With all the talk about how market research is changing radically, it’s somewhat comforting when age-old debates come up again and again. Like the topic of representivity.
Every so often I see a blog or infographic claiming that a particular panel is representative of the Census. This claim is puzzling, as it really doesn’t matter if a panel is representative of the population, as long as a representative sample can be drawn from the panel.
One of the world’s largest online panel providers, Lightspeed Research/GMI has more than four million double opt-in members in its MySurvey and Global Test Market panels. Using a wide range of sources, such as web ads, affiliate networks, communities, social networks and more, we continually recruit members. We don’t worry about quotas on our panels; we want as many people as possible.
From our large pool of respondents, we use proprietary software to build representative outgoing samples. The software balances samples on key demographics that match census or a target group to produce samples that are representative of the general population or of the particular demographic group that our client wants to survey.
Moreover, not only is it not necessary for a panel to be representative of Census, but it actually is a hindrance. Because certain demographic groups typically have lower survey response rates, it’s important to recruit more members of these groups to the panel.
At Lightspeed Research/GMI we use differential sampling to address typical response rates for key demographic groups. Differential sampling is the practice of sampling subpopulations out of proportion to their actual population in order to achieve a more representative sample. Essentially, differential sampling involves over and under sampling. We over sample in groups with lower response rates and under sample in groups with higher response rates. As a result, we can assure clients we can deliver their quotas on the harder-to-reach target groups.
It is important to keep in mind that a sample is only as good as its design. A poorly designed sample, at best, uses respondents inefficiently, causing feasibility issues and delays in field. At worst, it can result in a sample that is not representative of the target, thus leading to incorrect business decisions. Sampling is as much an art as a science. At Lightspeed Research/GMI we have experts who have spent years honing their craft.