pith. sign in

arxiv: 2604.12802 · v1 · submitted 2026-04-14 · 📊 stat.ME

Fundamental Limits and Optimal Methods for Sharp Analytical Causal Bounds in Instrumental Variable Models

Pith reviewed 2026-05-10 14:42 UTC · model grok-4.3

classification 📊 stat.ME
keywords sharpboundsanalyticalinequalitiesgrowsinstrumentalvariablecausal
0
0 comments X

The pith

Sharp analytical bounds for the average treatment effect in discrete IV models must use exponentially many terms; polynomial approximations cannot be sharp, with optimal code provided.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Causal inference often uses an instrument variable to bound the effect of a treatment on an outcome. Some methods solve this with slow optimization programs. Others try quick analytical formulas based on probability rules. This work shows that those quick formulas cannot give the tightest possible bounds unless they include an exponentially growing number of separate linear pieces when the outcome can take many different values. The authors also give ready-to-use computer code that computes the tightest analytical bounds without unnecessary extra work.

Core claim

any sharp analytical bound for the average treatment effect must be expressible as a maximum (minimum) over a collection of linear terms whose cardinality grows exponentially in the number of values taken by the outcome

Load-bearing premise

The instrumental variable model is discrete with finite support for all variables, and the target is the average treatment effect under standard IV assumptions.

read the original abstract

Bounding causal effects analytically, rather than numerically, is appealing for its interpretability and conceptual clarity. Existing sharp methods rely on optimization-based approaches such as the Balke-Pearl framework, whose computational complexity grows rapidly. An alternative line of work derives bounds heuristically using probability laws and generic inequalities, and some recent papers have claimed or conjectured that this approach can yield sharp analytical bounds with substantially lower complexity. In this paper, we show that this perceived advantage is illusory. In particular, in a discrete instrumental variable setting, we show that any sharp analytical bound for the average treatment effect must be expressible as a maximum (minimum) over a collection of linear terms whose cardinality grows exponentially in the number of values taken by the outcome. In parallel, we show that the number of instrumental variable inequalities itself also grows exponentially. Consequently, bounds and inequalities expressed using only polynomially many such terms cannot be sharp. As a constructive complement, the paper is accompanied by codes implemented in python and R to derive sharp analytical bounds and sharp inequalities with optimal efficiency, matching the lower bounds proven in this paper. These codes are available online.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The claim rests on the assumption of finite discrete support for all variables in the IV model and standard causal assumptions (instrument relevance, exclusion restriction). No free parameters or invented entities are introduced.

axioms (2)
  • domain assumption All variables have finite discrete support
    Required for the exponential cardinality result to hold in the discrete IV setting.
  • domain assumption Standard instrumental variable assumptions hold (relevance, exclusion, no unmeasured confounding)
    Implicit in the causal IV model used throughout the abstract.

pith-pipeline@v0.9.0 · 5509 in / 1169 out tokens · 37056 ms · 2026-05-10T14:42:28.771344+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.