Lord's Paradox¶

Show that two arithmetically correct analyses of the same pre-post data — raw change scores versus baseline adjustment — can reach opposite verdicts about an effect, because adjustment is a causal-modeling choice and the two answer different questions depending on whether baseline is itself caused by group membership.

Core Idea¶

Lord's paradox (Lord, 1967) is the phenomenon in which two defensible analyses of the same pre-post data — one using raw change scores, one adjusting for baseline (ANCOVA) — reach opposite conclusions about whether a treatment produced an effect. Both are arithmetically correct. The resolution (Holland-Rubin, Pearl) is that they answer different causal questions; which is right depends on whether baseline is itself caused by group membership. Adjustment is a causal-modeling choice, not a neutral cleanup.

Scope of Application¶

Lord's paradox lives across the empirical sciences that share the substrate of group comparisons over time on observational or quasi-experimental data.

Education and behavioural research — learning gains between groups with different starting scores.
Clinical-trial subgroup analysis — subgroups defined by baseline severity, where randomisation lapses.
Health-disparities research — change across groups whose baselines carry causal information.
Observational policy evaluation — difference-in-differences versus matched-on-baseline analyses.
Labor economics — wage changes versus wage levels conditional on starting wage.

Clarity¶

The paradox surfaces a normally invisible move: adjusting for baseline is a commitment to a causal model, not a neutral step. When competent analysts disagree, the instinct is to hunt for an arithmetic slip or argue over rigour; the paradox blocks both, showing the two analyses answer different estimands. The sharper question becomes not "should I control for baseline?" but "is baseline itself caused by group membership, and does conditioning on it answer the question I mean to ask?"

Manages Complexity¶

A recurring family of methodological standoffs — same data, opposite verdicts, no error to adjudicate — collapses onto a single structural recognition: the disagreement is never about the math but about the causal model. The whole dispute reduces to one yes/no — is baseline downstream of group? The analyst draws three nodes (baseline, group, outcome) and reads the correct adjustment off the structure by a fixed branch, relocating an irresolvable arithmetic quarrel to the causal assumptions where it can be settled.

Abstract Reasoning¶

The framework licenses a refusal-to-adjudicate move (stop hunting the arithmetic error; the analyses estimate different estimands), a diagnostic move keyed to one arrow (is baseline caused by group?), and a symptom-reading move run in reverse (a verdict that flips under adjustment signals baseline carries group information). A boundary-drawing move generalises the discipline: conditioning is always a causal commitment, so the adjustment set must be derived from the causal story, not analyst taste.

Knowledge Transfer¶

Lord's paradox is a named reasoning template, not a mechanism in the world, so it transfers as a diagnostic. Within causal inference the whole apparatus carries literally — the refuse-to-adjudicate move, the single diagnostic question, the three-node read-off, and the adjustment-is-a-causal-commitment discipline — because the substrate (a pre-post comparison where baseline may be a consequence of group) is fixed. Beyond it the general lesson that adjustment presupposes a causal story travels via the conditioning-reversal family — simpsons_paradox (of which Lord's is the continuous pre-post analogue), collider bias, selection_bias — and the parent causal_inference programme, not via this named construction.

Relationships to Other Abstractions¶

Current abstraction Lord's Paradox Domain-specific

Parents (1) — more general patterns this builds on

Lord's Paradox is a kind of, typical Simpson's Paradox Prime

Lord's Paradox is typically the continuous pre-post member of Simpson's conditioning-reversal family, with baseline adjustment replacing categorical stratification.

Neighborhood in Abstraction Space¶

Lord's Paradox sits in a sparse region of the domain-specific corpus (88^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (309 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12