It looks to me like the CA study was a preselected LGBT population and thus produces its 0.1% number based on an estimate of what portion of the total population are LGBT as well as on the assumption that the surveyed LGBT people are a good representation of all L, G, B, or T people in the state.
Meanwhile, though I don't know what happened in 2007 or 2009, I tracked down the fact that in 2012 and 2013 in MA
they called thousands of adults as part of their regular behavioral risk factor surveillance system, and both years 0.4% of those adults reported identifying transgender. (Which is to say, it doesn't have any of the sources of bias you claimed such studies usually have. It wasn't self-selecting for people who feel strongly on the issue, it wasn't self-selecting for people who are likely to have different demographics from the general population, it wasn't a particularly small sample size, etc.)
As such, I'd say that 0.4% study is quite a bit more reliable both than the 0.3% oft-reported estimate and the 0.1% CA result. And unless you have some specific knowledge of bias in the way the BRFSS research was done, or knowledge of a study that was equally well-done but that reports a significantly lower proportion, 0.4% is the number I'm personally going to accept as reliable going forward. (It may even be an underestimate for the reasons described in that fivethirtyeight article, where some people feel like they were assigned the wrong gender even though they wouldn't necessarily identify themselves as transgender on any but the most carefully anonymous surveys.)
Edit: CorruptUser's number for intersex people is off by quite a bit as well. Even if we only count those individuals who do not have strictly XX or XY karyotypes, Wikipedia's figures put the total at 0.2%. People who are XX or XY but can be considered intersex for other reasons (such as hormone over- or underproduction or insensitivity) seem to be less common, but raise the percentage slightly nonetheless. Even if there is considerable overlap between such people and those who identify as transgender, an estimate of at least
0.5% of people who can't be unambiguously described as "male" or "female" seems reasonable.
Nothing like the "majority" 12obin claims, but considerably more than HungryHobo's 0.1%.
(And I'm not at all sure why doogly thinks tenths-of-a-percent accuracy is unobtainable, when that works out to several dozen people in a large survey or up to several thousand in an analysis of hospital records.)