The added report sharpens the evidence base for both themes. Its central external source, Peñasco, Anadón and Verdolini’s systematic review, can be independently verified: the review covers 211 publications across ten decarbonisation instruments and explicitly reports sparse non-OECD coverage, limited instrument–outcome evaluations, and substantial disagreement in distributional findings. (Peñasco, 2021)

For geographic and instrument-specific green public expenditure, the report’s strongest defensible formulation is narrower and more precise than a generic call for “more spending research.” Most reviewed studies assess national policies in OECD countries; Sub-Saharan Africa, apart from South Africa, and the Middle East have very little coverage. The review also records only one distributional evaluation of government procurement, two of auctions, and one and three evaluations, respectively, for R&D funding and government procurement. (Peñasco, 2021) This supports a concrete study: evaluate one identifiable procurement, auction, or R&D programme in an under-studied jurisdiction using award-level or project-level administrative data, with outcomes for supplier participation, concentration, innovation, emissions, and household or worker exposure. The study should exploit a pre-existing eligibility threshold, scoring rule, or phased rollout rather than infer causality from aggregate green-budget totals.

For distributional contradictions and public acceptance of environmental taxes, the report’s framing is also well grounded. The verified review finds mixed evidence on the distributional effects of energy taxes and high disagreement across evaluations of emissions-allowance schemes; it concludes that instrument-design differences explain many contradictory results, beyond differences in methods, sectors, or contexts. (Peñasco, 2021) The complementary acceptance review finds stronger support where people understand policy content and revenue use, trust government, perceive fair cost sharing, and are concerned about climate change. It calls for research that disaggregates trust into responsiveness, reliability, integrity, openness, and fairness, and examines economic perceptions and tax morale. (Muhammad, 2021)

The most valuable refinement is therefore a linked incidence–acceptance design. Use household consumption or administrative data to model direct and indirect tax burdens under alternative revenue-recycling packages, then link those modeled net positions to a repeated survey or randomized policy-information experiment. Test separately whether support follows objective net incidence, perceived fairness, confidence in revenue administration, environmental concern, or the stated destination of revenues. This design can explain why a package that is distributionally progressive is not necessarily popular, and why earmarked environmental spending can gain support even when it offers less direct compensation.

The report’s practical sequencing is sound: begin the expenditure theme only after securing a specific programme and administrative-data partner; the tax theme is more immediately fieldable when a household expenditure panel and a policy-relevant survey frame can be linked. The prior report’s named literature must not be treated as self-validating, but its two central gap claims are supported by the independently verified systematic review and acceptance-review evidence above. (Peñasco, 2021) (Peñasco, 2021)

Preregistration-ready research designs for the eight gaps

Design architecture

The eight gaps require eight coordinated studies, not one omnibus regression. Studies 1–4 evaluate green public expenditure at the budget, contract, local-market, and long-horizon portfolio levels. Studies 5–8 evaluate environmental-tax incidence and acceptance at the household, policy-package, measurement, and institutional-trust levels. A common data spine makes the programme cumulative: each study records the fiscal instrument, geographic unit, eligibility rule, actual financial exposure, and a comparable acceptance or legitimacy measure.

This separation follows the evidence. The decarbonisation review reports sparse non-OECD coverage, few evaluations of several instrument–outcome combinations, and disagreement in distributional outcomes; it also identifies only one distributional evaluation of government procurement and two of auctions. (Peñasco, 2021) (Peñasco, 2021) The acceptance review finds that information, trust, perceived fairness, and revenue use are associated with support, while calling for more specific measures of trust, economic perceptions, and tax morale. (Muhammad, 2021) (Muhammad, 2021)

For every study, pre-register the exact jurisdiction, policy period, codebook, linkage protocol, treatment assignment, estimand, outcome hierarchy, covariates, subgroup definitions, missing-data rule, and stopping rule before accessing post-treatment outcomes. Administrative data should be pseudonymised and linked in a secure setting; survey linkage requires explicit consent and a separation between identifiers and analytic data.

1. Comparable expenditure measurement before effectiveness estimation

Question. Does a harmonised, programme-level classification of green expenditure change estimates of where spending occurs, who benefits, and whether public spending is associated with environmental outcomes?

Rationale and hypotheses. Environmental public-expenditure research has been concentrated in Europe and Latin America, often uses qualitative case studies, and calls for stronger transparency, accounting organisation, and indicators. (Pacheco, 2017) Pre-register H1: harmonised coding will reclassify a non-zero share of reported “green” expenditure relative to headline budget categories. H2: programme-level measures will show greater between-place heterogeneity than aggregate environmental-spending totals. H3: respondents receiving a disaggregated, independently validated spending account will report higher comprehension and more calibrated perceived effectiveness than respondents receiving an aggregate total. H3 concerns the information intervention, not causal effectiveness of spending.

Population, groups, and data. Select one country with at least 20 subnational governments and three fiscal years before fieldwork. Include all budget lines tagged by either the fiscal authority or an independently constructed keyword/classification rule. Code each line as operating expenditure, capital expenditure, grant/subsidy, procurement, public R&D, tax expenditure, or other; add mitigation/adaptation objective, beneficiary type, implementation stage, geography, and amount committed/disbursed. The comparison is (i) official aggregate category versus harmonised programme classification and (ii) jurisdictions with high versus low independently assessed coding completeness. Link the coded data to project locations and administrative environmental outputs only when identifiers permit.

Information intervention. Randomly assign a probability sample of adult residents within each jurisdiction to: aggregate green-spending total; harmonised instrument-by-objective dashboard; or the same dashboard plus an independent audit statement. Stratify randomisation by jurisdiction and baseline environmental concern. Primary survey outcomes are factual comprehension, perceived transparency, confidence in spending administration, and support for continuation; secondary outcomes are perceived household benefit and willingness to request more information.

Identification and analysis. The measurement estimand is the within-budget difference in expenditure shares under the two coding systems. Report inter-coder reliability from a double-coded holdout sample, disagreement by instrument, and sensitivity to alternative classification dictionaries. For the experiment, estimate intention-to-treat (ITT) effects with assignment indicators and jurisdiction fixed effects; use randomisation inference as the primary p-value method. Do not interpret cross-jurisdiction associations between spending and outcomes causally. Public expenditure reviews stress that low spending alone does not establish a need for more spending; effectiveness against environmental priorities is the relevant test. (Markandya, 2006)

Validity and preregistration. Pre-specify the codebook, adjudication rule, audit sample, primary outcome order, survey attention checks, and exclusions limited to duplicate or impossible records. Report dashboard effects separately from baseline trust. Threats include strategic budget labeling, non-comparable accounting systems, survey demand effects, and nonresponse; address them through blinded coding, a preregistered alternative-codebook analysis, survey weights, and a neutral control display.

2. Instrument comparison: procurement, grants, and public R&D

Question. At equal public cost and within one technology sector, which instrument produces the greatest environmental improvement, innovation, supplier entry, and equitable access?

Rationale and hypotheses. Existing procurement research has limited comparative evidence against other environmental policy tools and restricted geographic coverage. (Cheng, 2018) Theory cautions that procurement can be ineffective or counterproductive if its environmental standard does not exceed prevailing technology or lacks a defined environmental objective. (Lundberg, 2015) Pre-register H1: green procurement awards increase product innovation among eligible SMEs relative to comparable non-awarded bidders. This is consistent with a firm-level difference-in-differences study reporting a higher probability of environmentally friendlier products after green-award receipt, with no significant process-innovation effect. (Krieger, 2022) H2: instrument effects differ across procurement, grant, and R&D support after normalising by public cost. H3: more stringent criteria increase environmental outcomes but may reduce bidder entry or increase award concentration.

Population, groups, and data. Restrict to one sector—building retrofits, low-emission buses, or renewable-energy equipment—and to programme cohorts launched under the same regulatory regime. Link tender notices, bids, technical scores, award values, grant/R&D applications, firm registry data, patents or product certifications, employment, supplier ownership, and verified project-level energy/emissions outcomes. Define treated firms as winners of green procurement awards, grants, or R&D support; define comparison firms as eligible unsuccessful applicants or bidders within the same tender/application cohort. Measure public cost as disbursement plus procurement premium, not contract value alone.

Treatment/comparison design. The primary comparison is winner versus narrowly unsuccessful eligible applicant within each instrument. The secondary comparison is the difference between instrument-specific treatment effects. If a published scoring threshold determines award, implement a fuzzy regression discontinuity; otherwise use cohort-by-buyer fixed-effect event studies with matched applicants and pre-treatment outcomes. Do not pool instruments unless overlap in firm, product, and eligibility covariates is demonstrated.

Stakeholder survey intervention. Among bidders and local residents, randomise a standard contract-performance brief that reports verified lifecycle emissions, award premium, SME participation, and audit status versus a generic green-procurement description. Outcomes are perceived value for money, legitimacy, confidence in fairness, and support for repeated use. The intervention tests transparency effects, not the contract’s real environmental impact.

Outcome hierarchy and analysis. Primary outcomes: verified emissions or energy savings per public monetary unit; firm product innovation; bidder entry; and award concentration. Secondary: process innovation, employment, subcontractor participation, and support. Estimate event-time ITT coefficients, test pre-treatment leads jointly, cluster uncertainty at buyer or programme level, and use randomisation inference for the survey. A planned heterogeneity analysis separates SMEs from larger firms, given evidence that procurement’s product-innovation effect was concentrated among SMEs. (Krieger, 2022)

Threats and safeguards. Selection of better firms into applications, manipulation around thresholds, bundled subsidies, reporting incentives, and strategic bidding threaten inference. Pre-register balance diagnostics, density and covariate-continuity tests for any threshold, contract-specific treatment dates, exclusion of overlapping awards in the primary analysis, and a specification that re-includes them as a spillover analysis.

3. Local implementation conditions and spatial spillovers

Question. Do procurement implementation intensity and administrative capacity alter local environmental, innovation, competition, and legitimacy outcomes, including effects on neighbouring areas?

Rationale and hypotheses. The literature calls for evidence on localised factors in developing-country procurement and recognises tension between uniform policy requirements and local flexibility. (Carrasco, 2024) (Xi, 2026) Municipal green-building procurement has been evaluated using matching, panel data, and instrumental variables, with reported spillovers to private adoption and local green-building expertise. (Simcoe, 2014) Pre-register H1: a local mandate or rollout increases green-contract intensity and private adoption in the treated locality. H2: effects are larger where administrative capacity and a relevant supplier base are stronger. H3: neighbouring localities experience measurable spillovers rather than serving automatically as untreated controls.

Population, groups, and data. Build a municipality-by-year panel covering at least three pre-policy years. Collect mandates and rollout dates, tender texts, buyer capacity measures, e-procurement adoption, local supplier locations, firm outcomes, technology adoption or emissions measures, commuting/trade links, and local socioeconomic characteristics. The treated group is municipalities adopting or receiving the programme in each rollout cohort. Comparators are not-yet-treated municipalities matched on pre-policy trajectories and capacity; define neighbours before analysis using contiguous borders and a pre-specified commuting or supplier-network matrix.

Identification. Estimate staggered-adoption event studies using cohort-specific comparisons and report treatment effects by event time. Include direct-treatment and pre-specified spatial-exposure measures in the same model, rather than omitting neighbours. If rollout is determined by an administrative eligibility score, add a local regression-discontinuity analysis around that threshold. The estimand is the average direct effect among treated municipalities and the average spillover effect among pre-defined neighbours, not a national average.

Survey/intervention. Run a clustered survey of firms and residents in treated, comparison, and neighbouring municipalities. Randomise a factual local dashboard showing programme requirements, awarded suppliers, verified environmental results, and complaint channels versus a generic national-policy description. Primary outcomes: perceived procedural fairness, perceived local benefit, trust in procurement administration, and willingness to bid or support expansion. Stratify by firm status and household proximity to affected infrastructure.

Analysis plan and validity. Pre-register the adoption date, spatial matrix, treatment window, primary outcome sequence (contract intensity, private adoption, supplier entry, local support), fixed effects, and treatment-effect estimator. Assess parallel trends with leads; conduct placebo adoption dates; report estimates after excluding municipalities with contemporaneous climate programmes. Key threats are endogenous rollout, migration of suppliers, cross-border procurement, outcome measurement differences, and interference. These are addressed by rollout documentation, supplier-location tracking, explicitly modeled spillovers, and outcome definitions fixed before data linkage.

4. Long-horizon expenditure portfolios, additionality, and private investment

Question. Do sequenced green-expenditure portfolios generate delayed environmental and innovation effects and additional private investment, beyond annual spending totals?

Rationale and hypotheses. An EU analysis reports positive associations between absolute environmental-protection expenditure and green total factor productivity, but not expenditure intensity; it explicitly cannot establish definitive causality, has a short observation period, excludes private environmental investment, and may miss effects requiring more than two years. (Ibrić, 2026) Another country-panel study found no significant independent relationship between government environmental spending and its green-investment proxy and did not support a strong causal claim about complementarity with carbon pricing. (Abdulhasanova, 2026) Pre-register H1: commitment-to-disbursement timing predicts delayed project outputs rather than contemporaneous annual outcomes. H2: portfolios combining infrastructure, procurement, and R&D have larger medium-run effects than single-instrument spending at the same cost. H3: private investment response differs by carbon-price exposure and financial conditions; this is tested as a hypothesis, not assumed.

Population, groups, and data. Construct a programme–place–year panel with at least five pre-treatment years and five post-commitment years where feasible. Link commitments, disbursements, project completion dates, procurement awards, R&D grants, facility locations, private co-financing, firm investment, patents, emissions, energy output, and public cost. Define portfolio categories before outcome analysis from programme documentation. The comparison group comprises eligible places or firms not yet receiving an otherwise similar programme; never use completed projects as controls for projects not yet started without accounting for timing.

Identification. The primary specification is an event study around commitment, disbursement, and operational-completion dates, estimated separately. Where eligibility rankings or formula allocations exist, use the predetermined score or formula shock as the quasi-experimental source; otherwise label results associational. Estimate dynamic effects and report cumulative public cost, private investment, and environmental outcomes together. The causal estimand is the effect of programme eligibility or timing, not simply the coefficient on annual expenditure.

Survey/intervention. For sampled host communities and local firms, randomise a plain-language timeline showing commitment, disbursement, completion, expected benefits, and independent performance monitoring versus an annual-spending-only display. Outcomes are comprehension of policy timing, perceived additionality, and continued support. This tests whether transparent lag information changes interpretations of spending that has not yet produced outcomes.

Preregistered analysis and validity. Primary outcomes: verified project completion, emissions/energy outcome, and private co-financing. Secondary: patents, employment, local support, and green productivity. Fix lags, primary event time, portfolio rule, and data-winsorisation threshold in advance. Threats are anticipation, concurrent regulation, endogenous project siting, unobserved private investment, and attrition of delayed projects. Use pre-trend plots, negative-control outcomes, programme-place fixed effects, a documented co-policy inventory, and sensitivity bounds for incomplete private-finance observation.

5. Objective tax incidence, perceived burden, fairness, and support

Question. When households receive credible information about their own modeled carbon- or energy-tax burden and rebate, does objective net incidence, perceived incidence, fairness, or trust best explain support and subsequent behaviour?

Rationale and hypotheses. Carbon-tax distributional effects depend on energy use, measurement indicator, and household characteristics; lump-sum transfers are described as better suited than labor-tax reductions to mitigate regressive effects for lower-income households. (Köppl, 2022) Acceptance is associated with information, revenue use, trust, and fairness. (Muhammad, 2021) Pre-register H1: personalized, accurate net-incidence information reduces the gap between perceived and modeled household burden. H2: a rebate described and modeled as progressive increases support among net beneficiaries relative to an otherwise identical no-information control. H3: perceived fairness and trust explain residual variation in support after modeled net incidence is included.

Population, groups, and data. Draw a probability sample of adult household financial decision-makers, oversampling rural households, low-income households, high-energy users, and workers in exposed sectors. Link consented survey responses to household-expenditure microdata or bills, household composition, income band, residence, vehicle and heating characteristics, the tax schedule, and actual or simulated transfer eligibility. Produce a prespecified microsimulation of direct and indirect tax incidence and recycling; retain an uncertainty flag when expenditure data are insufficient.

Randomised information intervention. After baseline measures, randomise within pre-specified income–urbanicity–baseline-support strata to: (A) generic tax description; (B) personalised gross tax burden; (C) personalised gross burden plus modeled universal rebate; (D) personalised net burden plus targeted progressive transfer; or (E) net burden plus a verified account of environmental use of remaining revenue. Use identical tax rate and revenue total across arms. Include comprehension checks but analyse all assigned participants by ITT.

Outcomes and identification. Primary outcomes are binary support for the specified policy and absolute error between perceived and modeled net burden. Secondary outcomes are fairness, trust dimensions, maximum acceptable tax rate, stated coping choices, rebate take-up where administratively observable, and later energy use or voting/participation where lawful and feasible. Estimate arm-versus-control ITT effects, then pre-register a mediation analysis that treats perceived burden and fairness as post-treatment mediators and is explicitly exploratory unless its identification assumptions are satisfied.

Validity and analysis plan. Freeze the tax calculator, transfer rules, question wording, outcome order, and treatment text. Estimate weighted linear probability or logistic models with randomisation strata; report unadjusted ITT as the primary estimate and covariate-adjusted precision estimates second. Handle missing baseline covariates by prespecified multiple imputation; do not impute primary outcomes. Threats include incorrect self-reported expenditure, distrust of the calculator, treatment-induced demand effects, energy-price shocks, and selective linkage consent. Report simulation uncertainty, use a neutral third-party audit statement, retain non-consenting respondents for survey-only ITT, and repeat calculations under pre-specified price assumptions.

6. Revenue recycling as a policy-package contradiction

Question. Which revenue-recycling package jointly improves modeled distributional protection and policy support, and how does its effect depend on the prevailing tax-policy baseline?

Rationale and hypotheses. The fairness review finds concern about uneven burdens but frequent preferences for environmental projects rather than compensation, while recommending a combined vulnerable-household and environmental-project approach. (Maestre‐Andrés, 2019) Conversely, a Swiss conjoint experiment found no average acceptability increase from redistribution; progressive redistribution was favorable only under particular tax-policy baselines and among some political groups. (Prinzing, 2026) Pre-register H1: targeted transfers minimise modeled losses among the lowest net-income groups more than uniform tax cuts. H2: a mixed targeted-transfer plus visible environmental-investment package gains more support than either component alone. H3: estimated effects differ between respondents informed about the existing policy baseline and respondents not given that baseline.

Population, treatments, and data. Use the Design 5 household sampling frame and incidence model. Conduct a factorial conjoint in which every policy vignette fixes the tax base, tax rate, emission goal, and total revenue. Randomize four attributes: recycling package (universal dividend, targeted transfer, labor-tax reduction, environmental investment, or transfer-plus-investment mix); disclosure of respondent-specific net position; independent revenue audit; and policy baseline (status quo policy described versus absent). Randomize profile order and attribute levels; constrain profiles to budget neutrality.

Comparison and outcomes. Within respondent, compare support and fairness ratings across packages. The primary outcomes are choice of preferred package and predicted support for implementation. Secondary outcomes are perceived regressivity, credibility of revenue use, comprehension, and support conditional on a specified tax rate. Report distributional results separately from attitudinal results: a package is not labelled “equitable” solely because it is popular.

Identification and analysis. Randomisation identifies average marginal component effects and prespecified interactions for targeted transfer × environmental investment, net-position disclosure × package, and baseline × package. Estimate respondent-clustered standard errors and a hierarchical model that permits person-level variation without reclassifying exploratory interactions as confirmatory. Pre-register whether the policy choice task or a rating is primary, maximum number of profiles, attention checks, and removal only of mechanically invalid observations.

Threats and safeguards. Hypothetical choice, cognitive overload, social desirability, and policy distrust can distort effects. Use plain-language pretesting, a comprehension screen that does not determine primary-sample inclusion, randomized profile order, a no-policy status-quo option, and one incentivized information-choice task. Treat cross-country estimates as separate replications unless tax schedules and social-protection systems are demonstrably comparable.

7. Measurement of acceptability and prediction of behaviour

Question. Which acceptance measure—explicit support, maximum acceptable rate, stated willingness to pay, or a consequential action—best predicts later participation, consumption adjustment, or policy vote?

Rationale and hypotheses. Acceptability findings diverge with elicitation method, policy design, and context; explicit formats produced lower acceptance than implicit model-based measures in a within-subject study, and preferred and maximum acceptable tax levels differed. (Mölk, 2026) Earlier information interventions have yielded positive, negative, and null effects, with one panel analysis reporting an effect among less-educated respondents rather than across the full sample. (Ruprecht, 2023) Pre-register H1: explicit support, maximum acceptable rate, and willingness-to-pay measures are not interchangeable. H2: a consequential choice has stronger prospective association with a later observable behaviour than any single stated measure. H3: measurement differences vary by education and baseline policy knowledge.

Population, tasks, and comparison. Recontact the Design 5 panel at baseline, immediately after the information intervention, and after a predefined follow-up interval or policy event. Randomise task order among: referendum-style yes/no support; maximum acceptable tax-rate elicitation; budget-allocation/willingness-to-pay task; and a consequential action such as authorising a no-cost rebate-information sign-up, choosing an actual policy-information link, or allocating a small study incentive to a climate or household-support fund. The consequential action must be real, clearly explained, and approved by ethics review.

Outcomes and identification. Primary outcome: pre-specified observable follow-up behaviour, such as verified rebate enrolment, attendance at an information session, or consented energy-use change. Where a public vote occurs, self-reported vote is secondary unless record linkage is lawful and consented. The primary estimand is the difference in predictive performance for the follow-up outcome across baseline measures, assessed by preregistered out-of-sample cross-validation. Randomised task order identifies order effects, not the causal effect of attitudes on later behaviour.

Analysis plan. Pre-specify the prediction models, scoring rule, training/test split, calibration assessment, and handling of repeated measures. Estimate within-person differences across elicitation modes with respondent fixed effects. Declare the study unsuccessful for behavioural validation if the follow-up outcome cannot be observed at the preregistered date; do not substitute a different behavioural proxy after seeing results.

Threats and safeguards. Attrition, weak consequential stakes, learning across repeated tasks, and changes in energy prices threaten validity. Use panel-retention incentives unrelated to treatment, randomised task order, a documented minimum-stakes threshold, and inverse-probability weights for follow-up nonresponse. Keep policy information constant across elicitation modes.

8. Institution-specific trust, revenue administration, and acceptance

Question. Which verifiable institutional feature—transparent revenue accounting, independent audit, reliable compensation delivery, participatory rule-setting, or even-handed enforcement—changes trust and support for an environmental tax?

Rationale and hypotheses. The acceptance review calls for measures beyond generic trust, including responsiveness, reliability, integrity, openness, and fairness. (Muhammad, 2021) Qualitative evidence from the United Kingdom identifies distrust in revenue use, difficulty understanding a tax shift, and desire for incentives alongside penalties as design problems. (Dresner, 2004) Pre-register H1: verified information about timely compensation delivery and independent auditing increases reliability and integrity ratings relative to generic government assurances. H2: a participatory rule-setting opportunity increases procedural-fairness ratings. H3: effects on support are mediated descriptively by the specific trust dimensions targeted, not by a single omnibus trust score.

Population, intervention, and data. Use the same representative household frame, with a parallel sample of small firms where the tax materially affects business energy use. Verify all institutional claims against actual programme records before fielding. Randomise a 2 × 2 × 2 factorial vignette: independently audited public revenue ledger versus generic statement; verified on-time targeted-payment record versus no delivery record; and public consultation/enforcement explanation versus no process information. Hold tax rate, tax base, revenue recycling, and household net position fixed.

Outcomes and identification. Primary outcomes are five separately scored dimensions: responsiveness, reliability, integrity, openness, and fairness. Secondary outcomes are policy support, perceived effectiveness, willingness to accept a higher rate, and intended claim/take-up of compensation. Estimate factorial ITT main effects and pre-registered two-way interactions; control family-wise error within the five primary trust dimensions. Link, where possible, to actual compensation delivery and complaint-resolution data to test whether the treatment’s factual premise corresponds to administrative performance.

Analysis plan and validity. Pre-register wording, source documents, factorial cells, outcomes, multiplicity correction, and subgroup definitions (income, rurality, prior tax experience, political orientation, and exposure). Use randomisation-based inference and report every cell mean. Threats include disbelief in institutional claims, polarized prior beliefs, reputational shocks during fieldwork, and treatment spillovers. Measure baseline beliefs, timestamp fieldwork, include a manipulation check, and repeat the primary analysis excluding respondents who fail a basic factual comprehension item only as a sensitivity analysis.

Cross-study preregistration core

Register each design separately but deposit a shared protocol before outcomes are analysed. The protocol should include: (1) research questions and directional versus non-directional hypotheses; (2) exact policy and jurisdiction; (3) population, recruitment, consent, and inclusion criteria; (4) treatment texts and assignment code; (5) data dictionary and linkage keys; (6) a primary-outcome hierarchy; (7) estimands, model equations, fixed effects, clustering, and event windows; (8) multiplicity control within each outcome family; (9) missingness, attrition, and noncompliance handling; (10) heterogeneity analyses limited to pre-specified groups; (11) placebo, pre-trend, balance, and manipulation checks; and (12) publication of de-identified code, dictionaries, and a deviation log.

The programme should not claim that a survey experiment identifies the environmental effectiveness of a tax or spending programme, nor that an observational expenditure association identifies a causal effect. Its contribution is to connect verified fiscal exposure, policy design, real administrative outcomes, and legitimate public evaluation in designs that make each inferential step explicit.

Corrected, internally consistent research-design programme

Evidence audit and revisions

The programme should retain an eight-study architecture, but the completed design needs four corrections. The systematic review supports the broad claims of sparse non-OECD coverage, limited instrument–outcome evaluation, and disagreement in distributional evidence; it does not establish the causal performance of any proposed design. (Peñasco, 2021) Its detailed passage reports mixed energy-tax distributional results and high disagreement in emissions-allowance evaluations, and attributes much of the contradiction to policy-instrument design. (Peñasco, 2021) The acceptance review supports associations between acceptance and information, revenue use, trust, fairness, and climate concern, and explicitly calls for more specific trust measures, economic perceptions, and tax morale; it does not prove that any one information intervention will increase support in every setting. (Muhammad, 2021) (Muhammad, 2021)

Shared design rules

Each study is a separate preregistration with a shared data dictionary. The programme should use one jurisdiction and one sector or tax reform at a time; it should not claim cross-country comparability before harmonisation. The common fields are: fiscal instrument; legal eligibility; dates of announcement, commitment, disbursement, award, and implementation; location; recipient or household identifier; monetary amount; outcome definitions; and pre-specified exposure to concurrent policies.

Pre-register: population, recruitment and consent; data linkage and privacy controls; treatment assignment; the primary estimand; a ranked primary-outcome family; model equations; clustering level; covariates used only for precision; missing-data and attrition handling; multiplicity adjustment; subgroup definitions; placebo and falsification tests; and rules for any deviation. A survey experiment identifies the effect of assigned information or framed policy attributes on measured attitudes or choices. It does not identify the environmental effect of the underlying fiscal policy. A quasi-experimental allocation design identifies an effect only if its documented assignment rule and diagnostics support that interpretation.

Study 1. Comparable green-expenditure measurement

Research question. How much do expenditure estimates, geographic allocation, and beneficiary profiles change when headline environmental-budget lines are recoded into a harmonised programme-level taxonomy?

Evidence and hypothesis. Environmental public-expenditure research reviewed by Pacheco and colleagues was concentrated in Europe and Latin America, largely qualitative, and called for transparency, accounting organisation, and indicators. (Pacheco, 2017) H1: official and harmonised classifications produce different instrument shares and geographic allocations. H2: classification disagreement is larger for mixed-purpose and tax-expenditure lines than for direct procurement. H3: an audited, disaggregated spending display changes factual comprehension relative to an aggregate-total display; direction of the attitude effect is not pre-specified.

Groups and data. Select one country with at least 20 subnational governments. Include all relevant budget lines across three completed fiscal years. Double-code line items as operating spending, capital spending, grant/subsidy, procurement, public R&D, tax expenditure, or other, then attach objective, beneficiary, commitment/disbursement status, and geography. The measurement comparison is official category versus harmonised code. The embedded survey randomly assigns adult residents to: aggregate total; instrument-by-objective dashboard; or dashboard plus independent audit statement.

Identification and outcomes. The primary measurement outcome is the within-line and within-jurisdiction reclassification rate. Survey primary outcomes are a prespecified factual-comprehension score and perceived transparency. Secondary outcomes are perceived effectiveness and support. Estimate survey ITT effects with randomisation-stratum and jurisdiction fixed effects; use randomisation inference. Do not interpret correlations between coded spending and environmental outcomes as causal. This follows the public-expenditure-review principle that spending level alone does not establish effectiveness. (Markandya, 2006)

Threats and preregistration. Pre-register the codebook, independent adjudication process, held-out reliability sample, dashboard text, survey exclusions, and alternative taxonomy. Threats are strategic labelling, accounting non-comparability, coder drift, and survey demand. Report inter-coder reliability, blinded recoding, and sensitivity to the alternative taxonomy.

Study 2. Instrument comparison within a common sector

Research question. Within one sector and common eligibility population, how do green procurement, grants, and public R&D support differ in recipient innovation, verified environmental performance, supplier entry, award concentration, and public cost?

Evidence and hypotheses. Comparative discussion of procurement against other environmental tools remains limited. (Cheng, 2018) A firm-level cross-sectional difference-in-differences analysis reports that winning a green procurement award was associated with a higher probability of environmentally friendlier products, driven by SMEs, but not with a statistically significant process-innovation effect. (Krieger, 2022) H1: within each instrument, receipt increases the primary sector-specific output relative to eligible non-recipients. H2: treatment effects differ across instruments after standardising outcomes by public cost. H3: stricter environmental criteria affect both environmental outcomes and bidder entry; the direction for entry is not pre-specified.

Groups and data. Limit the study to one sector and one policy regime. Link tender and bid files, grant/R&D applications, eligibility scores, award values, firm registry information, patents or certifications, project energy/emissions records, employment, and supplier ownership. The primary group comparison is recipients versus eligible non-recipients within the same tender or application cohort and instrument. Measure public cost as disbursement plus documented procurement premium, not contract value alone.

Identification, intervention, and outcomes. If a known scoring threshold governs receipt, use a fuzzy regression discontinuity and prespecify bandwidth and manipulation tests. Otherwise use cohort-by-buyer or programme event studies, conditional on documented parallel pre-trends; label results associational if neither condition holds. Estimate effects separately by instrument before any cross-instrument comparison. Randomise a factual contract-performance brief versus a generic programme description among bidders and residents; its outcomes are perceived value for money, procedural fairness, and support, not emissions. Primary administrative outcomes are verified emissions or energy outcome per public monetary unit, product innovation, bidder entry, and award concentration.

Threats and analysis plan. Pre-register treatment date, eligibility window, cost definition, overlap rule for firms receiving multiple instruments, primary outcomes, and SME subgroup. Selection, threshold manipulation, concurrent support, reporting incentives, and strategic bidding are key threats. Report covariate continuity and density tests at thresholds, event-time leads, an overlap-excluded primary analysis, and an overlap-included sensitivity analysis. Procurement theory cautions that effectiveness depends on criteria exceeding prevailing technology and linking to defined environmental objectives, so contract requirements must be recorded rather than assumed equivalent. (Lundberg, 2015)

Study 3. Local implementation capacity and spatial spillovers

Research question. Does the local implementation of a green-procurement mandate change procurement practice, private adoption, supplier entry, and legitimacy in treated municipalities and their pre-defined neighbouring markets?

Evidence and hypotheses. Reviews identify an evidence gap on localised GPP factors in developing countries and limited attention to the tension between policy uniformity and local flexibility. (Carrasco, 2024) (Xi, 2026) A California study using matching, panel data, and instrumental variables reports spillovers from municipal green-building procurement to private LEED adoption and supplier expertise. (Simcoe, 2014) H1: mandate adoption increases the share of eligible tenders with verified green criteria in treated municipalities. H2: treatment effects vary with pre-policy administrative capacity and relevant supplier density. H3: neighbouring municipalities experience a non-zero spillover; direction is not prespecified.

Groups and data. Define the analysis population as all municipalities eligible for the mandate before looking at outcomes. The direct treatment group is each adoption cohort. The comparison group is municipalities not yet adopted at the same event time, matched only on pre-policy variables. Define neighbours before analysis using both contiguous borders and a commuting/supplier-network matrix. Link legal rollout records, tender text and awards, e-procurement capability, procurement staffing, supplier locations, firm entry, private technology-adoption or emissions records, and area socioeconomic characteristics.

Identification. The primary estimate is a staggered-adoption event study with cohort-specific treated-versus-not-yet-treated comparisons. Include direct treatment and pre-specified spatial exposure in the same model; do not treat neighbours as unexposed controls. If a documented administrative score determines rollout, add a local regression discontinuity. The causal claim is conditional on no differential pre-trends and on the validity of the rollout rule; otherwise report an adjusted association.

Information intervention. In a separate clustered survey, randomise a factual local dashboard—mandate requirements, awardee list, verified performance, and complaint channel—versus a generic national description. Survey groups are local firms and residents in treated, comparison, and neighbour municipalities. Primary outcomes are perceived procedural fairness and confidence in local procurement administration; secondary outcomes are bidding intention and support for expansion.

Outcomes, threats, and preregistration. Order outcomes: green-criteria uptake; private adoption; supplier entry; procurement concentration; then survey outcomes. Pre-register adoption dates, spatial matrices, event windows, capacity measure, clustering by municipality, and the treatment-effect estimator. Threats include endogenous rollout, supplier relocation, cross-border procurement, coincident climate policies, and spatial interference. Address them with pre-trend tests, placebo dates, documented concurrent-policy controls, supplier-location tracking, alternative spatial matrices, and a leave-one-region-out sensitivity analysis.

Study 4. Long-horizon portfolios, additionality, and private investment

Research question. Do programme portfolios with different timing and instrument composition generate delayed environmental outcomes and private co-investment beyond what is observed from annual aggregate spending?

Evidence and hypotheses. The EU study reports positive associations of absolute environmental-protection expenditure with green total factor productivity, but not expenditure intensity, and explicitly notes short observation, omitted private investment, and lack of definitive causal identification. (Ibrić, 2026) A 33-country panel finds no significant independent association between spending and its proxy for green investment and does not support a strong causal complementarity claim with carbon pricing. (Abdulhasanova, 2026) H1: operational completion, rather than commitment alone, precedes the largest environmental outcome change. H2: eligible portfolio exposure changes private co-investment relative to comparable ineligible exposure. H3: portfolio composition modifies effects; this is non-directional.

Groups and data. Build a programme–place–year panel with at least five pre-exposure years and five post-commitment years where records allow. Link commitments, disbursements, completion dates, procurement and grant awards, locations, private co-financing, company investment, patents, energy/emissions outputs, and costs. Define portfolios from programme documents before linking outcomes. Groups are places or firms above versus below a documented eligibility cut-off, or allocation-formula recipients versus comparably eligible non-recipients.

Identification and intervention. When a predetermined score or formula determines allocation, use it as the assignment mechanism; otherwise estimate an explicitly associational event study. Model commitment, disbursement, and completion as separate exposure dates and report all leads and lags. Randomise a timeline display—commitment, disbursement, completion, expected output, and independent monitoring—versus an annual-spending display among affected residents and firms. The intervention tests understanding of timing and perceived additionality, not portfolio effectiveness.

Outcomes, threats, and analysis. Primary outcomes are verified project completion, environmental output, and private co-financing. Secondary outcomes are patents, employment, and support. Pre-register the event window, completion definition, portfolio taxonomy, treatment of incomplete projects, co-financing measure, and missing-data bounds. Anticipation, endogenous siting, contemporaneous regulation, unobserved private investment, and cancellation bias are principal threats. Use pre-trends, negative-control outcomes, a concurrent-policy register, place fixed effects, and sensitivity bounds for missing private-finance data.

Study 5. Objective incidence, perceived burden, fairness, and support

Research question. For a specified carbon or energy tax and recycling rule, how do modeled net incidence, perceived net incidence, fairness, trust, and environmental concern relate to support and a pre-specified follow-up behaviour?

Evidence and hypotheses. The carbon-tax review states that distributional impact depends on energy use, indicator choice, and household characteristics, and describes lump-sum transfers as more suitable than labour-tax reductions for mitigating regressive effects among lower-income households. (Köppl, 2022) The acceptance review finds higher support where people are informed about effectiveness and revenue use, trust government, perceive fairness, and are concerned about climate change. (Muhammad, 2021) H1: personalised incidence information reduces the absolute gap between perceived and modeled net burden. H2: the effect of a targeted-transfer presentation on support differs by modeled net incidence. H3: fairness and specific trust dimensions remain associated with support conditional on modeled incidence; this is associational, not a causal mediation claim.

Groups and data. Recruit a probability sample of household financial decision-makers, with pre-specified oversampling of low-income, rural, energy-intensive, and affected-worker households. With consent, link survey responses to expenditure microdata or bills, household composition, income band, heating and transport characteristics, tax schedule, and transfer eligibility. Pre-build and externally audit a microsimulation of direct and indirect tax incidence; attach an uncertainty flag where data are incomplete.

Randomised intervention. Stratify by income band, rurality, and baseline support, then assign: (A) generic policy description; (B) personalised gross tax burden; (C) burden plus universal rebate; (D) burden plus targeted transfer; or (E) targeted-transfer information plus verified environmental use of residual revenue. Hold the tax rate, base, and revenue total constant. The personalised result must display uncertainty and calculation assumptions.

Identification, outcomes, and analysis. Primary outcomes are binary support for the exact policy and absolute error between perceived and modeled net burden. Secondary outcomes are fairness, the five trust dimensions, maximum acceptable rate, intention to claim a benefit, verified benefit enrolment where consented, and a pre-specified follow-up behavioural outcome. Estimate ITT effects against A with randomisation strata; report unadjusted estimates first and covariate-adjusted precision estimates second. Do not impute a missing primary outcome. Pre-register the calculator, treatment text, modelled-price assumptions, weights, and multiplicity adjustment across the two primary outcomes.

Threats. Self-reported expenditure error, model misspecification, distrust of personalised calculations, selective linkage consent, and energy-price shocks can bias interpretation. Retain a survey-only ITT sample, report simulation uncertainty, publish the calculator code, use neutral audited wording, and repeat results under pre-specified alternative price and pass-through assumptions.

Study 6. Revenue recycling and the policy-baseline contradiction

Research question. Which budget-neutral revenue-recycling package best meets a prespecified distributional target and, separately, receives the highest support when respondents are or are not informed about the existing policy baseline?

Evidence and hypotheses. The fairness review finds concern about unequal burdens but often a preference for environmental projects over direct compensation, and recommends combining support for vulnerable groups with environmental investment. (Maestre‐Andrés, 2019) A Swiss conjoint experiment reports no average acceptability increase from redistribution; progressive schemes became more favourable under particular policy baselines and for some political groups. (Prinzing, 2026) H1: targeted transfers reduce modeled loss among low-income households more than labour-tax reductions. H2: package effects on support depend on baseline information. H3: a transfer-plus-investment package has a different support profile from either component alone; direction is not pre-specified.

Groups, data, and intervention. Use Study 5’s sample and microsimulation. Every conjoint profile fixes tax base, rate, emissions objective, and total revenue. Randomise: recycling package (universal dividend, targeted transfer, labour-tax reduction, environmental investment, or transfer-plus-investment); disclosure of the respondent’s modeled net position; audit disclosure; and policy baseline (existing policy described or not). Enforce budget neutrality and randomise profile order. Include a status-quo/no-reform option.

Identification and outcomes. Randomisation identifies average marginal component effects and pre-specified interactions: package × baseline, package × modeled incidence, and transfer × investment. Primary outcomes are choice of preferred package and support for implementation. Secondary outcomes are perceived regressivity, revenue-use credibility, procedural fairness, and comprehension. Estimate respondent-clustered effects and report each attribute-level mean. Report modeled incidence results separately from preference outcomes; popularity does not demonstrate equity.

Threats and preregistration. Pre-register the full profile generator, number of tasks, prohibited implausible combinations, primary choice task, attention checks, hierarchical-model specification, and treatment of incomplete responses. Threats are hypothetical choice, cognitive burden, status-quo bias, social desirability, and policy-context non-comparability. Use cognitive pretesting, plain-language descriptions, random task order, a no-reform option, and replication rather than pooled estimates across incomparable countries.

Study 7. Acceptability measurement and behavioural validity

Research question. Which of explicit support, maximum acceptable tax rate, stated willingness to pay, and a consequential action best predicts a prespecified later behaviour?

Evidence and hypotheses. A within-subject carbon-tax study finds that elicitation method substantially affects acceptability, with lower explicit-survey acceptance than implicit model-based measures and a gap between preferred and maximum acceptable tax levels. (Mölk, 2026) H1: the four measures differ within person. H2: their prospective association with follow-up behaviour differs; no ranking is pre-specified. H3: order effects are non-zero, motivating randomised task order.

Groups, tasks, and data. Recontact Study 5 participants at baseline, immediately after treatment, and at a pre-registered follow-up date. Randomise the order of referendum-style support, maximum acceptable rate, a willingness-to-pay/budget-allocation task, and one consequential action, such as authorising enrolment for official rebate information. Use a real but ethically approved action, not an implied one. Keep policy description constant across tasks.

Identification and analysis. Task-order randomisation identifies order effects only. The primary validation outcome is one observable, consented behaviour chosen before fieldwork—for example benefit enrolment or attendance at an official information session. Compare out-of-sample predictive performance using a pre-specified train/test split, scoring rule, calibration measure, and repeated-measures model. If the outcome cannot be observed by the stated date, report behavioural validation as unavailable rather than substituting a new proxy.

Threats and safeguards. Attrition, weak consequential stakes, learning, and changing price conditions threaten validity. Pre-register retention procedures, minimum stakes, inverse-probability weighting for follow-up, and a sensitivity analysis for missing behaviour. Information-treatment effects should be reported by education and prior knowledge because previous evidence shows heterogeneous information responses. (Ruprecht, 2023)

Study 8. Institution-specific trust and administrative credibility

Research question. Which verifiable administrative feature changes specific trust dimensions and support for an environmental tax: transparent revenue accounting, independent audit, reliable compensation delivery, or participatory and even-handed administration?

Evidence and hypotheses. The acceptance review calls for analysis of responsiveness, reliability, integrity, openness, and fairness rather than generic trust. (Muhammad, 2021) UK qualitative evidence identifies distrust in revenue use, difficulty understanding tax shifts, and demand for incentives alongside penalties as policy-design problems. (Dresner, 2004) H1: independently audited revenue information changes integrity and openness ratings relative to a generic assurance. H2: verified payment-delivery information changes reliability ratings. H3: a participatory-process explanation changes procedural-fairness ratings. Effects on overall support are secondary and not assumed universal.

Groups, data, and intervention. Recruit households from Study 5’s frame and a parallel sample of small firms where business energy exposure matters. Verify every factual administrative statement from programme records before fielding. Randomise a 2 × 2 × 2 factorial: audited public ledger versus generic statement; verified on-time compensation delivery versus no delivery record; and consultation/enforcement explanation versus no process information. Hold policy price, tax base, recycling formula, and the respondent’s modeled net position constant.

Identification, outcomes, and analysis. The factorial ITT estimates main effects and preregistered two-way interactions. The five specific trust dimensions are co-primary outcomes, with family-wise error control across them. Secondary outcomes are policy support, perceived effectiveness, maximum acceptable rate, and intention to claim compensation. Link to actual payment and complaint-resolution records only to validate factual treatment content, not to attribute administrative outcomes to the vignette.

Threats and preregistration. Pre-register source records, wording, cells, multiplicity method, and subgroups by income, rurality, tax experience, political orientation, and energy exposure. Threats include disbelief, polarized priors, contemporaneous scandals, and spillovers. Measure baseline beliefs, timestamp fieldwork, include a manipulation check, report all factorial cell means, and treat analyses excluding failed comprehension as sensitivity analyses only.

Programme-level inference

The corrected programme makes three boundaries explicit. First, the systematic review justifies addressing sparse geographic and instrument-specific evidence; it does not identify a universal preferred instrument. (Peñasco, 2021) Second, the acceptance review motivates measuring revenue use, fairness, trust, and behaviour together; it does not license a claim that transparency or recycling will always raise support. (Muhammad, 2021) (Muhammad, 2021) Third, observational evidence on public expenditure and green investment remains associational unless a specific allocation rule supplies credible exogenous variation. (Ibrić, 2026) (Abdulhasanova, 2026) These boundaries preserve the original architecture while making the eight studies falsifiable, interpretable, and internally consistent.