `ggeffects` | Coefficient plots (`ggplot2`) | Coefficient plots (`ggplot2`) | Plots for effects, predictions, etc. | Experimental coefficient plot functions | Visualization of model parameters |
| **Model summary export options** | Coefficient plots, tables to PDF/html/Word via "huxtable" | Limited (console-focused) | Most extensive, customizable, easy to use (Markdown, LaTeX, HTML, etc.) | To HTML | To HTML and Markdown | Extensive (Word, PDF, HTML, etc.) | To HTML and Markdown |
| **Interface** | User-friendly, non-standard evaluation, customizable, bundled features in single function calls | Simple, tidy, pipeline-oriented | Simple, user-friendly, power user features | User-friendly, easy to integrate | Simple, user-friendly, pipeline-oriented | Pipeline-oriented, focus on table output | User-friendly |
| **Unique Features** | Extensive survey/weights support, plot intervals with robust errors, include partial residuals on plots | Detailed marginal effect analysis, flexible contrasts and hypothesis testing, extensive documentation | Also includes summaries of data frames | Support for psychometric item analysis | Plot intervals with robust errors, plot partial residuals on plots | Medical research-oriented summaries, focus on data summaries | Includes variable selection/data reduction tools |

\normalsize

# References

0 comments on commit 7b1e203

Please # to comment.