1. Are we giving BRT
passengers what they want?
User preference and market
segmentation in Johannesburg
Christo Venter
Dept of Civil Engineering
University of Pretoria
SATC July 2016
2. What do passengers (think they) want?
How does this differ across user groups?
What does this tell us about BRT design?
Grounded in actual BRT experience (RP & SP)
Mode captivity
Advanced modelling
3. 1. Data
2. Market segmentation
3. Choice model estimation
4. Implications for BRT design
4. 1. Data
- CAPI face-to-face surveys
- N= 1,208
(10,872 SP observations)
- All modes excl walking
5. 1. Data: Survey approach
5
Personal &
demographic
information
Mode access &
satisfaction
Revealed Preference
data: recent, frequent
trip
Stated preference
experiment:
Reference trip
(current mode)
vs
BRT alternative
7. 1. Data: Survey Design
Attribute Levels
Mode constant (current mode) Car, Gautrain, Taxi, Bus, BRT, Train
Number of transfers (PT only) No transfers; 1 transfer
Travel cost -20%; current; +20%
In-vehicle travel time -20%; current; +20%
Walk time to PT -50%; current; +50%
Wait time for PT -50%; current; +50%
Walk quality (BRT questionnaires
only)
Good (paved sidewalk & lighting);
Poor (no paved pavement or lighting)
Feeder mode (BRT & taxi
questionnaires only)
Walk to BRT; complementary bus to
BRT; taxi to BRT
7
10. 2. Market segmentation
10
Captives
Car captives
(car users with no PT
alternative at present)
Lifestyle
captives
(car users unwilling
to use PT under any
circumstances)
Availability
captives
(car users willing to
consider PT if
available in future –
become choosers)
PT captives
(PT users with no car
alternative at present)
All travellers
Choosers
(PT and car users with
both options at present)
NMT captives
(NMT users with no other
alternative at present)
11. 2. Market segmentation results
3 August 201611
Motorised modes
(PT & car)
2,336,029
(100%)
Car captives
644,424
(27%)
PT captives
1,183,839
(51%)
Choosers
507,765
(22%)
Lifestyle car captives
383,571
(16%)
Availability car captives
260,852
(11%)
POTENTIAL
BRT MARKET
13. 3. Mode choice model estimation
• Combined RP & SP data
– SP gives best indication of trade-offs between
attributes, e.g. time vs cost
– RP gives best information on current valuation of
mode attributes
• Reliability, comfort, safety, image, …?
• Separate coefficients for different captivity
groups
• Mixed logit used because of:
1. Correlations due to repeated observations for same
individual
2. Possibility of taste heterogeneity (random parameters)
13
14. VARIABLE UNIT
AVAIL CARCAP &
CHOOSERS PT CAPTIVES
BUS -0.9339
BRT (Reference) 0.0000
CAR +1.7641 --
GAUTR +19.7326
TAXI -0.9175
TRAIN +0.1082
COST Rands -0.1667 -0.0697
IN-VEH TIME Minutes -0.0050
WALK TIME START OF TRIP Minutes -0.0144
WAITING TIME Minutes -0.0072 -0.0195
SEAT AVAILABLE ON BRT* 1=Yes 0.0264
NO OF TRANSFERS Number -0.0914
14
3. Mode choice model estimation
Estimated (scaled) coefficients
Significant coefficients shown in bold
Scale parameter 0.3009
Log-likelihood = -4077.4
McFadden R2 = 0.66
Likelihood ratio test: Chi-squared=15742 (p-value= 0.000)
16. 4. Implications: WTP
3 August 201616
Willingness-to-pay measure
WTP: PT
captives
WTP: Choosers &
Availability car
captives
In-vehicle travel time (R/hour)
Walk time at start of trip (R/hour)
Waiting time (R/hour)
Value of each transfer
R4.30
R12.39
R16.78
R1.31
R5.98
R17.21
R8.56
R1.82
Willingness-to-pay: Trading off time for money
e.g. Value of in-vehicle travel time (IVT) =
𝛽𝐼𝑉𝑇
𝛽 𝑐𝑜𝑠𝑡
17. VARIABLE UNIT
AVAIL CARCAP &
CHOOSERS PT CAPTIVES
BUS -0.9339
BRT (Reference) 0.0000
CAR +1.7641 --
GAUTR +19.7326
TAXI -0.9175
TRAIN +0.1082
COST Rands -0.1667 -0.0697
IN-VEH TIME Minutes -0.0050
WALK TIME START OF TRIP Minutes -0.0144
WAITING TIME Minutes -0.0072 -0.0195
SEAT AVAILABLE ON BRT* 1=Yes 0.0264
NO OF TRANSFERS Number -0.0914
17
4. Implications: Mode constants
Estimated (scaled) coefficients
Significant coefficients shown in bold
Scale parameter 0.3009
Log-likelihood = -4077.4
McFadden R2 = 0.66
Likelihood ratio test: Chi-squared=15742 (p-value= 0.000)
BRTBUS (GAUTRAIN)
0.00-0.93
TAXI
-0.92 (+19.73)
CAR
+1.76
TRAIN
18. 18
4. Implications: Utility differences
Qualitative
factors,
73%
Travel cost,
26%
In-veh
time, 7%
Walk &
wait, 7%
% of actual
utility
differences
(BRT vs taxis)
explained
by…
20. 5. Conclusions
• Only about 25% of current car users are
“lost” to a good BRT
– Demand exists, but sensitive to BRT offer
• All other (potential) passengers value
short walks and waits more highly than
trunk speed
– Focus on dense supporting network rather
than many infrastructure-heavy trunks
– Keeping fares low is very important
– Transfers OK, but limit fare penalty
20
21. 5. Conclusions
• Choosers (people with a car available) are
willing to pay slightly more for better services
– Differentiated premium services?
• Qualitative factors (reliability, safety,
convenience?) are very important choice
drivers
– BRT competes more effectively on these than on
cost/time
– Need to understand better
– Protect and improve operating practices
• Conventional SP overstates value of time
21
24. 4. Implications: Utility differences
Average value of service
variable Average size of term (β.x)
TAXI BRT TAXI BRT difference
% of
difference
Alternative
specific constant
(ASC)
-- -- -0.918 0.000 0.918 73%
Travel cost
(Rands) R 18.72 R 13.93 -1.304 -0.970 0.334 26%
In-vehicle travel
time (minutes) 46.0 min 28.8 min -0.230 -0.144 0.086 7%
Number of
transfers 0.34 0.23 -0.031 -0.021 0.010 1%
Walk time at
start of trip
(minutes) 7.5 min 10.3 min -0.109 -0.148 0.039 3%
Waiting time
(minutes) 9.1 min 11.5 min -0.178 -0.224 0.046 4%
3 August 201624
Editor's Notes
Point of departure: We’ve been implementing IPTN systems now for 8 years already. Fact that we have running systems is an opportunity to take stock, ask what have we learnt, what have we done right and what can we do better?
Experience indicates that we can do indeed do things better. By and large, IPTNs have been complex and slow to implement, come at a higher cost than initially hoped, and have delivered lower ridership and benefits than initially thought. Why? One of possibilities is that we understand passengers less well than we think, and that our models tend to make us believe.
CoJ has been engaged in updating strategic PT network over last 3 years – took opportunity to ask these questions about user preference, in process of developing advanced mode choice model for use in planning the next phases of BRT.
We’ve been doing user preference studies for most of our recent pubic transport planning, but this is different in terms of:
Grounded in experience of actual BRT users (using revealed and stated preference)
Focus on understanding mode captivity, people’s ability and willingness to consider new modes
Used more advanced data collection and econometric techniques to help with some issues.
Sampled according to stratified random – geographical clustering.
About a fifth of sample was current BRT users, while many others in areas where BRT is currently operating.
(Deliberately oversampled in higher income areas as we were interested in willingness of that market to switch to BRT. Corrected during model estimation process).
BRT alternative automatically constructed as variations around reference trip – so-called pivot design. Thus preserved actual choice situation, rather than purely hypothetical choice which often bedevils SP results.
Explain BRT option, but helped that most people were already familiar with characteristics. Further contributed to realism of choice.
Tested these attributes. Not all in every game – 3 or 4 max. 9 replication per person.
Necessitated use of tablet computers, programmed to generate options on the fly. For instance, this example, taxi user that told us in first part of the interview, for most recent trip to work SAY, …
Made a priori decision to segment the market according to degree of choice that people have. Based on hypothesis that people behave differently when choosing, depending on what they are used to, particularly how much choice they have had in the past.
This leads us to differentiate between captive and choice travellers – captives have only limited set of options available.
Car captives (…), PT captives, walk captives
But this leads to problem of how enduring these categories are – particularly, some people who are currently car captive might become un-captive in future – say new BRT line comes close enough to their house that they now have that option. Would they continue to use only car? So we subdivided captives into two groups, namely lifestyle car captives, and availability car captives.
Survey allowed us to identify people in each group, based on current mode, what alternatives they had available for that trip (if any), and whether or not they were willing to switch to BRT option at all during SP game. If not, and they were current car captives, then defined as lifestyle captives.
We estimated models to predict people into each of these categories and allow us to expand it up to COJ using 2014 household survey data. Results:
Of the 2.3 million daily motorised trips in the COJ, about half are captive to public transport – made without the option of a car. Just more than a quarter are classified as car captive trips – so 1 in 4 motorised trips are currently made without the (perceived) alternative of public transport available. The remaining 22% of trips are classified as chooser trips. It follows that, of trips where the car is available (chooser plus car captive trips), just more than half feel they have no alternative but to drive.
However present car captives are not all persistently opposed to using public transport. About 4 out of every 10 car captives are classified as availability captive – these quarter of a million car users would be willing to use good public transport (BRT) options, should such options become available to them. Of course significant differences spatially – some areas captives are much more concentrated.
All in all the news is actually good: 84% of trips are potential market for BRT, PROVIDED service that is offered is sufficiently attractive. What is sufficiently attractive? That is what the next step of the analysis tried to find out: SP
SP: How do people trade-off travel time with cost, for instance, because experiment was designed that way.
RP: How do people value all those other things that are hard to specify, qualitative aspects like reliability, comfort, safety, image. Because people vote with their feet – used actual choice made previous trip to estimate Alternative Specific Constants (with some scaling)
Separate coefficients for different captivity segments – to allow for differences in way people trade off across different groups.
Mixed logit: more advanced model specification that addresses some of classical problems of estimating SP data, including correlations in error terms across individuals, and possibility that tastes can vary across sample, allowed for by estimating some coefficients not as fixed parameters but as random parameters. In this case as normal variates.
Model is highly significant
Significant coefficients estimated for all service variables, but not for all modes
Some coefficients differ between captivity groups and others do not. For instance, choice users have (more negative) cost coefficient, meaning higher willingness to pay for improvements.
Not easy to interpret, so let’s look at specific implications in terms of what people want and what we offer them.
Willingness to pay = how people trade off deterioration in one variable for improvement in another. Estimated from ratio of coefficients, e.g. VOT.
Few things to note:
Most important = VOT estimates are quite low, much lower than most previous studies in Gauteng. In particular, people value savings in travel time at between R4 and R6 – this is on average for all users. In line with values for LOW INCOME used in Gauteng Integrated Master Plan model, for instance, … OK, this excludes people who are lifestyle captives who might have higher VOT, but point is that on average those who could use BRT, are not willing or (more likely, able), to pay high fares in exchange for fast buses.
Secondly, the VOT is not a fixed value but varies across the population. In this case, people with car alternatives (choosers) are slightly more willing to pay for speed on average. Also remember that the VOT really varies according to the normal distribution. So there are niches of users with high values of time, so it might make sense to provide faster services at a premium price – express services – in some areas with lots of these choosers. We might be able to attract more people AND improve revenue if we differentiate our services better.
Thirdly, people value short walk times and wait times much higher than short travel times. This is in accordance with the literature. But have we taken it to heart? A heavily corridor-oriented BRT strategy relies on higher frequencies on the trunk, but longer walk distances and lower frequencies everywhere else. People away from the trunk want to be close to a high-frequency service, more than a fast service. Guess what, this is exactly what the taxi are giving them. Should we not focus more on the network rather than the trunk?
Lastly, people don’t mind transferring that much – one transfer is values equally to only about 10% of the average fare. So the average passenger is willing to transfer more (if convenient), in exchange for better coverage and higher frequency in an integrated system.
Mode constants tell us how people value all other aspects of the service, based on their actual use of the services, and already controlling for differences in travel time , cost etc.
BRT significantly outperforms minibus-taxi services in the passenger’s mind, among passengers who actively made a choice between taxi and BRT for their actual trip. This means that BRT does not simply compete with taxis on price, frequency and travel time, but that passengers take into account other qualitative advantages of the BRT. Rea Vaya really is perceived as better.
To what extent do the qualitative versus the service variables contribute to passengers’ current choices between taxi & BRT?
Are we right in building BRT for speed (with segregated bus lanes and enclosed stations), when people are not willing to pay for such speed? People away from the trunk want to be close to a high-frequency service, more than a fast service. Should we not focus more on the dense supporting network, especially the feeders, or have fewer or shorter trunk lines (which also cost us a lot), and rather make partnerships with the taxis, with Uber, with whoever, to provide that feeding function?
Are we right in building BRT for speed (with segregated bus lanes and enclosed stations), when people are not willing to pay for such speed? People away from the trunk want to be close to a high-frequency service, more than a fast service. Should we not focus more on the dense supporting network, especially the feeders, or have fewer or shorter trunk lines (which also cost us a lot), and rather make partnerships with the taxis, with Uber, with whoever, to provide that feeding function?