How to measure bias in a subgroup of people

31 Views Asked by At

I have a small fun problem that I gave myself to represent mathematically, but I got stuck very quickly.

Suppose there is a group of people (named A, B, C, D, E, F and G). Every week they meet:

  • every person speaks then nominates the next person to speak
  • $A$ always begins the sessions

Someone in the group suspects that B, C and F always seem to prefer nominating each other. Data has been collected for a number of sessions to try and test this hypothesis:

Week A B C D E F G
1 0 3 1 5 4 2 6
2 0 4 1 x 5 2 3
3 0 1 3 6 2 5 4
4 0 6 3 2 4 1 5
5 0 6 5 2 4 1 3
6 0 1 6 5 4 2 3
7 0 2 6 3 5 1 4
8 0 6 5 4 2 1 3
9 0 1 2 x x 4 3
10 0 5 6 3 1 2 4
11 0 2 3 x x 4 1
12 0 6 3 5 4 1 2
13 0 x 1 2 3 x 4

For each week's session (rows) the numbers represent the order in which each member spoke. Implicitly, a person speak in position $i$, nominated the person that spoke in position $i+1$. People who were absent on any particular session is marked as with an $x$.

In words I could start describing the problem as follow. (Please forgive my butchering of mathematical notations :) )

with $P(J)$ being the probability of $J$ being picked, and with $P(J|K)$ being the probability of $J$ being picked by $K$, we can state

$B$, $C$ and $F$ are bias if:

$P(C|B) + P(F|B) + P(B|C) + P(F|C) + P(B|F) + P(C|F) > P(B) + P(C) + P(F)$

However when going down this route, how do I represent:

  • the absentees
  • people going last on various weeks
  • the forced "choice" of the person who goes second to last
  • the fact that the choice of those outside the "bias group" to avoid nominating members of the bias group, will result in higher chance of the bias occurring (as B, C and F would end up being the last people to speak and have no choice but to nominate each other)

Then I started thinking whether there was a statistical threshold that defines the bias or whether it could be defined in a different way. So I started looking into graph theory and measures of modularity, but ended up confusing myself more.

Naturally, this is an example of a more generalised problem.

Given

  1. a set of people $G$, who take turns at an activity and then nominate the individual to go next,
  2. subset $B$, the bias subset, defined as a subset of $G$ where individuals are more likely to nominate from within $B$ than they are outside of $B$

How do we show that:

  • a given subset $B_1$ is bias
  • there exist a bias subset in $G$

Did I word a well known mathematical problem in the worst way possible (I did try searching for this online with not much success)?

Thank you very much.