2
$\begingroup$

When a set of data are binomial I have a lot of methods available to me (Wilson score interval, Wilson score interval with continuity correction, Jeffreys interval, Clopper-Pearson interval, Agresti-Coull Interval) other than the standard Wald interval estimation method (see Interval Estimation for a Binomial Proportion Author(s): Lawrence D. Brown, T. Tony Cai and Anirban DasGupta).

However, I can't use these methods for data that is itself proportional in the sense of it sitting in [0,1].

For example consider a sample of n respondents, giving a score for a question X as:

r_1 = 0.4
r_2 = 0.3
...
r_n = 0.2

How would I calculate the confidence interval for a mean based on this data? Is there any way other than bootstrapping?

$\endgroup$
  • 1
    $\begingroup$ Do you mean that the response to the question must lie between $0$ and $1$? Also, how large is the sample? If it's big enough (practically speaking $>30$ with a reasonable range of answers), asymptotics should allow you calculate it on something approaching normal... $\endgroup$ – dardisco Jan 15 '14 at 8:59
  • 1
    $\begingroup$ Yes it lies between 0 and 100 or 0 and 1 rather. The sample size is variable, we decided to only consider >30. For less than 30 we have considered a bootstrap. $\endgroup$ – dominic Jan 15 '14 at 9:12
  • $\begingroup$ I am wary of the CLT confidence interval interval for growing sample size - the paper above showed that the coverage probability varies greatly away from the nominal confidence level as n grows. So I'm really interested in a mean confidence interval method for non binomial data (apart from CLT CI). $\endgroup$ – dominic Jan 15 '14 at 10:41
  • $\begingroup$ You don't have bimomial data. You have discrete data on a scale of 0-100. Any scenario you think of for the actual population will have the sample means behaving like a normal distribution for n>30. You should run some simulations on your own to convince yourself. $\endgroup$ – SaraWong123 Jan 15 '14 at 16:38
  • 1
    $\begingroup$ Your answer echoes the very title: the O.P. says these are indeed not binomial. How, then, does this reply address his question about calculating a CI for the mean? $\endgroup$ – whuber Jan 15 '14 at 18:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.