I'm inclined to agree with David Poole. Collecting Census data has got many of the attributes of a "write only" or at least "write once" application. Once the form has been submitted the person filling it in has no need to access it again. In fact, from that point on the data is really only used as aggregates.

In Ireland they still distribute and collect the forms manually. It is expensive but it has benefits. This year I participated in a census as an "enumerator" - that's the guy who distributes the census form, collects it and helps with any questions people have. I did it because it was an opportunity to get out from in front of a computer screen! I looked at Steve's link to the US form, and it is pretty similar but the Irish one seems to have more classification data. My experiences may be of some interest for this discussion.

My experience of the form and the process I was following had the marks of "punched cards" and processing batches. That's not necessarily wrong but it was noticeable. I noticed that the problems were not so much systems as people related. At least some of my job was to persuade people to fill the form in at all! Another part was to help them fill it in correctly, and detect minor silly omissions - the so-called "doorstep check". And at least some of what I did involved dealing with edge-cases: people who were somewhere else on the night or questions like "do I tick this box or that box?"

For census data to be useful you want to get as close to 100% coverage as you can. That was where I came in (though I didn't realise it at first). There are some groups which tend to exclude themselves and you want to include them in the dataset. People like: foreigners, the illiterate, the elderly... you can add your own groups to the list.

From a systems point of view, a census is peculiar too. Would you like to be running a system which you got out of the box and ran once every 5 or 10 years? On the other hand would you like to develop it fresh every 10 years? Even though the data is so simple, systems have moved quite a bit in 5 years!

In conclusion: I think census systems have real potential for some kind of "package" solution. The core idea is dead simple. Every country runs them so there is a market. The issues are the fact they are run infrequently in any particular place, the problems around promoting completeness and accuracy and the political problems relating to privacy and security. It's a good one to puzzle over! 🙂