Food Shops and Population Are Linked Across Japan’s Municipal Boundaries

Food shopping does not stop at municipal borders. Lei Qin’s study examines how resident population is associated with food shops inside Japanese municipalities and in neighboring areas, while recognizing that residents support retailers just as retail convenience may influence where people live.

A two-way relationship

A nearby food shop can make everyday life more convenient, while a larger customer base can help keep that shop in business. A preprint by Lei Qin of the University of Niigata Prefecture examines this two-way relationship across Japan, including retail opportunities beyond municipal boundaries.

Its central finding is an association: changes in resident population are positively related to changes in both local and neighboring food retail supply. The analysis does not show that opening stores causes population growth, or that closing them causes population decline.

The paper’s theoretical framework explains why both directions are plausible. Food retailers can reduce shopping costs and offer convenience and variety, making a location more attractive. Population density can support retail entry through greater demand, provided that demand gains outweigh higher operating costs. Under specified stability conditions, these mechanisms can reinforce each other.

Neighboring stores complicate the picture. They can improve shopping opportunities for residents while drawing business away from local establishments. The model therefore leaves their overall population effect uncertain.

What the study measures

Qin combines Japan’s 2010, 2015 and 2020 Population Censuses with establishment counts from the 2012, 2016 and 2021 Economic Censuses. The balanced panel contains 1,725 municipalities and 5,175 municipality-wave observations, aligned to fixed 2020 boundaries.

The timing matters: population observations precede their paired establishment observations by eight to sixteen months. These are not measurements of store openings followed by subsequent population responses.

The food retail measure covers broad food and beverage categories, including grocery, produce, meat, fish, liquor, confectionery and bakery retailers. It excludes general merchandise retailers classified in a separate industry group, including department stores and general supermarkets. Consequently, it is not a count of every place selling food.

Neighboring supply averages the logarithm of one plus establishment counts in municipalities that share a boundary or boundary point, including across prefectural borders. It measures geographic adjacency, not actual shopping journeys or travel times.

The regressions compare changes within municipalities, accounting for characteristics that remain constant over time and for changes common to each observation wave. Additional specifications allow each prefecture to have its own wave-specific changes. Statistical uncertainty allows for correlated errors across time and nearby municipalities.

Positive associations on both sides of the border

In the main specification, the local food retail coefficient is 0.0893 and the neighboring coefficient is 0.1364. Both have 95% confidence intervals excluding zero. Allowing prefecture-specific wave effects produces similar estimates: 0.0904 and 0.1291.

For an accessible illustration, the paper reports that moving from the median count of 69 local establishments to 76 corresponds to approximately 0.85% higher fitted population in the main linear model. This is a regression comparison, not a forecast of what seven additional shops would achieve.

The positive associations also appear under alternative definitions of neighboring municipalities. However, accounting for population trends related to initial density reduces both estimates, especially the neighboring association. Restricting the original pairing to its first two waves also weakens the local estimate, whose confidence interval then narrowly includes zero.

Retail differs from prepared-food services

The study separately measures meal-oriented food services, including restaurants, coffee shops, takeout and delivery establishments. In a joint regression, the local food retail coefficient is 0.0928, while the local food-service coefficient is −0.0052 and imprecisely estimated.

To compare sectors with different amounts of variation, Qin standardizes their local associations. The difference favors food retail, with a 95% confidence interval excluding zero. That comparison persists with prefecture-specific wave effects, without the final observation wave, and after excluding 27 tsunami-inundated municipalities in Iwate and Miyagi.

It does not establish that restaurants are unimportant. Their association varies across specifications, and the local food-service estimate becomes positive with an interval excluding zero in one disaster-exclusion specification.

No reliable threshold for too many shops

A nonlinear model suggests that the local association weakens at higher retail counts. But its negative fitted slope at high counts moves to approximately zero after prefecture-specific wave effects are added. Excluding the final wave weakens the curvature further, and an alternative flexible model produces different patterns.

The defensible conclusion is that the relationship varies with retail scale in specification-dependent ways—not that municipalities reach a proven threshold beyond which shops reduce population.

Overall, the preprint highlights interconnected local markets. Establishment counts do not fully measure food access, and the analysis cannot separate retail’s influence on residents from residents’ influence on retail. Its evidence supports studying both sides of municipal borders, not a causal prescription for attracting residents.