Hong Kong–Shenzhen Transport Index Method and Anonymisation

One-line answer: The proposed formal version uses 2,400 valid events: 1,600 checkpoint observations and 800 airport pickup events, covering weekdays, weekends and at least four expo dates with random IDs and 15-minute time rounding.

Proposed sample and collection window

Module	Proposed sample	Stratification
Four land checkpoints	1,600	About 400 complete vehicle events per checkpoint
Hong Kong airport pickup	800	Arrival deviation, pickup/drop-off, time band and vehicle class
Total	2,400	Keep only de-duplicated events with complete timestamps

Proposed collection window: 1 September–31 December 2026, covering weekdays, weekends and at least four verifiable expo dates. If the budget is smaller, protect minimum coverage per checkpoint and date type before expanding the easiest daytime sample.

Definitions, cleaning and quality control

Calculate P50/P90 within checkpoint, direction, date type or airport segment using completed, timestamp-valid events;
Define checkpoint time from first queue point to completed both-side clearance and port-area exit; measure city access separately;
Remove duplicate IDs, reversed timestamps, impossible cross-midnight errors and incomplete trips;
Manually audit at least 5% of events and preserve data version, sample count, collection window and script version;
Do not delete outliers silently; document any trimming or sensitivity analysis.

Privacy controls

Use random event IDs and remove names, phone numbers, WeChat, WhatsApp and full order numbers;
Round published times to 15-minute bands;
Do not publish plates, driver names, flight numbers, exact pickup addresses or identifying field combinations;
Merge small groups or publish ranges when a segment is too small to protect privacy.

Follow-up questions

Why not derive clearance time from passenger-flow totals?

Passenger flow describes scale and peaks, but it cannot directly provide the minutes from a vehicle queue to port exit. A time index needs trip events with start and end timestamps.

Is 2,400 enough to represent a whole year?

It is a proposed pilot size, not an automatic guarantee of representativeness. The formal report should disclose coverage by checkpoint, direction, time, date type, vehicle class and unusual events.

Can dispatch teams still audit anonymised events?

Yes. A random event ID, time band, route dimensions and cleaning status are enough for operational review; raw access should be separated from the public report.

Sources and limits

This page is a proposed collection and governance plan; it does not claim that 2,400 field events have already been collected. Last updated: 2026-08-19.