What this gives you: for any one of the 1,511 neighbourhoods in Greater Jakarta, how much of the region's job-dense floor space is reachable by public transport in 30, 45 or 60 minutes on a weekday morning, how much of that the train network is responsible for, and how the place compares with the rest of the region.
What it does not give you: your commute. These are scheduled times read off published timetables, not observed ones, and they assume services run as advertised and that you will walk to a stop. Of the 3 checks this case declared in advance as must-pass, 1 passed and 2 failed. A fourth needed a commercial routing service and was never run, so no independent door-to-door comparison exists.
The middle of Greater Jakarta reaches almost none of it
Add up every square metre of non-residential building floor in the region — offices, shops, warehouses, workshops, factories — and you get 91.5 square kilometres of floor space spread over 6,858 square kilometres of land. That is the pot. The question is how much of it any given household can reach.
For the median resident of Greater Jakarta, leaving home in the morning window with an hour to spend, the answer is 0.21% of it — about 0.191 square kilometres of floor space. Inside DKI Jakarta the median is 4.3%, or 3.93 square kilometres. Outside it, across Bogor, Depok, Tangerang and Bekasi, the median is 0.049%. The gap between the two medians is a factor of 87.8.
Two places, both real, both on the map above. The 5 kelurahan of Menteng in central Jakarta hold 136,764 people, who can reach 10.0% of the region's workplaces in the hour and are a median of 9 minutes from a hospital. The 20 kelurahan picked to match them in Kabupaten Bekasi hold 1,182,104 people — nearly nine times as many — who can reach 0.31%, a factor of 32.2 less, and are 20 minutes from a hospital.
At the bottom of the distribution the number stops being a share and becomes a fact about a place. 430 neighbourhoods, home to 3,440,880 people — 9.3% of the region — can reach no job-dense floor space at all in the hour. Not a little. None. And 13,874,196 people, 37.6% of the region, cannot reach a hospital in that time either.
Now switch the map to the trains
Here is the result this case exists to publish, and it is not the one anybody expects. The KRL, MRT and LRT network is the visible, expensive, politically legible part of Greater Jakarta's public transport. Ask what it adds to what people can reach, and the honest answer is: for most of the region, nothing at all.
Of 1,511 neighbourhoods, the rail layer improves access for 292. With a 30-minute budget instead of an hour, it improves access for 68. Put the map on that view and the region goes matte, with the light left hugging the lines.
Where the gain does land, it lands on people who already had the most. 89% of everything the rail network adds falls inside DKI Jakarta, which holds 30% of the region's population. Sort everybody by how much the trains improve their own access, and the top tenth captures 80% of the total gain. Averaged over people, rail adds 0.0077 of the pot inside DKI and 0.0004 outside it.
The bus and minibus system — 240 TransJakarta routes, 98 Mikrotrans routes, 24 Bogor angkot routes, 8,568 stops between them, of which 7,517 see a scheduled bus at least every 15 minutes at peak — does something quite different. It raises the average share reachable by 0.0132 against 0.0026 for rail, and the top tenth captures 54% of that rather than 80%.
Summarised in the single number that describes how unevenly access is spread across the region — 0 would mean every neighbourhood reaches the same amount, 1 would mean one neighbourhood reaches all of it — walking alone gives 0.6293. Adding the buses and minibuses moves it by 0.1134. Adding the trains on top of that moves it by 0.0061. The unglamorous half of the system is doing nearly all of the work.
This is not an argument against building railways. It is a measurement of what the network as scheduled in 2026 delivers to whom, and the answer is that its benefit is concentrated where access was already best. A railway that ran to where the 3,440,880 people who can reach nothing actually live would produce a different measurement. That is the useful reading of this figure.
Is that bad, compared with other cities?
The number above cannot answer that, and it is worth saying why rather than reaching for a figure that looks like it fits. This case measures a share of job-dense floorspace; every published comparison measures jobs. The American series that does this best puts 1,156,929 jobs within an hour of the median New Yorker and 59,826 within an hour of the median Atlantan — a 19-fold spread inside one country using one method. Borrow a number across two methods and you learn nothing about either.
One measure does compare, because it was computed the same way in 26 cities: the share of residents living within a kilometre's walk of high-quality rapid transit. Jakarta the city scores 44%. Greater Jakarta scores 16% — against 23.7% for metropolitan regions across low- and middle-income countries, 37.3% for wealthy ones, and 17.2% even for the six American metros that study called among its worst performers.
The reason given there was that Jakarta's rapid transit did not reach past the city border. That is the same edge this page has been measuring all along, a decade later and by a different route: the middle of DKI reaches 4.3% of the region's job-dense floorspace in the hour, and the middle of Bodetabek reaches 0.05% — 88 times less. Two methods, twenty years of transit building apart, describing one boundary.
The poorer the area, the less it can reach
Joining each neighbourhood to the estimated poverty rate of its sub-district — 187 of them, from the same lab's small-area estimates for 2025 — the two rankings come out close to opposite. The measure of agreement is -0.509, where 0 would mean no relationship at all and −1 a perfect inversion: poorer places reach less, consistently enough that it is not a coincidence.
| Sorted by sub-district poverty | People | Typical share reachable | Reach essentially nothing |
|---|---|---|---|
| best-off fifth · poverty 1.08–3.69% | 12,811,286 | 0.979% | 1.6% |
| fifth 2 · poverty 3.69–4.80% | 9,785,552 | 0.173% | 18.5% |
| fifth 3 · poverty 4.83–6.06% | 6,079,183 | 0.071% | 33.1% |
| fifth 4 · poverty 6.07–7.00% | 5,502,210 | 0.182% | 33.4% |
| poorest fifth · poverty 7.04–23.08% | 2,762,278 | 0.002% | 56.7% |
In the best-off fifth of the region, 1.6% of neighbourhoods can reach essentially nothing. In the poorest fifth, 56.7% can. Both ends of that sentence come from the same travel times.
Cross the two directly and you get the group a government or an NGO would actually act on: neighbourhoods in the poorest fifth and in the bottom fifth for what they can reach. There are 170 of them, holding 1,057,344 people, 2.9% of the region. The third view of the map is that group.
Four choices that move the answer more than the data does
This is the part a government or an NGO should read before quoting any figure from work like this — including this work. Eleven checks were written down before the analysis ran, and the most damaging results are not about measurement error at all. They are about decisions somebody has to make and nobody has to disclose. Each one below leaves every journey time untouched.
- The clock. At 30 minutes the spread measure is 0.6547; at 60 minutes it is 0.7427. The ranking of places barely moves — the 45- and 60-minute orderings agree at 0.9899 — so who is badly off is robust. The headline number is not. The map's own time-budget buttons are that sensitivity test, run by hand.
- Who you count. Across all 1,511 neighbourhoods the spread is 0.7427. Keep only the 270 inside DKI Jakarta and it is 0.3392 — less than half — on the identical travel-time matrix. Inequality did not fall. The 1,241 neighbourhoods it was about were removed from the table. Almost every published transport-access figure for "Jakarta" makes this choice, and most do not say which one they made.
- What you divide by. DKI Jakarta contains 37.7% of the region's job-dense floor space. Measure a Jakarta resident against the whole region and the typical share reachable is 4.3%; measure the same person, on the same train, against Jakarta alone and it is 10.8% — 2.6 times larger on the average. The spread, meanwhile, hardly budges: 0.3392 to 0.3496. That asymmetry is the trap. A headline share is decided by its denominator; a headline spread is immune to it — and the two get quoted as though one caveat covered both.
- The size of the areas. The same travel times, summarised at neighbourhood, sub-district and regency level, give spreads of 0.7427, 0.7212 and 0.6603. The count of units reaching nothing falls from 430 to 9 to 0. Report this at regency level and the problem vanishes from the table without a single person's journey changing. The neighbourhoods themselves differ in area by a factor of 415, from 0.148 to 61.5 square kilometres, which is its own distortion.
The bug in the file this map is drawn from
One of the eleven checks asked whether rounding the published numbers had invented places that reach nothing. It had. At four decimal places 647 neighbourhoods looked like they reached no job-dense floor space; only 430 truly do. The other 217 — 3,289,289 people, 8.9% of the region — had been rounded there, and the map painted them as reaching nothing at all.
The file was rewritten at six decimals, which is where it stands now. It is still not enough. The published file this page's map is drawn from holds 439 exact zeros where the check counts 430: 9 neighbourhoods and 113,678 people are still rounded down to nothing on the map above. That is why its darkest band is labelled essentially nothing rather than nothing, and why the count in its legend (439) is larger than the 430 in the prose. The same two numbers are why the third view highlights 172 neighbourhoods while the text above says 170: two of the nine fall in the poorest fifth.
It is a small error and it is left visible on purpose. A rounding decision three stages upstream of a colour is exactly the kind of thing that reaches a reader as a fact about their neighbourhood.
All eleven checks, and how each came out
A Does the answer depend on the clock you pick? the ranking barely moves, the spread does
At 30, 45 and 60 minutes the places come out in almost the same order — agreement of 0.9899 between the 45- and 60-minute rankings, where 1 would be identical. The one-number spread, though, runs from 0.6547 to 0.7427. The ranking is robust; the headline figure is a choice.
B Could a circle drawn on a map have done this instead? most of the way, and then badly wrong in 34 places
Floorspace within a 10 km circle plus distance to the centre explains about 60% of the differences between places. But for 34 units — 739,629 people — the circle says plenty and the network says nothing reachable. Those are the places the routing exists to find.
C Workplaces, people or hospitals — does the choice change it? yes, and workplaces are the harshest of the three
Measured against where the region's workplaces are, the spread is 0.7427. Against where its people are, 0.6857. Against hospitals it becomes a different statement again: 38% of the region cannot reach one in the hour.
D Does the answer depend on how big the areas are? yes — and the count of cut-off places nearly vanishes
The same travel times, summarised at three administrative levels, give spreads of 0.7427, 0.7212 and 0.6603. The number of units reaching nothing falls from 430 to 9 to 0. Report this at regency level and the problem disappears from the table.
E Does cutting the map at the region's border distort it? a little, and it is bounded
Places within 10 km of the study boundary may have destinations just outside it that were never counted. There are 45 such units, 285,116 people, 0.8% of the region.
F Do the hand-typed train times match the published ones? FAILED — 3 of 4 journeys agreed
No operator publishes a machine-readable timetable for the KRL, MRT or LRT, so their times were entered by hand from published headways. Checked against four advertised journeys, 3 of 4 agreed. The one that failed is a bus corridor whose advertised speed the schedule does not support. Across the rail lines the schedule is optimistic by about 7% against observed running times.
G Do poorer areas have worse access? yes, clearly
Ranking places by poverty and by what they can reach puts them in nearly opposite orders — agreement of -0.509, where 0 would be no relationship and −1 a perfect inversion. In the best-off fifth, 1.6% of places reach essentially nothing. In the poorest fifth, 56.7% do.
H Did rounding the published file invent places that reach nothing? yes — 217 of them
Rounded to four decimals, 647 units looked like they reached nothing. Only 430 truly do. The other 217 — 3,289,289 people, 8.9% of the region — were rounded there. The file was rewritten at six decimals; 9 units are still rounded to zero even so, and this page's map is drawn from that file.
I Who captures what each layer adds? the trains: eight-tenths of the gain to one-tenth of the region
Sort everyone by how much the rail layer improves their access: the top tenth captures 80% of everything it adds. For the bus and minibus system the same figure is 54%. 89% of the rail gain lands inside DKI Jakarta, which holds 30% of the people.
J Does counting floors instead of footprints change it? it gets starker, not different
Swapping building footprint for building volume — an average of 3.68 storeys across the region — moves the spread from 0.7427 to 0.7807. The two rankings agree at 0.9988, so nothing about who is badly off changes; the gap between top and bottom simply widens.
K Does the choice of denominator decide the answer? it decides the level and leaves the spread alone
DKI Jakarta holds 38% of the region's job-dense floorspace. Measure a Jakarta resident against the whole region and the typical share reachable is 4.3%; measure the same people, on the same trains, on the same timetable, against Jakarta alone and it is 10.8% — 2.6 times larger on the average. Nothing was re-routed. Meanwhile the spread barely moves: 0.3392 to 0.3496, because a spread measure is blind to scale by construction. So a headline SHARE is decided by its denominator and a headline SPREAD is not — and the two are routinely quoted as though the same caveat covered both.
Two more limits belong here rather than in a footnote. The train times are hand-typed: no operator publishes a machine-readable timetable (GTFS) for the KRL, MRT or LRT, so their headways were entered from published figures and carry a stated 15% uncertainty; across the lines the schedule is optimistic by about 7% against observed running times. And the router found no destination whatsoever from 15 of the 1,511 origins — 141,626 people, 10 of them on the mainland — which is a gap in the walking network, not a fact about those places.
What a government or an NGO can actually do with this
Three things follow, and the first surprises most people.
- Buses do the distributional work, and it is not close. Adding the bus network moves the spread of access by 0.113; adding rail moves it by 0.006 — roughly 19 times less. Rail's gains land 80% in the top decile of areas against 54% for buses, and buses improve 1,013 areas against 292. That does not make rail wrong — it makes it a different instrument, bought for capacity rather than for equity — and a programme justified on fairness should say which one it is buying.
- Count the people reached, not the kilometres built. The median resident of this region reaches 0.21% of its job-dense floorspace in an hour. Inside Jakarta the median is 4.3% — about 21 times more. Length of line and access per resident rank projects differently, and only the second describes what a passenger gets.
- Look hardest at who is outside the surface. 94.4% of the gain from transit lands inside DKI, and 42% of the region's people reach almost nothing in an hour. A regional programme cannot borrow this evidence without checking whether it holds beyond the capital: the people the map leaves blank are the ones an equity argument is actually about.
What you must not conclude from this map. Not that a bright neighbourhood is a good place to live, or a matte one a bad one: this measures reachable workplace floor space, which is neither jobs nor wages nor whether anyone there wants a job in the centre. Not that anyone's commute takes the time shown — these are timetables, and a scheduled bus is not an arriving bus. Not that a place with no colour has no economy; a village with work in it needs to reach nothing. And not that the trains are wasted money — the finding is about who their benefit currently reaches, measured against a denominator this page had to choose, on one morning's schedule. Above all, do not quote a single spread figure from this work without saying which time budget, which denominator and which administrative level produced it. Three sections up is what happens if you do.
Could this run in daily operations?
FEASIBILITY ONLY — none of it yet
Every pipeline here is a batch backfill: it fetches a season or an archive, not a live feed. Anything operational means rebuilding the ingest for near-real-time arrival, whatever the verdict below says.
What stops it. Two of four checks failed, on the timetable and on the network. The external comparison never ran — it needs live Google Routes calls that were not authorised.
What it would need. The two failed checks passing, and an authorised routing comparison. GTFS feeds change without notice, so operational use also needs feed-drift monitoring nobody has built.
If you want the detail
The methods, the 11 checks in full, the travel-time matrix behind them and everything still uncertain are on the technical article, and the technical dashboard carries the whole access surface, the 5,766 destination cells and the route network. Every figure here is read from the same records as those, so the two cannot drift apart.
Words on this page, in plain language 10 terms
- GTFS
- The standard file format transport operators publish timetables in. It says when a bus is scheduled, not when it arrives.
- a look
- One pass of the satellite over a field. More looks means the crop's cycle is sampled more often; too few and a short stage — flooding, heading — can happen entirely between two looks and never be seen.
- fold
- One slice of the data in a repeated test. How the slices are drawn can matter more to the score than the model does.
- median
- The middle value: half are higher, half lower. Unlike an average, one extreme case cannot drag it.
- sensitivity (analysis)
- Changing an assumption on purpose to see how much the answer moves. If a conclusion depends on a choice nobody can justify, this is what exposes it.
- KRL
- Kereta Rel Listrik, Greater Jakarta's electric commuter rail. The oldest and by far the longest of the region's rail networks, reaching well past the city into Bogor, Bekasi and Tangerang, and the only one most commuters can actually reach.
- MRT
- Jakarta's metro, opened in 2019. Newer and far shorter than the commuter rail, so the number of people living near it is small next to the size of the region.
- LRT
- Light rail — two separate short systems in Greater Jakarta, one run by the city and one by the national operator. Smaller again in reach than the metro.
- regency
- The English name for a kabupaten: the administrative level below a province, and the level Indonesia publishes most official statistics at. A regency is not a city — cities are counted separately here, and on several of these cases they behave differently enough to be reported on their own.
- kabupaten
- An Indonesian regency — the administrative level below a province, and the level most official statistics are published at.
Travel times from published timetables plus the open street network, for a weekday departure between 07:00 and 09:00, taken at the middle of the window. Floor space and population from the European Commission's global settlement layers (CC BY 4.0); hospitals from Healthsites (ODbL); outlines from the humanitarian boundary set (CC BY-IGO (HDX COD-AB / BPS)); the street and rail network from OpenStreetMap (ODbL 1.0). Data vintage 2026-08-30. Outlines simplified to 200 m for the web, 37,374 points reduced to 16,587, from the same geometry the analysis used; no neighbourhood was dropped. Written to be read without a background in transport modelling; nothing was rounded to make a point, and the one place rounding did damage has a section of its own.
Found something wrong on this page? Report a correction — it opens a pre-filled issue — or email taufik.adi@openstudy.id. Corrections are credited by name in the errata, and one that changes a finding says so on the page.