E2SFCA · 2024 cross-section · village level · Taiwan

Taiwan's highest-risk villages
are not where you'd expect

Not the rural east, but Taoyuan (16 villages), Miaoli (13), Keelung (11) and Hsinchu County (11) lead the country.
County-level physician-density rankings systematically hide this.

Scroll down to see why ↓

The problem

What do current metrics miss?

Taiwan usually compares healthcare resources with county-level rankings such as physicians per 10,000 residents. These metrics have two structural blind spots that hide the places that are actually underserved.

Blind spot 1: travel time is ignored
Within one county, reaching the nearest hospital from a remote township can take more than an hour on mountain roads. A county average erases that distance. Hualien ranks 3rd nationally in physician density, yet villages in its Zhuoxi and Fuli townships sit around the lowest 11% of all villages in accessibility.

Blind spot 2: averages hide within-county inequality
Urbanized counties such as Taoyuan and Miaoli look fine overall, but older residents stay in fringe townships while care concentrates in city centers. These places are administratively "urban", so they fall outside the official remote-area designation, even though access there is poor.

Comparing metrics

What each metric can and cannot capture

Metric	Unit	Travel time	Within-county gaps
Physicians per 10,000	County	✗	✗
Facility density	Township	✗	Partly
E2SFCA (this study)	Village	✓	✓

Method

E2SFCA: village level, real drive times

The Enhanced Two-Step Floating Catchment Area (E2SFCA) method estimates how much physician supply each place can realistically reach, accounting for competing demand. Instead of straight-line distance it uses road-network drive times from a self-hosted OSRM routing engine (18.1 million village–facility pairs) for all 7,616 villages. Scores are shown relative to the national average (1.0x).

Villages in remote townships have only

—

of the accessibility of non-remote villages (median)

Remote = 86 of 368 townships under the Ministry of the Interior definition (population density below one-fifth of the national average, or offshore island counties). Zero-access villages excluded from this ratio.

▶ Distance-decay design

Drive time	Weight
0–10 min	1.00
10–20 min	0.68
20–30 min	0.22
30–60 min	0.10

The first three zones follow Luo & Qi (2009). The fourth zone (30–60 min) extends the halving logic of Taiwanese applications, which were metro-scale and stopped at 15 km. 41.7% of all drive-time pairs fall in the 30–60 min band, mostly in mountain villages, and the 60-min cap still covers 98.8% of pairs. Compared with a hard 30-min cutoff, zero-access villages drop from 78 to 14.

Accessibility by urban–rural class

Relative to the national average; the box marks the median and quartiles, width shows where villages concentrate. Zero-access villages (14) are excluded.

Results

Where is access actually lacking?

96 high-risk villages in two groups: 82 villages with low access (below the national 25th percentile) and a large elderly population (at least 797 residents aged 65+, the national 75th percentile), plus 14 villages with zero access within 60 minutes. The county breakdown contradicts the common assumption that the problem is concentrated in the east.

High-risk villages by county

Taoyuan leads, followed by Miaoli, Hsinchu County and Keelung, none of which are the eastern counties people usually picture

▶ Threshold sensitivity (why the 25th / 75th percentiles?)

Using national percentiles keeps the definition objective and easy to explain. The 25th-percentile access threshold is about 0.57x the national average. The table shows how the low-access × high-elderly group changes as both thresholds move.

Access \ elderly threshold	> Q70	> Q75 (base)	> Q80
< Q20	69	46	32
< Q25 (base)	113	82	50
< Q30	171	128	84

The baseline sits at the midpoint of the grid. Adding the 14 zero-access villages, the high-risk total ranges from 46 to 185 depending on thresholds.

Township accessibility ranking

Median village E2SFCA score per township, relative to the national average. Red = remote township, blue = non-remote.

Show top 20

Cluster analysis

K-means: six vulnerability profiles

353 townships clustered on three features: median E2SFCA score, share of residents aged 65+, and share of zero-access villages. K=6 was chosen by silhouette score (0.418, the highest for K≥3; K=2 only split remote from non-remote). Different profiles have different causes, so they call for different interventions.

Accessibility vs. share of elderly residents

Each dot is a township. Further left = lower access; higher = older population. Dot size reflects the share of zero-access villages. Note that zero-access mountain townships (large red dots) have relatively young populations.

Cluster summary

Cluster	Townships	Access	65+ share	Zero-access %	Remote

Lowest-access townships per cluster (top 5)

Policy implications

Different vulnerabilities need different responses

The 96 high-risk villages are not one group. All 82 low-access × high-elderly villages are in townships that are not officially remote, while all 14 zero-access villages are in mountain indigenous townships. Allocating resources by county or by the remote-area label alone will keep missing the urban fringe.

Suggested directions by profile

Profile	Examples	Main driver	Suggested direction
Mountain indigenous (zero access)	Jianshi (Hsinchu), Alishan (Chiayi), Wanrong (Hualien)	Terrain isolation, no local facility within 60 min	Increase mobile-clinic (IDS) coverage and frequency
Urban fringe (non-remote)	Longtan & Yangmei (Taoyuan), Zhushan (Nantou), Miaoli City	Care concentrated in city centers; not covered by the remote-area label	Base eligibility on measured access, not administrative status
General remote	Nanzhuang (Miaoli), Fuli (Hualien)	Out-migration, ageing population, long trips to care	Aging-in-place care and patient transport
Offshore islands	Penghu, Kinmen, Lienchiang	Not assessed: no road link to the main island (Dongyin excepted)	Needs a separate sea/air transfer analysis

All high-risk villages

High-risk villages

—

Of which zero access

—

Mean 65+ share

—

Village	Risk	Class	Access (vs national)	65+ share	65+ residents

▶ Limitations

1. Demand points are village centroids: the finest public resolution is the village, so polygon centroids stand in for population-weighted centers. This may understate travel times in large rural villages.

2. Geocoding misses are uneven: 7.3% of facilities in mountain indigenous townships could not be geocoded, vs 2.4% in non-remote townships. Missing rural facilities understate rural supply, so the urban–rural gap may be slightly overstated (impact on mountain-township supply < 8%).

3. Mobile clinics are not modelled: available mobile-clinic data does not match the 2024 analysis year. High-risk villages already served by mobile clinics may have better access than shown.

4. Offshore islands are excluded: 15 island townships have no road link to the main island. Their absence from the high-risk list reflects coverage, not safety.

5. Cross-sectional design: this is a 2024 snapshot and does not show change over time.

Data: Ministry of Health and Welfare (facilities & physicians, 2024-12-31), Ministry of the Interior (village population by age, 2024), National Health Insurance Administration, National Land Surveying and Mapping Center (village boundaries), OpenStreetMap contributors (road network). Place names follow official romanization where available.
Analysis & dashboard: Wei-Chi (Jill) Chiou

Taiwan's highest-risk villagesare not where you'd expect