Who worked at home before the pandemic?

Sun 18 September, 2022

Can we learn anything from pre-pandemic working-at-home patterns that will help us predict transport demand “after” the pandemic?

This post investigates work-at-home patterns from the ABS census 2016 for the six largest Australian cities, with some deeper dives for Melbourne and Sydney. I’ll answer questions such as: What occupations and industries were more likely to work-at-home? How did work-at-home rates vary by home and work locations? How many people had their home double as their workplace? Who was ‘remote working’ at home away from their regular workplace?

I’ve found the results quite interesting – and not quite what you might expect from the our current pandemic perspective.

What proportion of workers worked at home in 2016?

The following table shows between 3.3% and 5.2% of major city workers reported that they “worked at home” on census day in 2016 in Australia’s six largest cities:

This highest rate was in Brisbane, and the lowest in Canberra.

What occupations were more likely to work at home in 2016?

Here’s a chart showing 2016 journey to work mode shares across Australia’s six largest cities by main occupation category. Normally I exclude people working at home from mode share charts, but for this analysis I’m including “worked at home” as a “mode”:

Technical note: As usual on my blog, public includes all journeys involving a bus, tram, train and/or ferry trip, Active includes walk-only and cycle-only journeys, with all other journeys counted as Private.

You’ll notice the occupations with the highest rates of working at home were also the occupations with the highest public transport mode shares – professionals, clerical and administrative workers, and managers.

Here’s another view of that data, this time providing the occupation breakdown of commuters for each “mode”:

Again you can clearly see the same three occupation categories that dominated both working at home and public transport commuting.

No surprises there, right? These occupations generally spend a lot of time in offices either at computers or meeting with others – which can more easily be done online so are more likely to be amenable to working from home. The other occupation categories are more likely to necessitate working at a specific workplace.

But there are lots of different types of managers and professionals and they work in many different industries, so let’s dig a little deeper.

How did working at home vary by employment industry?

Here’s a look at the worked-at-home rates in 2016 by industry and occupation (highest level categories). I’ve sorted the occupations and industries such that the highest rates of worked-at-home are towards to the left and top of the table respectively.

The highest rates of working at home were found in Agriculture, Forestry and Fishing, Arts and Recreational Services, and Construction. Not exactly the sorts of jobs you would expect to fill multiple CBD office towers.

It’s also worth looking at the second level of occupation classifications:

Now we start to see working at home rates are very high for farm managers and arts and media professionals. For many of these people their workplace is quite likely to also be their home.

Many occupations that you might expect to be generally office-based had a working at home rate of around 9% – including HR, marketing, ICT, design, engineering, science and transport professionals.

How did working at home vary by home location?

Here’s a map showing working at home rates for SA2s across Greater Melbourne:

Working at home rates were highest in peri-urban areas, higher than the average in more advantaged suburbs of Melbourne, and the lowest rates of working at home were for employees from more disadvantaged areas.

Here’s the same for Sydney:

The highest rates were also seen in peri-urban areas of Greater Sydney, and the generally wealthy upper north shore.

How did working at home vary by workplace location?

Here’s a chart showing the worked at home share for the Melbourne and Geelong region, by workplace location.

The highest worked at home rate was 47.5% seen on French Island – a sparsely populated island south-east of Melbourne which contains many small farms and some tourist facilities. Other worked-at-home hotspots include the Point Cook East SA2 (which includes an air force base) and Panton Hill – St Andrews (which I understand contains many small farms). In fact, worked-at-home rates were again generally much higher in peri-urban areas and very low in suburban areas.

Some of the lowest rates of working at home were seen for employees in industrial areas and at Melbourne Airport. Many of these jobs are probably hard to do remotely.

2.0% of Melbourne CBD workers worked at home on census day in 2016. And the SA2s surrounding the CBD were all below 3%.

Here’s the same map for Sydney:

Again the highest rates of employees working at home were seen in peri-urban areas, and the Sydney CBD saw 1.9% of employees working at home.

These maps tell us that working at home in 2016 was most common in peri-urban areas, and relatively rare in dense employment areas such as CBDs. The COVID19 pandemic triggered significant levels of working at home during lockdown periods which emptied central city office towers and has remained quite common ever since. So it is likely that the profile of people working at home has changed significantly since 2016 to include a lot more white collar workers.

The fact that working at home rates were high in peri-urban areas when measured as both home location and work location suggests that for many people their home is their workplace. So…

How common was remote working in 2016?

You might have noticed that I’ve been referring to “worked at home” rather than the currently popular term “working from home”. That’s not just because its the wording used by the ABS in reporting the census, but because “working from home” is a little ambiguous as to whether people are working at home and away from their regular workplace, or whether their home is also their regular workplace. Perhaps a better term to describe people working at home and away of their regular workplace is “remote working”.

I have extracted worked-at-home workers’ home and work SA2 locations for people who lived and worked in Greater Melbourne and found that 89% of workers who worked at home, had their usual place of work in the same SA2 as where they lived (SA2s are roughly the size of a suburb).

So while 4.5% of workers who both lived and worked in Greater Melbourne worked at home in 2016, only 0.48% worked at home on census day when their regular workplace was in a different SA2. Remote working was an order of magnitude smaller than working at home.

Now it is also possible that some workers who lived and regularly worked in the same SA2 were actually working at home remote from their workplace on census day. However I expect this to be rare, and some further analysis (detailed in the appendix to the post) found that the almost every worker who worked and lived in the same SA2 had their home SA1 area intersect or overlap with their workplace Destination Zone (both the smallest census land areas available). This doesn’t guarantee that their home was their regular workplace, but it makes it quite probable. These workers would mostly not have had a very long commute, so there would be little incentive to remote work to avoid commuting effort. Also, I’ve found people who travelled to work in the same SA2 as they lived were slightly more likely to work in accommodation and food services, construction and retail trade – industries that are likely to require worksite attendance.

So I think I can fairly safely estimate the 2016 remote working rate in Greater Melbourne to have been 0.5%.

I’ve repeated this calculation for Australia’s six largest cities:

I’ve ordered the cities by working population, and you can see remote working rates decline across the chart for smaller cities. This might reflect there being a larger incentive to avoid longer and/or expensive commutes in larger cities by remote working.

Curiously Brisbane had the highest rate of workers whose home doubled as their workplace (4.9%), while the Australian Capital Territory (i.e. Canberra) had the lowest rates of both working at home and remote working.

I think these quite small estimated rates of remote working are an important finding, as several recent reports from the Productivity Commission, SGS Economics and Planning, Monash University, and iMove may have conflated working from home with working remotely at home, at least in their discussion of the topic. It’s critical that these metrics are not mixed up. And thankfully I’m not aware of any obvious miscalculations in their work.

How did rates of remote working vary across workplace locations in 2016?

The following maps exclude people who lived and worked in the same SA2, to get an estimate of remote working by workplace SA2:

The estimated remote working rate peaked for the Docklands SA2 at 1.8%, with Melbourne’s CBD at 1.7%, Southbank at 1.5%, and Albert Park at 1.4%. These are higher than the worked-at-home rates calculated above for all employees who regularly worked in the city centre, because they remove people who regularly work at their home in the central city.

There were also some seemingly random suburban locations with similar rates of remote working such as Forest Hill and Fawkner at 1.6%.

Here’s the equivalent map for Sydney:

There was a curious hot spot of West Pennant Hills at 5.5%, while the Sydney CBD area was 1.8%, North Sydney – Lavender Bay 2.3%, Macquarie Park – Marsfield 2.0%, and North Ryde – East Ryde 1.8%.

How did rates of remote working vary by home SA2?

Here’s a map estimating remote working rates by home location for Melbourne in 2016:

Generally higher rates were seen in peri-urban areas with Flinders at 2.0%, Mount Eliza at 1.4%, Gisborne at 1.2%, and Panton Hill – St Andrews at 1.6%. This may reflect “sea-changers” and “tree-changers” avoiding a long commute to work. The lowest rates were seen in the more disadvantaged areas of Melbourne, which probably reflects such employees being more likely to work in occupations that require attendance at their workplace.

And for Greater Sydney:

Higher rates of remote working were seen across the upper north shore, with Avalon – Palm Beach at 2.4%, and in many peri-urban areas. But the highest rate was seen at Blackheath – Megalong Valley (in the Blue Mountains) with 3.5%.

In what occupations and industries was remote working more common in 2016?

It’s stretching what you can do with ABS TableBuilder, but I’ve extracted counts of workers by home SA2, work SA2, industry main code, and whether the worker travelled to work or worked at home, for Greater Melbourne for 2016. I’ve then filtered for workers whose regular workplace is not in their home SA2. It’s a little problematic in that about one quarter of the non-zero records in this data were a value of 3, and ABS never reports counts of 1 or 2 as it uses randomisation to protect privacy for very small counts. So the totals are accumulating the impacts of lots of small random adjustments, but it’s not clear that this would introduce a bias to the overall estimate, but we should still treat these with caution and I’m not going to quote more than one decimal place. That said, the estimates do seem very plausible:

The industries with the highest estimated rates of remote working are mostly white collar jobs, whilst those industries with the lowest rates are more blue collar.

I did the same analysis for occupations, and again there are few surprises in the estimated rates of remote work across the categories:

What will the 2021 census tell us?

The 2021 census was conducted during a period of tight lockdowns in Victoria and New South Wales. Most other states had relatively few restrictions, but had experienced lockdowns in 2020, so were arguably in a “post pandemic” scenario – at least temporarily. So it will be very interesting to compare 2016 rates of remote working to those in different cities in 2021. For cities that were not in lockdown we will likely get a good sense of which occupations had high rates of (unforced) remote working, which will be very useful for modelling future rates of remote working and the ongoing impact on transport demand.

I expect the patterns across industries and occupations will be similar between 2016 and 2021, but with much higher rates of remote working in 2021.

The data will be released in October 2022 and I’ll be keen to calculate remote working estimates and share those on the blog.

There have also been several surveys that provide breakdowns of remote working by occupation and/or industry during the pandemic (Productivity Commission, iMove, University of Sydney ITLS).

Appendix: Did anyone live and work in the same SA2 but not have their workplace at their home?

To try to answer this I extracted data for “worked at home” cases in Greater Melbourne at the maximum available resolution – SA1 for home location and Destination Zone for workplaces, and determined whether their home SA1 intersected with their workplace Destination Zone. An intersection between these areas doesn’t guarantee the workplace is at their home, but the absence of an intersection does guarantee that the workplace is not at their home.

Here’s a map extracted from maps.abs.gov.au that shows 2016 destination zone boundaries in blue, and SA1 boundaries in red for part of the northern suburbs of Melbourne:

I dare suggest that if someone lived in an SA1 that intersected with their regular workplace Destination Zone, it’s pretty likely that they ordinarily worked at home.

This analysis is stretching the data, because when you extract small counts from ABS they apply random small adjustments to protect privacy and also you never see a count of 1 or 2 people. Problematic as it is, the sum of people living and working in the same SA2, but living in an SA1 that does not intersect the destination zone in which they work was just 95 for all of Greater Melbourne, out of around 70,000 people who lived and worked in the same SA2. This is a lower bound on the true number, but I expect the true number to still be very small. Hence I’m comfortable with an estimate of 0.5% remote working in Melbourne in 2016 (to one decimal place).

Another potential issue is that SA2s are not consistently sized across cities, and are generally smaller in Brisbane and Canberra. This means remote working from a nearby workplace would be more likely to be detected those cities. However I suspect these instances will still be tiny, and the estimated remote working rates in Brisbane and Canberra certainly don’t appear to be outliers.


What impact has the 2020 COVID-19 pandemic had on pedestrian volumes in central Melbourne?

Mon 25 May, 2020

[last updated 8 July 2020]

Consistent with reductions in vehicle and cycling traffic, central city pedestrian volumes measured by the the City of Melbourne also dropped considerably during the pandemic.

How much have volumes reduced? How has this varied by day types, locations, and times of day? Join me as I dive into the data.

The City of Melbourne have installed 64 pedestrian counters in and around the CBD. Here’s a map of the sites and some (arbitrary) groups (which I’ll use later):

The sensors are not evenly distributed over the city, with a bias towards the central retail core, so they are unlikely to be perfectly representative of central city pedestrian activity, but the data is available and is interesting.

Of course sensors fail from time to time, so we don’t have a complete time series for all sites for all days. There have also been many more sensors added over time. Here is a chart showing the sensors reporting for each day since counting began in 2009:

To get a reasonable comparison, the following chart aggregates data from 44 measuring sites that have complete or near-complete data for 2019 and 2020 (so far):

The gaps in the lines are due to public holidays, which have been excluded (I have not coded Easter Saturday as a public holiday).

You can see volumes drop significantly from around week 12 onward in 2020 (starts Sunday 15 March), as restrictions were introduced.

You can also see significant week to week variations in volumes in 2019, so when measuring the decline I’m going to compare volumes with those in the first two weeks of March (when universities had commenced on-campus teaching).

Here are daily volumes relative to the average of the first two weeks in March:

You can see volumes down over 80% by early April, followed by some small growth until late June when things went back into decline as the number of COVID-19 cases reports in Victoria rose.

The reductions were fairly consistent across all day types, until May when Saturday volumes grew faster. The variation between consecutive days in February reduced dramatically, suggesting perhaps there was a lot less discretionary pedestrian activity during the lock down.

During the first recovery phase there have been a few outliers:

  • Thursday 9 April was the day before Good Friday when most retail trading is restricted.
  • Wednesday 29 April was a very wet day (23.6 mm of rain)
  • Saturday 16 May was the first Saturday after restrictions were eased (also a fine sunny day of maximum 18 degrees).
  • Monday 1 June was cold and wet.

While the Sunday decline appears to be the largest, Sunday 8 March was during the Moomba festival on a long weekend, so there were many more people in the city than normal that night, inflating the baseline.

Likewise the first two Saturdays in March had quite different volumes, which may be related to special events as well. So I would suggest not getting carried away with the exact decline percentages.

How have volumes changed in different parts of the city?

Of course the pedestrian volume reductions have not been uniform by place or day of the week. Here are the reductions on weekdays for week 14 (29 March – 4 April), when overall volumes bottomed:

Volumes were down the most around Melbourne University, and reductions of around 85% were typical in the CBD grid. There were smaller reductions in Docklands (which might reflect many pedestrians being residents), and around Queen Victoria Market (one site only down 44%).

Here’s the same again for Saturday 4 April:

The largest reductions around the arts precinct in Southbank, the retail core of the CBD, and around Melbourne University. Lesser declines are again in Docklands and around Queen Victoria Market.

And here is Sunday 5 April:

Patterns are similar again.

What are the trends in different parts of the city?

The next chart looks at the volumes trends for my sensor groups over time for weekdays:

The relative decline was initially fairly consistent across the groups over the weeks, with the university sites showing the biggest declines, and the residential and retail sites showing the least decline. The retail precincts of Lygon Street, CBD central, and Melbourne Central (around the station) have shown the most growth since May, as cafe restrictions were listed.

The story is quite different on Saturdays:

There is historically a lot more week to week variation, and the numbers for Docklands have bounced around a fair bit – with 16 May close to normal levels of pedestrian activity (a dry day with maximum 18 degrees, lower days had rain). Saturday 23 May was a fairly wet day, so might have discouraged travel.

Queen Victoria Market has also shown considerable growth since early April – with volumes within the bounds of regular volumes.

The spike for CBD east on 7 June was due to the Black Lives Matter protest.

Saturday pedestrian volumes have declined from late June as concern rose about a second wave.

Sundays are similar:

Docklands, Queen Victoria Market and Melbourne Central all increased on Sundays during May (all with little or no rain).

The fact that the Southbank / River group has shown the largest decline is probably related to it having a high base – with the Moomba festival causing a spike in pedestrian volumes on 8 March.

How have volumes changed by time of day?

Here’s the profile of hourly volumes for sites with complete data for 2020 on weekdays:

You can see the normal AM peak, lunchtime peak, and PM peak, which have been largely flattened since the pandemic hit. Weekdays are now busiest in the afternoons – which is a profile much more like a normal weekend day (more on that in a moment).

If you follow the colours carefully you can see the rapid decline in late March, followed by slow growth, peaking in the week of 14 June, then declining again.

Here are hourly volumes relative to the first two weeks of March:

The biggest reductions have been in the AM peak and evenings, which reflects a reduction in commuters and hospitality activity. The reductions are slightly smaller mid-morning and mid-afternoon (between the regular peaks) reflecting a flattening of the profile.

The smallest percentage reductions have been at 4-5am in the morning, off a small base.

Here is Saturdays:

Reductions have again been largest in the evenings, just after midnight (Friday night), and least around dawn. You can see Saturday afternoons showing growth until mid June, but little growth in the evenings as restaurants, bars, and theatres remained closed.

Same again for Sundays:

Sunday 8 March is an outlier in the day and evening – with the Moomba festival on, and the following Monday being a public holiday.

Another way to visualise hourly data

Here’s a chart that shows pedestrian volumes for every hour of 2020 up to and including 7 July 2020. The rows are days, and the columns are hours of the day:

You can see how pedestrian activity very quickly became quiet in March. Before the shutdown you can also see the weekly patterns, with weekend activity starting later and finishing later.

The top row is New Years Day, and you can see high volumes in the first few hours from new year celebrations.

May 16th was the first Saturday after restrictions were eased and that shows up as the first spike in the recovery phase.

This can be filtered for locations. For example, here is the data for Queen Victoria Market sensors:

You can see clear stripes for days the market was open (including night markets). The first busy day after the shut down was the Thursday before Good Friday – perhaps people cramming shopping ahead of Good Friday (Easter Saturday was also busy). The market continued to trade throughout this time. Things started to quieten down in late June / early July.

I will try to periodically update this post during the recovery.

An aside: visualising activity over a long weekend

Nothing to do with the pandemic, but a bit of fun to finish. Here is an animation of pedestrian volumes over the Labour Day long weekend 6-9 March 2020 (Friday to Monday):

If you watch carefully you’ll spot some sudden surges from a Saturday evening event at Docklands Stadium.


What impact has the 2020 COVID-19 pandemic had on cycling patterns in Melbourne?

Sun 17 May, 2020

[fully revised 12 July 2020 with data up to 5 July 2020]

There has been talk about about a boom in cycling during the COVID-19 pandemic of 2020 (e.g. refer The Age), but has that happened across all parts the city, across lanes and paths, and on all days of the week?

About the data

In Melbourne there are bicycle counters on various popular bike paths and lanes around the city (mostly inner and middle suburbs), and so I thought it would be worth taking a look at the data. But I should stress that the patterns at these sites may or may not cycling patterns across Melbourne.

Here’s a map of the sites, including my classification of bike path sites into ‘recreational’ and ‘other’ (based on time of day demand profiles):

There’s also one bicycle counting site on Phillip Island, which I’m excluding this from this analysis as it is outside Melbourne.

But before plotting the data, it’s important to understand data quality. Since 2015 there have been 35 bicycle counting sites in Melbourne (most of them pairs counting travel in each direction). But for whatever reasons, data is not available at all sites for all days. Here is the daily number of sites reporting from January 2015 to 5 July 2020 (at least with data available as of 9 July 2020).

There are notable gaps in the data, including most of the latter part of November 2017, and around mid-2018.

So any year-on-year comparison needs to includes sites that were active in both years. For my first chart I’m going to filter for sites with complete data for 2019 (all) and 2020 (to 5 July). I’ve also filtered out a few sites with unusual data (very low counts for a period of time – possibly due to roadworks).

How do 2020 volumes compare to 2019?

Here is a chart showing average daily counts as a four week rolling average, dis-aggregated by whether the site was a bike lane (5 sites), recreational path (12 sites), or other path (9 sites) and whether the day was a regular weekday, or on a weekend/public holiday.

Weekday bike lane travel has reduced substantially, although in June the percentage reduction was less. This makes sense as most of these sites are on roads leading to the CBD, and most workers who normally work in the CBD have been working from home.

Volumes in bike lanes on weekends in 2020 have been very similar to 2019. This might reflect bike lanes not attracting additional recreational cyclists, or perhaps an increase in recreational cycling is offset by a decline in commuter cycling.

Volumes on recreational paths have been much higher than in 2019, most significantly on weekends. Again this makes sense, as people will be looking to exercise on weekends in place of other options no longer available (eg organised sports, gyms). In late June and early July, volumes did however reduce, perhaps as gyms reopened (22 June) and some sports resumed. This may change again during the second Melbourne lock down that commenced 9 July 2020.

Weekday traffic on paths not classified as recreational was down significantly in 2020. I’ll explore this more shortly.

How have volumes changed at different sites?

Here’s a look at the percentage change at each site on weekdays. I’m comparing weeks 14-19 of years 2020 and 2019 (33 sites have complete data for both periods). Weeks 14-19 mark the first four full weeks of the first full lock down.

You can see significant reductions near the CBD, and on major commuter routes (lanes and paths). The biggest reduction was 71% on Albert Street in East Melbourne.

The blue squares are mostly recreational paths where there has been massive growth, the highest being the Anniversary trail in Kew at +235%! However I should point out that these growth figures are often off very low 2019 counts. It may be that people working from home (or who have lost their jobs) are now going for recreational rides on weekdays.

You might notice one square with two numbers attached – the +27% is for the Main Yarra Trail (more recreational), and the -32% is for the Gardiners Creek rail (probably more commuter orientated at that point). The two counters are very close together so the symbols overlap.

Here is the same again, but with the changes in average daily counts:

Many of the high growth percentages were not huge increases in actual volumes. The bay-side trail experienced some of the bigger volume increases.

On weekends and public holidays, there were smaller percentage reductions near the city centre, and large increases in the suburbs:

The percentage increases on weekends are not as high because there was a higher base in 2019. The reductions in the central city are smaller, but still significant – this may reflect fewer CBD weekend workers with a downturn in retail activity.

Again, here is a map of the changes in volume on weekends:

How have volumes changed by distance from the CBD?

Here’s another way to view the data – sites by distance from the CBD:

Bike lane volumes are down significantly at most sites, particularly on weekdays. Bike path volumes are down on weekdays at most sites within 6 km of the CBD, but up at sites further out, and up at most sites on weekends.

How have volumes changed by time of day?

I’m curious about the volume changes on paths on weekdays, so I’ve drilled down to hourly figures. Here are the relative volumes per hour for the months April – June:

Bicycle volumes are down in weekday peak periods on all site types, which you might expect as a lot of peak period cycling trips are to/from the central city.

Daytime bicycle volumes are significantly higher on recreational paths, and to a lesser extent on other paths. For the warmer months of April and May there was a recreational peak around 4pm, while in June the profile was more of a single hump across the day.

So the changes in volumes on paths are certainly a mix of reduced peak traffic, and increased off-peak traffic. The middle of the day increase is perhaps people breaking up the day when working from home, or people who are no longer working.

On weekends bike lane volumes are slightly up in the middle of the day, but down in the morning and evening. There’s been a substantial increase in bike path volumes on weekends – suggesting people seeking recreational riding opportunities on the weekend are choosing the much more pleasant bike path environments.

What will happen post COVID-19?

Of course this data only tells us about what’s been happening during the first lock down and the gradual recovery thereafter (until just before the second lock down).

It will be interesting to see if there is an uptick in recreational cycling during the second lock down period, although it is in mid-winter.

Eventually there may well be a boom in cycling (particularly on bike lanes) when more people start returning to work – particularly in the central city – and look for alternatives to (what might be) crowded public transport.

I’ll try to keep an eye on the data over time and update this post periodically.


What impact has the 2020 COVID-19 pandemic had on road traffic volumes in Victoria?

Sun 3 May, 2020

[Last updated 25 July 2020, not all charts]

For the most recently analysis of road traffic volumes – see my twitter feed.

Roads in Victoria were noticeably quieter during the depth of the pandemic shutdown, but just how much did traffic reduced? Has it varied by day of the week, time of day, and/or distance from the city centre? How have volumes increased as restrictions have been eased? What has been the impact naming identifying hot spots and postcode lock downs?

To answer these questions I’ve downloaded traffic signal loop vehicle count data from data.vic.gov.au. The data includes vehicle detection loops at 3,760 signalised intersections across Victoria (87% of which are in Greater Melbourne).

I should state that it is not a perfect measure of traffic volume:

  • It may under-count motorway-based and rural travel which may cross fewer loop detectors.
  • There are occasional faults with loops, and I’m only able to filter out some of the faulty data (supplied with negative count values), so there is a little noise but I will attempt to wash that out by using median counts rather than sums or averages (although charts of averages show very similar patterns to charts of medians).
  • Some vehicles moving through an intersection might get counted at multiple loops, but I would hope this has minimal impact on overall traffic volume trends.

How have traffic volumes reduced during the pandemic?

Firstly, median 24-hour loop volumes for each day:

Note: the actual numbers aren’t very meaningful, it is the relative numbers that matter.

There are regular variances by day type (eg Fridays generally having the most traffic), so here is a chart looking changes by day of the week, relative to the first two weeks of March 2020. I’ve annotated various significant announcements and changes in rules.

At their lowest, weekday volumes went down around 40%, while weekend volumes went down more like 50%.

In late-June volumes were down only 10-20%, with significant growth on Saturdays. However volumes declined again as a second wave of infections hit, and more restrictions were reintroduced. The key turning point was Saturday 20 June when the first warnings were raised about outbreaks, increasing cases, and a slow down in easing of restrictions.

In the early part of the second lock down, volumes were similar to April, the bottom of the first lock down, but then they settled at higher levels (more on that shortly).

Some curious outliers:

  • Thursday 9 April – the day before Good Friday: there may have been some travel to holiday homes, and/or other travel that happens normally on the last workday of the week.
  • Wednesday 8 July – the day before Melbourne and Mitchell Shire re-entered stage 3 restrictions (lock down), suggesting many people brought forward travel activity that was about to no longer be allowed.
  • Saturday 16 May & Sunday 17 May: there was a surge in traffic volumes on the first weekend after restrictions where eased.

Have traffic trends been different in different parts of the state?

There have been many more COVID-19 cases in Melbourne than regional Victoria. Here’s a chart showing daily volume changes in Greater Melbourne:

There is very little difference compared to the whole of Victoria chart, as most signals are located within Greater Melbourne.

Here is a chart of only signals outside Greater Melbourne, showing much less decline in late June / early July.

A notable exception here is Sundays where there has been a decline in July – perhaps Sundays normally involve a lot of travel to/from Melbourne.

How has traffic changed during the second wave?

From late June, there were increasing warnings about outbreaks in LGAs, suburbs, specific postcodes entered lock down before all of Melbourne plus the Shire of Mitchell also went into lock down. This section looks at the impact of some of the responses to what has become a second wave.

On 25 June, 10 suburbs were announced as outbreak concerns, with door-to-door testing campaigns to be conducted. These suburbs were within 6 LGAs identified on 20 June, so this may have refined people’s concern.

It is possible to filter to signal sites in the listed hot spot suburbs, although there are only around 100 signalised such sites (and none at all fall into the small suburb of Albanvale) which makes for some noisy data. Also, I would dare say that a lot of traffic in these suburbs is through traffic rather than local traffic.

To overcome daily noise, I’ve calculated the rolling 7 day average volume – excluding public holidays with with some normalisation (see below chart explanation). That does mean that sudden daily changes in traffic are smoothed out over the following 7 days.

Boring but necessary technical notes: Many traffic signals are on roads that are LGA boundaries – and which LGA an intersection falls into is almost random – it depends on the coordinates of the intersection point. To normalise volumes, I have calculated the ratio of the average volume for each day of the week in February to the overall February average, and then adjusted daily volumes using these ratios to produce a relatively smooth daily time series. The rolling 7 day average then omits any public holidays. It’s not perfect, as you can see around Easter, but it was necessary to avoid having large gaps or blips in the above chart. For this analysis I used February as the baseline, as there was a public holiday in the first two weeks of March, complicating the normalisation.

Volumes immediately dropped more quickly in these suburbs compared to the rest of Melbourne, although they later settled at higher levels than the rest of Melbourne.

On 30 June there was an announcement that 10 postcodes would return to “lock down” (only four essential travel purposes allowed) from 2 July. Those postcodes mostly – but not entirely – lined up with the 10 warning suburbs. Here’s a similar chart that separates out those postcodes, from the rest of Melbourne (plus Mitchell Shire) that went into lock down on 9 July:

There was a step change from 2 July as the restrictions took hold (on top of a reduction from the school holidays), and the rest of Melbourne followed after 9 July.

During the first lock down, these 10 postcodes saw a slightly smaller traffic reduction compared to the rest of Melbourne, but in the second lock down other parts of Melbourne have not seen the same traffic reductions.

The 7 day averaging process hides a little of the behaviour change, so here is a daily volume chart for those 10 postcodes:

While volumes in these postcodes started declining from the first warning announcement on 20 June, if you look carefully you’ll see that on Wednesday 1 July there was little change in volume compared to the previous Wednesday. This was the last day before the lock down, and presumably some people made some extra travel that was about to become against the rules. Once the lock down had commenced, volumes were very similar to those experienced during the “stage 3” restrictions of early April. This is similar to the surge in traffic seen in Melbourne the day before the second lock down.

A more detailed look at Melbourne

The following animated map shows the change in weekday volume relative to the first two weeks of March, for each site, each week since the beginning of March. Note that there are anomalous sites for various reasons (eg faults, roadworks) – I’ve tried to filter out some sites with unusual data, but it’s difficult to get all of them.

If you ignore individual sites that look like outliers you can see some clear patterns:

  • Volumes haven’t reduced as much in industrial areas during lock downs, as freight and logistics largely keep operating, and factory workers continued to go to work.
  • Volumes didn’t recover in the central city as they have in the suburbs, which makes sense with so many office workers have continued to work from home.
  • Melbourne Airport volumes have been significantly below normal throughout, obviously due to national and international travel restrictions.
  • Volumes were slower to recover in the Clayton area – probably related to working from home, and Monash University not having on-campus teaching.
  • Volumes reduced from the week of 29 June, a mix of the school holiday impact, an increase in travel restrictions, and probably general fear about a second wave of infections.

I must apologise to the those with colour-blindness, it’s much more difficult to show the changes with only two-three colours.

This map doesn’t however explain the slightly smaller traffic reduction in Melbourne outside the initial 10 lock down postcodes.

The following map compares traffic volumes on Wednesday 22 July with those in the first two weeks of April (I’ve chosen a Wednesday to be clear of the Easter long weekend that happened in the second week of April). Note that the flip between orange and blue occurs at 110% (you might intuitively expect it to be at 100%).

This map pretty clearly shows that second lock down volumes were higher in the eastern and south-western suburbs, but much closer to April in the north-eastern suburbs. There have been fewer COVID-19 cases in the south-eastern suburbs, and this might reflect people’s self-regulation based on perceived local risk.

Indeed, here is a chart comparing active cases as at 19 July to traffic on 20 July relative to the first lock down:

Local government areas (LGAs) with higher numbers of active cases tend to have traffic levels closer to those in early April, while LGAs with fewer cases have seen higher traffic volumes in April. I might try to explore this relationship over time in future.

How does 2020 compare to 2019?

The above analysis hasn’t differentiated school days and school holidays, and any general seasonality across the year. Here is a chart comparing 2020 with 2019 for weekdays, Saturdays and Sundays (excluding public holidays):

I will emphasise that there will be week-to-week variations, particularly on weekends, due to short term factors such as weather and special events. Also, while school returned in week 16 of 2020, most students were not attending schools in person (ditto week 29).

The winter school holidays began in week 27, and traffic volumes in 2020 appeared to drop in proportion to the traffic reduction in the same week in 2019.

The following chart compares 2020 to 2019 on a daily basis (with 2019 days offset by -1 to align days of the week):

We can also look at the percentage difference between the years, but only for days that have the same day type in terms of school term or holidays, and public holidays where they fall on the equivalent day of the year. So there are some gaps in the following chart, plus some noise due to daily fluctuations:

This chart shows January to late July. There are gaps around the autumn school holidays and Easter as they didn’t perfect match days of the year perfectly.

You need to not get too excited about daily variations (the Tuesday in the second week of 2019 school holidays had unusually low volume in Melbourne for some reason, which shows up as a spike for 2020).

This chart gives a feel for variations from expected patterns. Traffic in the Melbourne was down a similar percentage in the first week of the winter school holidays compared to the previous week of school.

Melbourne traffic volumes began falling in the second week of winter school holidays with the rise in cases and commencement of some postcode lock downs, and then fell further with the Melbourne + Mitchell lock down from 9 July.

However in regional Victoria volumes were relatively higher in the winter school holidays – perhaps as Melbourne people were more likely to travel intrastate for holidays (interstate travel being heavily restricted, and travel not having been an option in the previous autumn school holidays). Regional Victoria travel volumes have been tracking around 10% below 2019 since early June.

The next chart compares each 2020 week with the same week 2019 for Melbourne LGAs plus Mitchell. However it is important to note that there was quite a bit of week to week variation in 2019, and the autumn school holidays started a couple of weeks earlier in 2020.

On this measure, weekdays bottomed out around 38% below 2019, but recovered to be ~10% down in week 24 (on weekdays and Saturdays). Weekends were down around 50%, but recovered to around 10-15% down before the second wave. However pre-pandemic volumes were around 5% higher than 2019, so you could perhaps add another 5% to the reduction figures.

How has traffic reduced by time of day?

The traffic signal data is available in 15 minute intervals, so it is possible to examine patterns in more detail.

Here’s a look at the traffic volumes by time of day for Wednesdays:

You can see a significant flattening of the traditional peaks from late March, although curiously the PM peak still commences around 3 pm, even during the school holidays. From late May there was a significant jump in peak period traffic, coinciding with the return to school of grades Prep, 1, 2, 11 and 12.

1 July was the first week of the winter school holidays and you can see substantial traffic reductions at school times, most notably in the AM peak. Meanwhile the PM commuter peak (around 5 pm) was very similar to late June.

There was a spike in traffic on 8 July – the last day before the second Melbourne full lock down.

Evening traffic was down considerably but it’s a little hard to gauge this reduction the chart. So here is a chart showing traffic volume changes relative to the first two weeks of March (with apologies to anyone with colour-blindness):

Volumes went down the most in the evenings (particularly around 9 pm) which might reflect the closure of hospitality venues, cessation of sports and reduced social activity. The AM and PM peak periods were down around 50% at the bottom, while the inter-peak period has held up the most – being only down around 30%.

Volumes recovered considerably over May and June, with volumes around 3pm back near pre-COVID levels (prior to the winter school holidays). The AM peak is interesting – at 7am, traffic on 17 June was still down around 28%, but at 8:45am is was only around 9% down – possibly reflecting the school peak, and/or a narrowing of the commuter peak (as lower congestion provides less incentive for peak spreading). As at mid-June, evening traffic was still down around 40%.

Again 8 July is an outlier – evening traffic was a lot busier, in fact traffic leading up to midnight was busier than early March, suggesting people cramming in travel activity that was about to become restricted.

I should point out that this analysis compares to a baseline of a two days in early March, and there may be some associated noise (eg weather or event impacts on particular days).

Here is the same for Fridays (excluding the Good Friday public holiday):

Late evening traffic was down even more than for Wednesdays, which probably reflects higher volumes of hospitality-related travel on Friday nights. Friday evening traffic jumped on 15 May when small social gatherings were allowed, and again on 5 June when restaurants and cafes were allowed to have dine-in patrons.

Here is Saturdays (excluding Anzac Day):

The Saturday profile shape hasn’t changed as much as weekdays, but the evenings were down most significantly.

Curiously there are several spikes in the curve in the morning – and they are the 15 minute intervals leading up to the hours of 7am, 8am, 9am, and 10am. Initially I wondered if it was a data quality issue, but I suspect it reflects a surge in travel just before work shifts and other activities that start on the hour.

For some reason traffic volumes were relatively low around 6 am on Saturday 7 March, which has resulted in other days showing less reduction.

Saturday night travel was down considerably – by over 70% by midnight at the depths of the shutdown, but jumped with restrictions easing, similar to Friday evenings. As of mid-June it was down around 25-30%.

You can also see early Saturday morning (Friday night) travel down around 60-70% at worst (discounting 11 April which was the Saturday morning following Good Friday).

Here is Sundays:

Sunday 8 March was on the Labour Day long weekend (including the Moomba festival), which probably explains the much busier traffic that Sunday night (not being a “school night”). You can more clearly see that on the following chart:

Another anomaly here is Sunday 7 June – which was another public holiday eve.

Here’s the profile by day of the week for each week since February (public holidays excluded):

This data suggests a roughly a one hour lag on Sunday mornings compared to Saturday mornings – ie travel volumes hold up an hour later on Saturday nights and ramp up an hour later on Sunday mornings. This pattern holds up for other weeks. It also shows the middle of the day on Saturdays to mostly be busier than the same time on weekdays.

Here’s another look at relative time of day traffic volumes for March through to July:

If you look closely (no, your eyes are not losing focus!) you can see:

  • Significant volume reductions after schools finished on 23 March
  • A surge in traffic on 9 April – the Thursday before Good Friday
  • Extremely quiet traffic on Good Friday (10 April)
  • Higher traffic volumes on 8 July (the day before the second lock down), particularly into the evening.
  • Generally higher traffic on the last weekday of the week, particularly in the afternoon and evening (including during the shut down period)

Have traffic impacts been different by distance from the CBD?

Here’s a chart showing year-on-year reduction in median traffic volumes at intersections by distance from the Melbourne CBD for weeks 14 and 15 (the lowest two weeks of the lock-down):

What is clear is that the central city experienced much larger traffic volume reductions than other parts of Melbourne, which makes sense as office workers stayed home, universities, cafes, restaurants and night-life closed, and (non-essential) retail activity slowed considerably.

There is some noise in the variations by distance from the CBD but I suggest not too much should be read into that as there will be various local factors at play.

The following animated chart shows median weekday volumes per week, by distance from the CBD, since the start of March 2020:

You can see the traffic decline has remained the largest in the central city. The reduction in traffic in the week of 28 June was mostly in the suburbs more than 3 km from the CBD.

Traffic signal data comes out daily, and so I will try to update this analysis at least once a week during the recovery period. There may be more frequent updates on Twitter.