I am currently working on a research problem where the existing literature proposes an algorithm that is tested on data generating 100,000 integer values within the range of 0 to 20. The common practice in these papers is to present the frequency of each value in tabular form, along with the average of the 100,000 produced values as the sole performance metric—where a lower average is considered better.

In my research, I have developed a new algorithm that produces 100,000 values with a lower average compared to the algorithms discussed in the relevant literature. However, I would like to extend the comparison beyond the average and incorporate additional metrics such as variance, range, and others.

My question is whether it would be ethically acceptable to compile the frequency data from the existing literature, as presented in their papers, and use it to calculate additional metrics for comparison in my research. I want to ensure that my comparison is comprehensive and provides a more holistic view of algorithm performance.

Any guidance on the ethical considerations and best practices for such data compilation would be greatly appreciated.

  • 4
    Why do you think this would pose ethical issues? As long as you correctly cite where you took previous data from, you can absolutely do this. In a meta analysis, something like this is the whole point. Commented Dec 7, 2023 at 9:29

2 Answers 2


It is always ethical to extend existing research and to use publicly available information. If you avoid such things as copyright infringement (and plagiarism) then you should be fine. Just cite what you use, and perhaps acknowledge those who developed the earlier data (or whatever).

In fact, academia in general and science in particular advances specifically on the work of earlier people. Just provide context and cite those on whose shoulders you stand.


Note that speed for a given data size/implementation/system is, while common, quite useless. There are many factors influencing the performance including the quality of the implementation, platform etc. Much more interesting is performance characteristics (different n) or at least re-implementing/remeasuring to reduce irrelevant differences.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .