Surface / Research / Methods / Research methodology

Michi Institute · Methods paper

Research methodology

One evidence foundation. Multiple analytical lenses. Conclusions that readers can trace back to the record.

Working draft 0.1 · October 27, 2026

Purpose

Michi Institute applies systematic inquiry to government, public policy, and civic participation. We begin with questions people encounter in their lives and trace them through public records, historical context, competing explanations, and measurable outcomes.

Our starting question was simple: why don't people vote? Case studies grew from that inquiry into a broader effort to help people understand what government does, how decisions are made, and what consequences follow.

The aim is to make public decisions easier to examine and to evaluate whether institutions raise the floor of opportunity, security, and participation. Public benefit and long-term sustainability matter in that evaluation. Profitability alone does not establish either.

One evidence foundation, multiple analytical lenses

The collection of evidence is separate from its interpretation. Original records form a shared foundation. Historical, economic, legal, political, and other analyses may examine those records through different lenses, using methods appropriate to their disciplines.

Competing theories remain visible when they offer materially different explanations. Each interpretation should identify its assumptions, methods, supporting evidence, limitations, and version. A change in interpretation should not erase the underlying record or the reasoning behind an earlier conclusion.

Nonpartisanship means applying the same evidentiary standards to every party, institution, and political position. It does not require equal conclusions when the evidence is unequal. Readers should be able to distinguish factual findings from judgments about which outcomes matter most.

The case-study sequence

Each investigation follows five connected questions:

1. What happened? Establish events, participants, decisions, and chronology from identifiable records.
2. Why might it have happened? Compare plausible explanations, stated intentions, incentives, constraints, and relevant historical conditions.
3. What did government do? Identify the actual policy, vote, rule, spending decision, enforcement action, or failure to act, including who had authority.
4. What changed? Examine implementation and outcomes over a defined period, including who benefited, who bore costs, and what remains difficult to measure.
5. What can the evidence establish? State the supported conclusion, its limits, competing explanations, and the questions still open.

The sequence is a discipline for inquiry, not a guarantee that every question can be resolved. An investigation may end with a documented uncertainty.

Defining the question and scope

Before drawing conclusions, specify the question, jurisdiction, time period, affected population, and relevant outcomes. Explain why the case was selected and whether it illustrates a broader pattern or only a particular event.

Where possible, define the outcomes and comparisons before evaluating results. Record meaningful changes in scope so readers can see when new evidence changed the investigation. A case study alone cannot establish how common a pattern is.

Finding and identifying sources

Seek original records wherever available: enacted text, official ballot materials, legislative proceedings, budgets, financial disclosures, court records, administrative data, and contemporaneous documents. Use scholarship and reporting to establish context, discover records, and test interpretations.

Label source types clearly: official record, peer-reviewed research, working paper, reporting, advocacy, personal account, or other material. Institutional affiliation does not by itself establish peer review. An official source can establish what an agency reported without settling whether that report is complete or correct.

For research, record the author, publication date, publication type, peer-review status, method, sample or study period, limitations, and available funding disclosures. Record retrieval or verification dates for material that can change. Describe search coverage honestly; a list of results is not evidence of a comprehensive search.

Assess each source for the specific claim it supports. Consider provenance, directness, methods, context, incentives, missing information, and independent corroboration. Several articles repeating one underlying report remain one evidentiary chain.

Preserving the evidence trail

The proposed preservation standard retains the original file where permitted, readable text, images of important pages when needed, the source address, and an archive reference when available. Record the document's date, retrieval date, relevant pages, and relationship to the claim.

Preservation copies protect access; they do not provide independent corroboration. File fingerprints can help establish that a retained file has not changed. They cannot establish that its contents are true.

Major findings should link directly to the relevant document, passage, table, or record. A source-linked evidence ledger should connect each finding to its support, qualifications, conflicting evidence, and review history.

Separating records, claims, and conclusions

Use explicit language to distinguish what a document records, what a participant alleges, what the analysis infers, and what remains unresolved. Describe a policy's stated purpose separately from its implementation and observed effects.

Agreement, shared interests, financial ties, or the existence of a plan may justify further investigation. None establishes coordination, intent, implementation, or causation on its own. Absence of a located record also requires care: report what was searched and what was not found without treating an incomplete search as proof of absence.

Testing explanations and evaluating outcomes

Examine evidence that challenges the initial explanation as deliberately as evidence that supports it. Compare alternatives and identify what evidence would strengthen, weaken, or overturn each explanation.

A change following a policy does not alone establish that the policy caused it. Where the evidence permits, examine baselines, comparison groups, timing, external events, selection effects, and alternative causes. Explain the limits of the chosen method.

Report practical effects and their distribution: benefits, burdens, access, costs, durability, and unintended consequences. State the time horizon and whose experience is represented. Preserve competing economic interpretations when the available evidence does not resolve them.

Uncertainty and scoring

Explain confidence at the claim level, including why a finding is well supported, contested, preliminary, or unresolved. These are proposed descriptive labels, not a validated numerical rating system. Confidence in a fact and preference for a policy are separate judgments.

Any future score must disclose its inputs, definitions, weights, missing-data treatment, and sensitivity to assumptions. Do not imply a scoring framework is established until its rules and validation are published.

AI assistance, human review, and member participation

AI may assist with locating material, transcription, extraction, comparison, organization, and drafting. Its output is a lead or analytical aid, not an independent source. Citations, quotations, calculations, and consequential interpretations require checking against the underlying evidence.

The proposed publication standard requires a named human reviewer for substantive conclusions, with a record of the version reviewed, date, checks performed, and remaining limitations. Automated assistance must not be presented as completed human review. Material AI involvement should be disclosed.

Member participation can help identify questions, priorities, and criticism. Evidence evaluation remains separate from member voting: popularity cannot establish a factual claim.

Publication, corrections, and continuity

Begin each public case with a concise account of the finding and its uncertainty. Provide a path into the timeline, evidence, methods, financial records where relevant, related cases, and unanswered questions. Explain consequences for different political priorities while making the assumptions behind those interpretations visible.

The publication standard should include authorship, review status, publication and revision dates, methods, source links, limitations, and relevant conflicts of interest. Preserve versioned interpretations and correction histories so another researcher can reconstruct the analysis.

Publish substantive criticism and the response, including why a proposed change was accepted or declined, with appropriate privacy protections. Correct factual errors visibly. Distinguish a correction from a conclusion revised because new evidence became available.

The research methodology belongs alongside the evidence. The Development Ledger serves a separate purpose: documenting website and design decisions.

Before adopting this draft

Confirm the actual review workflow and responsibility; agree on evidence labels; inventory which pages meet the preservation and citation standards; and define the correction process. Until those checks are complete, retain this document's working-draft status and identify incomplete practices accurately.

Back to the paper title · Research library