I don't use star as a reward for the developer, but rather as a bookmark.
I star a repository in case I have to do something with the code (or repository) itself: when I modify the code (either to scratch my own itch with a bugfix/missing feature, or to also upstream it), when I need to check the code itself for something, or when I create issues, either to ask for help due to missing documentation, or report a bug. (+1: neodrama repo is also starred, as it brings the nasty-tasty dev dramas. But it's more like a blog, in repo form)
In case a software just works as it should, and I don't have to mess with the code deeply, then I don't star - I see no reason to do so.
Regarding downloads, I would bet some CI systems / builds are the primary source. These don't really count as installations of it's just tests running in CI, imho anyway
An ancient post [1] for the defunct npm blog offers some details:
> So the count of “downloads” is much larger than the number of people who typed “npm install yourpackage” on any given day.
I'm sure the particulars have changed over time, but directionally, npm download numbers aren't a proxy for "# of developers using my package".
[1]: https://blog.npmjs.org/post/92574016600/numeric-precision-ma...