On 23/08/2014, john whelan <jwhelan0...@gmail.com> wrote: > In an ideal world the way to get a proper random sample would be to select > OSM mappers randomly then message them. Hopefully you'd get better than > 90% response rate to keep it statistically meaningful. > Reality is you might be lucky to obtain a 2% response. So the next best > thing is OSM-talk and hopefully he'll get the 1,000 responses which he > needs to make it statistically meaningful.
Remember that we're sending emails and that the task can be automated, so a 2% response rate isn't really an issue. And it's much better to individually contact a uniformly random sample than to globally contact a biased sample (only a particular kind of contributor follows mailing lists). As a added bonus of contacting individually, you already know the person's mapping profile. Here are a few proposed guidelines to keep things in check though : * treat a survey like an import: it should be community-reviewed and accepted before going ahead. Be transparent, be usefull, be well writen, be multilingual, etc. * set target request and response counts ahead of time, and stop sending requests whenever one of the counts is reached * provide a way to opt-out any future survey via you osm account Any other do's and don't ? _______________________________________________ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk