This is almost definitely a case for "not knowing what to search for", but am interested in how to refer to the following:

For a set of $n$ reals (say non-negative), I'm interested in characterizing how "concentrated" the values are in the top few (a la Pareto $80/20$). This could be for instance for thinking about percent of total variance of top few principal components.

Is there a way to refer to this function? I'd want a single value, similar to AUROC in statistics.

Context: I'm interested in optimizing over this function as a tradeoff, similar to penalties on norms like $L^1$ or $L^0$.

  • 1
    $\begingroup$ This may be relevant: en.wikipedia.org/wiki/Gini_coefficient#Definition $\endgroup$ Jul 2 at 0:14
  • $\begingroup$ What is the percent of total variance? And what are principal components of a set? There are like eight other phrases I don't understand but maybe they are economics terms. $\endgroup$
    – pancini
    Jul 2 at 0:17
  • $\begingroup$ Thanks @TravisWillse , I think Gini coefficient works! $\endgroup$
    – dashnick
    Jul 2 at 0:19
  • $\begingroup$ Cheers. In that case I'll write a short answer. $\endgroup$ Jul 2 at 0:58

1 Answer 1


One such metric is the Gini coefficient $G$, which quantifies how much a distribution of values deviates from perfect equality: Concisely, it is half of the relative mean absolute difference of a list of values, i.e., if the values are $x_1, \ldots, x_n$, we define $$\boxed{G = \frac{1}{n^2 \bar x}\sum_{1 \leq i < j \leq n} |x_i - x_j|} ,$$ where $\bar x$ is the mean of $x_1, \ldots, x_n$. Provided that all $x_i$ are nonnegative, the Gini quotient takes values between $0$ (perfect equality) and $1$ (perfect inequality). It is often used to quantify inequality in income or wealth.


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.