IE/EE Calculator - PMT Model Performance Benchmarking Portal

Dewan Ekonomi Nasional (DEN) | Indonesia

Welcome to the IE/EE Calculator, a benchmarking portal developed by the Dewan Ekonomi Nasional (DEN) to support the continuous improvement of Proxy Means Testing (PMT) models in Indonesia. This platform brings together PMT model developers from research institutions, universities, and government agencies to submit their model predictions and receive standardized, comparable performance evaluations against a common national test dataset.

Please read this carefully before continuing.

This portal enables structured submission of model predictions and automatically computes key performance indicators, including Inclusion Error (IE), Exclusion Error (EE), and the DPI-Friendly Score, against a common test dataset.

Test Dataset and Welfare Measure

All submitted predictions are evaluated against SUSENAS March 2025, the latest round released by BPS, which serves as the mandatory test dataset for all submissions. The welfare measure used is real household expenditure per capita, constructed by spatially deflating nominal per capita expenditure using the March 2025 poverty lines at the urban/rural-province level.

Inclusion Error and Exclusion Error

The portal computes both unweighted and weighted versions of IE and EE. Weighted estimates use BPS sampling weights, reflecting how many population units each observation represents. Inclusion Error measures non-targeted households incorrectly included among predicted eligible households. Exclusion Error measures targeted households incorrectly excluded from predicted eligibility.

Let:

• T_e = number of targeted households incorrectly excluded
• T_i = number of targeted households correctly included
• N_i = number of non-targeted households incorrectly included

The error rates are defined as:

In a symmetric budget-constrained assignment, T_e = N_i, so IE and EE will be equal. In practice they may differ depending on program targeting rules and budget allocation.

Model Performance Indicators

Each submission is automatically scored on four dimensions:

• Unweighted IE/EE - misclassification rates based on unweighted decile ranking
• Weighted IE/EE - population-representative misclassification rates using BPS sampling weights
• Number of variables - a measure of model parsimony
• DPI-Friendly Score - a composite indicator combining IE/EE and parsimony with equal weight: (1 - (0.5 * IE/EE + 0.5 * variables / 50)) * 100

Parsimony matters because a leaner model enables households to self-report their characteristics more easily through an on-demand application (ODA), as envisioned under Indonesia's Digital Public Infrastructure (DPI) framework.

Before You Continue - Prepare the Following Information

Before uploading your prediction file, please have the following ready. You will be asked to fill in all fields on the submission form:

• Model name (tip: include a version tag, e.g. PMT-ML-V1.01)
• Submitter name
• Institution
• Training dataset (select one or more: SUSENAS March 2020, 2021, 2022, 2023, 2024)
• Number of variables used in the PMT model
• Model regional specificity (National / Province / District)
• Short model description (plain text or PDF)

Your prediction file must contain exactly two columns: (1) the BPS household unique identifier as released in the original SUSENAS dataset, and (2) predicted real household expenditure per capita. No other variables should be included. The portal will match your predictions to the test dataset using the household ID and compute all performance indicators automatically.

When you are ready, click Submit Prediction to proceed.