It has to do with being a representative sample and the bias that self-selection introduces to the data collection process. [GenCon Attendees] won’t have nearly as random a selection of hobby participants as data collected from, say, a list of 1000 people who bought games from a set of websites or a similar number of people selected entering or leaving game stores.
In my 42 or so years of personal experience in the hobby across 3 states and 5 cities, I’ve only gamed with 2 other people who ever attended an RPG convention...and they went with me. If you expand the selection parameters to include CCG tournaments, I get to about a dozen.
I doubt that my experiences are typical, however. In fact, based on my time at ENWorld, I suspect they’re not.
So while the GenCon crowd may be an actual mirror to the hobby’s general demographics, odds are pretty good that it isn’t.
As a statistician, there is always the question "can I generalize from sample to population". I am thoroughly aware of the problem, and the pitfalls. It's impossible to be completely sure that the sample is representative. We don't have a lot of random sample data for gamers; it's nearly all self-selection of one form or another. One form of data we do have quite a bit of, and multiple years of, is the gen con event data. Yes, it's heavily self-selected. But that doesn't mean it is unrepresentative of game preferences. The default hypothesis is that is *is* representative, so let's look at why it might not be.
One poster suggested that Gen Con attendees have more money and/or free time. This makes a big difference in many other preference situations -- political affiliation being a prime example. However no-one seems convinced that there is a link between either money / free time and gaming preferences, certainly no-one seems willing to say it makes a
big difference, so that argument seems weak.
You state that a very small percentage of your associates go to conventions. That in itself is not really an issue -- what is on debate is whether they are
different sorts of people from the overall population. Unfortunately, you have only 2 cases to consider, so you can't really draw any conclusions -- too little data.
This may seem like a minor thing, but so many arguments and threads talk about "what gamers like" or "what is played" or industry trends and we have so little fact to go on. People who buy things online, as well as people who buy things from stores are also strongly self-selecting, and I could make a decent argument that they are more likely to be biased in tastes then convention-goers (e.g. people who prefer out of print games will have strongly different buying habits than those who like D&D 5E).
To do a great job, we would randomly phone people (using a known mix of cell-phone and landlines reflecting the overall US division) and ask simple questions like "have you played one or more roleplaying games this last month" and if they answer yes, collect information on what those games were and how frequently they played each. That would give us a pretty good unbiased sample (phone calls are currently the best randomizer available short of tracking people down census-style -- they're pretty good actually). But absent that info, I haven't seen anything other than "gut feeling" to suggest that Gen Con attendees are not reflective of the overall community in terms of game preferences.