humancentric

The argument is sound. The evidence is missing.

Everything on this site is built from published standards. What nobody has is a systematic record of what those standards produce on the ground. That is cheap to collect, and it is what turns a correct argument into a document an agency has to answer.

Five minutes a site. Consistency matters more than volume.

  1. Locate. Drop a pin and record exact coordinates, plus the nearest block number or road name. A vague location is a record an agency cannot act on.
  2. Measure nominal width. Tape measure across the paved footpath, edge to edge, perpendicular to the direction of travel. Metres, two decimals.
  3. Measure effective width. Deduct 0.20 m at a kerb edge and 0.20 m at a wall or building line, and measure around lamp posts, bins, drain covers and planters as obstacles. Record the narrowest clear width in the section. This is the number that matters and the one nobody publishes.
  4. Photograph three ways. Straight down the path at eye level; across the path with the tape measure visible and readable; and the boundary between footpath and cycling path in close-up, showing the kerb or verge profile.
  5. Record the boundary type. Flush, sloped kerb, vertical kerb, planted verge, paint only, or no cycling path. If there is a level difference, measure it in centimetres.
  6. Note observed behaviour, carefully. Count over five minutes: pairs that broke into single file, and people who walked on the cycling path. State the time and day, since both matter. Do not photograph identifiable faces.

A completed record looks like this:

SITE      SG-014
LOCATION  1.3521, 103.8198 — Blk 412 Ang Mo Kio Ave 10, north side
DATE      2026-08-06, 18:40 (weekday, evening peak)
NOMINAL   1.52 m
EFFECTIVE 1.12 m  (kerb 0.20, wall 0.20; narrowest at lamp post 0.94)
BOUNDARY  planted verge, 0.45 m wide, level difference 0.09 m
CYCLEPATH 2.0 m, colour-demarcated
OBSERVED  5 min: 7 pairs broke to single file; 4 pedestrians on cycling path
PHOTOS    SG-014-along.jpg, SG-014-across.jpg, SG-014-boundary.jpg
NOTE      Bus stop at 30 m compresses effective width further at peak
Aim for ten to fifteen sites before approaching anyone. A single site is an anecdote and will be treated as one. Ten sites across different towns, measured the same way, is a dataset. Include at least two that are compliant and comfortable — a submission reporting only failures reads as advocacy, while one reporting the range reads as survey work, and survey work gets a substantive reply.

Does footpath width change whether people can walk beside each other?

Every pedestrian design standard in the world assumes the answer is yes. None of them tested it. We searched: there is no published study measuring footpath width against the rate at which walking pairs break into single file. Not in Singapore, not in Europe, not in the United States.

This is a real gap, not a search failure. The walkability literature measures whether people walk at all, where width returns an elasticity of 0.00. Nobody measured what walking is like. Width standards worldwide are derived from body-envelope arithmetic and wheelchair-passing dimensions, and stop there.

Fifteen sites, a tape measure and about six hours of standing still would produce the first data on this that exists. It is genuinely within reach of one person over two weekends.

Register the prediction before you collect anything. Write it down, commit it to the repository, and let the commit timestamp be the record. That single step is what separates a study from motivated measurement, and it costs nothing. If the data comes out against you, you publish it anyway — and having pre-registered is what makes the rest of this site's claims believable.

The pre-registered prediction

The anthropometry makes two separate quantitative predictions, and they can fail independently. That is what makes this worth doing rather than merely worth asserting.

2.03 meffective width at which a pair should stop breaking for a single oncoming person
1.22 body + 0.20 clearance + 0.61 body
2.64 meffective width at which two pairs should pass without either breaking
1.22 + 0.20 + 1.22
Falsified ifbreak rate does not decline with effective width, or flow rate explains more of the variance than width does

The one design decision that matters

Count encounters, not pairs. Formation breaks are caused by meeting someone, not by width alone. If you count "pairs that broke formation" you will mostly be measuring how busy the path was that evening. The unit of observation is the pair-encounter: every occasion on which a walking pair meets someone travelling the other way. The measure is what proportion of those encounters forces a break. That holds flow constant, which would otherwise swamp the width signal entirely.

Definitions, fixed in advance

TermCounts asDoes not count as
PairTwo people travelling the same direction, side by side within about 1.5 m, evidently together: conversing, matched pace, or arriving togetherThree or more (record separately); two strangers who happen to be level; a parent carrying a child
EncounterThe pair meets at least one person travelling the opposite direction within the marked segmentBeing overtaken from behind; passing a stationary person; meeting a cyclist on the cycling path
BreakAt any point in the encounter, one member moves ahead of or behind the other so they are no longer abreastNarrowing the gap while staying level; one person turning their body without changing position
UncertainYou could not see clearly, or could not tell whether they were together

Record every uncertain case and report the count. A study that never reports ambiguity is a study where someone quietly resolved it in their own favour.

Method

  1. Choose the fifteen sites before you observe any of them. Pick them to span effective widths from roughly 1.1 m to 3.0 m, spread across at least four towns. Choosing sites after watching them is how you accidentally prove yourself right.
  2. Mark a 30-metre segment with two fixed reference points you can see clearly — a lamp post and a drain cover will do. Every encounter is counted only inside it.
  3. Measure the segment before you start: nominal width, effective width (deduct 0.20 m at kerb and 0.20 m at wall, and measure around obstacles), the narrowest point, obstacle count, boundary type, and whether a cycling path adjoins.
  4. Stand still, off the path, and be boring. A bus stop bench or a void deck edge is ideal. People change how they walk when they know they are being watched, so do not follow anyone, do not film faces, and do not stand in the segment you are measuring.
  5. Record 30 encounters per site, one row each. Note the time, what the pair met (single, pair, group of three or more), whether a break occurred, and who moved.
  6. Count flow separately. Every ten minutes, count total pedestrians passing the segment in one minute. This is the covariate that makes the analysis honest.
  7. Do the same site twice — one weekday evening peak, one weekend afternoon — if you can. Same-site repetition is cheap and tells you how noisy a single visit is.

The recording sheet

Two files, one row per site and one row per encounter. Templates with a worked example row are in the repository under /data.

SITES
site_id, date, lat, lon, location_name, observer, start_time, end_time,
weather, segment_length_m, nominal_width_m, effective_width_m,
narrowest_point_m, obstacle_count, cycling_path, boundary_type,
shade_estimate_pct, gradient

ENCOUNTERS
site_id, encounter_no, time, pair_direction, oncoming_type,
broke, who_broke, flow_per_min, notes

Analysis, decided in advance

Bin the sites by effective width in 0.25 m steps. Plot break rate per bin with confidence intervals. If you want the formal version, run a logistic regression with break as the outcome, effective width as the predictor and flow rate as a covariate, clustering by site.

Report the whole curve, including the bins that disagree with you. Report the excluded count. Report the two sites that behaved oddly and say you do not know why. Fifteen sites and 450 encounters is a pilot, powered to detect a large effect and not a subtle one — say that too, rather than waiting for a reviewer to say it for you.

Why this is worth two weekends. If the relationship exists, you have produced the first evidence for something every design standard on earth assumes without testing, using Singapore paths and Singapore pedestrians. If it does not exist, you have learned that your strongest intuition was wrong before an agency taught you in public — and you would still be the only person who has looked.

Two extra fields worth capturing.

For a leftover green parcel: before anything else, check its zoning on URA's Master Plan viewer. If it comes back Open Space or Park, the argument is airtight. If it comes back Residential with a plot ratio, then "why is this empty" has a boring answer, someone will give it to you in public, and you should quietly pick a different site. Then record: area, whether canopy already covers any of it, what fronts each edge, and existing seating in linear metres.

For a desire line: photograph straight down the worn track with something for scale, then record the two endpoints it connects and the length of the paved route between the same two points. The ratio is the finding.

Different arguments go to different doors.

OneService

Routes to the responsible agency and generates a case reference. That reference is the real value: it creates a paper trail you can cite later.

LTA and REACH

For anything about §2.1.2, effective width, or the design guide itself. Attach the dataset, not the complaint.

Your MP

At a Meet-the-People session. An MP can file a parliamentary question, and a written question on the record about §2.1.2 is worth more than any number of emails.

The sharpest single question available right now. Friendly Streets was announced in 2023 with "widened footpaths" among its headline features. It has since covered nine neighbourhoods with a billion dollars committed and is targeting all 24 towns by 2030. How many metres of footpath has it actually widened, and at which locations?

Corrections are worth more than agreement.

Every page here marks what is a sourced finding and what is an inference. If a figure is wrong, out of date, or superseded, it should be fixed before anyone acts on it. If you can improve a page, or you have measured sites to add, the repository is public and issues and pull requests are open.

The repository