Saturday, July 6, 2013

Condemn Snowball Samples to Hell

A snowball’s chance in Hell is mighty slim. Unless you have along with you a perpetually-powered refrigerator. Almost as slim are the odds of a snowball sample providing a retailer adequate guidance from consumer survey findings. Unless you have along some high-powered statistical expertise.
     A snowball sample is gathered by asking people who take your survey to convince their friends to also take the survey. Snowball samples are used most often when the types of people you’re aiming to question are difficult to find or would probably refuse to participate unless invited by a trusted pal. In criminal justice research, snowball sampling has been used to survey drug addicts and prostitutes.
     In the current issue of International Journal of Consumer Studies, researchers from University of Georgia report findings about people who regularly buy at private sale sites. The main conclusion was that these shoppers excel in innovativeness.
     But the 164 respondents were mostly obtained with a snowball sampling methodology, and there are two big problems with snowball sampling:
  • Bias which snowballs. You’d like your survey findings to represent what’s true for as broad a swath as possible of your current and potential shoppers. Each time you ask someone to participate in your survey, there’s some bias, since that person is almost certainly not completely representative of the whole population. To the degree that you can select survey respondents at random, you reduce this bias. But a snowball sample makes the bias progressively worse. People will invite others like themselves to accept your invitation to participate. 
  • Unknown population parameters. Statistical analyses of the replies adds to the validity and reliability of survey findings. The standard statistical analyses make assumptions about the size of the population, such as the approximate number of all consumers who use private sale sites. With snowball sampling, it’s difficult or impossible to accurately estimate the population size or important characteristics of the population aside from the one you’re most concerned about. 
     As a general rule, the larger your sample size, the less serious will be those two problems. So if you must use a snowball sample, go large. In addition, there are sophisticated statistical techniques—requiring the skills of an expert consultant—which can reduce the distortions from snowball sampling.
     But your best alternative is to consider findings from snowball sampling survey projects to be tentative and subject to confirmation by more rigorously conducted consumer surveys.

Click below for more: 
Decide on Your Segmentation Objectives Early 
Interpret Survey Results as a Retailer

No comments:

Post a Comment