I agree that many of the questions seem pretty scattered, with heterogeneous collections of available choices. In one case “Scikit-Learn” and “Sklearn” were listed as separate options although those are just two names for the same thing.
I also agree that brainstorming the survey creation here or in a similar forum would be useful. Can you share anything about the current survey design process?
The other, perhaps more controversial suggestion I would have is to have more questions that are about opinions rather than “facts” like “what IDE do you use”. There are quite a number of issues on which, shall we say, distinct viewpoints seem apparent in many discussions here. 
Packaging is a prominent one, but also (just off the top of my head) type hints, backwards compatibility concerns, the usefulness of a “batteries included” library in an era of widely available fast internet, etc. I think it would be valuable to leverage an annual survey to take the temperature of the userbase on such issues[1]. This could be done by having sequences of 2-3 related questions on each of a few such topics.
As an example, on this survey there were one or two questions about what version of Python is used and how/when it is upgraded. But it would be useful to have responses to questions like “How often do you refrain from upgrading Python due to concerns about backwards compatibility? (Multiple times a year/once a year/less than once a year)” or “How do you feel about the level of backwards compatibility maintained from one Python version to the next? (Python should move faster to remove cruft and support modern workflows, even if it means sacrificing some backwards compatibility/The level of backwards compatibility seems just right/Python should prioritize backwards compatibility more, even if it means delaying potential improvements)”.
The other suggestion I would have is that it would be nice if a form of the data as raw as possible (i.e., anonymized but not aggregated in any way) were made prominently available. This allows third parties to slice and dice the data in different ways than whatever may show up in official presentations of the results.
while of course remembering that the survey response group likely doesn’t represent the entire Python userbase ↩︎