One of the surprising things about the census fiasco is that of all the publicy-provided services that small-government advocates could target, the census is very near the bottom of the list of priorities. Many of the services provided by governments could and perhaps should be produced by the private sector. But the economics of databases such as the census aren't the same as the goods that inhabit most economic models, and so the standard market-is-best results of the First and Second Welfare Theorems don't apply. When it comes to things like the census, markets fail.
Standard public finance theory offers two dimensions in which goods can be classified: rivalry and excludability. If a good is a purely private - that is, rival and excludable - then the welfare theorems apply, and there's no obvious case for government intervention. But a database is different. Although intellectual property laws can provide a mechanism for owners to control access, it is non-rival: many users can benefit from a database just as easily as one can.
You can see why this means that markets would generate inefficient outcomes here: firms that require certain information would be obliged to produce multiple data sets, all serving pretty much the same purpose. The inefficiency takes the form of this duplication of efforts. It's more efficient to simply have one database and allow access to multiple users. In other words, the census is a natural monopoly.
But what pricing strategy should the census monopolist adopt? Usually, we'd try to set price equal to marginal cost. In this case, the marginal cost of providing access to a database is essentially zero: once it is set up, maintenance costs are trivial. So in an ideal situation, the census would become a pure public good: universal access at zero cost.
This is pretty much the model adopted by the United States. A trip to the websites of the Bureau of Economic Analysis or the Bureau of Labor Statistics will get you a wealth of data at the price of a couple of mouse clicks. And over at the Census Bureau, fees only seem to be incurred by those whose projects require working at their offices in Washington; I can't see any mention of fees for access to the data per se. (Readers more familiar with these matters will correct me if I've read things incorrectly.)
Things are different in Canada. If you go looking for data at the Statistics Canada website, you end up looking at a price list. For reasons going back to the Mulroney-era budget cuts, Statistics Canada is expected to generate a significant portion of its revenues. In 2009, some 20% of its $500m budget came from selling access to its data. Private-sector clients are not the only people who pay for StatsCan data: government departments and agencies at all levels must pay for access as well. (A significant chunk of the Parliamentary Budget Office's budget goes to Statistics Canada.)
The effect of this pricing model on research activity is considerable, setting the stage for a complicated shell game that would be comical if it wasn't so detrimental to the advancement of our understanding. A Canadian professor who works at a publicly-funded university must obtain research grants from publicly-funded institution in order to pay for data - that have already been collected! - from yet another government agency. Too often, the game ends with the researcher abandoning the project or choosing to use US data instead.
And now things will get much, much worse.