I can't see where I am going wrong.

There are two different ways of writing the standard deviation:

  1. $ \sigma = \sqrt{\frac{1}{N} \sum_{i=1}^N (x_i – \overline{x})^2}$

  2. $ \sigma = \sqrt{\frac{\sum_{i=1}^N x_i^2}{N} - \overline{x}^2 } $

If you have $N$ numbers with a mean $\overline{x}$, but then you add an additional number to the set which is equal to the mean, what happens to the standard deviation?

Looking at 1:

$N$ increases; $\sum_{i=1}^N (x_i – \overline{x})^2$ stays the same (since $(x-\overline{x}) = 0$);

Therefore $\sigma$ decreases. This makes sense to me.

However, looking at 2:

$\sum_{i=1}^N x_i^2$ increases by the amount $\overline{x}^2$; $N$ increases by one; $\overline{x}^2$ remains the same

Therefore, if $\overline{x}$ is greater than one, $\sigma$ increases?


The confusion seems to be over operations on fractions. In general, consider the two fractions

$$\frac AB \quad \mbox{and} \quad \frac{A+D}{B+1}.$$

Your question is essentially asking whether the fraction on the right is necessarily greater than the fraction on the left if $D > 1$. The answer is, no, it is not.

For example:

$$\frac{100}{10} > \frac{100+5}{10+1}.$$

As you can see, $5 > 1$, but $\frac{105}{11} < 10.$ In general, it's not just how much you increase each part of the fraction that matters, it's how much you increase each part of the fraction relative to (or as a percentage of) the old value. In the example above, if we add $10\%$ to the denominator but only $5\%$ to the numerator, the value decreases. If you start with a numerator that is much greater than the denominator, you have to add a proportionally larger amount to the numerator than to the denominator just to "keep up."

Applying this to the standard deviation formula, if the $x_i$ are not all exactly the same then the mean value of $x_i^2$ (which is what $\frac{\sum_{i=1}^N x_i^2}{N}$ represents) is greater than the square of the mean value of $x_i$ (which is $\bar x^2$). That's how we find that $\sigma > 0$. Now if we add one more observation to the set, but its value is equal to $\bar x$, we are adding $\bar x^2$ to the sum $\sum_{i=1}^N x_i^2$, but since $\bar x^2$ is less than the mean of $x_i^2$, the new observation "drags down the average". If $\bar x > 1$, then the sum $\sum_{i=1}^N x_i^2$ is already so much greater than $N$ that you would have to add more than $\bar x^2$ to the top to cancel the addition of $1$ to the bottom.

| cite | improve this answer | |

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.