Forgive me if this is poor-protocol but I am cross-posting from Cross Validated Meta because it was recommended there that this 'feature-request' belongs on this meta. I had done considerable work posing the arguments and I would rather reuse those arguments than start fresh.
I had posted it there because it contains anecdotes and examples that possibly only a statistics educated SE user can appreciate. The feature request portion however had broad application and I agree that it belongs here. Therefore here is the post for your thoughtful consideration.
I've noticed that it is difficult to get questions migrated from SO to CV. I've already read that this doesn't occur often 'in the wild' but as 'Big Data' and 'Analytics' become trendy in the software industry I would not be surprised to see an increasing trend of migration worthy question appearing on SO.
Anecdotally I have found that there is an increasing number of people in Silicon Valley/ Bay Area that did a quick MOOC, or read one of many machine learning/ data mining survey books (k-means/OLS/LogReg/DT/SVM/RF), that only describe the most common machine learning techniques as plug-and-play software tools in R and doesn't leave them with a sense of how much more they ought to learn. For instance I listened to a talk by a chief data scientist at a social gaming company that explained his investigation of a problem as 'trying a bunch of things' and settling on a linear regression fit because the R-Squared 'was good'. I have other anecdotes as well, for example the emerging trend of 'advanced analytics' vs 'analytics' to mean statistical/machine learning techniques because so much of what people call analytics in the software community is just average/max/min (without acknowledging population vs sample) and various elementary mathematical calculations called 'key performance indicators'(KPI).
I'm not sure how I would show that there is an inflation of claims to understand statistical/ machine learning topics given the data we have but I strongly suspect that it is occurring and if this is true, evidence should be mounting somewhere within SO. The challenge is that well-meaning SO moderators may not know how to identify these.
For example I have flagged 4 SO questions as migration candidates to CV with the most recent flag occurring more than a week ago.
- One of them, a feature selection, question was rejected because 'declined - a moderator reviewed your flag, but found no evidence to support it' and at the same time one of the close-voters, AGS, has commented on it 'This question appears to be off-topic because it is about machine learning theory / stats, and not programming related'. That looks like it is some evidence.
- One flag that has been open for 11 days now is about computing prediction intervals after using cross-validation.
- Another 11 day open flag is about hyperparameters in a Gaussian Process.
- The first flag of this kind that I have raised was about selecting an anomaly detection model. It wasn't until I demonstrated that it was CV worthy by answering it as one that it was migrated.
To recap, from my extremely limited sample size of 4 flags, the only successful migration was the one that was subsequently demonstrated to be a CV question by providing a CV answer first. I attribute this to the limited experience of SO moderators in CV topics.
As a solution I would propose the idea of a bridge moderator. Where I went to school for my undergraduate there was the concept of a bridge faculty. For instance there is a bridge faculty member for Math/CS and one for Stat/CS and both are able to liaison between their bridge disciplines. At my job which has many full time statisticians and full time software engineers, I act in a similar liaison capacity between Stat/Engineering, though this is a relatively new idea here. Since moderator level reputation may be rare enough in one discipline, let alone two, I would propose that a moderator level reputation on the migrate-to side and some lower threshold reputation on the migration-from side might be sufficient to strike a balance. Perhaps there are better ideas.
I believe that we can certainly continue to demonstrate the migration-worthiness of questions by first answering them but I also believe that it is an inefficient way to crowdsource this problem when cross topic contention will only increase with the increasing portfolio of SE topics. I would ask the community to consider more efficient options.