1

Context

Let say $S\in\mathbb{N}$ is the number of votes cast and $N\in\mathbb{N}$ is the number of seats in the parliament. We may assume $N\leq S$. The votes are distributed among $K$ political parties, so there exists $v_{1},\dotsc,v_{K}\in\mathbb{N}$ such that $\sum_{j=1}^{K}v_j=S$, where $v_j$ is the number of votes the $j$-th political party received.

We say that the $j$-th party is perfectly represented if its vote share equals its seat share, i.e. $\frac{e_j}{N}=\frac{v_j}{S}$, where $e_j\in\{0\}\cup\mathbb{N}$ is the number of seats the $j$-th party received. So, if $\frac{e_j}{N}>\frac{v_j}{S}$ the party is overrepresented, and, analogously, if $\frac{e_j}{N}<\frac{v_j}{S}$ the party is underrepresented. We may assume $\sum_{j=1}^{K}e_j=N$.

We want to define a measure of the over (under) representation of a party, one posibility is looking at the number

\begin{equation} a_j:=\frac{e_j}{N}-\frac{v_j}{S}\text{, } \end{equation}

so, if $a_j<0$, the $j$-th party is underrepresented and if $a_j<a_i$, the $i$-th party is more overreprented than the $j$-th one. Another possibility will be \begin{equation} q_j:=\frac{\frac{e_j}{N}}{\frac{v_j}{S}}\text{, } \end{equation} so, if $q_j<1$, the $j$-th party is underrepresented and if $q_j<q_i$, the $i$-th party is more overrepresented than the $j$-th one.

My question

Are these two measures compatible? In other words, if $a_j< a_i$, then $q_j<q_i$ and vice versa? Or, at least \begin{equation} \arg \min a_j=\arg\min q_j\text{?} \end{equation} Is it even true that $a_j=a_i$ if and only if $q_j=q_i$?.

I have been trying to prove this seemingly elementary facts for quite some time using elementary methods, but none of them seems to be fruitful. Any idea?

  • Not a direct answer, but your issue could hopefully be connected with the rather polyvalent Jaccard index. – Jean Marie May 13 '21 at 10:25
  • A general thought: hesitation between taking a difference or a quotient to measure variations is quite common. Occam razor principle is of no use: both are simple. But the difference is preferable because less sensitive to perturbations than the quotient in particular when the denominator in the quotient is small. – Jean Marie May 13 '21 at 10:38
  • A little googling led me to a certain number of works on this issue, for example (accessible as a Google book) the very interesting (United States) Congressionnal Record - Proceedings and Debates year 1929 page 1695 about "measure(s) of disparity" in representation. – Jean Marie May 13 '21 at 22:03
  • I am conscious that all these remarks are not answering you question. My opinion is that the quotient measure is almost unfathomable with the difference measure and should be dropped for reasons of unstability 2) A survey in a different field : income disparity.
  • – Jean Marie May 13 '21 at 22:20