One Rating, Multiple Services

Solving the 100-Service Dilemma: Automated Attribution for Multi-Service Public Satisfaction Surveys

🎤 Presented as a Lightning Talk at useR! 2026 — Virtual Presentation Room 📁 Topic: Applications → Case studies and applications

---

The Problem

Indonesian regulation (Permenpan RB 14/2017) requires every public service to be scored on 9 quality components (procedures, cost, staff competence, facilities, etc.). This is manageable for a single service — but government can offer hundreds of services under one roof.

Asking each visitor to rate every service they used, on all 9 components, separately, means hundreds of repetitive answers per person — leading to severe respondent fatigue and poor data quality.

The Approach

Instead of one form per service, respondents:

1. Check off every service they received in a single visit (e.g., Service A; Service B; Service E)
2. Give one holistic rating across the 9 components, based on their overall experience

This matches how people actually experience a visit — and shifts the challenge from survey design to a data engineering problem: turning one combined rating into accurate, per-service scores.

The Solution: Disaggregate → Attribute

Google Forms exports multi-select answers as a single semicolon-separated string. This repository provides a minimal {tidyverse} pipeline that:

1. Disaggregates — tidyr::separate_longer_delim() explodes one response row into multiple rows, one per selected service, copying the same ratings to each
2. Attributes — group_by() + summarise() averages every rating that touched a given service, across all respondents who used it (alone or combined with others)
3. Scores — applies the official Permenpan RB weighting (0.111 per component, scaled to 100) to produce a compliant satisfaction index per service

clean_data <- raw_data %>%
  separate_longer_delim(services, delim = "; ") %>%   # 1. disaggregate
  group_by(services) %>%
  summarise(across(U1:U9, mean), n = n()) %>%          # 2. attribute
  mutate(
    IKM = rowSums(across(starts_with("U")) * 0.111) * 25  # 3. score
  )

No respondent ever fills in more than one rating block, and the same 4-line pipeline scales from 5 services to 500+ without any code changes.

Repository Contents

File	Description
simulate_data.R	Generates example survey data (Google Form–style export, 15 respondents, Services A–E)
attribution_pipeline.R	The disaggregate → attribute → score workflow
visualize_results.R	Bar chart of satisfaction index per service
sample_data.csv	Example raw export with semicolon-separated multi-select responses

Requirements

install.packages("tidyverse")

Usage

source("simulate_data.R")
source("attribution_pipeline.R")
source("visualize_results.R")

Methodology Notes

• Validity relies on proper response tracking: each respondent's row is duplicated across every service they selected, so every service receives contributions only from respondents who actually experienced it.
• Cross-cutting components (e.g., facility quality, staff competence) are often perceived holistically by users, which supports rating them once per visit rather than once per service.
• This approach maintains regulatory compliance with Permenpan RB 14/2017 while substantially reducing respondent burden.

Presentation

• Format: Lightning Talk (5 minutes)
• Venue: useR! 2026, Virtual Presentation Room (Aula VI (SGH Warsaw School of Economics))
• Slides & speaker notes: see /slides in this repo

Authors

• Abdul Aziz Nurussadad — Badan Informasi Geospasial
• Akbar Rizki — IPB University

Citation

If you use this pipeline or reference this talk, please cite:

Nurussadad, A. A., & Rizki, A. (2026). One Rating, Multiple Services: Solving the
100-Service Dilemma — Automated Attribution for Multi-Service Public Satisfaction
Surveys. Lightning Talk, useR! 2026, Virtual Presentation Room.
Code: https://github.com/Nr5D/multi-service-attribution

License

MIT