Pretend for a moment you are asked by a person conducting a survey:
``Would you rather pay higher taxes or go hungry?''
It is probably fair to say that in response to this question, most would reply ``pay higher taxes,'' and the surveyor might then report ``Most respondents support paying higher taxes.''
The problem is that these two choices are presented as an or choice, but they are not mutually exclusive. It is possible to neither pay higher taxes nor go hungry, but this is not given as a choice. In this case, the surveyor has introduced a bias by (1) neglecting to list reasonable alternatives and (2) by equating the severity of not paying higher taxes and not eating.
Statistical analysis of measured data such as shot positions on a target are a bit more objective, but the danger of introducing bias remains. The researcher must be vigilant to remain objective and must constantly review the numbers from the point of view that they prove the opposite of what one really intends to 'prove.' This idea is so important in statistics that a special construct exists to aid the mathematical analysis: the Null Hypothesis (Section 10.3).
For example, suppose you shoot two five-shot groups, one from Load A and one from Load B, and these represent optimized loads using bullets by two manufacturers. You believe Bullet A is 'better,' but wish to test this using statistics (and ShotStat). In this case, your Null Hypothesis is that Loads A and B are the same, and it is the Null Hypothesis that you test. That is, you assume that A and B are the same, then set out to 'prove' your assumption 'correct.' Only when an analysis of the data suggests your assumption is incorrect do you safely conclude that A really is better than B (here 'better' means smaller group). This approach helps to lower the tendency to introduce your own bias into the analysis.