This package consists of a series of functions created by the author (Jacob) to automate otherwise tedious research tasks. At this juncture, the unifying theme is the more efficient presentation of regression analyses. There are a number of functions for other programming and statistical purposes as well. Support for the survey
package’s svyglm
objects as well as weighted regressions is a common theme throughout.
Notice: As of jtools
version 2.0.0, all functions dealing with interactions (e.g., interact_plot()
, sim_slopes()
, johnson_neyman()
) have been moved to a new package, aptly named interactions
.
For the most stable version, simply install from CRAN.
If you want the latest features and bug fixes then you can download from Github. To do that you will need to have devtools
installed if you don’t already:
Then install the package from Github.
You should also check out the dev
branch of this repository for the latest and greatest changes, but also the latest and greatest bugs. To see what features are on the roadmap, check the issues section of the repository, especially the “enhancement” tag.
Here’s a synopsis of the current functions in the package:
summ()
)summ()
is a replacement for summary()
that provides the user several options for formatting regression summaries. It supports glm
, svyglm
, and merMod
objects as input as well. It supports calculation and reporting of robust standard errors via the sandwich
package.
Basic use:
#> MODEL INFO:
#> Observations: 32
#> Dependent Variable: mpg
#> Type: OLS linear regression
#>
#> MODEL FIT:
#> F(2,29) = 69.21, p = 0.00
#> R² = 0.83
#> Adj. R² = 0.81
#>
#> Standard errors: OLS
#> 
#> Est. S.E. t val. p
#>     
#> (Intercept) 37.23 1.60 23.28 0.00
#> hp 0.03 0.01 3.52 0.00
#> wt 3.88 0.63 6.13 0.00
#> 
It has several conveniences, like refitting your model with scaled variables (scale = TRUE
). You have the option to leave the outcome variable in its original scale (transform.response = TRUE
), which is the default for scaled models. I’m a fan of Andrew Gelman’s 2 SD standardization method, so you can specify by how many standard deviations you would like to rescale (n.sd = 2
).
You can also get variance inflation factors (VIFs) and partial/semipartial (AKA part) correlations. Partial correlations are only available for OLS models. You may also substitute confidence intervals in place of standard errors and you can choose whether to show p values.
#> MODEL INFO:
#> Observations: 32
#> Dependent Variable: mpg
#> Type: OLS linear regression
#>
#> MODEL FIT:
#> F(2,29) = 69.21, p = 0.00
#> R² = 0.83
#> Adj. R² = 0.81
#>
#> Standard errors: OLS
#> 
#> Est. 2.5% 97.5% t val. VIF partial.r part.r
#>        
#> (Intercept) 20.09 19.15 21.03 43.82
#> hp 2.18 3.44 0.91 3.52 1.77 0.55 0.27
#> wt 3.79 5.06 2.53 6.13 1.77 0.75 0.47
#> 
#>
#> Continuous predictors are meancentered and scaled by 1 s.d.
Clusterrobust standard errors:
data("PetersenCL", package = "sandwich")
fit2 < lm(y ~ x, data = PetersenCL)
summ(fit2, robust = "HC3", cluster = "firm")
#> MODEL INFO:
#> Observations: 5000
#> Dependent Variable: y
#> Type: OLS linear regression
#>
#> MODEL FIT:
#> F(1,4998) = 1310.74, p = 0.00
#> R² = 0.21
#> Adj. R² = 0.21
#>
#> Standard errors: Clusterrobust, type = HC3
#> 
#> Est. S.E. t val. p
#>     
#> (Intercept) 0.03 0.07 0.44 0.66
#> x 1.03 0.05 20.36 0.00
#> 
Of course, summ()
like summary()
is bestsuited for interactive use. When it comes to sharing results with others, you want sharper output and probably graphics. jtools
has some options for that, too.
export_summs()
)For tabular output, export_summs()
is an interface to the huxtable
package’s huxreg()
function that preserves the niceties of summ()
, particularly its facilities for robust standard errors and standardization. It also concatenates multiple models into a single table.
fit < lm(mpg ~ hp + wt, data = mtcars)
fit_b < lm(mpg ~ hp + wt + disp, data = mtcars)
fit_c < lm(mpg ~ hp + wt + disp + drat, data = mtcars)
coef_names < c("Horsepower" = "hp", "Weight (tons)" = "wt",
"Displacement" = "disp", "Rear axle ratio" = "drat",
"Constant" = "(Intercept)")
export_summs(fit, fit_b, fit_c, scale = TRUE, transform.response = TRUE, coefs = coef_names)
#> Registered S3 methods overwritten by 'broom':
#> method from
#> tidy.glht jtools
#> tidy.summary.glht jtools
Model 1 
Model 2 
Model 3 

Horsepower 
0.36 ** 
0.35 * 
0.40 ** 
(0.10) 
(0.13) 
(0.13) 

Weight (tons) 
0.63 *** 
0.62 ** 
0.56 ** 
(0.10) 
(0.17) 
(0.18) 

Displacement 

0.02 
0.08 

(0.21) 
(0.22) 

Rear axle ratio 


0.16 


(0.12) 

Constant 
0.00 
0.00 
0.00 
(0.08) 
(0.08) 
(0.08) 

N 
32 
32 
32 
R2 
0.83 
0.83 
0.84 
All continuous predictors are meancentered and scaled by 1 standard deviation. *** p < 0.001; ** p < 0.01; * p < 0.05. 
In RMarkdown documents, using export_summs()
and the chunk option results = 'asis'
will give you nicelooking tables in HTML and PDF output. Using the to.word = TRUE
argument will create a Microsoft Word document with the table in it.
plot_coefs()
and plot_summs()
)Another way to get a quick gist of your regression analysis is to plot the values of the coefficients and their corresponding uncertainties with plot_summs()
(or the closely related plot_coefs()
). Like with export_summs()
, you can still get your scaled models and robust standard errors.
coef_names < coef_names[1:4] # Dropping intercept for plots
plot_summs(fit, fit_b, fit_c, scale = TRUE, robust = "HC3", coefs = coef_names)
And since you get a ggplot
object in return, you can tweak and theme as you wish.
Another way to visualize the uncertainty of your coefficients is via the plot.distributions
argument.
These show the 95% interval width of a normal distribution for each estimate.
plot_coefs()
works much the same way, but without support for summ()
arguments like robust
and scale
. This enables a wider range of models that have support from the broom
package but not for summ()
.
effect_plot()
)Sometimes the best way to understand your model is to look at the predictions it generates. Rather than look at coefficients, effect_plot()
lets you plot predictions across values of a predictor variable alongside the observed data.
And a new feature in version 2.0.0
lets you plot partial residuals instead of the raw observed data, allowing you to assess model quality after accounting for effects of control variables.
Categorical predictors, polynomial terms, (G)LM(M)s, weighted data, and much more are supported.
There are several other things that might interest you.
gscale()
: Scale and/or meancenter data, including svydesign
objectsscale_mod()
and center_mod()
: Refit models with scaled and/or meancentered datawgttest()
and pf_sv_test()
, which are combined in weights_tests()
: Test the ignorability of sample weights in regression modelssvycor()
: Generate correlation matrices from svydesign
objectstheme_apa()
: A mostly APAcompliant ggplot2
themetheme_nice()
: A nice ggplot2
themeadd_gridlines()
and drop_gridlines()
: ggplot2
themechanging convenience functionsmake_predictions()
: an easy way to generate hypothetical predicted data from your regression model for plotting or other purposes.Details on the arguments can be accessed via the R documentation (?functionname
). There are now vignettes documenting just about everything you can do as well.
I’m happy to receive bug reports, suggestions, questions, and (most of all) contributions to fix problems and add features. I prefer you use the Github issues system over trying to reach out to me in other ways. Pull requests for contributions are encouraged. If you are considering writing up a bug fix or new feature, please check out the contributing guidelines.
Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.
The source code of this package is licensed under the MIT License.