Net-based education has since the dawn of the Internet been presented as a solution for many educational obstacles. The American National Academy of Engineering 2008 ranked “Advance personalized learning” as the fifth most important “Grand Challange”. It comes after fusion energy and clean water, but before preventing nuclear terror. Personalized learning often use computerized instruction via the Internet for example “intelligent” Web-based education systems. There are numerous pedagogical aids available to enhance learning in digital environments. However, the effects of these aids are rarely evaluated on a quantitative scale. Therefore we have analyzed data from the web-based courses given at KTH’s recourse centre for net-based education (RCN) than enrol more than 10 000 students yearly. Net-based courses have many similarities with on-line surveys: the tests and exams are really just a questionnaire, they both potentially have a large number of respondents/students, drop-out rates are a significant problem and the limited possibilities to observe the respondents/students make evolution of the instruments problematic. Due to the large number of respondents/students involved, even small improvements in completion rates are valuable. I will present some examples of statistical analyses of student behaviour in hope of an interesting discussion on similarities with on-line surveys and remedies for the observed problems.