← Projects · Metro Effect · causal inference on Opal data

What did the new metro change?

The Sydney Metro City & Southwest opened on 19 August 2024. Did it change how many people travel from the centres on the line? Opal tap-on data, three causal inference methods, and a placebo test to check the answer isn't noise.

+9.2%Evening tap-ons at Chatswood and Macquarie Park, relative to Parramatta and Strathfield
-0.9%At Sydney CBD and North Sydney, where the new stations opened. Not different from zero
0 of 26Fake opening dates that gave an effect as big as the northern one

Every centre grew, so raw numbers fool you

Comparing the 18 months before opening with the year after, evening tap-ons in the CBD rose +8.0%. That looks like a metro effect until you see Parramatta, which isn't on the line, rose +11.3%. People were still coming back to the office. So every estimate here is relative to Parramatta and Strathfield: what changed at the treated centres beyond what changed everywhere.

New metro stations On the line, no new station Not on the line (controls)

Mean weekday tap-ons between 3pm and 7pm, Feb 2023 to opening against opening to Aug 2025. Public holidays and data outages removed.

The northern end of the line rose after the metro opened

Chatswood and Macquarie Park already had metro trains from 2019, but they ended at Chatswood. From August 2024 they ran through to the city. Relative to the controls, evening tap-ons there rose +9.2% (95% interval +6.8% to +11.5%). The gap had no trend in the 78 weeks before opening, and every month after it is up. Single months bounced around before opening too, so it's the whole year after that stands out, not any one month. Chatswood stepped up in the opening week. Macquarie Park had started rising a few weeks earlier, so not all of its rise can be put down to the metro. At the CBD and North Sydney there's no step at all.

-10%-5%0%+5%+10%+15%+20%-12-9-6-30+3+6+9+12Metro openMonth -12: +2.7% (95% interval +1.2% to +4.2%)Month -11: -2.1% (95% interval -4.2% to +0.1%)Month -10: +1.5% (95% interval -0.4% to +3.5%)Month -9: -4.7% (95% interval -6.6% to -2.7%)Month -8: +0.0% (95% interval -2.0% to +2.0%)Month -7: -3.3% (95% interval -4.6% to -2.0%)Month -6: +2.5% (95% interval -1.1% to +6.3%)Month -5: +5.1% (95% interval +3.7% to +6.6%)Month -4: +0.7% (95% interval -1.7% to +3.1%)Month -3: reference periodMonth -2: reference periodMonth -1: reference periodMonth 0: +6.9% (95% interval +4.9% to +8.9%)Month +1: +7.0% (95% interval +4.7% to +9.4%)Month +2: +8.7% (95% interval +6.3% to +11.1%)Month +3: +5.8% (95% interval +4.4% to +7.3%)Month +4: +10.5% (95% interval +8.6% to +12.5%)Month +5: +8.6% (95% interval +6.8% to +10.5%)Month +6: +6.2% (95% interval +2.0% to +10.6%)Month +7: +15.4% (95% interval +13.8% to +17.0%)Month +8: +13.5% (95% interval +12.0% to +15.1%)Month +9: +12.3% (95% interval +10.8% to +13.8%)Month +10: +7.1% (95% interval +4.9% to +9.3%)Month +11: +10.0% (95% interval +8.2% to +11.9%)Month +12: +13.4% (95% interval +11.7% to +15.3%)

Months from opening (August 2024 = 0). Each dot is the treated-minus-control gap that month, against the three hollow reference months just before opening. Lines are 95% intervals.

What would have happened without the metro?

Synthetic control builds a stand-in for each centre from a weighted mix of the controls, fitted to the weeks before opening. After opening, the stand-in shows the path the centre would have taken if nothing changed. Chatswood: +6.8% against its stand-in, and the fit is 3.0x worse after opening than before, a clear break.

10k12k14k16k20242025Metro open
Chatswood, actual Synthetic Chatswood (dashed): 66% Parramatta, 34% Strathfield

Could it be chance?

The test: pretend the metro opened on a different date, before it really did, and run the same estimate. Do that at 26 dates from Feb 2023 to Feb 2024, each with 26 weeks either side, and at the real date with the same window. If fake openings often give effects as big as the real one, the real one isn't convincing. For Chatswood and Macquarie Park, none of them do. For the CBD and North Sydney, 24 of 26 do.

-10%-5%0%+5%+10%Placebo date 1: -1.0%Placebo date 2: -1.4%Placebo date 3: -2.0%Placebo date 4: -2.6%Placebo date 5: -2.0%Placebo date 6: -1.5%Placebo date 7: -2.2%Placebo date 8: -2.3%Placebo date 9: -2.5%Placebo date 10: -2.4%Placebo date 11: -1.9%Placebo date 12: -1.7%Placebo date 13: -1.5%Placebo date 14: -2.1%Placebo date 15: -1.7%Placebo date 16: -1.2%Placebo date 17: +0.7%Placebo date 18: +0.3%Placebo date 19: -0.2%Placebo date 20: +0.4%Placebo date 21: +0.9%Placebo date 22: +2.0%Placebo date 23: +1.6%Placebo date 24: +1.4%Placebo date 25: +1.9%Placebo date 26: +2.6%Real opening, same window +6.6%

Each grey dot is the effect at one fake opening date. Every estimate here, the real one included, uses 26 weeks either side of the date, which is shorter than the main window, so the real-date number differs from the headline. Placebo p = 0.037.

Does it survive changing the setup?

Each row after the first changes one thing. The northern effect stays between +7.0% and +12.8%, and the new-station effect between -2.3% and +0.6%. Individually, Chatswood is +7.3% and Macquarie Park +11.0%, with a wider interval since it's a smaller, noisier centre.

CheckCBD + North SydneyChatswood + Macquarie ParkChatswood
Main specification-0.9%+9.2%+7.3%
6 months after opening+0.6%+7.0%+6.8%
Everything after opening (to Sep 2026)-1.0%+10.7%+9.3%
Longer before-period (from Jul 2022)-1.2%+8.6%+6.7%
Christmas to January dropped-1.1%+8.9%+7.1%
Tue-Thu only-1.2%+8.8%+7.3%
Fridays only+0.2%+12.8%+9.2%
Parramatta as the only control-2.3%+7.7%+5.8%
Strathfield as the only control+0.5%+10.7%+8.8%

What this can't tell you

  • Why the north rose. A one-seat ride to the city making Chatswood and Macquarie Park more useful fits the pattern. Region-level tap-ons don't separate modes or trips, so this data can't confirm it.
  • Whether the CBD result means nothing happened. Someone who switched from Wynyard to Barangaroo is in the same centre both times. A real shift between stations could net out to zero here.
  • Anything with only two controls. Parramatta and Strathfield are both in the west, so something that hit the north but not the west around August 2024 would look like an effect. North Sydney's own gap was also drifting before opening (pre-trend p = 0.04), so its single-centre estimate is the weakest here.

How it works

Difference-in-differences on the weekly gap between the treated and control averages, with Newey-West standard errors for the correlation between nearby weeks. With six centres, clustering by centre isn't meaningful, so the uncertainty comes from the time series instead. The window runs from Feb 2023, once the post-COVID return had settled, to a year after opening: 78 weeks before and 47 after. Tests check that each method recovers an effect planted in made-up data, and CI rebuilds every number on this page and fails if any of them change.

Data: Transport for NSW Open Data (Opal Tap Data), CC BY 4.0, via the Commute Pulse pipeline (1 January 2020 to 3 October 2026). Opening: Monday 19 August 2024.