Bureaucratic Deliberation and Performance: Evidence from a Field Experiment in Benin

Treated municipalities received a structured deliberation intervention: three sessions held at monthly intervals (September, October, and November 2016), convened by the municipal administration and led by the Secretary General. Each session lasted approximately 90 minutes and followed the same format across municipalities — a presentation of performance-relevant information followed by open collective discussion. The three sessions covered, respectively: (1) the municipality's most recent CoNaFiL audit report and the auditors' findings and recommendations; (2) results of a survey of the municipality's bureaucrats; and (3) results of a survey of citizens. All information was presented as aggregate, organization-level summaries rather than assessments of individual bureaucrats. All bureaucrats in the municipality were invited and attendance was voluntary; bureaucrats retained control of the agenda; mayors were not permitted to attend, though bureaucrats could invite additional participants (e.g., councilors, civil-society representatives, central-government officials, or media) where relevant. The aim was to make objective, performance-relevant information about the bureaucracy as a whole public within the organization and open to collective discussion. Control municipalities held no such sessions.

2016-09-01

2016-11-30

Municipal performance, measured as the annual CoNaFiL audit-based performance score (the municipality's total score as a percentage of the maximum attainable score based on rules), built from three components---record keeping, transparency in public procurement, and overall administrative management (municipality-year panel, 2014--2019).

The score aggregates three components: record keeping (maintenance of accounting records, functioning archives, quarterly and annual statements of transferred resources, updated physical registers); transparency in public procurement (functioning of procurement oversight bodies, compliance with procurement procedures, completeness of published reports, traceability of FADeC transfers); and overall administrative management (the Secretary General's coordination of local services — regular documented meetings with department heads, follow-up on assigned tasks, internal review of financial vouchers and administrative files — and management of civil-status documents). Auditors score individual criteria within each component; we compute the performance score as total points obtained ÷ maximum attainable that year × 100 (the full list of criteria is in the paper's appendix). We analyze it as a municipality-year panel (2014–2019), with municipality and year fixed effects, standard errors clustered at the municipality level and complemented by randomization-inference p-values.

From the audit reports (municipality-year, 2014–2019):
(a) Financial irregularities — any financial irregularity; value of irregularities as a % of FADeC received; any ineligible project; and an aggregate irregularity index.
(b) Procurement — the number and share of contracts awarded through open vs. restricted procedures (restricted = restricted tender + restricted consultation); and a procurement openness index.
(c) Project implementation — the shares of works projects halted, abandoned, and ongoing (overall and split by carried-over vs. new contract); and an implementation index.
(d) Investment composition — sectoral shares of FADeC investment; the share allocated to service-delivery sectors vs. the municipality's own administration; and a social allocation index.
(e) Funding and budget execution — total FADeC received; the commitment (engagement) rate; the payment (mandatement) rate.

From the endline bureaucrat survey (December 2016):
(f) Beliefs about the rule of law — perceptions of whether rules are followed (for equipment procurement, public procurement, and recruitment) and of corruption (among councilors, city-hall vendors, and colleagues), combined into a single index (corruption reverse-coded).
(g) Internal meeting practices — the number of meetings held/attended within the administration, and the topics discussed (in particular, whether meetings covered the audit report and corruption).
(h) Interpersonal trust — trust in councilors, office colleagues, colleagues within the same department, and city-hall colleagues, combined into a single index.

Downstream (citizens):
(i) Citizen-experience index, constructed from Afrobarometer rounds 6 (2014) and 7 (2017).

Audit-report outcomes. Financial irregularities: "any financial irregularity" is a constructed indicator equal to one if auditors flag any financial-management irregularity for the municipality-year; "financial irregularity (% of FADeC)" is a constructed ratio, the value of the flagged irregularities divided by total FADeC received; "any ineligible project" is a constructed indicator equal to one if the municipality spent FADeC funds on a project ineligible for FADeC financing. Procurement: the report records the procedure used for each contract awarded — open tender, restricted tender, restricted consultation, or other procedures. "Open" is the open-tender category as reported; "restricted" is constructed by us by combining restricted tender and restricted consultation. For each we compute the number of contracts and their share of all contracts awarded that year — the denominator includes contracts awarded through other procedures too, so the open and restricted shares need not sum to 100. Project implementation: the report lists each of the municipality's works (infrastructure) projects and its execution status; we construct "% halted," "% abandoned," and "% ongoing" as the number of projects in each status divided by the municipality's total number of works projects that year, computed overall and separately for carried-over contracts (marchés non soldés) and new contracts (marchés conclus). Investment composition: the report records the FADeC investment amount allocated to each expenditure sector; we compute each sector's share of the municipality's total FADeC investment. "Social (service-delivery) sectors" is constructed as the combined share of the service-delivery sectors — education, health, water and energy, sanitation, urban roads and rural paths, markets, and agriculture — i.e., all sectors other than the municipality's own administration and a residual community-managed (PSDCC) category. Funding/execution: "total FADeC received" is the transfer amount from the report; the commitment (engagement) and payment (mandatement) rates are constructed as the shares of available FADeC funds committed and paid.

Index method. Every index — the four domain indices (irregularity, procurement openness, implementation, social allocation), the overall index that aggregates them, and the survey indices — is a standardized aggregate constructed following Anderson (2008): each component is normalized to the control group and combined into a single index, signed so that higher values denote better performance (measures running the "wrong" way, e.g. irregularities and work stoppages, are reverse-coded before aggregation).

Survey outcomes. The belief-about-rule-of-law and interpersonal-trust indices are Anderson indices of the survey items listed above; meeting practices are measured from survey reports of the number of meetings and their content.

Citizen-experience index. Constructed from Afrobarometer rounds 6 (2014) and 7 (2017); aggregates items on perceived living conditions, perceptions of corruption, direct experiences accessing services (official documents, water, sanitation), and whether the respondent paid a bribe; worse-outcome items reverse-coded, standardized to the control group. Full construction details are in the paper's appendix.

The experiment was conducted across 20 municipalities in Benin (of the country's 77), spanning 8 of 12 departments, with 10 municipalities assigned to treatment and 10 to control; randomization was at the municipality level. In treated municipalities, the municipal administration convened three structured deliberation sessions (September, October, and November 2016), each approximately 90 minutes, presenting respectively: (1) the municipality's most recent CoNaFiL audit report and auditor recommendations; (2) results of a representative survey of bureaucrats; and (3) results of a representative survey of citizens. Performance information was presented as aggregate, organization-level summaries rather than individualized assessments. All bureaucrats in the municipality were invited; attendance was voluntary; bureaucrats retained control of the agenda; mayors were not permitted to attend, though bureaucrats could invite additional participants (councilors, civil-society representatives, central-government officials, media). Control municipalities held no such sessions. Outcomes are measured from a municipality-year administrative audit panel (2014–2019; treatment effects estimated with municipality and year fixed effects) and an endline bureaucrat survey (December 2016).

Randomization in office by a computer

Randomization was at the municipality level: 20 municipalities, drawn from the country's 77 municipalities, were randomly assigned to treatment (10) or control (10).

Yes

20 municipalities (10 treatment, 10 control)

Administrative audit panel: 120 municipality-years (20 municipalities × 6 years, 2014–2019). Endline bureaucrat survey: 600 bureaucrats (30 per municipality).

Clusters: 10 treatment municipalities, 10 control. Bureaucrat survey: 299 treatment, 296 control.