Is there any social inequality measure for categorical or ordinal variables, based on a goodness of fit chi squared test (distance from an uniform distribution)?

Example: part of my data is unemployment rates by 4 age groups. I know that inequality is minimum if the rates are equal for the 4 age groups (uniform distribution), for example [0.04, 0.04, 0.04, 0.04].

N.B. The sum of these rates is $\neq$ 1.

  • 2
    $\begingroup$ Can you explain your question more fully? What type of data do you have? What are you measuring? What is your research question? $\endgroup$ – Joel W. Aug 20 '18 at 23:53

For a possibly useful general discussion of 'Diversity indexes' you might want to look at Wikipedia, perhaps starting with the Simpson index $$\lambda = \sum_{i=1}^R p_i^2,$$ where $R$ is the number of different 'types' of individuals and $p_i$ are their respective proportions in the population.

For a large population, roughly speaking, $\lambda$ can be viewed as the probability that two individuals randomly chosen from the population are of the same 'type'. Thus, smaller $\lambda$ corresponds to greater diversity.

  • $\begingroup$ Please see my edits. I am not sure if with the Simpson index $\sum_{i=1}^R{p_i}$ has to be $=1$. Thank you for the link to 'Diversity indexes'. $\endgroup$ – sbac Aug 21 '18 at 15:32
  • $\begingroup$ In your unemployment example: conditionally on being unemployed, age groups are equally likely [.25,.25,.25,.25] and you could use the standard chi-squared test statistic for adherence to that proper probability distribution (with probabilities summing to 1). $\endgroup$ – BruceET Aug 21 '18 at 16:38
  • $\begingroup$ And the Simpson index also works in this case? Even if the unemployment rates are for example [0.04, 0.08, 0.16, 0.32]? In this case should I force (compute) the sum of probabilities to be 1 as in [0.066, 0.133, 0.267, 0.533]? $\endgroup$ – sbac Aug 21 '18 at 17:20
  • $\begingroup$ For the Simpson index, I think it would be $p_1 = p_2 =p_3=p_4 = 0.25,$ so that $\sum_{i=1}^4 p_i = 1.$ $\endgroup$ – BruceET Aug 21 '18 at 17:50
  • 1
    $\begingroup$ You want the proportion of all unemployed subjects who happen to be members of each age group. 'Condition on unemployment` means $P(G_i|\text{Unemployed}).$ Not $P(\text{Unemployed}|G_i).$ Not $P(\text{Unemployed}\cap G_i).$ $\endgroup$ – BruceET Aug 21 '18 at 22:20

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.