d-social-kld

This repository contains replication code for D-SOCIAL-KLD scores (Carroll, Primo, & Richter 2016) using KLD data that runs through the year 2018.

Files

Files in this directory include:

• README.md: this file;
• 01-cleanKLD.R: a script for converting off-the-shelf KLD data into workable form for the MCMCpack routine;
• 02-getIdeals.R: the script calling MCMCpack::MCMCdynamicIRT1d, which actually gets more than just the ideal points;
• 03-pullThetas.R: the script pulling/cleaning the D-SOCIAL-KLD scores;
• fn-get_start_values.R: a helper function for starting values in the MCMCpack::MCMCdynamicIRT1d call; and
• d-social-kld_summary-stats.csv: summary statistics for the posterior distributions.

The data with all 2500 pulls from the posterior are too large to be added here, so please check socialscores.org for updates.

Installation

The most recent analysis was conducted in April 2021. The analysis was conducted in R (Version 4.0.5). The only high-level package dependencies are:

• tidyverse (Version 1.3.0);
• foreach (Version 1.5.1); and
• MCMCpack (Version 1.5-0).

Of course, these establish their own dependences, most notably MASS and coda for MCMCpack and all the usual suspects (dplyr, tidyr, etc.) for tidyverse.

Notes

• There are well-known and oft-lamented classes between dplyr::select and the MASS package called in MCMCpack. We've tried to minimize errors, but the occasional sequencing error may have slipped through here or there.
• Several firms from distinct KLD "domiciles" share the same ticker symbol. In these cases, the newer-to-the-dataset firm is assigned a random four-character tag placed after a double-tilde separator. So, if two firms both have ticker symbol Z, then the older-to-the-dataset firm remains Z and the newer-to-the-dataset firm is assigned ticker Z~~abcd, where abcd may be any random tag of four lowercase letters. Those that read the code in 01-cleanKLD.R carefully will see that several precautions were taken to ensure continuity across these tags. However, should you find any errors, please do not hesitate to drop a line or issue a pull request.