Before you spend money collecting data, find out whether your study can answer the question.

The problem

Most failed studies do not fail because the hypothesis was wrong.

They fail because the design never had a realistic chance of answering the question. The sample was too small for the effect. The instrument could not register the change. The recruitment frame missed the population. The power calculation rested on an assumption that did not hold.

Funders review budgets. Ethics committees review participant protection. Statistical review often comes after data collection.

That is why forskai exists. We move that question to the planning stage, when the design still costs nothing to change.

An estimated $28 billion a year goes to US preclinical research that does not reproduce. Much of it was settled at the design stage. Freedman et al., 2015

What can fail

A trial is powered for a moderate effect. The intervention produces a smaller one. The study concludes “no difference.” Years later, another group finds the effect with a better design.

That study did not fail on its hypothesis. It failed in one of four familiar places.

• Measurement

  Ceiling and floor effects. Instrument sensitivity. Whether the measure covers the construct you named.

• Design

  Sample size. Missing data. Recruitment. Follow-up long enough to see the change.

• Analysis

  Model identification. Confounding. Estimator behaviour. The assumptions the analysis quietly relies on.

• Data pipeline

  Coding. Transformation. Calibration. Whether the numbers survive the trip from collection to result.

How it works

Every review answers one question.

If the effect is real, can this study recover it?

1. STEP 1

   Understand the study

   We read the research question, the estimand, the measurements, the assumptions, and the planned analysis.

2. STEP 2

   Stress-test the design

   We simulate how the design behaves under realistic uncertainty: measurement error, missing data, effect sizes at the low end of plausible.

3. STEP 3

   Deliver the review

   You receive a documented read of the design's strengths, its risks, and the changes worth making, before data collection begins.

• ●Recoverable
• ▲At risk
• ■Not recoverable
• ◆Not resolvable at this n

What forskai can stress-test

The designs a recoverability test covers.

Each estimator is cross-checked against an independent R implementation, so the recovery a design earns is the same in Python and in R.

• Two-arm trial, continuous outcome

  A treatment and a control compared on a continuous measure — the t-test workhorse.

• Two-arm trial, binary outcome

  A yes/no or event outcome between two arms, analysed by logistic regression.

• Factorial 2×2 / interaction

  Two factors crossed — whether the design recovers a moderation effect, not only the main effects.

• Cluster-randomised / multilevel

  Randomisation by clinic, class, or ward — recovery after the intracluster correlation eats effective sample size.

• Regression with correlated predictors

  Multicollinearity — whether a coefficient stays identifiable, or the design cannot separate the effects.

• Observational study with confounding

  A non-randomised comparison — adjusted versus crude bias, and whether the effect survives adjustment.

• Meta-analysis

  Pooling across studies — whether the combined evidence recovers the effect given the spread between studies.

A verdict is design evidence under stated assumptions — never validity about people. More design families (latent-variable and survival models) are in development.

The offer

Two design pilot moments to check a design. Both before the data.

Applying now? Write the review into your budget as external methods consulting, and we deliver after the award.

What you’ll receive

A written, signed report.

The Research Design Review is a written, signed report. Clients buy the review; the stress test is how we produce it.

Pricing

Priced by the decision it protects.

We price the decision, not a bundle of features. What you are buying is protection against an expensive mistake, on a study that will cost far more than the review.

Funding

The grant can pay for this.

External methods consulting is an eligible cost in Horizon Europe (“other goods, works and services”), in NIH budgets, where statistical consultation is a named example, and in the Nordic national funders. A design review at this price is one to three percent of a typical national project grant, and it sits below the direct-purchase threshold in Finland, Sweden, Norway, and Denmark.

A sole-source justification memo is available on request.

A stress test is only as strong as the numbers you feed it.

Every recovery test rests on assumptions: the effect size you expect, the variance you assume, the base rates you take for granted. The Assumption Review grounds those numbers in what the current literature actually supports, so your design test starts from defensible inputs instead of hopeful ones.

Proof

We tested the method on our own published research before offering it to anyone else.

We took our own mixed-methods study, a cross-sectional survey of how healthcare professionals in Finland understand palliative care and their readiness for early integration, and ran the design review on it retrospectively. The report shows what the review would have caught at the design stage, the pitfalls we found only later, and the changes that now carry the work toward validation of a clinical tool.

About

Every report carries a signature.

Research design should be auditable in the same way as statistical analysis. Every review forskai delivers is written, signed, and justified, so you can show a co-investigator, a supervisor, or a funder exactly how the verdict was reached.

forskai was founded by Heidi Helena Andersén, MD, PhD, a thoracic-oncology clinician-researcher who designs, analyses, and publishes clinical studies, and who builds open research software for transparent, reproducible workflows. The judgement in every review comes from work like this:

• Construct ambiguity, at registry scale. An asthma–COPD overlap study on decades of Finnish hospital discharge records, where prevalence moved with the operational definition.

  Andersen H, Lampela P, Nevanlinna A, Säynäjäkangas O, Keistinen T. High hospital burden in overlap syndrome of asthma and COPD. The Clinical Respiratory Journal, 2013. doi:10.1111/crj.12013

• Harmonisation across countries. A population-based Nordic study of the social gradient in respiratory symptoms, harmonising variables and controlling confounding across national cohorts.

  Andersén H, Bhatta L, Bashir M, Nwaru B, Langhammer A, Krokstad S, Piirilä P, Hisinger-Mölkänen H, Backman H, Kankaanranta H, et al. Is there still a social gradient in respiratory symptoms? A population-based Nordic EpiLung study. Respiratory Medicine, 2024. doi:10.1016/j.rmed.2024.107561

• The limits of real-world causal claims. A guideline-adherence study in elderly non-small-cell lung cancer, testing where causal interpretation of observational data has to stop.

  Lindqvist J, Jekunen A, Sihvo E, Johansson M, Andersén H. Effect of adherence to treatment guidelines on overall survival in elderly non-small-cell lung cancer patients. Lung Cancer, 2022. doi:10.1016/j.lungcan.2022.07.006

ORCID 0000-0001-5923-5865·LinkedIn

The boundary

The boundary

forskai does not guarantee successful results, does not validate scientific truth, and does not replace statistical judgement.

It gives you evidence about whether the planned design is likely to answer the question, under the assumptions you state. When the sample cannot settle that question, the verdict says not resolvable at this sample size instead of guessing. A “not recoverable” verdict is not a failed project. It is the cheapest early warning you will ever get.