Projects & Writing

Beyond the papers

Independent research built on public data, and a series on learning and teaching psychometrics with AI. The sites and the series are in Traditional Chinese.

← Back to home

Independent research

  • 01

    Legislative Yuan Issue Atlas 立法院議題關注地圖

    Which legislators attend to which issues in Taiwan’s Legislative Yuan, how that attention shifts from year to year, and how far bills progress.

    Data
    Legislative years 2016–2025 (each runs Feb 1 to Jan 31): rosters, bills, meetings, gazette transcripts, and the IVOD video index from LYAPI (a third-party compilation of Legislative Yuan open data), bill-number mappings from the Legislative Yuan open-data platform, and winner lists from the Central Election Commission.
    Methods
    Every record is classified by keyword rules; a legislator attends to an issue in a year with at least 3 substantive speeches or primary proposals on it that make up at least 5% of their weighted records. Each release is replayed offline from raw snapshots and published with SHA-256 manifests; every figure links back to official records.
    Limitation
    The issue classification is exploratory, with no human validation, and the figures describe observable activity, not policy credit or causal effects.
  • 02

    Taipei–New Taipei Mobility Atlas 流動的雙北

    Does a nearby bus stop make daily life accessible? How bus routes, the MRT, and walking access shape everyday reach in Taipei and New Taipei.

    Data
    1,051 bus routes (2,568 directional and branch records) across 41 districts, a 6,047-cell residential grid, MRT origin–destination trips (July 2026), workplace employment (2021), and weekday daytime activity population (November 2023).
    Methods
    Route geometry compared by dynamic time warping and shared stops by Jaccard similarity; UMAP, diffusion maps, and MDS for route maps; walking distance measured both in a straight line and along the street network.
    Finding
    In 509 of the 6,047 residential cells, the 90th-percentile distance to a stop is within 500 m in a straight line but exceeds 500 m along the streets.
    Limitation
    District-level figures are not stop-level counts, and the frequency scenarios compare operating conditions rather than the effects of implemented policies.
  • 03

    New Taipei Support Access Map 新北身心支持資源地圖

    How easily can residents of New Taipei’s 1,039 villages reach mental-health services, institutional respite care, and residential care beds?

    Data
    New Taipei directories of mental-health services, institutional respite care, and elderly welfare institutions (November 2025 to August 2026), village boundaries and single-year-of-age population (September 2026), a national digital-access survey, and an OpenStreetMap Taiwan snapshot (September 4, 2026).
    Methods
    Walking time (at 4.8 km/h) and congestion-free road time from each village, with facilities geocoded to OpenStreetMap address points, and enhanced two-step floating catchment area (E2SFCA) accessibility; digital use is a scenario estimate that applies age-specific internet use to each village’s age structure.
    Finding
    For residents aged 65 and over, 70% live within a 15-minute walk of any mental-health clinic or counselling service and 90% within 30 minutes; 107 villages (35,042 people aged 65+) are in the lowest fifth on both physical access and estimated digital use.
    Limitation
    The directories record no capacity, waiting times, or prices, so each service point counts as one unit of supply; the map does not estimate anyone’s mental health, and a gap means only relatively low access.
  • 04

    Referendum Atlas 公投與在地

    Did villages closer to nuclear, coal, and gas plants vote differently in Taiwan’s national referendums, and how do district results line up with legislators’ issue attention, a nuclear roll-call vote, and the 2025 recalls?

    Data
    16 national referendum questions (2018, 2021, 2022, 2025) across 7,751 villages from the Central Election Commission, matched across boundary changes with village boundary histories; party-list votes (2016, 2020, 2024); a 2025 roll-call vote and recall results; legislators’ speech data from the Legislative Yuan Issue Atlas; and 11 energy facilities located from OpenStreetMap.
    Methods
    Yes share by straight-line distance band to the nearest facility (0–5, 5–10, 10–20, 20–50, 50+ km) with township-resampled intervals; village fixed-effects panel models for Cases 16, 17, and 21, with the unrelated Cases 10, 13, and 18 as checks; results shown with and without controls for party-list leaning.
    Limitation
    All results are village or district aggregates and say nothing about how individuals voted; distance may also stand for urban–rural and other local differences, since the unrelated referendums show cross-sectional gaps of similar size.
  • 05

    Taiwan Fertility Atlas 臺灣的孩子為什麼變少?

    Why are fewer children being born in Taiwan? Births, fertility rates, maternal age and birth order, and how the change in the number of births breaks down, nationally and by county and township.

    Data
    Household-registration statistics from the Ministry of the Interior: national series from 1947 to 2025 and county and township series from 2007 to 2025, covering births, age-specific and total fertility, birth order, marriage timing, and marital status; township income from tax returns (Fiscal Information Agency, 2012–2023).
    Methods
    Age-specific and total fertility rates computed from births and female population, with three-year pooled township rates; a demographic decomposition of the change in births into the number of women of childbearing age, age structure, and age-specific fertility; county and year fixed-effects models for associated factors (2013–2024, factors lagged one year).
    Finding
    Births fell by 107,417 between 2015 and 2025: −21,101 from fewer women of childbearing age, −8,220 from age structure, and −78,096 from lower age-specific fertility. The official total fertility rate in 2025 was 0.695.
    Limitation
    Figures are household-registration counts assigned to the registered residence; the decomposition is an accounting identity, not a causal analysis, and the factor associations are descriptive.

Writing