I have a consistent data about a decision making problem where I need to decide whether I choose A or B to compete, the winner is the one with larger number per round. If I remove one point, A beats B each time. But the sample is so small that removing a data point makes the data much smaller.

The goal of this competion is to maximize the result with low standard deviation. So I have here a non-dominating situation because player B has the largest record 86 while A has the smallest SD (about 150% while B 250%).

If I make a conclusion based on smaller sample, I feel a mistake -- how can I say this statistically? How much less significant will a conclusion be if I use smaller sample?

R stands for round.

R   A   B
1   85  86
2   83  83
3   83  82
4   83  82
5   83  81
6   81  80
7   80  78


The player I choose have to compete in the same game again against other players. I want to maximize the future rounds. Look B has a rising trend while A has historically more stable results. Will do some analysis soon.


1 Answer 1


Much of what you say is not very clear...it seems that, whether you omit the first round or not, you should still prefer A to B. But I'll take a stab at an answer: don't omit any data, and choose A over B.

I would only omit that first round if it were clearly an error (and I'd prefer to fix it rather than omit it).

Perhaps you're thinking: go with the one with the maximum value? I would avoid the maximum as a statistic, as it's too much influenced by outlying values. So I'd suggest reformulating your statistic for comparing A and B.

  • $\begingroup$ but B has a rising trend, it is not that easy thing to choose A over B. I want that the player will get high results in future competitions. A has more stable results but historical indicator not forward-looking indicator. Dilemma...have to investigate. $\endgroup$
    – hhh
    Sep 23, 2011 at 22:54
  • $\begingroup$ What do you mean by a "rising trend" - B's scores have been falling steadily over time. I also agree with @KarlBroman - omitting data because it doesn't give you what is, as far as I can tell, the "results you want" is a flawed way to go about any analysis. $\endgroup$
    – Fomite
    Sep 24, 2011 at 0:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.