Open Science Bootcamp

Description

Research standards in the social sciences are rapidly shifting and increasingly emphasise transparency and reproducibility. Yet, early-career researchers still receive little formal training in open-science practices. The goal of this summer school is to fill this gap and give participants a firm grasp on practices that will make their own research more transparent and reproducible. This will not only make it easier to adhere to new open-science norms and requirements, but also serves ‘selfish reasons’ (Markowetz, 2015): learning how to work transparently and how to minimise bias and error in one’s research helps prevent catastrophes, can greatly increase efficiency in the long term, and makes collaboration easier.

The course has three core subjects: Data sharing, bias prevention, and error prevention.

Data sharing: Research data should be FAIR – findable, accessible, interoperable, and reusable. But data from human subjects often contain sensitive or identifiable information that prohibit open sharing. We will address both of these problems by covering 1) how to document data properly so that others (and future you) can reuse it, how to find a reliable repository, and how to make your data discoverable, and 2) strategies for dealing with sensitive data, including different methods for de-identification and different options for restricted access.

Bias prevention: Research results should be free from researchers’ conscious and unconscious biases. We will cover three techniques that help avoid fooling oneself and others: preregistration, Registered Reports, and analysis blinding. For each of these, we will not only look at how they work in principle, but also at common problems and how to address them.

Error prevention: Research results should be as free of error as possible. But typical research workflows in the social and behavioural sciences are highly error-prone, involving numerous manual steps—each susceptible to human error—very few routine checks to catch these errors, and minimal documentation for back-tracing. Not surprisingly, computational reproducibility (verifying results by re-running the same analyses on the same data set) is shockingly low. To address this issue, we will cover reproducible practices including effective project documentation, dynamic documents in RMarkdown (mixing code and text to eliminate all copy-pasting), and version control with GitHub (backing up every step of the project).

Participants are requested to bring their own laptop and install R and RStudio before the workshop (these are freely available online).

Ideally, participants will also bring an own research project that they are currently working on or have worked on in the past (but this is not strictly required).

Target audience

The course is targeted at researchers in the social and behavioural sciences who want to make their work more transparent and reproducible, or who have experienced problems with implementing open-science practices. The primary audience are early-career researchers—PhD students, postdocs, and potentially advanced Master’s students—though participants from all career stages are welcome.

We will be using R and RStudio. Prior knowledge in R is not required, but helpful.

Aim of the course

After the course, you will be able to:

• write effective preregistration plans for your own research projects and handle common problems (such as post-hoc analysis decisions)
• document research projects and data (e.g., using codebooks) to a level that allows them to be independently understood and reused
• find solutions for sharing sensitive data without sacrificing data safety
• design workflows that remove common sources of error and/or help identify errors quickly
• produce dynamic documents with RMarkdown
• understand the basic principles of version control (using Git) and create version-controlled projects in RStudio

Tags

• M&S
• Open Science
• reproducibility
• replication
• open data
• FAIR
• data sharing
• data management
• preregistration
• Registered Reports
• analysis blinding
• reproducible code
• version control
• Project Management
• meta-data
• codebooks
• GitHub
• R Studio