Cognitive Social Structures

Perceptions, Slices, and Accuracy in Social Networks

Introduction to Cognitive Social Structures

Social networks are not just objective, behavioral realities; they are also cognitive representations that exist in the minds of the people who navigate them.

  • The Cognitive Turn (Krackhardt, 1987):
    • Pioneer of the Cognitive Social Structures (CSS) framework: David Krackhardt (1987).
    • Standard network surveys ask: “Who do you go to for advice?” (Ego’s direct outbound ties).
    • CSS surveys ask: “For every pair of people \(i\) and \(j\) in this group, do you think \(i\) goes to \(j\) for advice?”
    • This captures each individual’s perception of the entire network structure, going beyond their own immediate ego-network.

The 3D CSS Network Matrix (\(R_{N \times N \times N}\))

While standard network data is represented as a 2D adjacency matrix \(\mathbf{A}_{N \times N}\), CSS data is represented as a three-dimensional matrix \(\mathbf{R}_{N \times N \times N}\).

  • Matrix Formulation:
    • \[R_{ijk} = 1 \quad \text{if person } k \text{ perceives a tie exists from } i \text{ to } j\]
    • \[R_{ijk} = 0 \quad \text{if person } k \text{ perceives no tie exists}\]
  • Cognitive Slices (\(\mathbf{R}_{..k}\)):
    • Slicing the 3D data cube along the third dimension (\(k\)) yields \(N\) distinct 2D adjacency matrices.
    • Each slice represents the complete group network from the specific, subjective mental lens of individual \(k\).

Locally Aggregated Structures (LAS)

To study the “actual” underlying network using subjective reports, Krackhardt defined methods to aggregate individual perceptions.

  • Locally Aggregated Structures (LAS) combine the direct reports of the dyad directly involved (\(i\) and \(j\)):
    • 1. Intersection LAS (Row/Agreement LAS):
      • A tie \(i \rightarrow j\) is considered to exist only if both \(i\) and \(j\) agree that \(i\) sends a tie to \(j\):
      • \[LAS^{\cap}_{ij} = R_{iji} \times R_{ijj}\]
    • 2. Union LAS (Or LAS):
      • A tie \(i \rightarrow j\) exists if either \(i\) or \(j\) (or both) say it exists:
      • \[LAS^{\cup}_{ij} = \max(R_{iji}, R_{ijj})\]

Consensus Structures (CS)

We can also aggregate perceptions across the entire group to find a “public consensus” of who is connected to whom.

  • Consensus Structures (CS):
    • Also called Consensus Networks or Central Limit Theorem structures.
    • A tie \(i \rightarrow j\) is considered to exist if a threshold proportion (\(\alpha\)) of all group members perceive it:
    • \[CS^{\alpha}_{ij} = 1 \quad \text{if} \quad \frac{\sum_{k} R_{ijk}}{N} \ge \alpha, \quad \text{else} \quad 0\]
  • Majority Consensus:
    • Commonly setting \(\alpha = 0.50\) (the tie is “real” if more than half of the organization agrees it exists).

Key CSS Metrics: Perceptual Accuracy

By separating cognition from structure, we can measure how accurately individuals perceive the social world around them.

  • Cognitive Accuracy:
    • Computed as the QAP correlation (Quadratic Assignment Procedure) or Hamming distance between an individual’s slice \(\mathbf{R}_{..k}\) and the group Consensus Structure (\(\mathbf{CS}\)) or \(\mathbf{LAS}\).
  • Network Cognitive Biases:
    • Ego Bias: Over-perceiving ties to oneself, or over-reporting one’s own popularity/centrality.
    • Reciprocity Bias: Perceptually symmetrizing asymmetric relations (if I like \(j\), I assume \(j\) likes me).
    • Balance/Transitivity Bias: Subjectively “closing” open triads in our minds (perceiving a friend’s friend as our friend, even if they aren’t).

Applied Research: Structure, Cognition, and Power

Do more accurate cognitive maps of informal networks make you more powerful? This was tested empirically by Krackhardt (1990).

  • The Landmark ASQ Study (Krackhardt, 1990):
    • Analyzed a small manufacturing firm (Silicon Systems) of \(N = 36\) employees.
    • Collected complete CSS surveys for both advice and friendship networks.
    • Reputed power was measured by having every employee rate the influence of all other employees.
  • Key Findings:
    • Individual cognitive accuracy regarding the advice network was highly and positively correlated with reputed power (\(r = 0.51\), \(p < 0.001\)).
    • This effect held strong even when controlling for actual network centrality (indegree, outdegree, and betweenness).
    • Conclusion: Power belongs to those who have an accurate map of who is connected to whom.

Applied Research: The Ties That Torture

CSS data also helps researchers understand the psychological stress of structural constraints, leading to Simmelian Tie Theory (Krackhardt (1999)).

  • Simmelian Ties:
    • A tie between \(i\) and \(j\) is Simmelian if:
      1. The tie is mutually reciprocated (\(i \leftrightarrow j\)).
      2. Both \(i\) and \(j\) are mutually connected to a third party \(h\), forming a co-members clique of size \(\ge 3\).
  • Perceptual and Behavioral “Torture”:
    • Simmelian ties are heavily constrained by third-party observation and norms, leading to strong expectations of conformity and consensus.
    • Simmelian cliques are highly stable, but individuals in them experience reduced autonomy, as conflicts cannot be easily negotiated bilaterally.

Additional Dimensions: Graph Connectedness and Efficiency

In addition to CSS, Krackhardt (1994) developed fundamental structural metrics of informal organizations based on reachability (\(D^r\)).

  • 1. Graph Connectedness (\(C\)):
    • The proportion of all possible directed pairs of nodes that are actually reachable via a path of any length:
    • \[C = \frac{\sum_{i \neq j} D^r_{ij}}{N(N-1)}\]
  • 2. Graph Efficiency (\(F\)):
    • Measures how “economically” the network is connected (trees: \(F = 1.0\); complete graphs: \(F = 0\)):
    • \[F = 1 - \left[\frac{2(m - n + 1)}{(n-1)(n-2)}\right]\]

Summary of David Krackhardt’s Key Contributions

David Krackhardt’s work is foundational to the cognitive and structural analysis of informal organizations.

  • Key Contributions to SNA:
    • CSS Methodology (1987, 1990): Pioneer of 3D matrix networks and cognitive social structures.
    • Structure, Cognition, and Power (1990): Proving that cognitive accuracy is an independent source of power.
    • E-I Index (1988): Standard metric for group homophily and boundary ties.
    • Informal Dimensions (1994): Reachability-based measures (Connectedness, Hierarchy, Efficiency, LCA).
    • Simmelian Ties (1999): Analysis of triadic consensus and structural constraint.

References

Krackhardt, David. 1987. “Cognitive Social Structures.” Social Networks 9 (2): 109–34.
Krackhardt, David. 1990. “Assessing the Political Landscape: Structure, Cognition, and Power in Organizations.” Administrative Science Quarterly 35 (2): 342–69.
Krackhardt, David. 1999. “The Ties That Torture: Simmelian Tie Analysis in Organizations.” Research in the Sociology of Organizations 16 (1): 183–210.