Structure your projects purposely

• Always keep raw datasets! Save temporary “clean” datasets if needed.

• Common project structure:

my_project/
├─ raw/    # raw datasets
├─ temp/   # "clean" datasets
├─ output/ # figures, tables, model results
├─ 01_script_first.R
├─ 02_script_second.R # ...
├─ README.txt

project_example/
├─ raw/
│  ├─ qog_bas_cs_jan22.csv
├─ temp/
│  ├─ 01_qog_clean.csv
├─ output/
│  ├─ 02_model_coefplot.pdf
│  ├─ 02_model_results.rds
├─ 01_clean_data.R
├─ 02_run_model.R
├─ 99_paper_example.qmd
├─ 99_paper_example.pdf
├─ project_example.Rproj
├─ README_project_example.txt

Use RStudio projects

• A project can be anything: a paper, a homework assignment, your CV…

• RStudio Projects are folders. Just like the ones we just saw. There’s an .Rproj file indicating that the folder is an RStudio Project.

• Never worry about working directories again. Simply reference your files within the folder. For example: read_csv("raw/qog_bas_cs_jan22.csv").

• Start afresh every time you open RStudio with the following options (Tools > Global Options… > General):

  • Uncheck “Restore .RData into workspace at startup”
  • Set “Save Workspace to .RData on exit” to “Never”

Use Quarto (or R Markdown)

• Keep code and output in the same document! No more copy-pasting tables.

• You can write short reports, presentations (with revealjs or Beamer), or papers! Let’s see an example on 99_paper_example.qmd.

• It compiles automatically to .html or .pdf (via \(\LaTeX\): everything works!).

  • Installing \(\LaTeX\) locally is fairly easy. Just run the following code one time and you’ll be able to compile to PDF:

install.packages("tinytex")
tinytex::install_tinytex()

Make publication-ready tables with modelsummary

m1 <- lm(mpg ~ hp, data = mtcars)
m2 <- fixest::feols(mpg ~ hp | cyl, data = mtcars, vcov = "cluster")

modelsummary::modelsummary(list(m1, m2))

Make publication-ready tables with modelsummary (2)

modelsummary::modelsummary(
  list("(1) OLS" = m1, "(2) FE" = m2), 
  gof_omit = "Log|RMSE|[AB]IC", 
  coef_map = c("(Intercept)" = "Intercept", "hp" = "Horspower")
)

Make publication-ready tables with modelsummary (3)

• If compiling to PDF/\(\LaTeX\):

Save plots with ggplot2::ggsave()

p_mpg <- ggplot(mtcars, aes(x = mpg)) +
  geom_histogram()
p_mpg

Save plots with ggplot2::ggsave() (2)

• Don’t use “Export” from RStudio. It’s not reproducible!

ggsave(plot = p_mpg, filename = "output/plot_mpg.png",
       width = 8,  # other values to try: 10, 12
       height = 6, 
       scale = .8) # lower values will make text and other elements larger

Write abstract code when possible

• We want our code to be flexible enough to withstand minor changes upstream. In general, use objects/names instead of numbers!

# bad
mtcars[, 1]

# good
mtcars$mpg
mtcars[["mpg"]]

# bad
sum(mtcars$mpg) / 32

# good
sum(mtcars$mpg) / nrow(mtcars)

Write code in a consistent style

• I highly recommend the tidyverse style guide (even if you don’t use the tidyverse). But here’s the most important part, in my opinion.

• Use descriptive object names. df_anes is better than data.

• Follow the snake_case convention for object names (including columns). Don’t use capital letters or spaces. df_anes is better than dfAnes.

• Use janitor::clean_names() to clean messy variable names:

readxl::read_excel("raw/example_excel.xlsx") %>% 
  janitor::clean_names()

# A tibble: 2 × 3
  id_variable gdp_agr_fish x2010_production_log
        <dbl>        <dbl>                <dbl>
1           1         3000                   30
2           2         2000                   20

Don’t trust copy-pasting; write iterations!

• Let’s say that we want to multiple all of our variables by 100.

• This gets old fast… and it’s prone to error.

mtcars$mpg  <- mtcars$mpg * 100
mtcars$cyl  <- mtcars$cyl * 100
mtcars$disp <- mtcars$disp * 100
# ...

mtcars2 <- mtcars
for (var in colnames(mtcars)){
  mtcars[[var]] <- mtcars[[var]] * 100
}

mtcars <- mtcars %>% mutate(across(everything(), ~ . * 100))

mtcars <- mtcars %>% mutate(across(where(is.numeric), ~ . * 100))

( mtcars <- mtcars %>% mutate(across(mpg:disp, ~ . * 100)) )

                     mpg cyl  disp  hp drat    wt  qsec vs am gear carb
Mazda RX4           2100 600 16000 110 3.90 2.620 16.46  0  1    4    4
Mazda RX4 Wag       2100 600 16000 110 3.90 2.875 17.02  0  1    4    4
Datsun 710          2280 400 10800  93 3.85 2.320 18.61  1  1    4    1
Hornet 4 Drive      2140 600 25800 110 3.08 3.215 19.44  1  0    3    1
Hornet Sportabout   1870 800 36000 175 3.15 3.440 17.02  0  0    3    2
Valiant             1810 600 22500 105 2.76 3.460 20.22  1  0    3    1
Duster 360          1430 800 36000 245 3.21 3.570 15.84  0  0    3    4
Merc 240D           2440 400 14670  62 3.69 3.190 20.00  1  0    4    2
Merc 230            2280 400 14080  95 3.92 3.150 22.90  1  0    4    2
Merc 280            1920 600 16760 123 3.92 3.440 18.30  1  0    4    4
Merc 280C           1780 600 16760 123 3.92 3.440 18.90  1  0    4    4
Merc 450SE          1640 800 27580 180 3.07 4.070 17.40  0  0    3    3
Merc 450SL          1730 800 27580 180 3.07 3.730 17.60  0  0    3    3
Merc 450SLC         1520 800 27580 180 3.07 3.780 18.00  0  0    3    3
Cadillac Fleetwood  1040 800 47200 205 2.93 5.250 17.98  0  0    3    4
Lincoln Continental 1040 800 46000 215 3.00 5.424 17.82  0  0    3    4
Chrysler Imperial   1470 800 44000 230 3.23 5.345 17.42  0  0    3    4
Fiat 128            3240 400  7870  66 4.08 2.200 19.47  1  1    4    1
Honda Civic         3040 400  7570  52 4.93 1.615 18.52  1  1    4    2
Toyota Corolla      3390 400  7110  65 4.22 1.835 19.90  1  1    4    1
Toyota Corona       2150 400 12010  97 3.70 2.465 20.01  1  0    3    1
Dodge Challenger    1550 800 31800 150 2.76 3.520 16.87  0  0    3    2
AMC Javelin         1520 800 30400 150 3.15 3.435 17.30  0  0    3    2
Camaro Z28          1330 800 35000 245 3.73 3.840 15.41  0  0    3    4
Pontiac Firebird    1920 800 40000 175 3.08 3.845 17.05  0  0    3    2
Fiat X1-9           2730 400  7900  66 4.08 1.935 18.90  1  1    4    1
Porsche 914-2       2600 400 12030  91 4.43 2.140 16.70  0  1    5    2
Lotus Europa        3040 400  9510 113 3.77 1.513 16.90  1  1    5    2
Ford Pantera L      1580 800 35100 264 4.22 3.170 14.50  0  1    5    4
Ferrari Dino        1970 600 14500 175 3.62 2.770 15.50  0  1    5    6
Maserati Bora       1500 800 30100 335 3.54 3.570 14.60  0  1    5    8
Volvo 142E          2140 400 12100 109 4.11 2.780 18.60  1  1    4    2

Add sanity checks with assertthat::assert_that()

• Sometimes you perform an operation and then manually check whether it worked well. Formalize this so you don’t have to do it every time something else changes!

• This is very common in joins. For instance, you might want your output to conserve the number of rows of the X data frame. If not, you know something went wrong…

( df_x <- data.frame(id = c("A", "B", "C"), x_val = 1:3) )
( df_y <- data.frame(id = c("B", "C", "C"), y_val = 11:13) )

  id x_val
1  A     1
2  B     2
3  C     3

  id y_val
1  B    11
2  C    12
3  C    13

df_merged <- left_join(df_x, df_y)

assertthat::assert_that(nrow(df_x) == nrow(df_merged))

Error: nrow(df_x) not equal to nrow(df_merged)

assertthat::assert_that(sum(is.na(df_merged)) == 0)

Error: sum(is.na(df_merged)) not equal to 0

Set seeds when randomizing

• Every time you run this, you’ll get a different result:

rnorm(n = 5, mean = 0, sd = 1)

[1]  0.53327428 -1.92674383 -0.09157776 -1.27632757  1.04407709

set.seed(123)
rnorm(n = 5, mean = 0, sd = 1)

[1] -0.56047565 -0.23017749  1.55870831  0.07050839  0.12928774

set.seed(123)
rnorm(n = 5, mean = 0, sd = 1)

[1] -0.56047565 -0.23017749  1.55870831  0.07050839  0.12928774

Use renv to make a package vault

• Keeping track of package versions is important for future reproducibility.

• Using renv makes the most sense when making a replication package, or sharing an advance project with colleagues.

• Run renv::init() in your project to create a record of all packages used and their versions. This will create a couple of folders in your project (most notably, renv/).

• If you ever need to update this, run renv::snapshot().

• When someone else wants to run your code, they start with renv::restore(). This will scan renv/ and the other files and install all needed packages in their appropriate versions.

Automate your entire workflow with targets

• When projects grow large, it’s hard to keep track of how everything’s related.

• Say that you just caught a bug in one of your data cleaning scripts. What analyses do you need to rerun downstream?

• targets builds a dependency graph for your datasets/scripts. If you edit a node anywhere in the flow, targets knows what other nodes become out of sync, and can rerun them automatically.