In the absence of blinding, observers may err in measuring data towards the expected outcome. (For example, when I lose a ball in golf, I assume it is in the water. When my partner loses a ball, I assume it is in the grass.)
When researching thrombolytics for ischemic stroke, a researcher performing a telephone interview with a patient whom they know received tPa might expect the patient to be performing better than patients who received placebo. Consider the patient statement: “I notice some changes, but overall I think I am doing pretty well. I mean, I don’t cook or clean for myself – my wife has to take care of all that – but I never did any of the cooking or cleaning before, so I guess I can carry out all my usual activities; at least as much as I was doing before the stroke”. This could be interpreted as MRS 1 (no significant disability despite symptoms; able to carry out all usual duties and activities) or MRS 2 (slight disability; unable to carry out all previous activities, but able to look after own affairs without assistance), or maybe even MRS 3 (moderate disability; requiring some help, but able to walk without assistance). The expectation that patients receiving the new “miracle” drug will perform better is likely the influence the category that the researcher assigns the patient to.
Krishna R et al. Research bias: A review for medical students. Journal of Clinical and Diagnostic Research. 2010; 4:2320-2324.