Factors Associated with Return-to-Fishing Inflows in South Korea

1. Demographic changes in fishing villages

Fishing villages in South Korea have become smaller and older over time. As shown in <Table 1>, both the number of residents and the number of households in fishing villages have steadily declined over the past two decades. The share of residents aged 70 and above rose sharply, from 6.62% in 2000 to 22.43% in 2020. Average household size also fell, from 3.08 persons in 2000 to 2.25 persons in 2020.

<Table 1>

Demographic changes by age group

Against this backdrop of demographic decline, a distinct migration pattern has emerged. The number of urban-to-rural migrants moving to fishing villages increased until around 2015 and then declined thereafter, as did the number of migrant households [Fig. 1].

[Fig. 1]

The number of migrants and households.

Migrant household size fell rapidly after 2017, reaching 1.33 in 2020, and the share of migrants aged 50 and above rose steadily [Fig. 2].

[Fig. 2]

The ratio of migrants by age group.

These trends indicate that urban-to-rural migration is largely composed of small households—typically single individuals or couples—with growing prevalence among middle-aged and older adults.

Yet despite this inflow, fishing villages continue to face significant quality-of-life constraints. Government reports (The Ministry of Ocean and Fisheries, 2021a) show that average quality-of-life scores are lower in fishing villages (5.2) than in agricultural villages (5.8) or urban areas (6.4). Remoteness exacerbates these challenges, limiting access to transportation, medical services, and cultural amenities. For instance, the Second Master Plan for Development of Fisheries and Fishing Villages for 2021-2025 reported that the average travel time from a fishing village to a medical facility was 89.5 minutes.

2. What do migrants consider?

To identify factors that may matter for migration decisions, we first review the 2020 Fact-Finding Investigation on the Status of Return to Farming and Fishing and Return to Rural Communities(The Ministry of Agriculture, Food and Rural Affairs, 2020). Although the evidence from this survey focuses primarily on migration to agricultural villages, it provides a useful benchmark for migration to fishing villages.

The survey distinguishes two groups of rural migrants. The first consists of individuals who relocated to work in agriculture, while the second includes those who moved without engaging in farming.

<Table 2> summarizes responses from the first group. Environmental quality and economic opportunity emerged as key motivations: 29.3% moved for a better natural environment, 19.5% to work on or manage a family farm, and 18.6% for agricultural economic opportunities. When asked why they chose a specific town, 42.2% cited family origins, 26.8% proximity to their former city, and 8.2% better infrastructure.

<Table 2>

A survey on urban-rural migration for farming

<Table 3> reports responses from the second group. Here, 28.5% emphasized a less stressful lifestyle and improved environment, while 27.2% pointed to non-agricultural job opportunities. Affordable housing was noted by 8.5%. Regarding town selection, 19.6% mentioned workplace proximity, 17.5% family ties, and 14.0% affordable housing prices.

<Table 3>

A survey on urban-rural migration for living

Overall, the findings highlight three broad categories shaping migration decisions: environmental amenities, economic opportunities, and settlement conditions. This framework underpins the empirical specification adopted in the study.

Taken together, the survey evidence suggests that migration decisions are shaped by a combination of economic motives, living conditions, and environmental amenities. This motivates our empirical framework, which groups the explanatory variables into economic, infrastructure, and environmental dimensions.

3. Government supports

In 2018, the Ministry of Ocean and Fisheries introduced the first master plan to encourage urban-to-rural migration into fishing villages. The plan was driven by structural changes such as a shrinking working-age population, rapid aging, and the retirement of the baby-boomer generation.

The policy rationale also reflected concern that existing support systems did not adequately meet the needs of prospective migrants. Many were more interested in service sector or tourism-related opportunities than in traditional fisheries, and differences in norms and lifestyles between urban and rural communities often posed barriers to settlement and integration.

To address these challenges, the government established migration information and education centers nationwide. These centers provide consulting services for potential and current migrants, along with training programs in fishing, aquaculture, and the customs and culture of fishing villages.

The master plan further introduced concrete support measures across aquaculture, housing, education, health, and welfare. Examples include reorganizing illegal aquaculture facilities to create opportunities for migrants to obtain licenses, property tax deductions and renovation subsidies for housing, partial tuition support for households with middle- or high-school students, subsidies for health insurance and childbirth grants, and welfare measures such as support for fishermen’s annuity insurance and labor replacement costs.

Taken together, these initiatives underscore the government’s view that economic opportunity, housing, education, health, and welfare are central to migrants’ settlement decisions, and highlight the importance of local service conditions in the context of fishing villages.

4. Estimation strategy

For the empirical analysis, we build on the key themes identified in the above three subsections as follows:

Specifically, economic conditions are proxied by primary-industry gross regional domestic product. Infrastructure is proxied by doctors per 1,000 residents and fishing-village population (a proxy for settlement size and related local conditions). Environmental quality is measured by particulate matter 10 (air quality) and coastal total phosphorus (water quality), where lower values indicate better conditions. Thus, X includes primary-industry GDP, doctors per 1,000, fishing-village population, PM10, and coastal TP. The rationale for including these variables is presented in the following Data section. In addition, building on Kwon and Yeo(2018), we exploit province-year variation and account for unobserved time-invariant heterogeneity across provinces. Province indicators α_i account for time-invariant differences across provinces, and year indicators δ_t summarize common time patterns relative to 2013. We interpret β as partial correlations conditional on these indicators.

Because the dependent variable is a nonnegative count and includes zeros, a standard linear fixed-effects model does not match the distributional properties of the outcome. A Tobit model may be considered when the dependent variable is limited at zero. Count-data models such as Poisson or negative binomial are also natural candidates when the dependent variable is an integer count (Cameron and Trivedi, 2010; Kim et al., 2020).

In principle, negative binomial models may be useful when overdispersion is present. However, there is a debate over the interpretation of fixed effects in panel negative binomial models and the extent to which such models fully remove unobserved heterogeneity (Allison and Waterman, 2002).

Thus, we estimate several alternative specifications-including a linear fixed-effects model, a panel Tobit model, a pooled Poisson model, and a fixed-effects Poisson model- and compare their qualitative results. We place emphasis on fixed-effects Poisson estimation with robust standard errors, which remains valid under more general conditions (Allison and Waterman, 2002; Wooldridge, 1999).

1. Data

We construct a province-year panel for 2013–2020 using data from the Korean Statistical Information Service. South Korea includes both metropolitan cities and provinces at the same administrative level. For simplicity, we refer to both types of jurisdictions as provinces. We restrict the sample to jurisdictions with a coastline and exclude inland areas without fishing villages, such as Daegu, Daejeon, Gwangju, Sejong, Seoul, and North Chungcheong. The final sample consists of eleven coastal provinces from 2013 to 2020.

The dependent variable represents the number of return-to-fishing migrants and their accompanying household members. A return-to-fishing migrant is defined in the source data as a person who had lived in an urban area for at least one year, moved to a rural area during the reference period, and was registered in an official list that recognizes fisheries participation. Accordingly, the analysis focuses on return to fishing rather than broader rural relocation.

We combine this with province-level information on gross regional domestic product of primary industries, the number of doctors per 1,000 residents, fishing-village population, PM10, and coastal total phosphorus. These variables are selected to proxy the three categories identified in the background discussion: economic conditions, infrastructure, and environmental quality.

First, we use gross regional domestic product of primary industries—including agriculture, forestry, and fisheries—as our main economic proxy. Alternative measures such as GRDP per capita or the value of fish production could also be considered. However, GRDP per capita reflects the structure of all industries in a province and may therefore be less informative about economic conditions relevant to fishing villages. We also prefer primary-industry GRDP to the value of fish production because a common feature of fishing-village livelihoods in South Korea is the joint engagement of households in both fisheries and agriculture (Cha and Shin, 2013).

Second, we use doctors per 1,000 residents and fishing-village population as infrastructure-related proxies. Doctors per 1,000 residents captures local access to medical services, which may be especially relevant if migrants are disproportionately middle-aged or older. Fishing-village population is included as a broad province-level indicator of settlement scale rather than as a direct measure of infrastructure. Because it may simultaneously capture population decline, aging, and other demographic shifts, we interpret this variable cautiously.

Third, PM10 and total phosphorus (TP) are included as objective proxy indicators of environmental quality. PM10 captures air quality relevant to everyday living conditions, while coastal TP captures water quality that may matter both for residential amenity and for fisheries-related local conditions. These variables do not directly observe migrant preferences, but they provide objective proxy indicators of air quality and coastal water quality that may affect the attractiveness of fishing villages as places to live and work. <Table 4> shows descriptive statistics.

<Table 4>

Descriptive statistics(n=88)

2. Estimation results

<Table 5> reports the estimated relationships between migration inflows and the explanatory variables. Columns (1) through (4) present results from the linear fixed-effects model, panel Tobit model, pooled Poisson model, and fixed-effects Poisson model, respectively. All models are estimated with robust standard errors, and the panel specifications include province and year fixed effects.

<Table 5>

Estimation results

Although the alternative models differ in functiona form and distributional assumptions, the air quality proxy shows a consistent directional relationship across specifications. In contrast, the results for the other variables vary by model. Differences in statistical significance and coefficient signs across columns likely reflect the distinct ways in which each model handles zeros, nonlinearities, and the count nature of the dependent variable.

We emphasize the fixed-effects Poisson model as the main specification because the dependent variable is a nonnegative count with zeros.

First, the number of doctors per 1,000 residents is positively associated with migration inflows. The coefficient is statistically significant in the fixed-effects Poisson model and suggests that migration inflows are higher in province-years with better medical service availability. This pattern is substantively consistent with the age composition of recent migrants, who are increasingly concentrated among middle-aged and older households.

Second, fishing-village population is negatively associated with inflows in the preferred specification, although the estimated magnitude is small. Because the coefficient is identified from within-province changes over time, it may reflect broader demographic contraction, aging, or compositional shifts rather than infrastructure per se. We therefore interpret this estimate cautiously and do not treat fishing-village population as a clean measure of infrastructure at the province level.

Third, both PM10 and coastal TP are negatively associated with migration inflows, suggesting that worse measured air quality and worse coastal water quality are related to fewer migrants. These results should be interpreted as relationships with objective province-level environmental conditions rather than with directly observed household perceptions. The stronger association for PM10 may indicate that air quality is more linked to perceived livability, whereas coastal TP may be a noisier proxy for the environmental and economic conditions relevant to prospective migrants.

The coefficient on primary-industry GRDP is not statistically significant in the preferred specification. This may reflect limited within-province variation over the sample period, the aggregated nature of the measure, or offsetting mechanisms that weaken the net association between broad economic activity and migration inflows.

For interpretation, we focus on incidence rate ratios (IRRs), exp(β), from the fixed-effects Poisson model. Because marginal effects in nonlinear count models depend on the conditional mean and the values of covariates, we do not emphasize a single marginal-effect estimate as the main summary measure.

In <Table 6>, the IRR for doctors per 1,000 residents is 3.19 for a one-unit increase. However, because the doctor variable is measured at the province level and a one-unit increase is large relative to the sample distribution, this scale is not intuitive. In addition, doctor availability within a province may be concentrated in large cities rather than evenly distributed across fishing villages. For this reason, it is more informative to interpret the estimate using a 0.1-unit increase. On that scale, a 0.1 increase in doctors per 1,000 residents is associated with about a 12.3% increase in expected migration inflows, which is calculated by (exp(0.1 × ln(3.19)) - 1) × 100.

<Table 6>

Incidence rate ratios

By contrast, a one-unit increase in TP and PM10 is associated with about a 0.7% and 1.6% decrease in expected inflows, respectively. The stronger association for PM10 than for TP is notable. One possible interpretation is that air quality more directly affects perceived livability for prospective migrants, whereas coastal water quality may be a noisier proxy for the environmental and economic conditions that matter to settlement decisions. Initially, one might expect water quality to matter at least as much as air quality in fishing villages because of its potential link to fishing productivity as well as residential amenity. The estimates, however, suggest that the more direct and observable amenity channel through air quality may be more salient in migration decisions.