aliquote.org

Develop good habits from the start

July 10, 2024

Last month I came across a blog entry posted on HN: A Bunch of Programming Advice I’d Give To Myself 15 Years Ago. I couldn’t help thinking that some of the advices discussed here are good ones even for managing a statistical project:

All the above apply equally well to statistical analysis. Even if should select the best statistical package for the task at hand (while running a t-test is a common task that most software offer nowadays, analyzing survey data is not something to be taken lightly). As someone who value good software and the command-line (not much because I like typing commands but because I can record them for later use or review) I’ve almost always relied on R, Stata and a few dedicated software (e.g., jags, stan, xgboost). Writing code also means you should value your text editor (Emacs first, then Vim, then Neovim in my case). About the second point, I faced it both when teaching medium-to-hard techniques to graduate students or during corporate training, and by presenting my own analysis results, most of the time resulting from multivariate analysis. I was using Git mostly for mono repos so that I could track changes along project timeline (my changes, and their – i.e., the client or colleagues – changes as well). As for debugging, especially making it easier to detect errors or misbehaving programs, I believe it comes with experience and good practices.

Other rule-of-thumbs I adopted over time:

To sum up, developing good habits from the start is always worth it. I can’t count the number of times I’ve been asked to produce a result with slightly updated data 6 or 12 months after the end of a project: having a solid, well-tested script has made things easier every time.

♪ Passenger • Wicked Man’s Rest

See Also

» Cochran-Mantel-Haenszel test » How many permutations » Algorithms for statistical computing » Permutation test in Lisp » Revisiting Random Forests