If I were to run the experiment 100 times, I’d get 100 slightly different confidence intervals.

If they are 95% confidence intervals, then 95 out of those 100 CIs would contain the true mean. Or, conversely, 5 of those CIs would NOT contain the true mean. The indivdual CIs either have the true mean or not. There is no probability for an individual CI.

So a 95% CI is erroneous 5% of the time.

]]>I read lots of paper for Confidence interval. I understood the theory behind that, but I have one practical problem.

If I have 95% confidence interval of mean weight of one school (55kg-75kg) of 240 students. Now please let me know what should I interpret with respect to (55kg-75kg). Total school students are 3500.

Thanks

Hitesh ]]>

I found the original blog and found the link you provided there to this site. I didn’t really want to waste my time looking at this for my classes, but figured it would be a good chance to get into R a little bit more.

Anyhow, I have one online course where I was suspicious of some cheating. The students are taking the test on their honor not to work together. I thought probably a good majority of students are honorable, and I think that’s still the case. But I had a good number of student falling on the perfect match line for the first exam I’ve looked at. Not cool! When I dug into other aspects of their tests, my worries were confirmed (they took the exam at the same time and even had the same wrong choice on the wrong answer). I’ll be presenting this to my chair or dean to see what they want to do with it.

In my other large class that meets in person and has well proctor exams I only saw one problem on the one exam I checked (and I had been suspicious of this student from before – he failed anyway so I’m not too concerned).

Thanks for posting your more detailed analysis. I have not tried it yet, but I may have to if I need to convince people more rigorously of the problem.

]]>I just now looked the article up in Google Scholar, and it looks like there has been a reasonable amount of work since then.

If I get a chance, I’ll edit out identifying information, and send you the report I wrote. I

]]>It’s a shame that people cheat and make us worry about them. I hope talking more about it will have some deterrent effect.

]]>I think I understand your question. If you have N=30 participants, get p < .05 and then subsequently add 10 more subjects, what happens then? It's a little tricky to handle with my function because we would need to know if you plan on adding 10 more subjects regardless of the results. If the plan all along was to add more subjects no matter what, then there is no worry about p-hacking. You can just report the final results. However, if there was never any plan to add more subjects until after the data were gathered, there would probably need to be some adjustment done. The proper adjustment might depend on what made you decide to collect more. This function assumes that thing that made you decide that was p > .05. But in your example case, it is just the opposite, p < .05. This makes me think something like sequential analysis may be useful: http://daniellakens.blogspot.com/2014/06/data-peeking-without-p-hacking.html

I hope that helps. :-/

]]>