add "other" field to selection questions

[fricklerhandwerk]

there was a lot of feedback where people couldn't submit their outlier data, and a major value of the survey is to give people an opportunity to get heard

[GuillaumeDesforges]

Copy of my thoughts from the previous PR:

I'm strongly against it. It circles back to open-ended questions being costly and hard to analyze.

IMO one of the reasons we could make the most of the survey in 2024 was that we put more thought in the question prompt and answer choices. If anything is missing, it should be added to everybody.
A very probable risk I foresee is that many people will prefer using the "other" choice to input a choice that would usually fall into one of the provided category. While I agree that it is important to be open and inclusive, I think it is our responsibility to provide usable data as well. Basically, if people feel a question needed an "Other" field, we need to work on the question instead, e.g. provide more choices or clarify the prompt of the question.

But in the end, whoever implements the survey is free to do what they want, so I'll let the PR open until next survey is designed.

[fricklerhandwerk]

Providing more choices is not always an option. People rightfully complained about questions that had dozens of options were confusing, and finding all the possible enum answers is an uphill battle. Do you really want to list 50 languages people could be working with? Most of them will be hyper niche

[GuillaumeDesforges]

Fair point. Then the question is "which questions should have an 'Other' choice?" which I find much more reasonable.

[GuillaumeDesforges]

Also, is it an 'Others' choice with a free text input or without?

[fricklerhandwerk]

I suggest that all single and multi select questions that have an unbounded domain should have a freeform "other" field.

Example: Which languages do you program in?

Counter-example: Which Nix experimental features do you use?

The rule of thumb should be to have a fixed number of most-often selected answers from past years be hard coded and leave the tail end open. We can throw ML at the rest to do a clustering, and whatever is left is almost always a curious outlier worth an honorable mention.