Selective attrition can be minimized by (1) designing experiments in such a way that selective attrition is highly unlikely and then shows that the data on “arrivals” is consistent with random attrition, and (2) driving down inference-relevant attrition to make it negligible. For most applications, Method 2 is superior, though it comes at the cost of a more elaborate experimental design.
By providing strong incentives to continue, such as forfeiting their “show-up” fee if the entire experiment is not completed, or requiring the user to first perform a tedious but well-payed task that is then forfeited if the remainder of the compensated experiment is not completed.
To use these “hooks” ethically, it is important to provide as much information upfront as possible, such as approximately what they will be doing and the minimum compensation for the task. By providing plenty of information from the onset and using appropriate hooking tasks, it is possible and reasonable to consistently drive attrition to zero.