The "all you see is all there is" fallacy is probably in play here; if respondents were asked about each minority individually, simply the fact that other minor

The "all you see is all there is" fallacy is probably in play here; if respondents were asked about each minority individually, simply the fact that other minorities not named didn't come readily to mind could account for the obviously poor math here. If all the minorities were listed and the respondents were asked to estimate all of them at the same time, you'd see a dramatically different result. I suspect that similar over estimations would happen if you exposed a test subject to a chart showing the relative proportion of ten different quantities (eg, a chart partitioned into different sized regions, distinctly colored) then asked them to estimate one of the quantities without referring back to the chart. It's not so much an estimation problem as it is a limited working memory problem (the old "seven, plus or minus two" rule).