2

Variance is

$$\dfrac{\sum_{i=1}^{n}(x_i-\bar x)^2}{n-1}$$

But why square the difference? Why not cube it, or any other exponent?

Related question.

Ethan T
  • 123
  • 2
    If you cube it then some errors will cancel others due to negative signs and that is not good. – Mohammad Riazi-Kermani Sep 27 '19 at 20:44
  • 1
    First of all, your formula is not the variance, but rather the sample variance. The variance of a random variable $X$ is defined as $$\mathbf{Var}(X)=\mathbb{E}[(X-\mathbb{E}[X])^2].$$ Now you may still ask what is the advantage of squaring the deviation $X-\mathbb{E}[X]$. My personal impression on this question is two-fold: (1) Variance is one of the two parameters that characterize the normal distribution. (2) Variance comes from covariance, which gives rise to the (degenerate) inner-product structure on the space of square-integrable random variables. – Sangchul Lee Sep 27 '19 at 20:53

2 Answers2

2

You can also use other exponents, but then it is not called variance anymore. See this wikipedia page, for example.

Variance comes up quite naturally in various contexts, and carries a lot of information about the distribution.

Daniel Robert-Nicoud
  • 29,795
-1

Still, I am not convinced by this answer. The naming convention can be separate topics to discuss. Here what is the issue if we take cubes or quad etc?

Asad
  • 1