The general data we found range between 10 - 20 min in young and healthy individuals, but the more accurate data relative to this tasks (visual attention) will be great.
Start here: Wilkinson, R. T., & Houghton, D. (1982). Field test of arousal: a portable reaction timer with data storage. Human Factors, 24(4), 487–493. Then follow up with Dave Dinges' articles about the PVT and also the vigilance work of the late Raj Parasuraman.
Wilkinson, using the UART with WWII radar operators, showed a 30% decrement in vigilance performance across 20 minutes. That is still viewed as being the "classic" vigilance decrement. The PVT is not a vigilance task, per se, as it has no interfering background events. Thus, its signal probability is 1.0 and signals are not "rare." However, the PVT provides good info about lapses in attention, some of which may be related to the "Bills Block," another well-known phenomenon characterized in older literature. If you can deal with the poor reproduction quality, reading this report will give you an idea about vigilance tasks: https://www.researchgate.net/publication/261436919_Vigilance_Research_and_Nuclear_Security_Critical_Review_and_Potential_Applications_to_Security_Guard_Performance
Technical Report Vigilance Research and Nuclear Security: Critical Review and...
From my Chapter 2 in the Handbook of Operator Fatigue
Contemporary studies of vigilance are most often based on measurement of ‘miss’ and ‘false positive’ errors in line with the theory of signal detection (Green & Swets, 1966), but two alternative approaches should be noted. Wilkinson and Houghton (1982) developed a simple, portable 10-min test of “arousal” and “continuous, concentrated attention”, called the unprepared simple reaction time test (USRT), and based upon a cassette tape. Dinges then introduced (Dinges & Powell, 1985) a solid-state version of the USRT called the Psychomotor Vigilance Task (PVT), which is now used widely and successfully in sleep loss studies (Lim & Dinges, 2008). However, the USRT/PVT has a signal probability of 1.0, much unlike many classic vigilance tests that have signal probabilities of less than 0.05 (Miller & Mackie, 1980). While the USRT/PVT is sensitive to fatigue (especially sleepiness), it is difficult to classify it as a “vigilance” task because it is such a short task and its signals are not embedded in a background of high-frequency, non-meaningful events. Thus, though the USRT/PVT addresses some aspects of sustained attention, and captures errors of omission in the form of lapses (i.e., the Bills Block), it fails to address the visual search and decision-making components of vigilance performance.
Right now i´m working in a very pre-experimental design, although my main interest isn't vigilance or attention i'd think is a relevant variable to take account when you're trying to achieve high control over the experimental settings or conditions.
A related control issue is sleep quantity, quality and circadian timing preceding the experiment. We instructed subjects to sleep eight hours per night for the three nights preceding an experimental session, retiring and arising at the same times of night/day. Also, we always tested at the same time(s) of day/night.