Jyotiram Pasupalak

Hi Madan, If you simply consider population size and confidence level (CL) you can only calculate confidence interval (CI). Hence your calculation of 467 sample size tells that you can be 95% certain population picks the defined mean value within the confidence interval. If I have understood your question correctly, with your 10% sample your mean anyway is same or very close to the population and why do we need 467 samples. Why not 211. It's quite possible all your data points are very close to the mean which means variation is quite low. But if you take any random sample without calculating, in this case let's say 211 not 467 then you need to be clear about one more factor called percentage. Accuracy also depends on the percentage of your sample that is how close to a particular value/answer. In this case population mean. What I see is nearly 90% of your sample (211) are closed to population mean. The chances of error are remote, irrespective of sample size. However, if the percentage is 50%, which means 50% of your 211 are away from population mean, the chances of error are much greater. In your case if you change your percentage to 50% you will get CI of 6.4 for 211 sample size. I hope this explains. Regards Jyotiram

Thanks for your input. It's quite useful. Let me give my realtime scenario where I have the challenge. I run the IT support service center. My team handles simple to very complex customer technical problems. I track the time to resolution of each of the problem. Due to the complex technical issues, mean is always very high. This is due to few outliers which take long time to resolve or has huge external variables involved. Hence I track trend of p75 performance and use the control chart for this trend. I really can't control the external variables neither the very complex issues which will definitely take time. I don't track the standard deviation neither the mean. Do you think what I am following is the right approach?

Hi, In what all scenario P75/3rd quartile measurement is relevant? For example in Pizza delivery case if we measure mean as the central tendency and try to bring in improvement, there will be extreme scenarios which will have large delivery time due to unavoidable causes. This will shift the mean from required mean. In this case should I measure and monitor P75? and if my P75 is always within required level then does this conclude that my delivery performance is stable? or I should consider variance and deviations? Regards Jyotiram