Literature notes — grounding the causal diagram in research
Purpose: justify every node/arrow in our DAG with published evidence
(not intuition), per the course requirement that DAGs come from “theory
and domain knowledge to think through the data-generating process”
(Class 1, slide 38). The dataset is the World Management Survey (WMS),
so the WMS research programme is the directly relevant literature.
Core sources (World Management Survey programme)
- Bloom, N. & Van Reenen, J. (2007). “Measuring and Explaining
Management Practices Across Firms and Countries.” Quarterly Journal
of Economics 122(4). — the foundational paper; defines the 18-item
management score and its determinants.
- Bloom, Genakos, Sadun & Van Reenen (2012). “Management Practices
Across Firms and Countries.” Academy of Management
Perspectives. — accessible summary of determinants.
- Scur, Sadun, Van Reenen, Lemos & Bloom (2021). “The World
Management Survey at 18: lessons and the way forward.” NBER WP 28524 /
IZA DP 14146.
- Lemos & Scur; Bloom et al. on family/dynastic ownership and
primogeniture.
Finding 1 — the treatment effect itself (founder/family ownership →
management)
- Founder-owned and family-owned-and-family-run firms are consistently
among the worst managed. Family firms that bring in an
external (non-family) CEO score as well as
dispersed-shareholder / PE firms.
- Mechanism (theory for the arrow D→Y): reluctance to
delegate/professionalise; CEO selection by
primogeniture (eldest-son succession) rather than
talent. This is a governance/incentive channel, NOT merely a size or
country artefact.
- Implication: there is a strong, theoretically-motivated
direct effect of founder/ family ownership on
management quality — exactly the estimand the VC fund wants.
Finding 2 — determinants of management quality (causes of the
outcome Y)
Per WMS papers, management score is higher with: - Stronger
product-market competition (compet_* dummies). - Larger
firm size (employees). - Multinational status
(MNEs are better managed everywhere they operate). - Higher
human capital / workforce skills & education (degree_nm). -
Country factors (GDP/capita, labour-market flexibility)
and industry/skill-intensity.
Finding 3 — implications for variable ROLES in our DAG (to be
FINALISED after Week 2 slides)
The literature tells us these variables cause Y. The open question
for each is whether it also causes the treatment (→ confounder,
control) or is caused by the treatment (→ mediator, do NOT
control for the total effect). Working hypotheses, with reasoning:
- country → CONFOUNDER. Family-ownership prevalence
varies strongly by country (institutions, inheritance norms,
capital-market development) AND country drives management (GDP/capita,
regulation). Common cause of D and Y. Handle as fixed effects (Class 1
slides 31–32; nested data).
- industry → CONFOUNDER (similar logic): some sectors
are structurally more family-dominated, and sector drives
management/competition. Fixed effects.
- firmage → CONFOUNDER. Pre-determined; older firms
are likelier to have transitioned ownership away from the founder AND
age relates to management maturity. (Watch reverse story: better mgmt →
survival → age; treat as pre-treatment control.)
- **competition (compet_*)** → CONFOUNDER / cause-of-Y. Market
structure is largely exogenous to a single firm’s ownership; strongly
drives management. Safe to control (closes a backdoor and/or adds
precision).
- employees (size) → CONTESTED (likely MEDIATOR).
Family/founder firms tend to stay smaller (capital access, control
retention) → D affects size; size strongly drives management. So size
plausibly sits on D→size→Y. Controlling it estimates the DIRECT effect
and removes a real mechanism. Decision: primary spec = total effect
(exclude), secondary spec = include to quantify the size channel. TEST
empirically (regress size on D).
- multinational → CONTESTED (likely MEDIATOR). Family
firms internationalise less → D affects multinational; multinational
drives management. Same treatment as size.
- degree_nm (workforce education) → CONTESTED. Could
be a mediator (ownership shapes hiring) or a confounder/cause-of-Y
(local labour supply). Weakest prior; test empirically.
Method implication (consistent with course)
The WMS papers themselves estimate management score via OLS with
controls for country, industry, size, etc. — i.e. regression adjustment
guided by which factors are confounders. This supports our Week 1–2 +
Week 4 plan: causal-diagram-guided control set, then regression
adjustment and propensity-score matching for robustness. Our
methodological contribution beyond a naive regression is the explicit
confounder/mediator separation and the matching/weighting robustness
checks.
Still to do before finalising the DAG
- Read Week 2 slides (the dedicated causal-diagram week) for the
precise course rules on confounder / mediator / collider and the control
criteria.
- Optionally fetch Bloom et al. (2012) PDF to confirm how they treat
size/MNE/competition as controls vs mechanisms.