For my Master's Paper at UNC, I researched change blindness in the context of web applications. Bill Scott (Director of UI Engineering at Netflix at the time) made an interesting claim during his presentation at the User Interface Engineering 16 conference: webpages create a change blindness effect due to the way they load page-by-page. I set out to research this claim by creating a web application that served up one of 3 "real" web pages (code, not image maps) then randomly altered one of 5 things on the page in one of 3 ways.
The study design was a 3x3x5 within-subjects design where participants went through 30 trials. Each trials consisted of viewing one of 3 page types that would undergo one of 3 change types in one of 5 locations. The combination of page type, change type, and change location were randomized throughout to mitigate learning effects.
Results showed that the "one-shot paradigm", where a 0.5-second flicker occurred between pages, lowered performance compared to normal full page refreshes, as expected. Users detected changes 5% more often and with response times 0.1 second faster when things instantly changed on the page. This was not a significant difference between a full page refresh and instantly modifying a UI component on screen. So, single page applications that minimize full-page refreshes may help reduce the incidence of change blindness on the web, but full page refreshes aren't the scary change-blindness-inducing aberration that some designers fear they are.