REAL
Keystone sector methodology applied to Portugal
We tested for the hypothesis that any single actor played such a role. The
procedure consists in removing one at a time, each central entity found in precedent
tests from all the three socio-matrices after which we conduct another component
analysis with the resultant matrix. There is a cut point when the removal of that entity
from the system will be able to increase the number of components, i. e., dividing the
graph in two or more separate components, leaving them without connections.
We ran this test for several entities one at a time within each of the matrices and
none of them was positive, i.e., the whole matrix always maintained just one
component. In the case of INFORMATION and ORGANIZATION we tried yet to
remove the top 6 entities but the result did not changed. Though, when we excise both
Private Construction Firm and the Private Insurance Company in MONEY relations,
the matrix will spread in 5 components. In our opinion this is due to the extremely
weak density of the MONEY matrix rather than to the vital importance of those actors
and, at the same time, reveals the general low propensity to risk while instead of make
other applications, entities prefer to guarantee the future in insurance and real estate
assets.
When any entity appears to be critical we say that the single component
structure in all the matrices is robust, i.e., there is no single actor that does not have a
substitute as a vector in each network considered. As we found that excising 2 of the
actors from the MONEY matrix could be critical, we followed tests through the Block
Modeling procedure for the structural equivalence test.
4.2.5 Block Modeling - Redundancy test
The UCINET software provides a procedure called CONCOR (from
CONvergence of iterated CORrelations) that delineates structurally equivalent
entities. The test identifies sets of entities among the all sample with different behavior
patterns. Formally we say that actors’ n and n+1 are structurally equivalent if n relates
only with k and k+1 (n ≠ k), then actor n+1 relates only with k and k+1 and the
sociomatrix will turn in rank N-1. Simplifying in other words this is the same than
considering that the information of two rows (or columns) is equal; one of them could
be removed without altering the network connectivity (redundant nodes).
This test proceeds in two steps: first by a square case-by-case correlation
matrix; the second step is a clustering procedure to group the cases into stochastically
19