AN-096: Bipartite structure of the shell–pollster relationship

From the shell side, the dominant pattern is tight 1-to-1 with one pollster (8 of 14 shells route ≥80% of their polls through a single pollster; median 85%). From the pollster side, only 2 of those 8 pairings are reciprocally tight (VS Publicidade ↔ Publi. QC, FacUnicamps ↔ IPOP) — the other 6 pollsters operate diverse sponsor portfolios with 8–34 distinct sponsors per pollster, using the shell as one of several channels. One shell (Nivaldo Galindo) has the inverse asymmetry: shell uses 2 pollsters, pollster is 93% captive to the shell. No shell operates across many pollsters: there is no intermediary or partisan-aggregator pattern in the data.

Hypothesis
H13: Shell-contratante polls show larger residual β
Confidence
green
Type
descriptive
Design
Sample
14 PROBABLE_SHELL entities from AN-094, plus the pollster each shell most frequently hires
Specification
For each AN-094 shell: shell_top_pollster_share = (polls via top pollster) / (shell total polls) pollster_share_via_this_shell = (polls from this shell) / (pollster total polls) Classify shells by shell-side concentration: TIGHT ≥ 80 % top pollster share MODERATE 50 % – 80 % DIFFUSE < 50 % The 2 × 2 (shell-side × pollster-side) cross-tab gives the bipartite typology.
Comparator
paper §2 anchor pairs (FacUnicamps ↔ IPOP-Cidades, VS Publicidade ↔ Publi. QC — both reciprocally tight)
Script
source/analysis/an-096-shell-pollster-bipartite.py
Target
build/intermediate/an-096-shell-pollster-bipartite.csv
Status
interpreted · 2026-06-17
Created
2026-06-17

Question

Two structural hypotheses about how shell contratantes work predict very different (shell × pollster) bipartite distributions:

The two predict different concentration on the shell side: tight 1-to-1 in the first case, diffuse in the second.

Results

Shell-side concentration

Concentration Shells Median top-pollster share Pattern
TIGHT (≥80 % via one pollster) 8 of 14 92 % Shell paired to one pollster
MODERATE (50–80 %) 5 of 14 72 % Primarily one pollster + backup
DIFFUSE (<50 %) 1 of 14 40 % Distributed

Across all 14: median 2 distinct pollsters per shell, median 85 % via the top one. No shell uses more than 8 pollsters. The intermediary / partisan-vehicle hypothesis predicts the opposite — there should be at least some shells with 5+ pollsters at <30 % top concentration. None appear.

Pollster-side reciprocity (only for the 8 TIGHT shells)

Shell (top pollster) Pollster's total polls Pollster's distinct sponsors Pollster's share via this shell Reciprocal?
VS Publicidade → Publi. QC 219 3 100 % ✓ Tight both sides
FacUnicamps → IPOP-Cidades 68 1 100 % ✓ Tight both sides
Estação I → IPPI Pesquisas 107 33 49 % ✗ Pollster operates network
ABC Publicidade → J J Coelho 51 8 55 % ✗ Pollster operates network
Tres Marias → Instituto Franca 110 21 24 % ✗ Pollster operates network
Gledson Lopes → Ivani Mota de Araujo 60 14 40 % ✗ Pollster operates network
Hyago Cavalcante → Severino de Araujo 41 10 49 % ✗ Pollster operates network
Assoc. Marketing MG → Tony Brand 33 12 52 % ✗ Pollster operates network

Only 2 of the 8 tight-paired pollsters reciprocate the exclusivity — the two paper-§2 anchors. The other 6 paired pollsters operate diverse sponsor portfolios (8–33 distinct sponsors per pollster), and the shell is just one of several registration channels they use. The non-shell sponsors of those pollsters are some mix of real candidate committees (Routes A–D), other shells, and occasionally real media; the shell isn't the only cover but it's a meaningful component.

A useful inverse case

The MODERATE-concentration shell Nivaldo Galindo / N. R. Estúdio Multimídia (CNPJ 7257404000104, R$ 0 capital, 26 polls in PE) is interesting: from the shell side, it routes only 54 % of its polls through Ronald Dias Falabella; from the pollster side, 93 % of Ronald Dias Falabella's 15 mayoral polls in 2024 come from Nivaldo Galindo. The pollster is captive to the shell, not the other way around. Plausible reading: Ronald Dias Falabella is a small pollster whose main client is Nivaldo Galindo's operation, and Nivaldo also uses another pollster (the remaining 12 polls).

Bipartite typology

Type Shell-side concentration Pollster-side reciprocity Count Examples
1. Tight 1-to-1 ≥80 % ≥80 % 2 VS Publicidade ↔ Publi. QC; FacUnicamps ↔ IPOP
2. Shell-tight, pollster-portfolio ≥80 % <80 % 6 Estação I, ABC Publicidade, Tres Marias, Gledson Lopes, Hyago Cavalcante, Assoc. Marketing MG
3. Shell-moderate, pollster-captive 50–80 % ≥80 % 1 Nivaldo Galindo (pollster 93 % captive)
4. Both-moderate-or-diffuse <80 % <80 % 5 G S Negreiros, SX Empreendimentos, José Vanderlúcio, DDD91, Ramon Margiolle (diffuse)

Interpretation

  1. No shell is an intermediary or partisan vehicle. The hypothesis that some shells would aggregate polls across many pollsters — the signature of a funder-side middleman — is refuted by the data. The maximum number of pollsters used by any of the 14 shells is 8, and the lowest top-pollster share is 40 %.

  2. The dominant pattern is shell-as-pollster-cover, not shell-as-funder-vehicle. 8 of 14 shells route ≥80 % through one pollster. The shell exists for the pollster's operational convenience — to register polls without naming the candidate funders behind them.

  3. Reciprocal exclusivity is rare (2 of 8 tight pairs). The cleanest cases (VS Publicidade ↔ Publi. QC, FacUnicamps ↔ IPOP) are the ones already in paper §2, and they look the way they do because the pollster in each case is also single-sponsor — turning the bipartite relationship into a private channel between two CNPJs. Most other shell-pollster pairs are asymmetric: the shell is paired, but the pollster also takes work from other sources (some real, some shell, some mixed).

  4. Each shell-pollster pair is a separable operational unit. No cross-pair coordination is visible from the registry — VS↔Publi.QC in São Paulo and FacUnicamps↔IPOP in Goiás look the same not because they're connected, but because each is a local instance of the same operational template.

  5. Implication for paper framing. §2's "the cover vehicle changed across cycles; the pollster's volume did not" punchline now generalizes more carefully: it's the pollster that has continuity across cycles; the shell is the cover vehicle for one pollster's output. In the asymmetric cases, only part of the pollster's output goes through the shell, but each shell is still a pollster-specific cover, not a funder-side aggregator.

Follow-ups

  1. Audit the Type-2 pollsters' other sponsors (extension): for the 6 shells with tight shell-side concentration but asymmetric pollsters, the remaining 51–76 % of those pollsters' polls go to other sponsors. Are those other sponsors candidate committees (Routes A–D), other shells, or real media? Auditing them would tell us whether each pollster is "real media + shell as side channel" or "shell network with the registered one being just one of many".

  2. Type-4 shells (both diffuse) (blind spot): 5 shells have neither tight pairing nor tight reciprocity. These could be (a) opportunistic shells used by candidates for one-off contracts across the market, (b) shells in transition between paired-pollster operations, or (c) genuinely smaller-scale aggregators worth a closer look (Ramon Margiolle uses 6 pollsters — the highest in the audit).

  3. Cross-cycle bipartite check (extension): did the same shells appear paired with the same pollsters in 2022? AN-095 documents the 2020 → 2024 self → shell shift for Publi. QC; the 2022 federal cycle has its own data and could show whether the shell-pollster pair is stable or constructed for each cycle.

Caveats