I want to develop a tool which makes it easy for teachers and professors to write tests: multiple version, multiple choice, long answer, scantron, the whole works. The question is, would I be better off doing this as a web service (subscription model) or a desktop application (buy-once model)? The attractive thing about this market is the target is academia, which often has a lot of money to spend on this sort of software. I'm just not sure which way to go with it.
Then you'll need to get in touch with the professors and teachers then ask them about it:
If you haven't already I suggest you read some of Patio11 blog posts, he created a piece of software called Bingo Card Creator which is aimed fairly squarely at academia, initially this was a standalone desktop application, then it became a web and desktop application and now I believe it is web only. On his blog he writes a lot of justification for what steps he has taken and the pros and cons to both.
The education market is huge, it's also incredibly difficult to break into at a the organisational level even at a single school rather then a district. Consequently you will initially have to aim at individual teachers, people who will adopt your software, or put a lot of effort into sales and business development to target the larger market. The problem I suspect depending on which route you take will probably influence you answer.
If you are targeting influencing academic staff, then a web application makes it easier to reach a large percentage of these people, it also reduces support costs in effect you only have to deal with the site, not the app plus the potentially unlimited different setups, or simply reduces "does this work on my phone" questions.
If you are targeting the organisation, they will have their own IT policies, these policies may mean web apps are simply a no no, or they might make it the only feasible solution depending on how tight they are.
One other factor that will make a large difference, is if personal student data is being stored, if the solution is simply to create tests, which the student get's printed and handed to them then this is not an issue. If you expect students to take the test online or through the application then where and how data is handled will make a large difference. Most schools will not be happy to store their student data away from their own servers (indeed in some countries it would actually be illegal for them to do so).
Last thing to think about is potential hybrid, so have a online solution with all the bells and whistles and an offline version for those without an internet connection. Using Adobe Air and similar the amount of work required wouldn't be like starting from scratch.