Interestingly, exactly half of participants answered the question without scrolling, using elements on the static page to make their estimates. I hypothesize that more participants would have chosen to scroll if they had been using a newer version of Mac OS, which hides the native scrollbar when stationary. At any rate, although most non-scrollers who were shown a custom indicator looked at it, four of ten looked at the native scrollbar in addition to or instead of the custom indicator. (Note: Non-scrollers didn't notice the flag indicator, which only appears during scrolling.)

The half of participants who did scroll to assess the remaining length of the article scanned the column of text (not individual words) and scrolled all the way or most of the way to the bottom of the page. Most glanced back and forth to the scroll indicator as they scrolled, but three participants ignored the indicator completely. Users seemed to rely on the spatial dimensions of the page to get a visceral "feel" for its length, even with scroll indicators present. I call the resulting behavior "scroll-scanning:" when users quickly scroll to get a sense of the physical length of a page.

Speaking only for myself, I do this all the time. Reading on the web is less a linear process than a series of scroll-scans, which help me evaluate whether to keep reading. I'm constantly asking, "How long and dense does this look?" "Are there any pictures that break up the text?" "Do I have time to read one more section before lunch?" Today's custom scroll indicators are too "dumb" to answer these questions.