Reproducible Analytical Pipelines

Is RAP relevant to me?

Would it be a nightmare to have to go back and rerun your process from the beginning if you found a mistake?
Do you have to make a lot of manual edits to code before each run?
Is there a lot of repetition in your code?
Would a new person find it difficult to understand the process?

Automated statistical and analytical processes that are:

In order to achieve these benefits, at a minimum a RAP must:

Developed by Government Analysis Function to help analysts and analyst leaders understand why RAP is important and how to deliver it.
Vision: RAP is the default approach to analysis
Three goals
- Tools
- Capability
- Culture
Scottish Government RAP Strategy Implementation Plan 2023

School Information Dashboards
10 data sources
Data cleaned, linked and analysed manually in Excel
Dashboards created in Tableau
Updated twice a year; each update took three weeks of work for three statisticians - longer if errors were found

RAP Support team; find people who can help you or are willing to be a critical friend
What are we aiming for, whilst also being open to this changing as we go
Work with data providers; don’t feel restricted by the current setup. If something doesn’t make sense and you’re having to do extra work to accommodate, talk to people. Often you’ll find they’re also having to do extra work to produce the output in the format they think you need it in!
This links to an important point - RAP isn’t about replicating / shoehorning an existing process into new software. Step back - look at what you have to start with, what you’d like to end up with and the most logical way of getting from one to the other. It’s easy to assume processes are complex and convoluted for unavoidable reasons, but often they’re just a result of years of different analysts adding new layers on top of old ones until it doesn’t make sense anymore!

Organised folder structure

Writing manual processes as code

Standardised naming conventions

Functions

Open-source software

Version control using git

Relative file paths

Open code on GitHub

Faster
- Previously took three weeks for three statisticians (twice per year)
- Now takes at most one day for one statistician (twice per year)
More accurate
Reduced risk
Well documented
Developed skills to apply to other projects

Would it be a nightmare to have to go back and rerun your process from the beginning if you found a mistake?
- Git

Do you have to make a lot of manual edits to code before each run?
- Setup script