Prior to today’s lab, make sure that you have installed Latex and papaja (see Sara’s announcement on Canvas). You can download the Rmd file here and the example papaja Rmd file here.

Purpose

The following lab covers papaja, a package for creating reproducible, APA-style manuscripts in Rmarkdown. We will go over how to write a manuscript in papaja, including creating tables, figures, citations, and in-text statistical reporting. This page has some broad (overview) information and links to useful resources.

papaja

papaja is an R package for Preparing APA Journal Articles. It contains several functions and perhaps most importantly, an RMD template that formats the output document (.docx or .pdf) in APA format. Here is the link to papaja’s documentation, the most helpful and comprehensive papaja resource out there.

Installing papaja

First, we need to install papaja. papaja is not yet on CRAN, and so we have to download it from github rather than using install.packages(), which requires the devtools package. Let’s install both.

# Install devtools package if necessary
if(!"devtools" %in% rownames(installed.packages())) install.packages("devtools")

# Install the stable development verions from GitHub
if(!"papaja" %in% rownames(installed.packages())) devtools::install_github("crsh/papaja")
# call papaja
library(papaja)

In addition to papaja, the citr package is going to be helpful when it comes to citing works in text (within our RMD) and the corx package is going to make it easier to create a table of correlations.

#devtools::install_github("crsh/citr")
#install.packages("corx")

OK, now we’re ready to go! Below I’ll briefly review some of the functions we’ll use today before we dive into hands-on work in a papaja template RMD.

Tables

There are a few options for making apa-style tables in papaja, each of which provide different pros and cons.

We’ll start by going over apa_table from papaja.apa_table() can be used to format a dataframe as an APA-style table. FYI, it is built on top of knitr::kable(). When using apa_table(), the workflow generally goes something like:

  1. Create a dataframe that contains everything you want in the table.
  2. use printnum() (from papaja) to format the numeric columns.
  3. Pass this formatted dataframe to apa_table(), set the caption and note (if applicable).
  4. Always set the chunk option results = "asis"

Table of Descriptives

Let’s start by getting a table of descriptive stats. We’ll use the bfi data from psych, and use the bfi.keys to score the 5 scale scores.

b5 <- psych::bfi %>% 
  cbind(psych::scoreVeryFast(psych::bfi.keys, psych::bfi))

First, we create a dataframe with the descriptives of our Big Five scales. We’ll use some of the tidyverse functions for this.

descriptives <- b5 %>%
  select(agree:openness) %>% 
  gather(scale, score) %>% 
  group_by(scale) %>% 
  summarize(
    Mean = mean(score), 
    Median = median(score), 
    SD = sd(score), 
    Min = min(score),
    Max = max(score))
## `summarise()` ungrouping output (override with `.groups` argument)
descriptives

Next, we format the numerical columns using printnum() function from papaja for this. In essence, this puts everything to 2 decimal points (you can change this), pads with zeroes when necessary, and turns those to strings so they print correctly. “-1” indicates that the first column is non-numerical so does not apply.

descriptives[, -1] <- papaja::printnum(descriptives[, -1]) 

descriptives

Then, we run apa_table() on the dataframe, and we can add a caption and note. You have to set results = "asis" in the chunk options for it to render correctly:

apa_table(descriptives, 
          caption = "Descriptive statistics of Big Five Scale Scores.", 
          note = "This table was created with apa_table().", 
          escape = TRUE
)
(#tab:descrips_tab)
Descriptive statistics of Big Five Scale Scores.
scale Mean Median SD Min Max
agree 4.65 4.80 0.90 1.00 6.00
conscientious 4.27 4.40 0.95 1.00 6.00
extraversion 4.15 4.20 1.06 1.00 6.00
neuroticism 3.16 3.00 1.20 1.00 6.00
openness 4.59 4.60 0.81 1.20 6.00

Note. This table was created with apa_table().

 

Now we have a nicely formatted APA-style table that is fully reproducible. We can also reference the table within the papaja rmd with \@ref(tab:descrips_tab) as you’ll see when we walk through the example script.

Table of Correlations

Next, let’s get a table of scale inter-correlations for the same dataset. Correlation tables are a little difficult in papaja, but the corx() function from the corx library makes it a little easier.

Basically, the workflow for this is:

  1. Use corx to create the correlation table
  2. Use apa_table() to add a caption, note, and print it
library(corx)
## Warning: package 'corx' was built under R version 4.0.3
cor <-  b5 %>%
  select(agree:openness) %>% 
  corx(triangle = "lower",
       stars = c(0.05, 0.01, 0.001),
       describe = c(`$M$` = mean, `$SD$` = sd)) 

cor

corx(data = ., stars = c(0.05, 0.01, 0.001), triangle = “lower”, describe = c($M$ = mean, $SD$ = sd))

1 2 3 4 \(M\) \(SD\)
1. agree - 4.65 0.90 2. conscientious .26*** - 4.27 0.95 3. extraversion .46*** .26*** - 4.15 1.06 4. neuroticism -.19*** -.23*** -.22*** - 3.16 1.20 5. openness .15*** .20*** .21*** -.09*** 4.59 0.81

Note. * p < 0.05; ** p < 0.01; *** p < 0.001

papaja::apa_table(cor$apa, 
          caption = "Example corr matrix",
          note = "* p < 0.05; ** p < 0.01; *** p < 0.001",
          escape = TRUE)
(#tab:corr_tab)
Example corr matrix
1 2 3 4 $M$ $SD$
1. agree - 4.65 0.90
2. conscientious .26*** - 4.27 0.95
3. extraversion .46*** .26*** - 4.15 1.06
4. neuroticism -.19*** -.23*** -.22*** - 3.16 1.20
5. openness .15*** .20*** .21*** -.09*** 4.59 0.81

Note. * p < 0.05; ** p < 0.01; *** p < 0.001

 

Regression Tables

Now we’ll get a regression table using another method available in papaja. We’ll use the apa_print() function. This function takes statistical results as input and produces a list containing an APA-style table and several objects useful for printing results in-text (covered below).

The workflow for this method is:

  1. Conduct statistical test(s)
  2. Create the apa-style results with apa_print()
  3. Extract the table from the apa_print object with $table

Let’s do that with a couple of regression models, regressing Conscientiousness on age (model 1), and also education (model 2).

First, conduct the regressions

b5 <- b5 %>% 
  mutate(
    education = as.factor(education),
    education = case_when(
      education == 1 ~ "Some HS",
      education == 2 ~ " (Finished HS)",
      education == 3 ~ " (Some College)",
      education == 4 ~ " (College Grad)",
      education == 5 ~ " (Grad Degree)")) %>%
  filter(!is.na(conscientious),
         !is.na(age),
         !is.na(education))

# Regression models
model_1 <- lm(conscientious ~ age, data = b5)
model_2 <- lm(conscientious ~ age + education, data = b5)

Next, use apa_print() to create those apa-style results objects:

# model 1 results
model_1_results <- apa_print(model_1)
# model 2 results
model_2_results <- apa_print(model_1)
# model comparison results
model_comp_results <- apa_print(list(model_1, model_2),
                              boot_samples = 0)

Note: We set boot_samples = 0 otherwise it creates a bootstrapped CI for delta R^2. We will cover bootstrapping next week.

Then, we can pull the tables out of them.

You could get a table of just the Model 1 results:

model_1_results$table %>% 
  apa_table(caption = "Model Regressing Conscientiousness on Age",
            note = "* p < 0.05; ** p < 0.01; *** p < 0.001",
            escape = TRUE)
(#tab:reg_tbl_1)
Model Regressing Conscientiousness on Age
Predictor \(b\) 95% CI \(t(2575)\) \(p\)
Intercept 4.07 \([3.96\), \(4.17]\) 75.42 < .001
Age 0.01 \([0.00\), \(0.01]\) 4.71 < .001

Note. * p < 0.05; ** p < 0.01; *** p < 0.001

 

Or we could get a table of just model 2’s results:

model_2_results$table %>% 
  apa_table(caption = "Model Regressing Conscientiousness on Age & Education",
            note = "* p < 0.05; ** p < 0.01; *** p < 0.001",
            escape = TRUE)
(#tab:reg_tbl_2)
Model Regressing Conscientiousness on Age & Education
Predictor \(b\) 95% CI \(t(2575)\) \(p\)
Intercept 4.07 \([3.96\), \(4.17]\) 75.42 < .001
Age 0.01 \([0.00\), \(0.01]\) 4.71 < .001

Note. * p < 0.05; ** p < 0.01; *** p < 0.001

 

And finally, we can get them together with their comparison:

model_comp_results$table %>% 
  apa_table(caption = "Model Comparison: Conscientiousness by Age and Age + Education",
            note = "* p < 0.05; ** p < 0.01; *** p < 0.001",
            escape = TRUE)
(#tab:reg_tbl_3)
Model Comparison: Conscientiousness by Age and Age + Education
Model 1 Model 2
Intercept \(4.07\) \([3.96\), \(4.17]\) \(3.89\) \([3.75\), \(4.04]\)
Age \(0.01\) \([0.00\), \(0.01]\) \(0.01\) \([0.01\), \(0.01]\)
Education Finished HS \(0.02\) \([-0.12\), \(0.16]\)
Education Grad Degree \(0.04\) \([-0.09\), \(0.17]\)
Education Some College \(0.22\) \([0.12\), \(0.33]\)
EducationSome HS \(-0.02\) \([-0.18\), \(0.13]\)
\(R^2\) [90% CI] \(.01\) \([0.00\), \(0.02]\) \(.02\) \([0.01\), \(0.03]\)
\(F\) 22.16 10.93
\(df_1\) 1 5
\(df_2\) 2575 2571
\(p\) < .001 < .001
\(\mathrm{AIC}\) 6,979.81 6,955.69
\(\mathrm{BIC}\) 6,997.38 6,996.68
\(\Delta R^2\) \(.01\)
\(F\) 8.06
\(df_1\) 4
\(df_2\) 2,575
\(p\) < .001
\(\Delta \mathrm{AIC}\) -24.12
\(\Delta \mathrm{BIC}\) -0.70

Note. * p < 0.05; ** p < 0.01; *** p < 0.001

 

The last thing I’ll mention here is that we can use the chunk labels for our table to reference them in text. For example, we could reference this last one with by adding Table \@ref(tab:reg_tbl_3) in the papaja RMD.

Figures

You can create figures using basically any of the options we’ve learned about. I personally like ggplot, which can be put into APA format using the theme_apa() theme (which comes from papaja).

Let’s get an APA-formatted plot for the relation between age and conscientiousness.

ggplot(b5, aes(x = age, y = conscientious)) +
  geom_point(alpha = .5) +
  geom_smooth(method = "lm") +
  labs(x = "Age",
       y = "Conscientiousness") +
  theme_apa()
## `geom_smooth()` using formula 'y ~ x'

The best way to give that figure a caption is to use preface it with:

(ref:fig-ageXconsc-cap) Conscientiousness by Age.

You then set the chunk option fig.cap = "(ref:fig-ageXconsc-cap) ", There is an example of this in the example papaja manuscript, so we will see it in action in a moment.

Just to be perfectly clear, you should be able to expand this to more complicated plots. For example, let’s take a look at a figure that also has education in it:

ggplot(b5, aes(x = age, y = conscientious, color = education)) +
  geom_point(alpha = .5) +
  geom_smooth(method = "lm", se = FALSE) +
  labs(x = "Age",
       y = "Conscientiousness") +
  theme_apa()
## `geom_smooth()` using formula 'y ~ x'

Or one that has age by each of the Big Five:

b5 %>% 
  select(age, agree:openness) %>% 
  gather(trait, score, -age) %>% 
ggplot(aes(x = age, y = score)) +
  geom_point(alpha = .5) +
  geom_smooth(method = "lm") +
  labs(x = "Age",
       y = "Scale Score") +
  facet_grid(~trait) + 
  theme_apa()
## `geom_smooth()` using formula 'y ~ x'

In-text Summaries of tests

One of the most useful aspects of writing manuscripts in RMD and papaja is that we can report our results directly from the statistical models we run in R, reducing the likelihood of copy and paste errors.

We can use apa_print() from papaja to report the results of our statistical tests in text. Abbreviated results (estimate and CI) can be called with $estimate from the apa_print() object. For example, model_1_results$estimate$age will print a nicely formatted printout of the slope for age on conscientiousness:

Age significantly predicted conscientiousness, \(b = 0.01\), 95% CI \([0.00\), \(0.01]\).

We can instead get the full result (including t, df, and p) by calling the $full_result object, so model_1_results$full_result$age will be rendered like so:

Age significantly predicted conscientiousness, \(b = 0.01\), 95% CI \([0.00\), \(0.01]\), \(t(2575) = 4.71\), \(p < .001\).

We could get the results of other slopes by changing out $age to reference other terms. For example, we could report the result of the difference between some high school and having finished high school with model_2_results$estimate$education_Finished_HS:

Finishing highschool (vs. not finishing highschool) was not associated with Conscientiousness, .

Regression model results

And we could get model fit results too, both from each model and the comparison object. For example

  • model_1_results$full_result$modelfit$r2 will print the \(R^2\) for model 1.
  • model_2_results$full_result$modelfit$r2 will print the \(R^2\) for model 2.
  • model_comp_results$full_result$model2 will print the \(\Delta R^2\) value (difference in \(R^2\)), and the F test for that difference.

Putting it all together, we can report those results with something like the following:

The model with just age explained less variance, \(R^2 = .01\), 90% CI \([0.00\), \(0.02]\), \(F(1, 2575) = 22.16\), \(p < .001\), than the model that also included (dummy-coded) education \(R^2 = .01\), 90% CI \([0.00\), \(0.02]\), \(F(1, 2575) = 22.16\), \(p < .001\), which was significant \(\Delta R^2 = .01\), \(F(4, 2,575) = 8.06\), \(p < .001\)

papaja has various print methods, so be sure to check that part of the documentation.

Bibliography

The final ingredient we will talk about before getting into our example manuscript and the (singular) hack for the lab is the bibliography. All RMDs can include a bibliography: entry in the YAML (the header portion of an RMD), where you can link to a bibtex (i.e., .bib) file. Bibtex entries look something like this:

@article{goldberg1990alternative,
  title={An alternative" description of personality": the big-five factor structure.},
  author={Goldberg, Lewis R},
  journal={Journal of personality and social psychology},
  volume={59},
  number={6},
  pages={1216--1229},
  year={1990},
  doi = {https://doi.org/10.1037/0022-3514.59.6.1216},
  publisher={American Psychological Association}
}

Creating a BibTeX file for references

One convenient thing, is that you can get them from google scholar. Let’s take a look at the scholar results when we search for McClelland & Judd (1993).

From there, click on the quotes, and then on the BibTeX link, which takes you here

You should see this:

@article{mcclelland1993statistical,
  title={Statistical difficulties of detecting interactions and moderator effects.},
  author={McClelland, Gary H and Judd, Charles M},
  journal={Psychological bulletin},
  volume={114},
  number={2},
  pages={376},
  year={1993},
  publisher={American Psychological Association}
}

You would then want to copy that into a .bib file (you can edit it within RStudio or choose another text editor), and then make sure you reference that .bib file in the references: part of your YAML (we’ll see this in the example).

You can also use popular reference managers like mendeley, zotero, etc. to create bibtex files.

In-text citations

Then, we could reference this article by referring to its name. There are basically three options:

Type code how it renders
Parenthetical citation [@mcclelland1993statistical] (McClelland and Judd 1993)
Non-parenthetical @mcclelland1993statistical McClelland and Judd (1993)
just the year [-@mcclelland1993statistical]. (1993)

Note that you can also put additional text in the brackets. For example,

[see @mcclelland1993statistical]

becomes…

(see McClelland and Judd 1993)

You can also add multiple citations by separating them with a ;. For example,

[@mcclelland1993statistical; @cohen2013applied]

becomes…

(McClelland and Judd 1993; Cohen et al. 2013)

One of the coolest things about using BibTeX is that references will only appear in your reference list that you cite in text! You can see that here, because the ‘lab-9.bib’ file has four entries and the reference list below has just the two we reference above. That means no more last-minute checks on the reference list to make sure you removed the articles you references in a previous draft (that are no longer referenced). This alone is a huge time saver.

Citing R packages (& Helper functions from papaja)

papaja makes it easy to cite any R package you’ve used with the r_refs(file = "r-references.bib"), which creates a .bib file (named r-references.bib in that example code) with bibtex entries for all R packages you’ve called in the script. It is super handy!

You can combine this with the cite_r() package from papaja, which will create an in-text citation of all the packages you’ve used - we’ll see this in the example papaja Rmd.

That is all for this tutorial. Now let’s turn to the example RMD, and finally work on the hack (time permitting)!

The Singular papaja hack

Today there is just one hack. For this hack, I want you to get started on your final project RMD.

  1. Create a new RMD file for your final project using the papaja template.

  2. Edit the YAML to have the title and author of your project

  • Don’t worry about the abstract for now, unless you’ve already written it

  1. Create a .bib with at least one reference you know you’ll need for your final project.

  2. Cite the reference from your .bib (part 3) in text.

  3. Run at least one statistical test.

  4. Report the statistical test from part 4 in-text, in a table, and in a figure.

  • If you don’t have your data yet, run a similar test with a built-in dataset or one from a prior lab. The more similar the better - it will make it easier for you to adapt the code for your own data.
  • Bonus points if you can reference the table and figure in-text.

References

Cohen, Jacob, Patricia Cohen, Stephen G West, and Leona S Aiken. 2013. Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences. Routledge.

McClelland, Gary H, and Charles M Judd. 1993. “Statistical Difficulties of Detecting Interactions and Moderator Effects.” Psychological Bulletin 114 (2): 376.