Research has a strange hangup with sampling. The selection of the right people is kept separate from the content you want to get from them.Which is understanable but has become a limitation. Because it has conventionally been kept physically separate. In this post I want to suggest that sampling could become more dynamic. Which would move research a lot closer to the gathering of analytical content which has almost no sampling discipline applied to it at all. One of the reasons research is so different is because of this sampling discipline which is part of the bedrock.

Quantitative research surveys tend to collect the information about who you are at the end of the survey because its dull and being factual you may be too wearied from the survey to bother to embroider. It is very clear which are the classification questions to be used for weighting and cross tabulation. And which is the content which will be carved up.Qualitative research takes a lot of trouble to make sure you are the right person before you start the research at all. And in analysis much is made of the distinction between process data and content data. The bits of the research interview (or group) that are identified as being linked to the dynamics of the research process itself - these are discarded as contaminated. It could be the warm up for the group or something a respondent agrees with which is reckoned to be a part of the group dynamics than a sign that person B agrees with person A.

want to suggest that using two parallel tagging systems that we could make sampling dynamic and apply it to web content. Sample tags would be used to identify who is the author of the content. Content tags would be used to identify the content. Although a lot of data might carry both tags the same data item should not be used for sampling and content because that would be bootstrapping. This approach would have to be probablistic - because it is unlikely that we would ever have enough information from sources of internet content to identify them  (apart from asking them to complete a sample frame questionnaire of course) - but I''m trying to avoid this because it would turn content analytics into a different way of recruiting online respondents - great content - can you fill in this survey? Similar to scoring models used in direct mail which use detailed information about a sample to make intelligent guesses about the propensities of people living in similar types of houses or buying similar branded goods.

The sampling bottleneck which we need to get past is analogous to the Van Neuman bottleneck which was a major step forward in computing. Before Van Neuman computer designers used two different types of data - control/instruction data and content which the instructions worked on. . The old knitting machines programed with cards are a hangover from that period. Even though Alan Turing's original concept paper showed that a symbol processing machine could handle both in the same medium. What Van Neuman suggested was that data be stored in memory and be acted upon using a CPU which applied programming instructions using the same medium. No more knitting cards. And computers have been designed the same way every since. The instructions and the data have been in the same format stored in the memory and processed in the CPU. I am suggesting that sampling and content needs to go the same way. Drawn from the same source and handled as part of the same medium