Representations and the symmetric group - Diaconis data

Posted on 2024-02-07

character table of S5

Let’s use GAP to find the character table of S5.

gap> G:=SymmetricGroup(5);
Sym( [ 1 .. 5 ] )
gap> tab:=CharacterTable(G);
CharacterTable( Sym( [ 1 .. 5 ] ) )
gap> Display(tab);
CT1

     2  3  2  3  1  1  2  .
     3  1  1  .  1  1  .  .
     5  1  .  .  .  .  .  1

       1a 2a 2b 3a 6a 4a 5a
    2P 1a 1a 1a 3a 3a 2b 5a
    3P 1a 2a 2b 1a 2a 4a 5a
    5P 1a 2a 2b 3a 6a 4a 1a

X.1     1 -1  1  1 -1 -1  1
X.2     4 -2  .  1  1  . -1
X.3     5 -1  1 -1 -1  1  .
X.4     6  . -2  .  .  .  1
X.5     5  1  1 -1  1 -1  .
X.6     4  2  .  1 -1  . -1
X.7     1  1  1  1  1  1  1
gap> 

Diaconis example – survey data

This data is taken from the paper (Diaconis 1989)

It describes 5,738 completed ballots rank-ordering 5 candidates.

View a rank-ordered ballot as an element of the symmetric group S5; we want to study the frequency function f.

first ranking table

the regular representation

This diagram shows the decomposition of the regular representation into isotypic components.

Be careful: the notation Diaconis is using here does not match that used by GAP above. For example, the representation Diaconis writes as V3 is the isotypic component determined by the irreducible representation labeled X.5 by GAP.

The second row reflects the decomposition of the frequency function f. Namely, write f=i=17fiwith fiVi.

The second row entries are the “sums of squares” fi,fi.

Remember that we can compute the fi using the idempotents in C[G].

For example,

f1=15!σS5σ.f

More generally, if χi denotes the character of the irreducible representation Li with Vi=C[G](Li) then fi=15!σS5χi(σ1)σ.f

Note that f3,f3=459 is relatively large (ignoring f1,f1 since f1 is trivial).

normalizing the first-order data

THe i,j entry in this table is the number of votes ranking candidate i in the j-th position, minus the sample size over 5.

In particular, rows and columns sum to 0.

This normalization can also be achieved as follows:

Let f2 be the projection on V2, and consider the functions σδi,σ(j).

The i,j entry of the preceding table is f2,δi,σ(j)

Interpretation in this last table:

Compute the projection f3 of f into the component V3 of M=C[S5].

Now, consider the easily understood functions σδ{i,i},{σ(j),σ(j)} in C[S5] for distinct i,i and distinct j,j.

The space of these functions is a 100 dimensional subspace of W C[G].

The entries in the table are the inner products f3,δ{i,i},{σ(j),σ(j)}

Summary observations

The data were to elect a president for the American Psychological Association. Candidates 1 and 3 were clinicians while candidates 4 and 5 were academicians, two groups within the association with somewhat divergent perspectives.

In the second-order table, we see a preference for candidates 1 & 3 witnessed by the entry 376 corresponding to the entry for candidates {1,3} and ranks {1,2}.

And we see a (slightly smaller) preference for candidates 4 and 5 witnessed by the entry 296 corresponding to the entry for candidates {4,5} and ranks {1,2}.

Bibliography

Diaconis, Persi. 1989. “A Generalization of Spectral Analysis with Application to Ranked Data.” The Annals of Statistics 17 (3): 949–79. https://www.jstor.org/stable/2241705.