A student decide to ignore any results that are too far above or below their prediction. He/she repeats their trials until a value closer to their prediction is obtained. What has the student done by doing this?
He has committed a mortal sin , according to the Ten Commandments of Science.
He adapted measurements to tally with his prediction, by leaving all contradicting measurements out.
Just to echo MeneerNask, if the student had no reason to reject the
outlying data, for all he or she knows, the outlier might be the true
value, and the grouped measurements subject to some error.
Of course, the measurements might NOT be drawn from the same population of data. That is, the experimenter was cack-handed when taking that particular measurement, or there was trouble with the scales or the instrument of measurement, or that measurement was a botched job, or it was the wrong phase of the moon, or some entirely random reason. Any of these reasons (maybe not a particular one), if they actually occurred, might be legitimate reason to reject a particular data point, but not otherwise.
Moreover, if someone were paying you to make a measurement, you would want to quote a large enough error measurement, so that the true value, whatever it is, lay somewhere in the range of measurements you quoted. And clearly, the bigger the error measurement, the more likely that the true value would fall somewhere in the given range. An outlying measurement
extends the range, and gives the experimenter more confidence that his/her quoted measurement reflects the true value.