0

I want to find confidence intervals for my scenario and I see this sentence in many research papers related to my work. " Confidence interval of less than 1% of the average value"? What does it mean? You can also explain this with the example given below.

E.g. Say for x=100, y= 0.01. This 0.01 is obtained as an average of running the code 10 times for x=100. I will use those 10 values to find the confidence interval at this x=100 point. Hope this is the right way to compute the interval.

Context as asked: I am plotting some blocking value (y-axis) against traffic load (x-axis). X-axis values from 100,200, upto (say) 800. And y-axis values are of the form 0.006 or 0.01 or 0.2 and so on. I run the code to generate y value for the given x and I do this 10 times and take the average. So, each blocking value point vs a given x is an average of 10 times. I want to plot confidence intervals for each x point. So I assume I have to use those 10 values as the sample values to calculate the interval. Is it correct? If so, then I want to know the meaning of the statement I asked at the top in this context.

knowledge_seeker
  • 11
  • 1
  • 2
    It would be really helpful if you included some context or examples. – Matt Krause Jan 14 '22 at 18:51
  • 1
    @MattKrause I have added the context. Please re-open. I agree that the context wasn't detailed but don't know why it was downvoted to indicate no research effort. The example I gave itself is self-made (research effort) in order to let people find it easy to explain my first statement. Hope it fits the criteria for reopening now. – knowledge_seeker Jan 14 '22 at 21:17
  • I appreciate your effort. However, the quoted statement, because it is just a fragment, makes no sense in English, which is one reason we are looking for context and clarification. – whuber Jan 14 '22 at 21:20
  • @whuber I understand. Thanks. I hope now it is understandable and someone is able to throw light. – knowledge_seeker Jan 14 '22 at 21:21
  • I won't be able to understand it without further explanation. – whuber Jan 14 '22 at 21:22
  • @whuber then maybe you can consider asking me whatever missing elements/questions are coming to your mind when you are reading it. Because I don't think it is needed to go into the whole telecommunications specific explanation as to why I need them as I feel the question is more about statistics here. The papers that mentioned the asked statement refer to the graphs whose plotting I have already mentioned in the context scenario. – knowledge_seeker Jan 14 '22 at 23:22
  • @whuber Also, as a first guess I was even feeling maybe the 'average' word in the statement relates to the mean of the samples considered to calculate the intervals. And what those samples can be I mentioned in the context i.e. getting some 10 values for the same x=100 point on the x-axis. But I wasn't sure if it is the right way and so I expressed doubt on the manner of my calculating the confidence interval which in turn might make sense with the asked statement. – knowledge_seeker Jan 14 '22 at 23:27
  • @Whuber And please, if it's possible please reopen my question. I have added context as asked. If not you, at least maybe someone else be able to ask or even answer something. If I hadn't edited it then you shouldn't reopen but now I think I can request you. – knowledge_seeker Jan 14 '22 at 23:32
  • With only ten values one cannot identify the distribution type for those values. Read about [confidence interval methods](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6294150/). – Carl Jan 15 '22 at 00:31
  • @Carl Thanks for the resposne. I will read it. But would 30 be ok? I remember seeing 30 in one my research related paper that's why. Because if it's not then maybe the way I am thinking of calculating the CI for my scenario is wrong.... – knowledge_seeker Jan 16 '22 at 03:25
  • That depends on the method used to calculate CI, what the desired intervals are and what the density function is. For example, an 80% confidence interval using quantiles from a quasi-normal distribution needs fewer samples (e.g., 20 might be enough) than a 99% confidence interval from a Cauchy distribution (might need 1000 samples). How many are needed should be determined by testing, – Carl Jan 17 '22 at 05:23
  • @Carl Thanks for your input. I understand the concept a bit more better now. – knowledge_seeker Jan 20 '22 at 03:53

0 Answers0