Akshay Vij: Choice modeller
Vij, can you tell me where you work now and was it is that you do?
I work at the Institute for Choice which sits within the School of Commerce at the University of South Australia. I do a lot of choice modelling, developing statistical models that can be used to understand and predict human behaviour. My work’s mostly been applied to transportation contexts – understanding why people drive, how we can get them to drive less, why people engage in dangerous driving behaviours, etc.
Presently, a lot of my research is trying to understand consumer preferences for emerging mobility services and technologies, the implications of driverless cars and shared mobility services and electric vehicles, and how that’s going to disrupt existing practices within the transport sector.
Are you doing mostly research or are you also taking on teaching?
No, I’m in a full-time research role.
And do you have a notion yet of just how big this area of research is? Are you surprised on a day-to-day basis of the sheer volume of things that are coming our way?
I guess I’m not surprised. I am quite happy to be in this space at this moment in time. Transportation for the last 50 years has been very dry. All we’ve been talking about is how can we get people to drive less, and it’s really only in the last five to ten years that suddenly you have electric vehicles and shared mobility, and connected and autonomous vehicles are on the horizon. There’s not been this much disruption since literally Henry Ford invented the Model T.
Indeed. And have you found that that the horizon has vastly increased in the last two or three years?
Absolutely. There’s always been the core transportation researchers who’ve been working in the field, but now we’re also seeing a lot of cross-fertilisation because other peripheral disciplines are getting more and more interested in transportation because of these new challenges that have arisen.
There’s also a lot of folks from machine learning and the computational sciences and electrical engineering coming in. We’re also seeing more interest from theoretical economics and related fields.
When was it you began to move towards working in the transport space, and why?
All my degrees are in civil engineering. It was really during my undergrad that a couple of courses in transportation really made a strong impression on me, just because of the scope of impact. Anybody who lives in a city understands the importance of transport systems, and how they shape and influence the structure of cities, and how people feel about their cities, and really has a big impact on overall well-being. And so that scope of impact was something that always really attracted me to the field.
Where was that undergrad study, Vij?
At the Indian Institute of Technology Bombay, between 2004 and 2008.
And after Bombay?
After Bombay, I did a Master’s and a PhD in Civil Engineering. Specifically in the area of transport at the University of California, Berkeley. From 2009 to 2010, I finished my Master’s, from 2010 to 2013, I finished my PhD, I hung around for a post doc, for a year and a half, and in 2015 I moved to the University of South Australia.
Terrific, okay. And now let’s get away from you for a minute and visit hypothetical land. Someone has come to you with a very large budget, and a generous timeframe, which to do it. With such a budget and timeframe, what would you like to take on to make a big impact in transport? It could be in Australia, anywhere in the world, any project that you want to do.
I have quite a boring answer for this. I gave it a lot of thought, and I was tempted to go for something futuristic and sexy, that gets people’s attention, but I feel that sort of research is getting plenty of traction already.
A lot of strategic transport planning is done using these very large model systems that try to predict when people are travelling, where they’re going, who they’re travelling with, what mode of transport they’re using. There’s a lot of moving pieces, a lot of little sub-models, and these model systems can get very big and very expensive quite quickly. A single model system for, say, a city like Adelaide, could cost between one and three million dollars, depending on the functionality.
However, because these models are so expensive, nobody has the money to compare the performance of these more complex models against comparable simpler frameworks. They have typically been sold on the grounds of conceptual arguments – they are more behavioural, more realistic, etc. They’ve also been sold in terms of capability, so you can get certain outputs from these models that you can’t with simpler frameworks. But in terms of predictive performance, there really hasn’t been any robust benchmarking.
And yet, smaller cities such as Adelaide are being pressured by national and international bodies to adopt these very expensive model systems. And I don’t think there is a compelling commercial case. Either in terms of capability, as not every city needs to have the same detailed level of functionality that, say, the model system for Sydney or Melbourne would have. And like I said, there’s little evidence that shows that these models do any better in terms of predictive performance.
Because of those two reasons, I feel like that would be money well spent. It would, in my opinion, help ease a lot of this pressure off these smaller cities, allowing them to allocate resources to other things in a more efficient way.
In short, you’d spend the money to come up with a better model for smaller places; which will be translatable to lots of other smaller places.
That sounds great, yeah. Thank you for paraphrasing it so succinctly!
So still a big spend, but it would save the smaller cities some money. Sounds like a good idea. Given that you’ve budgetised your first big ticket item, what would be your choice with a smaller budget, in a smaller time frame, that would make a big impact?
That’s a good question. At this point I feel like it has to necessarily be in the space of emerging mobility services. One study that I would really like to undertake in that space is in the area of cost benefit analysis.
Current methods for transport investment appraisal implicitly view any potential investment as a now or never proposition – if the government does not invest in the project now, they will never be able to do so. The ability to delay an investment is especially important when the investment is irreversible, and even more so when outcomes from the investment are uncertain, which is particularly relevant now, given the many disruptions facing the transport sector.
Now Sydney’s building its light rail, and it’s supposed to open anytime this year. When the cost benefit analysis was done, when they were looking at potential ridership, they didn’t account for the arrival of shared autonomous vehicles. Shared AVs could provide door-to-door transport at a fraction of the cost of existing public transport services. There’s a very real possibility that they will cannibalise the trips that Sydney’s light rail project is expecting to attract, and the light rail system might be obsolete within the next five to ten years, well before its planned expiration.
In such cases, it may be in government’s best interest to postpone their decision, until more information about the future is available and a more considered decision can be made. But existing appraisal methods do not explicitly allow for postponement. Given the long lifespans of most transport infrastructure projects, these methods can lock cities such as Sydney into suboptimal patterns of growth that are difficult to undo at later stages.
In finance, in energy, they often use what’s referred to as real options analysis, which is the idea that you can defer some of these irreversible large investment decisions to a future date, in order to have more information about the future. So there is essentially an opportunity cost to the information that you’re foregoing if you’re making that decision today.
If we can apply these same ideas to transport investment appraisal, they could potentially allow us to undertake infrastructure planning in a more flexible and nimble way that doesn’t lock you into these very expensive investment decisions, and allows some room for flexibility.
What I would like to do with a limited budget and a limited scope study is focus on the Sydney light rail, look at how the cost benefit analysis was done in the past. Then redo the accounting to cover all the future uncertainty about shared AVs and future technologies and services and see how that changes the economic case. Accounting for that uncertainty and the opportunity cost of future information, should we still be building that light rail, should we have deferred that decision to a future date, or should we have altogether abandoned the project?
Now, let’s get back to you. Of the things you’ve done so far, what project have you been most proud of?
I was really proud of the project we did through iMOVE CRC on assessing consumer preferences for Mobility as a Service (MaaS) and on-demand transport in Australia. It was one of the first projects with industry that I had led from start to finish, and it was a great experience to have.
I feel like we did a good job at defining the scope of the study, but then also being very thorough in seeking feedback from a wide group of project stakeholders, external stakeholders, and government. Then being able to serve some really practical and tangible findings that anybody who’s interested in providing MaaS or on-demand transport service in Australia can use, in terms of what should be the service attributes that they should be focusing on, and who should they be targeting their services at.
It’s going to be very interesting to see when a real MaaS project gets up and running and the difference it makes.
Yes, it will be. I feel like the challenge is more on the supply side than the demand side. I think there is definitely a market for these services. Getting over the institutional hurdles to provide that level of integration across different services, that’s really hard.
Perhaps the educational process of it? To talk the public into using it?
Yes, but I feel that there is definitely a sizeable segment of the population that is very tech savvy. They’re already using Uber and Lyft and Ola. And that has that appetite and the knowledge of how to access these things. I absolutely agree there is an important role for public education. Particularly to ensure that access is provided to some of the more transport-disadvantaged sections … older adults, lower socioeconomic groups, people in regional and remote areas, etc.
But I don’t feel like it’s essential to the commercial viability of either MaaS or on-demand transport.
Right. Now you worked across a few different areas in your time in transport. Is there any area within the ecosystem that you haven’t yet done any work on that you would like to?
Hmm, that’s a good question. Obviously transport planning overlaps a lot with urban planning. So there’s a lot of talk about land use and transport integration and things like that. I would love to be more involved on the urban economic side of things and really understanding how transport and urban planning interact with each other. How housing and housing choices influence the provision of transport services. How the provision of transport infrastructure itself influences urban form.
That area of overlap between transport and urban planning, that would be a really interesting area for me personally to get more into.
Well I hope some of that work comes your way. Final question, in the next three to five years, what is it in transport or smart mobility technology that you’re most excited about?
I want to see the first commercial driverless car on the road in the next three to five years. I’m really curious to see how consumers actually respond to them. My sense is that there will be more or less a collective shrug, almost like what we saw in response to that first fatality from a driverless car. I think a lot of us in the transport community were dreading that moment when that would happen, and we were expecting all sorts of backlash that would follow once that first accident did happen.
I was quite surprised at how little people seemed to care. Everybody was more or less of the opinion, ‘Well, people engage or cause accidents all the time, why shouldn’t a machine?’ The reasonableness of that collective response to that first fatality was very reassuring.
I’m curious to see how people respond to those first few fleets of commercial shared autonomous vehicles on the road, and how long it takes for people to adopt those services.
Yes, I think transport wise we are living in interesting times. Thanks for your time, Vij.
In regards to your Symposium subject, how do travellers know what they want, particularly in instances where they may not yet know what advances to transport systems are on the way? How might they account for the gap between present and future options/modes/etc?
Great question that cuts to heart of the matter! There are two broad ways in which we can elicit consumer preferences for new transport modes and services, and their consequent impacts on extant patterns of travel and activity behaviour. Traditionally, these effects have been estimated using discrete choice experiments (DCEs), also known as stated preference (SP) data, where potential consumers are asked to state their choices under a wide variety of hypothetical scenarios.
The appeal of DCEs can be ascribed to their ability to measure preferences for goods and services not yet in existence, by designing scenarios that closely mimic choices that would be faced by consumers in potential real-world future markets. The trick here is to make the hypothetical situations as credible and realistic as possible, so that study participants can provide meaningful responses. Conventional DCEs might not be sufficient in cases where consumers have little knowledge or experience with the product in question. In such cases, additional audio-visual devices and multimedia technology may be used to address the gap between the decision-maker’s current experiences and the functionality of the new product. For example, can we use driving simulators to give potential consumers a hands-on experience of what it might feel like to be in a car with self-driving controls?
Alternatively, we can create real-world situations that mimic some of the essential features of these new transport technologies and services. For example, a study being led by researchers at the University of California, Berkeley is using naturalistic experiments to project consumers into a world with self-driving vehicles. The study “mimics potential life with a privately-owned self-driving vehicle by providing 60 hours of free chauffeur service for each participating household for use within a 7-day period”. Such studies offer a version of revealed preference (RP) data, based on actual observable behaviours in the real world, that offer a higher degree of credibility than analogous data from SP studies. However, such exercises are significantly more expensive to conduct than SP studies.
How do you capture the data in this exercise?
Through surveys, most frequently. In some cases, through observational studies (e.g. traffic counts, pedestrian counts, etc.). Passively collected sources of data, such as cell phone data records, Bluetooth traces, public transport smart card data, etc., are increasingly being used to offer insight on travel preferences and behaviours.
How do you measure success in this?
Ideally, through comparison with what is known about these behaviours in the real world. Do the estimated trade-offs between different product features or service attributes make sense? Do predicted adoption rates for different technologies and services match our prior expectations, or deviate from them in sensible and defensible ways?
For example, in a study on Australian consumers’ preferences for Mobility as a Service, we looked at how willingness to use such a service varied in the Greater Sydney region. As expected, residents in richer car-dependent neighbourhoods, such as Northern Beaches and Sutherland Shire, had the lowest potential demand for MaaS. Residents in inner city neighbourhoods had high potential demand for MaaS. However, and perhaps surprisingly, the greatest potential demand for MaaS was not in the inner-city neighbourhoods, but in the suburbs surrounding Sydney, such as Parramatta and South West Sydney, which have lower average household incomes and depend more heavily on the public transport network, in particular the rail network, to fulfil their mobility needs.
This is the kind of result that I always look for – one that you wouldn’t necessarily have expected (most transport experts cite inner-city neighbourhoods as the most likely early adopters of MaaS), but one that still makes sense (of course lower-income suburban neighbourhoods that are heavily dependent on the public transport network would like to have access to a service that improves their ability to use the public transport network).
How do you trust that the questions asked, and the answers received, are the ‘right’ ones?
Through lots of testing. Our survey questions are usually developed through qualitative discussions with consumer focus groups. We beta-test the survey instrument with subject experts who weren’t directly involved in the survey design. And we typically also pilot the survey with a small sample, to make sure we’re getting the responses that we need.
What do you see as the Transport of Tomorrow in the short-term? (5 years out)
First fully driverless car on the road, accelerating the adoption of shared mobility services, and continuing the decline of private car ownership. Electric vehicles start gaining traction in Australia. First drone-based urban delivery system takes off.
What do you see, if your crystal ball/wish list extends this far out, in the Transport of Tomorrow in the medium-term? (20 years out)
Private car ownership has declined substantially, to the point where most of us own at most one car, and we rely on a fleet of publicly operated shared autonomous vehicles for most of our day-to-day mobility needs. Cars running on petrol have been completely phased out, and the electrification of our transport system is complete. Humans are still allowed to drive, but legislation seeking to outlaw human-driven vehicles is in the pipeline. A system of drones has taken over urban delivery services, and the accompanying air space over most cities.