Should you manipulate your primary data to accept the right hypothesis is an extension to the above question, where we could bring in ethics, beyond the technique. In the informal lingua of research, the practice of creating false data or creating selective reporting of data according to requirement is called as “data fudging”. One of the most common example of this, that we see commonly in research, involves choosing that section of the data that shows consistency in that direction of the hypothesis that is preferred by the researcher. The remaining data is ignored or eliminated.
In general, the norms of research say that the validity of those results is always under question whose results cannot be reproduced by the investigators. With the fear of getting caught over fudging of data, some scientists keep away from publishing their data and the methods. They only give out the results and the interpretation.
By now we know that this practice is prevalent in the field of research and you may see its presence in all disciplines. But data manipulation is quite a serious concern that challenges the honesty and integrity of ethics. The outliers, missing data and the non normality of the data adversely affects not just the validity but the reliability as well of the data.
Researchers should not confuse removing of outliers from the data before analysis as a practice of data manipulation. Rather that is more appropriate, to study the real problems through scatter diagrams and remove those points that appear far away or detached from the main cloud. These points are removed only for a cause.
It is agreeable in research to accept the null hypothesis, how much ever trivial it be, as long as you are able to justify the reasons for this acceptance. These could vary from finding error in the theory or the observation, your interpretation of the theory could be the reason or some external interference with the experiment faltered the results. As long as you are able to give an explanation for the failure to support your thesis, there is some learning and it can become the base to form an alternate hypothesis that can be created on the grounds of failure of the first experiment.
Always remember that the searches for trivial techniques to play around with data is far from science. A good scientist is the one who strives to disapprove the hypothesis so that its acceptance can be ensured. Statistical significance isn’t the only proof.