Reproducibility, Replicability & Reliability made easy (hopefully!)
After attending a series of workshops at the British Neuroscience Association conference, I realised how little I really knew about Open Science. I have decided to curate a list of resources to help me (and hopefully others), learn more about how to make our research more open. This list is a work in progress (and by no means exhaustive) and suggestions for resources to be added are more than welcome (cemail me or tweet @AleLautarescu)
Why open science
A good starting point is reading Munafo et al. (2017) - A manifesto for reproducible science https://www.nature.com/articles/s41562-016-0021.pdf
"Open science refers to the process of making the content and process of producing evidence and claims transparent and accessible to others."
Open science publications get more citations & media coverage
Badges: A number of journals have started to offer Open Science Badges to signal and reward when underlying data, materials, or preregistrations are available. (https://cos.io/our-services/open-science-badges/)
Practicing open science is quickly becoming a requirement from funding agencies and journals (e.g. https://cos.io/our-services/top-guidelines/)
Open science practices in hiring and promotion (there are several fellowships specific for researchers who practice open science
Read more @How open science helps researchers succeed https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4973366/
What can you do
It is important to understand that open science is not an all-or-nothing approach. You can start by implementing only one or two of the suggestions below.
You can join your local UK Reproducibility Network hub. A list of hubs can be found at https://www.bristol.ac.uk/psychology/research/ukrn/networks/
Planning your research project
Conduct a systematic review when formulating your hypotheses. Systematic reviews reveal uncited studies, in the way that reading reference lists may not.
Conduct a sample size estimation (Low statistical power increases the likelihood of obtaining both false-positive and false-negative results!). You can also find and use existing datasets (https://toolbox.google.com/datasetsearch)
Preregistration of study design, primary outcome and analysis plans. You can pre-register on the Open Science Framework (http://osf.io/) or AsPredicted.org. Journals have also started adopting Registered Reports(i.e. peer review before results are known). This helps eliminate the bias against negative results & prevent HARKING (i.e. hypothesising after results are known) and P-hacking (i.e. analysis decisions made with knowledge of the observed data). You can find a list of journals accepting Registered reports at https://cos.io/rr/. Lastly, there are other options for publishing your peer-reviewed protocol (such as https://bio-protocol.org/default.aspx)
Analysing your data
Attend training courses in methodology and stats (this will help you understand what p-values actually mean, the importance of statistical power and effect sizes etc). There are some good free online courses on Coursera (such as https://www.coursera.org/learn/statistical-inferences) and other resources that you can implement, such as StatCheck - which checks for errors in statistical reporting (http://statcheck.io/)
Use open-source software (such as R) for data analysis. As the analysis takes the form of scripts, you already have a full, reproducible record of your analysis path (which you can then share). For info on how to make your data reproducible with R, see online courses on Open Science & Reproducibility (https://github.com/cbahlai/OSRR_course & rmarkdown (https://github.com/libscie/rmarkdown-workshop)
Make sure you are explicit about what analyses are hypothesis-based and which are exploratory (pre-registering helps with this).
Publishing / Sharing your data
Publish in open access journals
Share your study protocol (https://www.protocols.io/, or Nature's protocol exchange)
Share your code (https://github.com/) . For a friendly intro to github see https://kirstiejane.github.io/friendly-github-intro/
If possible/applicable, share your data (https://datahub.io/)
Improve the quality of reporting to allow replication (reporting guidelines on http://www.equator-network.org/)
Preprints (such as https://www.biorxiv.org/). These have pros and cons so make sure you read up on it before deciding.
Publish negative findings! Don't selectively report. There's a good article in Nature talking about how research cannot be self-correcting when information is missing (https://www.nature.com/articles/d41586-017-07325-2). Several journals have started to publish negative findings (e.g. European Journal of Neuroscience, Brain and Neuroscience Advances, BMC Research Notes). I have yet to find an exhaustive list but will update this if I do.
Be clear about what each author has contributed to the paper. A good resource with examples of roles to allocate (e.g. conceptualization, resources, formal analysis) can be found on https://www.casrai.org/credit.html
Great list of resources on https://opensciencemooc.eu/resources/
Image by https://commons.wikimedia.org/wiki/File:Knowledge-sharing.jpg