I don't want to come across as rude, but I am fascinated how easy people are buying into these numbers. From my perspective the survey is in its current form far less informative than the original anecdote.
There is absolutely zero information on the data source he or Survate was using. The population seems to be selected based on the single factor age - meaning they might be getting the data from the easiest, cheapest accessible data sources available. In the end most of the responses could be coming from 30 y/o guys in India clicking on random answers to earn their 5c on mTurk.
In contrast to that, the original anecdotal evidence had the full story, provided the exact context on why and how the data was generated. But data without context is no data.
Survata is a survey-wall -- meaning you have to answer a survey before you see a given article. They have about 20 publishers signed up, and they're all non-spammy content. Users are all US-based, which can be verified via geo-ip, and there's little reason for people to spoof that here.
I never claimed to have more anecdotal evidence -- the whole point is to try to validate the claims via some form of data collection. It was cheap and fast to do.
Garry's explanation is a good one. The data for this survey was collected via surveywalls (example at [1]), which let visitors access premium content online for free in exchange for answering a few questions.
All respondents here have US IP addresses and self-report age in the 13-25yr range. We generally see honesty rates of 90% or higher to questions for which we can verify the answer (e.g. "Which OS are you currently using?" or "Who is the President of the US?").
There is absolutely zero information on the data source he or Survate was using. The population seems to be selected based on the single factor age - meaning they might be getting the data from the easiest, cheapest accessible data sources available. In the end most of the responses could be coming from 30 y/o guys in India clicking on random answers to earn their 5c on mTurk.
In contrast to that, the original anecdotal evidence had the full story, provided the exact context on why and how the data was generated. But data without context is no data.