I guess I need to disagree with my colleague in terms of what the literature actually shows. One of the leading experts on non-response is also the head of the U.S. Census Bureau and therefore my colleague. He's also a member of Statistics Canada's advisory committee on statistical methods. Recently, he wrote an excellent paper on the whole topic of non-response, which I would invite members to consult.
He said a number of very important things that I think are very relevant to this debate. He said that non-response can but need not induce non-response bias in survey estimates. He said the non-response rate of a survey alone is not a very good indicator of the magnitude of the bias. He said the risk of non-response bias, not non-response bias itself, is reduced with decreasing non-response rates. He said that ultimately there is very little empirical support for the notion that low response rate surveys de facto produce estimates with high non-response bias.
I just want to come back to the point that we cannot predict what the outcome will be before we start.