Talk:People and feedback are hard to do right

Judging specific cases doesn't really belong to an essay. --Nemo 05:55, 27 August 2014 (UTC)

Dunno about this

So to avoid bias, you need to let people write whatever they want in one big box, which you'll then process and interpret—which adds an enormous opportunity for bias right back into the system. It's also harder to release it publicly, because (as anyone who processed AFT comments knows) people frequently add personal information like e-mail addresses, so a dishonest surveyor could even cover up his biased reporting by declaring anything that disagreed with him as being "personal". By contrast, the data for anyone replying on a w:en:Likert scale can be released immediately, since there's nothing potentially private about choosing a radio button.

It's also expensive and slow: a thousand text responses means that you have to read and classify a thousand text responses. If you care about reliability, then you need to have every response read and categorized by at least two people, so that one person's (conscious or unconscious) bias doesn't cause problems. A scale that doesn't require free-form text scales to millions, and you can get some answers immediately.

Text has a problem with self-selecting responders: people who don't feel competent in the local language are less likely to reply at all. The same is true for people who are busy at the moment or on a device that makes typing inconvenient.

I think that the right format for feedback depends on what you're trying to learn. A blank textbox might work if you want to know all the semi-random things people want to say. A blank textbox is not the best approach if what you want to learn is, "Do you want to turn off this feature?" A single button, "Click here to turn this off" would be more effective for that purpose. Whatamidoing (WMF) (talk) 22:04, 4 September 2014 (UTC)