4 ms·

It is indeed a metric, but it's never once been a metric that I've used when considering github projects. To me, important metrics are: 1. Project maintenance

It is indeed a metric, but it's never once been a metric that I've used when considering github projects. To me, important metrics are: 1. Project maintenance -- when were the last commits and how important were they? 2. Issue tracking -- how active are the maintainers in covering issues? 3. Actual code/test quality As far as stars go, I've only ever looked at them as the number of people who decided to "like" something, but that's never a factor in my consideration since I don't know when or why they did it, whereas the other three things I listed are much, much more direct ways of viewing the overall "health" and usefulness of a project. My perspective is apparently an unusual one, though, which is why I wanted to ask and make sure I was understanding the linked article properly.

anonytrary 8y ago
> My perspective is apparently an unusual one It's not unusual, it's just very high-effort and it makes no sense to ignore stars when N is large. In other words, when time is limited, low-resolution information becomes very valuable. > As far as stars go, I've only ever looked at them as the number of people who decided to "like" something, but that's never a factor in my consideration You might think this, but I am skeptical. For example, if you had to review hundreds of projects and you only had time to run your fair, thoughtful analysis on 10 of them, you would effectively be forced to look at stars to sort the list before processing it, as it's the next-best metric for judging a project. This is the same reason why job and college applications go through shallow sorting methods before they are reviewed by humans. You might initially think that GPA, years of experience, and other numerical metrics are shallow and do not tell you anything about somebody, but they are very useful when you have tens of thousands of applicants and you need to filter the list down to hundreds. My point is that sometimes you need to lower the resolution of your analysis because you don't have a lot of time. In practice, most people (and even computers) don't have time to look at everything, so they use any useful information they can find (whether it's conscious or not). I like to think that the number of points on an HN post doesn't affect my probability of reading the article, but if I took the data and did the analysis, I'd probably find that isn't the case and that I am biased towards spending my limited time on posts with more points.

anonytrary 8y ago

> My perspective is apparently an unusual one It's not unusual, it's just very high-effort and it makes no sense to ignore stars when N is large. In other words, when time is limited, low-resolution information becomes very valuable. > As far as stars go, I've only ever looked at them as the number of people who decided to "like" something, but that's never a factor in my consideration You might think this, but I am skeptical. For example, if you had to review hundreds of projects and you only had time to run your fair, thoughtful analysis on 10 of them, you would effectively be forced to look at stars to sort the list before processing it, as it's the next-best metric for judging a project. This is the same reason why job and college applications go through shallow sorting methods before they are reviewed by humans. You might initially think that GPA, years of experience, and other numerical metrics are shallow and do not tell you anything about somebody, but they are very useful when you have tens of thousands of applicants and you need to filter the list down to hundreds. My point is that sometimes you need to lower the resolution of your analysis because you don't have a lot of time. In practice, most people (and even computers) don't have time to look at everything, so they use any useful information they can find (whether it's conscious or not). I like to think that the number of points on an HN post doesn't affect my probability of reading the article, but if I took the data and did the analysis, I'd probably find that isn't the case and that I am biased towards spending my limited time on posts with more points.