Another Online Survey on Online Action Last week's "Numbers Guy" column simply by Carl Bialik of the Wall Roads Journal Online seen another online survey a good online activity, in this case about online beer sales to adolescent children. Bialik noticed something inside release that the several other outlets should have at a minimum checked. The "online" example was not drawn from just about all teenagers, but rather through teenagers who obtained volunteered to participate for online surveys. It was a good non-random online survey of an on the net activity, something together with the potential to create a major bias. I will have Bialik take it from there: People who admit participate in online surveys seem to be, by definition, Internet surfers, something that not all kids are. (Also, people that actually take the time to carry out such surveys may be more likely to be energetic, or heavy, Online surfers.) It's safe to say that kids who use the Internet consistently are more likely to shop online than those who don't. Teenaged Research Unlimited said to me it weighted laptop computer results to adjust for age, sex, race and geography regarding respondents, but found no way to adjust with respect to degree of Internet wearing. Regardless, the survey found that, after weighting, just Some.1% of the 1,001 participants bought alcohol on the web -- compared with 56% who previously had consumed alcohol. Making the questionable forecasts that their trial was representative of most Americans aged 17 to 20 with internet access -- and not just those with any time and inclination to participate in online surveys -- the study concluded that 551,000 have been buying alcohol web based. Bialik goes on to raise all the more fundamental problems with the making from the survey bring in -- the Wine and Alcohol Wholesalers of United states of america -- a group that consistent with Bialik, "has long fought endeavors to expand online revenue of alcohol.In . For one, their subject claims that "Millions of Kids Buy Internet Beer," while even an questionable survey appraisal adds up to just 551,000. One point not grown in Bialik's excellent element is that the survey reports a margin with error ("plus or devoid of three percentage points"). All the margin of mistake is a measure of occasional sampling error, which inturn applies only high is a random some sample. This survey was based on a sample drawn from your volunteer panel, not really random sample investigation. I have raised this challenge before (here and also here). Yes, occasional sample surveys confront challenges of their own, if a sound statistical schedule exists for coverage a "margin of error" intended for non-probability samples, I am not yet aware of it.