FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteNo-Evidence version preferred126 / 162 · 7171379234dd322f

Causal inference methods for combining randomized trials and observational studies: a review

Statistics · 2011.08047v4

FLOWING EVIDENCE BENCHMARK

How do we tell whether Evidence helps?

At the same writing position, with the same model and task, how does supplying retrieved paper passages change the first output? We compare matched versions and retain ties, unusable outputs and incomplete reviews.

SAME DRAFT · EVIDENCE ON OR OFF

01 · FIXED WRITING POSITION

MANUSCRIPT

Same writing position ▌

Matched autocomplete at the same draft position

02 · TWO MATCHED INPUTS

SHARED BY BOTH

Manuscript context, model, task and prompt

A · With Evidence

Retrieved paper passages supplied

B · Without Evidence

No retrieved passages supplied

LLM

Same model and version

A → first output

B → first output

Blind judge agent

First outputs are anonymized as X and Y

Continuation review: accuracy, fit to the writing task and usability
Returns: X preferred / tie / Y preferred / both unusable

Order check: X / Y → Y / X

HOW DOES BLIND JUDGING WORK?

① Anonymize both outputs
The judge sees the same draft and both first outputs without knowing which received Evidence.

② Compare and swap order
The judge applies task-specific criteria in X/Y and then Y/X order.

③ Review disagreements
A third pass resolves disagreements. Incomplete reviews remain in the denominator.

How are autocomplete positions stratified?

Before seeing generation outcomes, we check whether a retrieved passage contains a specific proposition that directly supports the next writing move. Those positions appear in the left opportunity group; the rest are ordinary positions on the right. We select a balanced sample from admitted papers in each field. The 50/50 split is experimental, not a measure of how often either type occurs in writing.

AUTOCOMPLETE · 110

Biology, statistics and astrophysics

55 positions on each side; statistics uses the ten-paper rerun.

AUTOCOMPLETE · 52

Psychology and climate science

26 positions on each side; psychology includes nine papers and climate science four.

How are the table percentages calculated?

Across five fields there are 81 source-opportunity positions. The original blind review preferred Evidence in 51; a task check moved one empty Evidence continuation to both unusable, leaving 50 in public counts. In the ordinary group, another pair of empty outputs moved from no winner to both unusable. Original verdicts remain visible on case pages.

50Evidence version preferred
÷
81All positions in this group
=
62%Evidence preference in this group

Source contribution is a separate review: 16 of 21 Evidence wins entered into source review directly used retrieved papers; another 29 wins await review.

These are development-stage model judgments pending independent human review. They are not formal Benchmark conclusions and do not, on their own, establish causality.

Manuscript writing position

Text excerpt · not a PDF page

Research manuscript · excerpt

Causal inference methods for combining randomized trials and observational studies: a review

1 Introduction

…n criteria which can lead to a trial sample that differs markedly from the population potentially eligible for the treatment. Therefore, the findings from RCTs can lack generalizability to a target population of interest. This concern is related to the aim of external validity , central in medical research ( Concato et al.,, 2000 ; Rothwell,, 2005 ; Green and Glasgow,, 2006 ; Frieden,, 2017 ) policy research ( Martel Garcia and Wantchekon,, 2010 ; Deaton and Cartwright,, 2018 ; Deaton et al.,, 2019 ; Jeong and Namkoong,, 2022 ) , and other fields such as advertising ( Gordon et al.,, 2019 ) .

…er assumptions such as unconfoundedness it is possible to estimate a causal treatment effect from observational data. In practice, methods such as matching, inverse propensity weighting (IPW), or augmented IPW (AIPW) are used ( Imbens and Rubin,, 2015 ) . Even when a confounder is unobserved, solutions exist at the price of additional assumptions, for example the front-door criterion ( Pearl,, 1993 ) , instrumental variables ( Angrist et al.,, 1996 ; Hernán and Robins,, 2006 ; Imbens,, 2014 ) , and sensitivity analysis ( Cornfield et al.,, 1959 ; Rosenbaum and Rubin,, 1983 ; Imbens,, 2003 ) .

↓ Next paragraph was generated here

The highlighted text is near the writing position. This is a layout of the evaluation input, not a PDF screenshot.

A matched comparison at the same writing positionJudgment: No-Evidence version preferred

Both arms use the same model, task and manuscript context. Only the retrieved passages differ. The first recorded output from each arm is shown below.

Evidence

Same model and position · retrieved passages supplied · first output

Not preferred in this comparison

A related line of work develops identification strategies for settings in which treatment assignment has unobserved causes, including the front-door criterion and the edge g-formula for mediation and unobserved confounding.

Passages supplied to the Evidence version

Causal Inference with a Graphical Hierarchy of Interventions ↗

Our results motivate the study of estimation theory for the edge g-formula, since we show it arises both in mediation analysis, and in settings where treatment assignment has unobserved causes, such as models associated with Pearl’s front-door criterion.

No Evidence

Same model and position · no retrieved passages · first output

Preferred by blind review

A complementary strand of the literature has sought to combine these two sources of evidence rather than treat them as substitutes. By anchoring an observational analysis to the randomized estimate, such designs aim to retain the internal validity of the trial while borrowing information from the broader, more representative observational sample.

Judgment and task check

The blinded model review preferred the no-Evidence version. Its original rationale appears below.

Read the original model judgment

In this rationale, Output A = No Evidence; Output B = Evidence.

Output A provides a coherent, well-supported continuation that bridges the two draft paragraphs: it moves from discussing RCT generalizability limitations and observational methods to a natural next topic—combining RCT and observational evidence. This is a standard, well-established theme in causal inference literature that follows logically from the setup without requiring source support. Output B, however, makes a specific unsupported claim: it states that 'the edge g-formula' is developed 'for mediation and unobserved confounding' as a related line of work. The source only mentions the edge g-formula in passing as arising in mediation analysis and front-door settings, and explicitly says 'Our results motivate the study of estimation theory for the edge g-formula'—meaning it is not yet developed, and the source does not present it as an established identification strategy. Moreover, the draft already covered the front-door criterion and unobserved confounding; Output B merely repeats this with an unsupported specific mechanism (edge g-formula). The phrase 'related line of work develops' falsely presents this as an established literature, when the source says it motivates future study. This is an unsupported factual claim about the state of research.

The passages are input material; their presence does not mean an output used them or that they support every claim. Draft and source passages are excerpted; no simulated scores or PDF appear here.