BRIDGE Trial: Behavioral Research for Informed Government Decision-Making Using Evidence — A 2×2 Factorial Randomized Controlled Trial Testing Framing and Messenger Effects on Evidence Use Among Senior Ugandan Government Officials

How to cite

Babigumira, Joseph; Waiswa, Peter; Matovu, Fred; Asiimwe, Stephen; Namirembe, Susan Kavuma; ACODE; Makaaru, Jacklyn Arinaitwe; Babiiha, Patricia Munabi. 2026. "BRIDGE Trial: Behavioral Research for Informed Government Decision-Making Using Evidence — A 2×2 Factorial Randomized Controlled Trial Testing Framing and Messenger Effects on Evidence Use Among Senior Ugandan Government Officials." AEA RCT Registry. AEARCTR-0018388. https://doi.org/10.1257/rct.18388-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Joseph Babigumira* , Peter Waiswa , Fred Matovu , Stephen Asiimwe , Susan Kavuma Namirembe , ACODE , Jacklyn Arinaitwe Makaaru , Patricia Munabi Babiiha

* Primary investigator

Registry ID
AEARCTR-0018388
DOI
https://doi.org/10.1257/rct.18388-1.0
First registered
Last updated
Status
On going
Trial dates
2026-01-12 – 2027-12-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/18388

Abstract

Background. Evidence-informed policymaking (EIPM) is widely advocated but poorly understood experimentally, particularly in low- and middle-income country (LMIC) economic policymaking contexts. A systematic review identified only 18 counterfactual impact evaluations of EIPM interventions globally, with just 6 rated low risk of bias. No prior factorial randomized controlled trial (RCT) has simultaneously tested multiple behavioral mechanisms among senior government officials in LMIC economic ministries. The relative contributions of how evidence is framed and who delivers it remain experimentally untested. Objectives. The BRIDGE Trial estimates the main effects of (i) evidence framing — technical versus narrative with citizen voice — and (ii) messenger identity — peer civil servant versus external academic expert — on evidence use among senior Ugandan government officials, and tests for an exploratory framing-by-messenger interaction. Methods. BRIDGE is a pragmatic, 2×2 factorial, individually randomized, controlled superiority trial. Two hundred and forty senior civil servants (grades U1, U1E, U2, U3, and U4; at least 35 percent women) from the Ministry of Finance, Planning and Economic Development (MoFPED), the Ministry of Energy and Mineral Development (MEMD), the Ministry of Trade, Industry and Cooperatives (MTIC), and the National Planning Authority (NPA) will be randomized in equal numbers (n=60 per arm) across four arms: technical/peer, technical/expert, narrative/peer, narrative/expert. Allocation uses constrained randomization stratified by institution, seniority (U1, U1E, and U2 versus U3 and U4), and gender (16 strata), with a public randomization ceremony in Month 3. Participants receive up to 18 evidence briefs on fiscal, energy, and trade policy topics over an 18-month delivery window (Months 4–21), aligned to Budget Framework Paper cycles. The primary outcome is the Evidence Use Index (EUI), a 0–30 composite of citations in policy documents, follow-up evidence requests, and budget allocation alignment, scored at endline by two blinded coders. Secondary outcomes capture conceptual, attitudinal, procedural, and process domains. Analysis is intention-to-treat (ITT) using a factorial linear model with HC2 robust standard errors; the trial is powered at 80 percent to detect main-effect minimum detectable effects (MDEs) of 0.30 standard deviations (SDs), assuming an intracluster correlation coefficient (ICC) of 0.10 and attrition under 20 percent. A pre-specified adaptive design review at Month 9 may modify arm pooling or brief frequency if the interaction effect is negligible. Significance. BRIDGE will produce the first factorial experimental evidence on optimal modes of evidence delivery to LMIC policymakers and inform the design of scalable EIPM interventions in Uganda and comparable settings. Funder. UK Foreign, Commonwealth and Development Office (FCDO) via the International Initiative for Impact Evaluation (3ie), grant RCC037ODAGHE, total £399,751.

Intervention

Up to 18 evidence briefs of 4–6 pages each, delivered over 18 months (Months 4–21) to 240 senior Ugandan civil servants. Briefs cover fiscal consolidation, oil and gas revenue management, energy transition, AfCFTA tariffs and trade facilitation, domestic resource mobilization, gender budgeting, debt sustainability, climate finance, industrial policy, and NDP IV monitoring, aligned to Budget Framework Paper cycles. Each brief contains a one-page executive summary with two to three recommendations, an evidence synthesis (30–40 percent visual), implementation pathways, and a gender implications section. Sources are restricted to systematic reviews, Uganda-specific modeling, and regional comparatives from 2020 onward, with GRADE confidence ratings applied.
Briefs are quality-assured by an Independent Quality Assurance Panel (IQAP) comprising senior academics, ex-government officials, and a gender expert, through a three-stage review (technical accuracy, policy feasibility, user testing). Delivery uses WhatsApp Business API, tracked email, a secure portal, and SMS reminders, complemented by printed briefs at quarterly policy dialogues. Each brief carries a unique digital watermark enabling contamination monitoring; sensitivity analyses are triggered if cross-arm sharing exceeds 10 percent.
All participants receive evidence briefs; the experiment tests which combination of presentation and delivery features maximizes evidence uptake, not whether evidence provision matters. An election pause protocol is retained as a contingency should any political sensitivity arise; Uganda's January 2026 general election concluded during the mobilization phase before active delivery began.

Experimental design

Pragmatic 2×2 factorial individually randomized controlled trial with superiority design. Factor A (Framing) compares technical presentation — statistics-forward with effect sizes, confidence intervals, regression tables, and university-level readability (Grade 14+) — against narrative presentation with citizen voice — story-driven case studies, infographics, plain language, and Grade 10–12 readability. Factor B (Messenger) compares peer civil servant messengers — senior U1–U2 officials from comparable ministries, trained over two days — against external expert messengers — academic researchers from Makerere University, EPRC, ACODE, and other qualified Ugandan research institutions, given a one-day orientation.
Four arms with 1:1:1:1 allocation: Arm 1 technical/peer (n=60), Arm 2 technical/expert (n=60), Arm 3 narrative/peer (n=60), Arm 4 narrative/expert (n=60); total N=240, with at least 35 percent women. Main effect of framing pools Arms 3+4 versus Arms 1+2 (n=120 vs 120); main effect of messenger pools Arms 2+4 versus Arms 1+3 (n=120 vs 120). The framing-by-messenger interaction is exploratory.
Allocation by constrained randomization (Moulton 2004 [see note above]) stratified by institution (4 levels: MoFPED, MEMD, MTIC, NPA), seniority (2 levels: U1, U1E, and U2 grouped together versus U3 and U4), and gender (2 levels: male versus female), yielding 16 strata. Outcome assessors and analysts are blinded to treatment assignment; participants and the intervention team cannot be blinded due to the nature of the intervention. Primary analysis is intention-to-treat using a factorial (at-the-margins) linear model with HC2 robust standard errors; multiarm pairwise comparisons with Tukey-adjusted confidence intervals are reported as a sensitivity analysis.

Design details

Randomization method
Constrained randomization following Moulton (2004): all possible allocation sequences are generated; those failing pre-specified covariate balance criteria are discarded; and one acceptable sequence is randomly selected. Balance criteria require differences across arms of no more than 10 percent in officials per institution, mean years in position, and baseline evidence use score, and no more than 5 percent in gender composition. The randomization list is generated by an independent statistician, stored in a password-protected file, and revealed only after baseline completion. A public allocation ceremony in Month 3 announces assignments before independent witnesses from NPA and participating ministries, following Khan, Khwaja and Olken (2016).
Randomization unit
Individual (senior civil servant, grades U1, U1E, U2, U3, or U4). For stratification purposes, U1E is grouped with U1 and U2 in the upper seniority stratum.
Clusters
000
Observations
240 senior civil servants at baseline. To preserve statistical power (≥80% for main effects) and maintain arm balance, a pre-identified alternate pool of at least 15 percent additional eligible officials per institution (~36 alternates total) will be enrolled as replacements for withdrawals. Replacements are assigned to the same arm and randomization stratum as the withdrawn participant and are flagged separately in the study database. Sensitivity analyses will compare results with and without replacement participants; Lee bounds will assess potential attrition bias
Arms
60 per arm across four arms (technical/peer, technical/expert, narrative/peer, narrative/expert); total N=240. For factorial main effects, each comparison pools n=120 versus n=120. An alternate pool of approximately 36 pre-identified officials is available to replace withdrawals while maintaining arm balance.
Primary outcomes
Primary outcome: Evidence Use Index (EUI), a 0–30 composite score with three equally weighted components, each scored 0–10, measured at endline. Component 1 — Citations (0–10): Documented references to brief content in Cabinet papers, Budget Framework Papers, and ministerial briefs. Scoring: 0 = none; 1–3 = indirect reference without attribution; 4–6 = direct citation without recommendation adoption; 7–10 = citation with partial to full recommendation adoption. Component 2 — Requests (0–10): Follow-up requests for evidence or engagement with the research team. Scoring: 0 = none; 1–3 = passive engagement (downloads only); 4–6 = active requests (questions, meetings); 7–10 = sustained engagement including champion behavior (multiple requests, referrals to ministry leadership). Component 3 — Budget Allocations (0–10): Changes in budget proposals consistent with brief recommendations. Scoring: 0 = none; 1–3 = under 5 percent of budget lines; 4–6 = 5–15 percent of lines; 7–10 = over 15 percent with explicit evidence justification. Two blinded coders score all documents independently; at least 20 percent are double-coded; target inter-rater reliability Cohen's kappa ≥0.80. Discrepancies greater than 2 points trigger third-coder adjudication.
Secondary outcomes
Conceptual: problem reframing (survey items) and evidence sophistication (text analysis of policy documents). Attitudinal: trust in evidence sources and self-reported willingness to request evidence. Procedural: standard operating procedure (SOP) integration, request formalization, and routinization of evidence use. Process: Acceptability of Intervention Measure (AIM), Feasibility of Intervention Measure (FIM), and System Usability Scale (SUS). Conceptual and attitudinal outcomes are measured at baseline, midline (Month 8), and endline (Month 23); process outcomes (AIM, FIM, SUS) are measured at midline and endline only. Engagement analytics are logged continuously throughout the delivery period.

Cited by

No citation data available for this registration yet.

Keywords