Say we have a dataset that contains 1 million values and Every single value is equal to $ 5$.

Say we take take a large number of samples from this set, and work out the mean of the sample each time. According to the central limit theorem, as the sample size approaches $\infty$, the distribution of sample means should tend to a normal distribution.

However, thinking about this logically, the sample mean is going to be $5$ every single time, and therefore the distribution of sample means will not be normally distributed.

How does this work?


You have a dataset whose "random" variables are drawn from the pdf $p(x) = \delta(x-5)$ (I'm infering this, you've provided 1 million samples with such a rigid structure and asked a question which implies your data is 'certain' to be 5).

The delta "function" is the distribution you get if you take a Gaussian distribution and send its variance to 0 (in the appropriately careful mathematical sense).

One nice, somewhat physical, way to see this is to note that the Gaussian distribution with variance $\sigma^{2} = t$ solves the Heat Equation $\partial_{t}p(x,t) = \partial_{x}^{2}p(x,t)$ with initial condition $p(x,t) = \delta(x)$. From this point of view a standard Gaussian is a diffused 'melted' version of the delta distribution.

  • $\begingroup$ Sorry, I did not understand much of what you said there as I'm an early-level maths student. Does the central limit theorum hold true in this case? $\endgroup$ – Katie Chandling Aug 15 '16 at 12:28
  • $\begingroup$ @KatieChandling what version of the central limit theorem are you using and under what circumstances are you allowed to apply it ($iid$ requirement etc.)? $\endgroup$ – user190080 Aug 15 '16 at 17:14

Your Answer

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

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