Finding Economic Articles With Data

In my view, one of the greatest developments during the last decade in economics is that the Journals of the American Economic Association and some other leading journals require authors to upload the replication code and data sets of accepted articles.

I wrote a Shiny app that allows to search currently among more than 3000 articles that have an accessible data and code supplement. Just click here to use it:

http://econ.mathematik.uni-ulm.de:3200/ejd/

One can perform a keyword search among the abstract and title. The screenshot shows an example:

One gets some information about the size of the data files and the used code files. I also tried to find and extract a README file from each supplement. Most README files explain whether all results can be replicated with the provided data sets or whether some results require confidential or proprietary data sets. The link allows you to look at the README without the need to download the whole data set.

The main idea is that such a search function could be helpful for teaching economics and data science. For example, my students can use the app to find an interesting topic for a Bachelor or Master Thesis in form of an interactive analysis with RTutor. You could also generate a topic list for a seminar, in which students shall replicate some key findings of a resarch article.

While the app performs well for a single user, I have not tested the performance for many simultaneous users. If it is too sluggish or you don’t get connected there are perhaps currently too many users. Then just try it out a bit later.

If you want to analyse yourself the collected data underlying the search app, you can download the zipped SQLite databases using the following links:

• Main database (should suffice for most analyses):
  http://econ.mathematik.uni-ulm.de/ejd/articles.zip
• Large database containing names and sizes of all files in the data and code supplements:
  http://econ.mathematik.uni-ulm.de/ejd/files.zip

I try to update the databases regularly.

Below is an example, for a simple analysis based on that databases. First make sure that you download and extract articles.zip into your working directory.

We first open a database connection

library(RSQLite)
db = dbConnect(RSQLite::SQLite(),"articles.sqlite")

File type conversion between databases and R can sometimes be a bit tedious. For example, SQLite knows no native Date or logical type. For this reason, I typically use my package dbmisc when working with SQLite databases. It allows to specify a database schema as simple yaml file and has a lot of convenience function to retrieve or modify data that automatically use the provided schema. The following code sets the database schema that is provided in the package EconJournalData:

library(dbmisc)
db = set.db.schemas(db,schema.file=
  system.file("schema/articles.yaml", package="EconJournalData"))

Of course, for a simple analysis as ours below just using the standard function in the DBI package without schemata suffices. But I am just used to working with the dbmisc package.

The main information about articles is stored in the table article

# Get the first 4 entries of articles as data frame
dbGet(db, "article",n = 4)

The table files_summary contains information about code, data and archive files for each article

dbGet(db, "files_summary",n = 6)

Let us now analyse which share of articles uses Stata, R, Python, Matlab or Julia and how the usage has developed over time.

Since our datasets are small, we can just download the two tables and work with dplyr in memory. Alternatively, you could use some SQL commands or work with dplyr on the database connection.

articles = dbGet(db,"article")
fs = dbGet(db,"files_summary")

Let us now compute the shares of articles that have one of the file types, we are interested in

# Number of articles with analyes data & code supplementary
n_art = n_distinct(fs$id)

# Count articles by file types and compute shares
fs %>% group_by(file_type) %>%
  summarize(count = n(), share=round((count / n_art)*100,2)) %>%
  # note that all file extensions are stored in lower case
  filter(file_type %in% c("do","r","py","jl","m")) %>%
  arrange(desc(share))

Roughly 70% of the articles have Stata do files and almost a quarter Matlab m files. Using open source statistical software seems not yet very popular among economists: less than 3% of articles have R code files, Python is below 1% and only 2 articles have Julia code.

This dominance of Stata in economics never ceases to surprise me, in particular when for some reason I just happened to open the Stata do file editor and compare it with RStudio… But then, I am not an expert in writing empirical economic research papers – I just like R programming and rather passively consume empirical research. For writing empirical papers it probably is convenient that in Stata you can add a robust or robust cluster option to almost every type of regression in order to quickly get the economists’ standard standard errors…

For a teaching empirical economics with R the dominance of Stata is not neccessarily bad news. It means that there are a lot of studies which students can replicate in R. Such replication would be considerably less interesting if the original code of the articles would already be given in R.

Let us finish by having a look at the development over time…

sum_dat = fs %>% 
  left_join(select(articles, year, id), by="id") %>%
  group_by(year) %>%
  mutate(n_art_year = n()) %>%
  group_by(year, file_type) %>%
  summarize(
    count = n(),
    share=round((count / first(n_art_year))*100,2)
  ) %>%
  filter(file_type %in% c("do","r","py","jl","m")) %>%
  arrange(year,desc(share))  
head(sum_dat)

library(ggplot2)
ggplot(sum_dat, aes(x=year, y=share, color=file_type)) 
  + geom_line(size=1.5) + scale_y_log10() + theme_bw()

Well, maybe there is a little upward trend for the open source languages, but not too much seems to have happened over time so far…