While working on a paper over the weekend it suddenly hit me that "necessary but not sufficient" is the perfect way to describe the whole array of techniques that have emerged over the last few years in the name of improving panel data quality. Not to be confused with the Goldratt "business novel" (now there's a concept!) of the same name I mean it here in the legal sense of required but not enough. Yet from most of what I read the goal of online seems to continue to be to produce representative samples, a goal we can never reach as long as Internet use is less than universal and we continue to generate samples from databases of volunteers who are themselves a tiny franction of those online. Nonetheless, we keep talking about cleansing and routing and weighting as the path to representative samples. What's especially troubling about this is that deep down we know better. But we can't seem to stop. It's not that we can't use online to do very good work and to generate a lot of really smart insights. It's just that creating representiative samples is not on the list.
At the recent ESOMAR Congress Frederic John described online panels as "a Trojan horse that we have gleefully brought to our clients as the solution to all our problems that ultimately may damage our credibilty beyond repair." So please let's stop ourselves before we make it worse. Let's ban the R word.