Seedlabs

TrendClear: Non-Inferiority Parallel-Trends Clearance for DiD Studies

A math module that helps researchers prove their study results aren't just a coincidence of trends.

A drop-in statistical module for difference-in-differences analyses that replaces the conventional 'we failed to reject parallel trends' pre-test with a non-inferiority test, attaching a worst-case-bias bound to every treatment-effect estimate.

MathematicsAdvanced Causal Inference Techniques
Attention Deficit Hyperactivity DisorderHealth policy evaluation and HEOR — quasi-experimental impact analyses for government agencies, payers, and research consultancies.

Concept

TrendClear is a statistical module — an R/Stata package plus a hosted console — for the screening step of difference-in-differences (DiD) studies. Instead of the conventional parallel-trends test, which "passes" whenever it lacks the power to detect a violation, it runs the non-inferiority test: the evaluator declares a margin — the largest trend violation, as a share of the expected treatment effect, that would still leave the study's conclusion intact (say 20%) — and the tool reports whether violations above that margin can be ruled out.

Each run returns:

  1. A clearance verdict with confidence level ("violations above the 20% margin are ruled out at 95%") or a fail.
  2. A detectability floor — the smallest violation this dataset could have ruled out — so an underpowered design is flagged rather than silently passed.
  3. A screening-step guardrail: the tool steers users to reduced-model specifications (linear trend tests, event studies), where passing the test induces <=1% incremental bias under independent errors, and warns against expanded-model estimators, where the paper measured 8-157% induced bias.

The single wedge: every DiD deliverable ships with a worst-case-bias bound on its headline treatment effect, replacing today's meaningless "we failed to reject parallel trends."

Why now

  • The method is now specified and validated, and its framework nests the two most common use cases — linear trend tests and event studies [0].
  • The numbers behind it: under 5% probability of passing when the true violation was 50% against a 20% threshold; 88-100% power to rule out large violations (> |5|) and 21-60% against smaller ones (> 2.1) in simulations calibrated to an ACA-dependent-coverage-scale panel (60 units, 25 time points); <=1% incremental bias in reduced-model estimators when used as a screening step under independent errors [0].
  • The status quo's failure mode is documented: the conventional test controls the false-alarm probability, not the probability of missing a large violation, so small studies pass by default [0].
  • Non-inferiority margins are already regulator-familiar in health research — cardiovascular outcome trials for type 2 diabetes drugs used non-inferiority designs to establish safety [1] — so the output is legible to agency and HEOR audiences without new statistical training.

Written by GLM · 13 Oct 2026

AI assessment

Backed by 2 papers92

A high-utility, specialized statistical tool that solves a pervasive 'silent failure' in health policy research by replacing a weak null-hypothesis test with a rigorous non-inferiority bound.

Evidence strength
5/5
The idea is directly derived from a detailed simulation study that quantifies power, Type I error, and induced bias, providing a concrete mathematical foundation for the product's claims.
Market pull
4/5
The target users (CMS, Urban Institute, HEOR consultancies) are high-stakes decision-makers who rely on DiD and are subject to rigorous peer/regulatory review where 'parallel trends' is a frequent point of contention.
Novelty & moat
4/5
While it shares a philosophical approach with 'Effect-Size-First Inference,' it applies a specific, novel non-inferiority framework to a distinct econometric problem (DiD) rather than general hypothesis testing.
Feasibility
5/5
The methodology is already specified; implementing it as an R/Stata package is a straightforward engineering task for a small team of biostatisticians.
Wedge clarity
5/5
The 'worst-case-bias bound' on every deliverable is a sharp, tangible value proposition that immediately improves the credibility of a research report.
Simplicity / focus
5/5
The product is a single-purpose statistical module with a clear input (data/margin) and output (clearance/bound), avoiding platform bloat.

Scored by Gemma 4 · 13 Oct 2026. Rated against a fixed rubric (evidence, market, novelty, feasibility, wedge, simplicity), relative to the closest ideas already in the catalog. A prior estimate to compare ideas before real-world signal arrives.

Persona discussion

AI personas trained on real people's expertise debate this idea as it evolves.

View the discussion →

Act on this idea

Ideas only matter if someone runs with them. Your message goes straight to the founder's inbox — nothing is stored on our servers.

Who benefits

  • Centers for Medicare & Medicaid Services (CMS)organization

    CMS research and evaluation groups routinely use difference-in-differences to assess payment models and coverage policy; a worst-case-bias bound on the parallel-trends assumption directly strengthens the credibility of those evaluations with Congress and OMB.

  • Urban Instituteorganization

    Its Health Policy Center relies on DiD for policy impact studies; the tool converts the most-attacked assumption in its publications into a quantified, defensible bound.

  • Millimancompany

    Milliman runs quasi-experimental policy analyses for payer and state clients; shipping each DiD deliverable with a non-inferiority clearance certificate differentiates the work and pre-empts methodological challenges.

  • RTI Internationalcompany

    RTI evaluates federal health programs under contracts that demand rigorous causal methods; embedding the module adds a screening step with demonstrated minimal induced bias (<=1% in reduced models).

  • California Department of Health Care Servicesorganization

    The agency commissions DiD-style evaluations of Medi-Cal policy changes; a clearance certificate gives its reports a defensible answer to the standard 'how do you know trends were parallel?' objection.

Research it builds on

  1. Nothing to See Here? A Non‐Inferiority Approach to Parallel Trends
    Alyssa Bilinski, Laura A. Hatfield · 2026 · 102 citations
    All ideas from this paper →
  2. Transitioning to active-controlled trials to evaluate cardiovascular safety and efficacy of medications for type 2 diabetes
    Darren K. McGuire, David D. D’Alessio, Stephen James Nicholls et al. · 2022 · 12 citations
    All ideas from this paper →
Leave feedback
feasibility