Sleep Cycles (2)
or "Why you shouldn't try and design surveys in under an hour"
Well, I've finally given up hope on more results for my sleep survey. I'm still not quite certain whether 74 is above or below what it was expected, especially considering the number of people I spammed with emails.
Discussion of the results is going to be split across 2 or more posts, as there are some nice natural groupings of thoughts, and I'm still rather busy this month.
Anyway, the first thing the results showed were the sheer number of floors in my survey. The first, and most obvious, was the lack of resolution in most of the fields. This was mainly an artefact of the way Google Forms works (in that it can only have 5 columns in a grid) and that I didn't want to have to spend time processing text inputs. There are a variety of solutions to this problem, from drop downs, to writing my own survey form from scratch. As it is, the data I got is pretty much useless for proper analysis - the sleep durations might be normally distributed, but I could probably also show that they follow the coefficients of a binomial expansion.
So, what would be the way I tackle this now, with the benefit of hindsight? Well, I'd probably not do an electronic survey for preference, but rather have a large sample of people from a variety of different backgrounds participate in a month long study where a little wrist-watch or something records when the fall to sleep, and when they wake up. Interestingly, I still haven't been able to find any papers performing a study of this sort; I haven't looked too hard, and there's a good chance my keywords are using the wrong terminology.
Regardless, have a lot more flexibility for the user; text boxes where they can type in most of the numeric values, and spend a little while working through the results manually to validate them all. I mean, on the interwebs, no-one is going to deliberately try and break my results, right? Right?