Talk Aims

• Describe political science/panel-series data problems.

• Introduce psData, our first attempt at a solution.

• Thoughts for future framework.

• Get ideas for going forward/avoiding effort duplication.

What is panel-series data?

Good

Many political scientists are creating/making publicly available panel-series data sets!

• Database of Political Institutions

• Polity

• Reinhart and Rogoff banking crisis varieties

Issues (1)

• Data tied to particular studies.

• Data posted and maintained haphzardly.

Issues (2)

• Variety of formats (SPSS, Stata, Excel, CSV).
  • Excel data is often a mess.

• Variety of panel/series identifiers, even for regularly used panels (especially countries).

• Some variables are suggested by the literature (e.g. winset) are composed of other variables, but aren’t regularly updated.

Consequences

• Effort Duplication: Political scientists (or RA’s) waste a lot of time downloading/cleaning/transforming/merging commonly used data sets.

• Errors introduced by data import and transformation scripts that are written individually and never shared across researchers
  • Don’t benefit from code review.

Solution

R package psData

Goals (1)

• Avoid effort duplication
  • Standardised functions to download/clean/transform political science/panel-series data sets.
  • Easy to contribute new functions.

Goals (2)

• Standardised output
  • Makes it easy to merge multiple data sets with the same panel-series structure.

Goals (3)

• Viral/quick error fixes
  • Updates and errors are (inevitably) made they can be found and corrected by anyone.
  • Development hosted on GitHub
  • Corrections automatically propagated via CRAN.

Master Build: Version 0.1.2

• Master build on CRAN

• Silos each data set into its own getter or creator function.
  • DpiGet
  • PolityGet
  • RRCrisisGet
  • WinsetCreator

• Creator functions source getter functions for raw data.

Examples (getter)

# Load package
library(psData)

# Download/transform polity2 variable
PolityData <- PolityGet(vars = "polity2")

head(PolityData)
##  iso2c     country year polity2
##    AF Afghanistan 1800      -6
##    AF Afghanistan 1801      -6
##    AF Afghanistan 1802      -6
##    AF Afghanistan 1803      -6
##    AF Afghanistan 1804      -6
##    AF Afghanistan 1805      -6

Examples (getter)

# Download/transform Reinhart and Rogoff (2010)
RRData <- RRCrisisGet()

head(RRData)[1:5]
##  iso2c country year RR_Independence RR_CurrencyCrisis
##    AO  Angola 1800               0                 0
##    AO  Angola 1801               0                 0
##    AO  Angola 1802               0                 0
##    AO  Angola 1803               0                 0
##    AO  Angola 1804               0                 0
##    AO  Angola 1805               0                 0

Example (creator)

# Create winset and selector variables
WinsetData <- WinsetCreator()

head(WinsetData)
##  iso2c     country year    W ModS
##    AF Afghanistan 1975 0.25    0
##    AF Afghanistan 1976 0.25    0
##    AF Afghanistan 1977 0.25    0
##    AF Afghanistan 1989 0.50    0
##    AF Afghanistan 1990 0.50    0
##    AF Afghanistan 1991 0.50    0

Dev builds

• Development at rOpenGov.

• Ideas we are playing with:
  • Creating an S4 R method for handling the downloaded data in a standard format.
  • Homebrew-style recipies, specifying how to download the data and including meta-data (e.g. BibTex citations).
  • Improve testing (e.g. testdat).
    
  • Integration with data version tracking (minimum SHA hashes, maybe dat).

Audience

Ideas for avoiding effort duplication with others building open data frameworks?

Contact

Twitter: @chrisgandrud

GitHub: rOpenGov/psData