Differences between Statistical programmers and Software developers

[EllieSparling]

We know there is a tried and tested way to utilize git within software development, and many of these approaches are also used in open source package development. To understand potential future ways of working in git for statistical programmers, lets discuss the differences between the two roles, requirements and workflows.

[langkabh]

I think the main difference between the work of software developers and statistical programmers are the respective products: Software developers work towards a single product (e.g. software, application, website) with a single code stack of which there is - ideally - a production version at every point in time and of which there is no "final" version. Sstatistical programmers on the other hand work towards multiple products (e.g. data monitoring reports, DSUR, CSR) with a single code stack of which there only needs to be a (partially) production-grade, final version at specific delivery time points. Ok, maybe it's product and publishing lifecycle (continuous vs. one-off publishing). In my opinion, the main consequence of this is a difference in how work branches: In software development, you have anabranching, i.e. you start with the master, work branches out and lateron merges back into master - like anabranches/side arms in a river which join the main stream again and then continue the general journey. In statistical programming, you have off-branching, i.e. you start from a master/template copy, work branches out and then completes as a separate end - like branches of a tree which spread out and each come to a different end and never join the stem again.

[aassuied-ps]

I'm not a software developer, so I'm not sure I can speak about all the differences. 😊 However, I think I can say that statistical programming using Git is different from working on an open-source package, which can certainly benefit from the same approaches as software development as you say. And I agree with @langkabh on the differences between one/several products.

Speaking of the fact that statistical programming activity often means having the same program running (sometimes differently) across several products/milestones (CSR, DMC, DSUR, etc.), I'm wondering if considering CI/CD (GitHub Actions) could be beneficial to have automation (run all tests on several environments/products or make it specific). It's not purely Git related but since we're planning to talk about tools, why not this one?

A difference may be that some statistical activities are like "task based" and needs to be delivered very fast (ie. HA requests), and the git procedures (branches, merges, CI/CD) could slow the process down.