📗 Reproducibility in Neuroimaging

📗 Reproducibility in Neuroimaging#

This chapter aims to highlight the importance of reproducible research practices and briefly explain what will be covered in the rest of the book.

🤨 Why Care about Reproducibility?#

  • Science builds on science, without reproducibility, findings can’t be trusted or extended.

  • In computational neuroscience and neuroimaging, findings are often results of complex pipelines, built with multiple software tools, and parameter choices.

  • Without access to code, even small analytical tweaks can become untraceable.

“Science is a social enterprise: independent and collaborative groups work to accumulate knowledge as a public good.”

Munafo et al. (2017) [1]

🤔 What’s at Stake?#

  • Credibility of research outputs.

“The authors of research papers have no obligation to share their data and code, and I have no obligation to believe anything they write.”

Andrew Gelman (professor of statistics and political science at Columbia University)

  • Wasted time taken to re-implement procedures reported in previous works.

  • Reduced impact of research findings, as they cannot be easily verified or built upon.

🔀 Categorizing Reproducibility:#

Botvinik-Nezer and Wager[2] identify three types of reproducibility:

  1. Analytical reproducibility: Reproducing findings using the same data and same methods.

  2. Replicability: Finding similar results in independent datasets using similar methods.

  3. Robustness to analytical variability: Obtaining consistent results using different analytical approaches.

Tip

The goal of this session is to introduce practices for ensuring analytical reproducibility. This can serve as a foundation for achieving replicability and methodological robustness.

Gorgolewski and Poldrack[3] cover 3 major topics in open science (see Fig. 2), with implications for reproducibility:

  1. Data: Access to the original data is required to examine analytical reproducibility.

  2. Code: Access to the implementation scripts is also needed.

  3. Papers: Access to documentations of methods, and interpretations of results.

../../_images/pillars.svg

Fig. 2 Three pillars of open science.[3]#

Tip

In this session, we’ll cover topics addressing code openness.

🚩 Let’s Start#

Now that we’ve covered why reproducibility matters and what this session will include, let’s jump in. We’ll focus on two main topics:

📑 References#

Contents