Over the past 20 years, the political science community has increasingly pursued transparency through encouraging or requiring authors to publish “replication files” intended to make each step the research process as explicit as possible. Beginning with the recommendations of King (1995) the research community and publishers have adopted a series of guidelines (for example DA-RT, 2015) culminating in the implementation of in-house and third-party certification workflows by top journals (Christian et al, 2018).
The American Political Science Association’s (ASPA) A Guide to Professional Ethics in Political Science states that:
Top journals, including the American Journal of Political Science (AJPS) and the American Political Science Review (APSR) require authors reporting empirical and quantitative results to deposit data, software and code, and other information needed to reproduce findings (ASPR, 2019). As discussed in detail below, AJPS is the only journal to implement a third-party certification of replication packages.
As highlighted by Dafoe (2014), the replication standard in political science is in part motivated by a number of high-visibility controversies in the social sciences. He cites the example of an influential paper in economics that was discovered three years later to have errors, arguing that the availability of the replication file for the study would have at least accelerated the identification of potential errors.
In January 2016, 27 political science journal editors signed the “Joint Statement on Data Access and Research Transparency” (DA-RT, 2015) that includes a number of requirements related to the APSA ethics guidelines for authors centered around data citation, transparency of analytic methods (e.g., code), and improving access to data and other research materials.
Example: American Journal of Political Science¶
Christian et al (2018) describe an operationalization of the replication standard implemented by the American Journal of Political Science (AJPS) in collaboration with the Odum Institute for Research in the Social Sciences and the Qualitative Data Repository (QDR). AJPS is one of the top-ranked political science journals with an ISI ranking of 1/169 in 2017. In 2012, AJPS adopted guidelines for authors to deposit replication packages in Harvard’s Dataverse. According to Jacoby (2017), due to concerns about the quality of deposited materials, AJPS implemented the third-party certification process starting in 2015.
Christian et al describe the basic workflow is as follows:
The author submits a manuscript to AJPS for peer-review. If accepted, the author is required to submit the replications materials to AJPS Dataverse.
Once the replication materials are available, the editor contacts Odum/QDR to begin the curation and verification process.
Data is reviewed per a data quality review framework. Statistical experts perform verification by executing the analysis code and comparing the output to tables and figures reported in the manuscript.
A “Verification Form” is returned to the editor including the results of the review process and any errors. The editor notifies authors to correct problems.
Once the data review and verification process is complete, the editor issues the acceptance notification and the materials are published in Dataverse (including DOI).
The paper and replication package are linked via DOI.
The authors further note that only 10% of submissions pass review without the need for revision and that, as of 2019, the process requires roughly 6 hours of effort for a single manuscript.
In response to his presentation to the NAS Committee on Replication and Reproducibility in the sciences, Jacoby (2017) notes that:
Odum archive staff handle both data curation and verification (statistical)
Errors are generally not serious (e.g., lack of documentation or tables that don’t reproduce exactly).
Mean number of resubmissions is 1.82
The verification process is paid for by the Midwest Political Science Association
AJPS requires only the data used in analysis (i.e., not all of the data collected)
Anecdotally, he has had feedback that the resource is invaluable for methodology courses (See also Janz 2016)
In 2018, the Odum Institute was awarded a $500,000 grant from the Sloan Foundation to improve and automate the verification process.
Jacoby (2017) notes that other political science journals have in-house verification processes, typically relying on graduate students. In these cases, it’s likely that the focus is on re-runnability of the code without necessarily comparing the reported results. In response, an example was raised from the field computer science where reproducibility reports are written by community reviewers, notably Information Systems journal (Chirigati, 2016).
The AJPS provides a “Quantitative Data Verification” checklist for the preparation of replication files that includes:
README file containing the names of all files with a brief description and any other important information regarding how to replicate the findings (i.e., the order files need to be run, etc.)
Includes a Codebook (.pdf format) with variable definitions for all variables in the analysis dataset(s) and value labels for categorical variables
Includes clear information regarding the software version used to conduct analysis
Includes complete references for source datasets Includes the analysis dataset(s) in a file format readily accessible to the social science research community (i.e., text files, delimited files, Stata files, R files, SAS files, SPSS files, etc.)
Includes a unique case identifier variable linking each observation in the analysis dataset to the original data source
Includes software command file(s) for reconstructing the analysis dataset from the original data source and/or extracting and merging multiple original source datasets, including information on source dataset(s) version and access date(s)
Includes commands needed to reproduce all tables, figures, and other analytical results presented in the article and supplementary materials
Includes commands/instructions for installing macros or packages
Includes comment statements used to explain the analysis steps and distinguish commands for tables, figures, and other outputs Includes seed values for any commands that generate random numbers (e.g., Monte Carlo simulations, bootstrap resampling, jittering points in graphical displays, etc.)
Includes any additional software tools needed for replication (e.g., Stata .ado files and R packages)
Harvard’s Dataverse includes hundreds of Political Science replication packages, including those verified through the Odum/QDR workflow.
AJPS replication policy https://ajps.org/ajps-replication-policy/
AJPS Quantitative Data Verification Checklist. 2016. https://ajpsblogging.files.wordpress.com/2019/01/ajps-quant-data-checklist-ver-1-2.pdf
AJPS Guidelines for Preparing Replication Files, https://ajpsblogging.files.wordpress.com/2018/05/ajps_replication-guidelines-2-1.pdf
APSA Guide to Professional Ethics, Rights and Freedoms https://www.apsanet.org/portals/54/Files/Publications/APSAEthicsGuide2012.pdf
ASPR. (2019). Submission Guidelines. https://www.apsanet.org/APSR-Submission-Guidelines. Accessed February 8, 2019.
Barba, Lorena A. (2018). Terminologies for Reproducibly Science. https://arxiv.org/pdf/1802.03311.pdf
Christian et al. Operationalizing the Replication Standard: A Case Study of the Data Curation and Verification Workflow for Scholarly Journals https://osf.io/preprints/socarxiv/cfdba/
Dafoe, 2014. Science Deserves Better.
DA-RT. (2015). Data Access and Research Transparency (DA-RT): A Joint Statement by Political Science Journal Editors. https://doi.org/10.1177/0010414015594717
Jacoby, William. 2017. Presentation to National Academy of Sciences Committee on Replication and Reproducibility in the sciences. https://vimeo.com/252434555
Janz, 2016. Bringing the Gold Standard into the Classroom: Replication in University Teaching. https://doi.org/10.1111/insp.12104
Fernando Chirigati, Rebecca Capone, Rémi Rampin, Juliana Freire, Dennis Shasha. (2016). A collaborative approach to computational reproducibility. Information Systems, Volume 59, 2016, https://doi.org/10.1016/j.is.2016.03.002.
TOP guidelines (https://cos.io/our-services/top-guidelines/)