24 Feb, 2016

Normal Approximation


Drawing stuff from a box

    EV SE
Proportions sum $p\times n$ $\sqrt{p(1-p)}\times \sqrt{n}$
  mean $p$ $\sqrt{p(1-p)} / \sqrt{n}$
Means sum $\bar{x} \times n$ $\displaystyle\sqrt{\frac{(x_1-\bar{x})^2+…+x_n-\bar{x})^2}{n}} \times \sqrt{n}$
  mean $\bar{x}$ $\displaystyle\sqrt{\frac{(x_1-\bar{x})^2+…+x_n-\bar{x})^2}{n}} / \sqrt{n}$

Note that the Standard Errors (SE) are computed from the standard deviation of the box (one draw from the box).


Confidence Intervals

A confidence interval is computed as follows:

\[\text{CI} = \text{estimate} \pm z^* \times \text{SE}\]

So, the $95\%$ confidence interval for a mean is:

\[\bar{x} \pm 1.95 \times \frac{s}{\sqrt{n}}, \text{ where } s = \displaystyle\sqrt{\frac{(x_1-\bar{x})^2+...+x_n-\bar{x})^2}{n}}\]

and the $95\%$ condifence intercal for a proportion is:

\[p \pm 1.95 \times \frac{\sqrt{p(1-p)}}{\sqrt{n}}\]