Estimating multidimensional poverty under incomplete data in North Macedonia
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "Estimating multidimensional poverty under incomplete data in North Macedonia".
Warren: Here's the gist of the paper. Multidimensional poverty indices are available for more than 100 countries, but rarely below the national or regional level We show how to measure local multidimensional poverty under such incomplete data, using North Macedonia as a case study Combining settlement-level access to services with municipal census data, we pair an independence-based point estimate with sharp bounds over all joint distributions consistent with the data We find that, nationally, 5.01 percent of households are multidimensionally poor, with bounds of 2.01 to 17.44 percent The authors address the challenge of combining settlement-level access to services with municipal census data to measure multidimensional poverty in North Macedonia under incomplete data.
Ruth: They construct a thirteen-indicator household MPI from two sources The first source contains six indicators of geographic access to basic services, which are computed jointly for every settlement in the country The second source contains seven census indicators of education, employment, and housing, reported as municipal marginal rates in the tabulations used.
Warren: They compute a point estimate by assuming independence wherever the joint distribution is unobserved Uncertainty Quantification via Bounds The method employs linear programming to find the lowest and highest values of the poverty headcount and the MPI that are consistent with all observed data.
Ruth: This approach yields sharp bounds, which are not confidence intervals, but rather identification intervals The bounds are informative in their own right, showing what the available data can settle and whether some places remain poorer than others across the whole range.
Paper summary: Ruth: To wrap up this discussion on "Estimating multidimensional poverty under incomplete data in North Macedonia", the paper really shows how you can create meaningful local estimates even when you don't have the perfect household-level data linking everything together. The authors used North Macedonia to show that combining settlement-level access with municipal census information allows for this estimation.
Warren: The main point of the study is that by pairing an independence-based point estimate with sharp bounds over all consistent joint distributions, they are giving us a way to quantify the uncertainty in their local poverty findings. It’s about showing what the available data can settle and what remains ambiguous.
Iris: I think it matters because it gives policy makers a tool to guide where they might want to act, like identifying places that are far from kindergartens or schools. But as we discussed, the paper is limited by not being able to assess how deprivations overlap inside households themselves.
Ruth: Right. So the implication is that this method can help pinpoint geographic hotspots for poverty based on what's available at a municipal level, even with incomplete data structures. It helps us narrow down the search area before we try to get into household surveys.
Warren: And it reinforces that the location of poverty is relatively certain, even if the exact numerical measure of poverty has some wide bounds around it. That certainty about location versus uncertainty about magnitude is a key distinction they make here.
Iris: So in simple terms, this work demonstrates a way to use imperfect data sources to create useful local maps of multidimensional poverty where perfect data is unavailable. It's an estimation tool built on making assumptions about independence when the true relationships are unknown.
Conclusion: Ruth: So we've been digging into how they measured multidimensional poverty in North Macedonia even when they didn't have all the perfect household data to make a single number for everyone.
Warren: Yeah, that’s the big idea here, trying to get something useful out of incomplete information.
Ruth: The authors are tackling this problem head-on by combining settlement-level access data with municipal census information. It’s not about getting one perfect poverty rate for every single family in the country.
Iris: What they did was construct a thirteen indicator household multidimensional poverty index from these two sources and then used linear programming to find the lowest and highest possible values consistent with all that observed data.
Warren: That method gives them sharp bounds, not just some vague confidence interval, which is interesting for seeing what the data actually settles on versus what stays ambiguous.
Ruth: Exactly. So they found that nationally, about five point zero one percent of households are multidimensionally poor under this framework, with those bounds running from two point zero one percent up to seventeen point four four percent.
Iris: And the authors highlight that poverty is really concentrated in small rural municipalities, which are poorer than almost all places in the capital city regardless of how you structure the dependence.
Warren: It’s not just about how poor a place is overall; it’s about where you actually have to look when making policy decisions.
Ruth: And they point out that the most common deprivations there are distance to a kindergarten, low adult education levels, and being far from a secondary school.
Iris: The socio-economic sub-index showed stronger links to deprivation than the geographic sub-index did when compared against other measures of hardship.
Warren: So the location of poverty seems pretty certain, even if the exact level has that wide range around it. It’s about pinpointing where to focus effort rather than just getting a single national figure.
Ruth: It really shows how you can use imperfect data structures to map out geographic hotspots for poverty when perfect household-level data isn't available.
Iris: But the authors are clear, this index can't replace a full household assessment because it doesn't capture how different deprivations actually overlap inside those same homes.
Warren: So they can tell you where to look first, but they leave the detailed understanding of what’s happening inside the house for later. That makes sense for a first pass at identifying targets.
Ruth: Precisely. This work sets up a framework that helps guide policy by showing where the most significant gaps are located geographically based on available census data.
Iris: Next time, we'll look at how they test if this whole approach holds up when you change the way you weight those different indicators or change the distance thresholds they use in their model.
Maja Gerovska Mitev, Viktor Stojkoski
Ss. Cyril and Methodius University in Skopje · Ss. Cyril and Methodius University in Skopje, Faculty of Economics-Skopje
econ.GN, q-fin.EC
Submitted: 2026-11-10
Updated: 2026-11-10
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 84/100
The gist: Multidimensional poverty indices are available for more than 100 countries, but rarely below the national or regional level We show how to measure local multidimensional poverty under such incomplete…
Key concepts
- Multidimensional Poverty Index (MPI)
- A composite index created by combining six indicators of geographic access to basic services with seven census indicators about education, employment, and housing. This index helps measure poverty beyond just income by looking at multiple deprivations simultaneously.
- Identification Intervals
- Sharp bounds derived from linear programming used instead of confidence intervals. These bounds show the range within which the poverty headcount and MPI can realistically fall given the available data, indicating what is certain about the measurement.
- Poverty Concentration
- The finding that poverty is heavily clustered in small rural municipalities, especially those far from kindergartens and schools. This suggests that geographic location plays a significant role in determining where poverty exists within the country.
Terminology
Summary
Multidimensional poverty indices are available for more than 100 countries, but rarely below the national or regional level We show how to measure local multidimensional poverty under such incomplete data, using North Macedonia as a case study Combining settlement-level access to services with municipal census data, we pair an independence-based point estimate with sharp bounds over all joint distributions consistent with the data We find that, nationally, 5. p. 001 percent of households are multidimensionally poor, with bounds of 2. p. 001 to 17.44 percent
Methodology for Data Fusion
The authors address the challenge of combining settlement-level access to services with municipal census data to measure multidimensional poverty in North Macedonia under incomplete data. They construct a thirteen-indicator household MPI from two sources The first source contains six indicators of geographic access to basic services, which are computed jointly for every settlement in the country The second source contains seven census indicators of education, employment, and housing, reported as municipal marginal rates in the tabulations used. p. 1 They compute a point estimate by assuming independence wherever the joint distribution is unobserved
Uncertainty Quantification via Bounds
The method employs linear programming to find the lowest and highest values of the poverty headcount and the MPI that are consistent with all observed data. This approach yields sharp bounds, which are not confidence intervals, but rather identification intervals The bounds are informative in their own right, showing what the available data can settle and whether some places remain poorer than others across the whole range. p. 2
Key Findings on Poverty Concentration
The study finds that poverty is concentrated in small rural municipalities, which are poorer than nearly all municipalities of the capital city under any dependence structure The most common deprivations are distance to a kindergarten, low adult educational attainment, and distance to a secondary school The socio-economic sub-index shows stronger associations than the geographic sub-index when compared against external measures of deprivation
Robustness and External Validation
The results are robust to variations in measurement choices, such as domain weights, distance thresholds, and indicator sets. The municipal ranking hardly changes when these choices are varied, suggesting that the location of poverty is much more certain than the level of poverty itself Validation against official statistics shows that the socio-economic sub-index behaves as expected across all four demographic indicators, while the geographic sub-index only weakly correlates with them
Policy Implications and Limitations
The municipal estimates can guide where to act, identifying places far from kindergartens, schools, and health facilities. However, the index cannot replace household-level assessment because it lacks information on how deprivations overlap within households Future work should constrain the joint distribution with empirical information on co-deprivation to narrow the bounds and allow local estimates to inform which combination of deprivations to target The study also suggests that statistical offices could publish census cross-tabulations reporting how pairs of deprivations overlap. p. 0
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved AI systems can produce local multidimensional poverty estimates by combining settlement-level access to services with municipal census data, pairing an independence-based point estimate with sharp bounds over all joint distributions consistent with the data.
-
Improved AI systems can identify which deprivations matter for policy by examining which indicators are most common and how the index relates to external evidence, such as finding that
distance to a kindergarten, low adult educational attainment, and distance to a secondary school are the most common deprivations.
-
Improved AI systems can guide localized action by identifying municipalities where
more than 30 percent of households are poor even under the most favorable dependence structure,
allowing policymakers to focus resources effectively. -
Improved AI systems can perform robust policy analysis by determining which municipal comparisons
hold under every joint distribution consistent with the data,
providing stronger statements than point estimates, such as finding that517 of the 6,320 ordered pairs are ranked in the same way under any dependence structure.
-
Improved AI systems can compare different poverty measures to understand their strengths by showing that
the socio-economic sub-index behaves as expected across all four indicators,
helping to clarify which dimensions of deprivation are captured by which index.
Related papers
- Huge If True: Learning from the Size of the Surprise
- The Rebundling of Freelance Work in the AI Age
- Dynamic investment and coordination in teams: Theory and experimental evidence
- Does Frontier AI Strengthen Cyber Defense? Evidence from Project Glasswing
- Generative AI at Work
- What is Work For? Labor, Politics, and Meaning in the Age of AI