This document summarizes a research paper that evaluates the efficiency of public transportation systems in larger cities using data envelopment analysis (DEA). Four DEA models are used to measure efficiency based on services used and space used. The analysis finds that medium-sized European cities like Bern, Munich, Prague and Zurich demonstrate high efficiency. Mega-cities perform poorly in models measuring efficiency of services used. Lowest performance is seen in some Spanish cities, Athens, and cities in the Middle East and North America. Regression analysis also found higher DEA efficiency results in lower private car usage in large cities. The research is limited by only using 2001 data and being limited to the DEA method.
1. The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1463-5771.htm
Efficiency
Benchmarking efficiency of public
of public passenger transport transport
in larger cities
23
Olli-Pekka Hilmola
Kouvola Research Unit, Lappeenranta University of Technology,
Kouvola, Finland
Abstract
Purpose – The purpose of this paper is to evaluate public transportation efficiency in larger cities.
Global agreements to decrease environmental emissions in the future (CO2), world-wide decreasing
reserves of oil, and growing population in larger cities is the main motivation to develop efficiency
benchmarking measurement models for public transportation systems, and gives reason for this
research work. Also, from the point of view of the city, data envelopment analysis (DEA) based
efficiency measurement has not been researched earlier, which is another motivation for this study
from the method development perspective.
Design/methodology/approach – Four different DEA-based efficiency benchmarking models are
used to evaluate public transportation efficiency in larger cities. Data are from year 2001, and amount
of analyzed cities in smaller DEA model is 52 and in larger 43. This gives statistical significance and
efficiency measurement confidence over the results.
Findings – Medium-sized, old and central European cities such as Bern, Munich, Prague and Zurich ¨
show frontier performance in all four models. Mega-cities fail to reach frontier and/or good performance
in small “services used” DEA model. However, some other medium-sized cities show contrarian
behaviour for “space used” DEA model. Lowest performance is more divergent in the analyses, but is
found from Spanish cities, Athens, Middle East and North America. The author also found support from
regression analysis that higher DEA efficiency results in lower share of private car use in large cities.
Research limitations/implications – This research work uses only year 2001 data, and should be
repeated in the future as public transportation data18base is being updated. The research is also
limited on the use of DEA method, and other efficiency measurement methods should be used to verify
the results further.
Originality/value – According to the author’s knowledge, this research work is seminal from the
city-level DEA efficiency benchmarking studies concerning public passenger transportation systems.
Earlier research works have concerned actors (e.g. bus companies or rail operators), but the overall
picture from the city level has not been researched before.
Keywords Cities, Passenger transport, Benchmarking, Europe, Middle East, North America
Paper type Research paper
1. Introduction
Typically public passenger transport is significantly dependent on the amount of potential
users in its sphere of influence (Lao and Liu, 2009; Karathodorou et al., 2010; Karttunen
et al., 2010), and therefore it is not surprise that mega-cities (Jain et al., 2008) or larger
entities (Odek, 2008) have been analyzed to be most efficient in previous benchmarking Benchmarking: An International
studies. Although, privatization and deregulation processes are catching up in global scale Journal
Vol. 18 No. 1, 2011
in transportation industry overall, e.g. rail-based passenger transport is still having rather pp. 23-41
minor share of private companies operating (Approx. 13 per cent from produced volume q Emerald Group Publishing Limited
1463-5771
based on Amos and Thompson (2007)). Thus, research findings are giving their support DOI 10.1108/14635771111109805
2. BIJ for more deregulated and privatized transportation systems (Jain et al., 2008; Cowie, 1999),
18,1 particularly in bus industry (Cowie and Asenova, 1999; Odek, 2008). However, making
public passenger transportation system profitable one is extremely difficult task, and
based on our knowledge Guangshen Railway Company in Hong Kong is among the few to
have achieved this (also listed in stock exchange). For example, passenger operator
Amtrak in the USA has produced massive annual losses for decades (Amtrak, 2008;
24 Rhoades et al., 2006) and does not have any end in horizon with this regard. Similarly,
George and Rangaraj (2008) concluded from Indian railways that passenger transport
hurts regional efficiency considerably, and those regions transporting more freight were
having higher performance. Veolia’s (2009) transportation segment (including passenger
and freight transport both in road and rail; one of the largest private operators in the world)
has produced Approx. 2.5 per cent operating income in year 2008, not too flourishing result
as thinking about investments needed for operations.
It should be noted that public transportation is only increasing its importance, due to
continuing urbanization and for the need to connect suburbs and regions into centers
(Qin, 2008). Among this, increasing environmental pressure from road transports
(CO2 emissions), road transports’ very significant dependency on oil (especially private
cars; Sandalow, 2008), and estimated decline in oil availability in the world scale
(Maggio and Cacciola, 2009) are all increasing the reasons to investigate the efficiency of
public short-distance passenger transportation systems. This is particularly concern in
larger cities, and population concentration centers of prospering emerging economies
(Kenworthy, 2002; Hu et al., 2009); frightening scenario is that these emerging cities will
adapt to use within large-scale private cars by following the examples of West
(Cameron et al., 2003, 2004). It should be highlighted that transportation in general has
nothing but increased its CO2 emissions within previous two decades time period
(generally in other sectors contrarian development has been reached) – for example,
even in EU (2010), which has showed proactive role in emission prevention, have
recorded 30 per cent increases from year 1990 levels. In general, increased emissions are
caused by road transportation and aviation.
Public transportation in cities has been subject of some number of earlier data
envelopment analysis (DEA) based efficiency studies (Cowie and Asenova, 1999;
Jain et al., 2008; Odek, 2008; Lao and Liu, 2009), but these have been concentrating on
service production through actors in some selected cities or inside of one country.
However, big picture from the perspective of a city has not been considered, and this is the
main motivation behind this research work. Short-distance public transport is typically
only concentrated on passenger transport, and problematic incorporation of joint inputs
with freight segment is therefore avoided (George and Rangaraj, 2008; Yu and Lin, 2008).
Among passenger transportation service production efficiency, we are interested from
used amount of land, which is also scarce resource in ever enlarging cities. We could
assume a priori that efficiency benchmarking based on DEA method will reveal great
differences between cities, since analyzed area involves significant public influence, but
also private sector presence. In previous studies, e.g. gas distribution in Ukraine
(Goncharuk, 2008), beer production in Ukraine (Goncharuk, 2008) and railway sector in
larger Europe and entire world (Hilmola, 2007, 2009; Yu, 2008) has illustrated such results
in efficiency benchmarking studies. In comparison, intensively competed service sector
consisting only private companies has showed much lower differences between analyzed
actors (Keh and Chu, 2003; Joo et al., 2007; Debnath and Shankar, 2008).
3. This manuscript is structured as follows: in Section 2, literature research is completed Efficiency
from public transportation issues in larger cities. Thereafter, in Section 3 used research of public
methodology, data and public passenger transport measurement models from city
perspective are being presented. Section 4 shows the results of four used efficiency transport
measurement models in larger cities. Results of these models are altogether discussed in
Section 5. Finally, Section 6 concludes our work, and proposes further areas for research.
25
2. Literature review – urban transport
Major concern of larger cities, but also countries within passenger transport, is the
increasing popularity of private car-based road transports. For example, if China does
not bother to do nothing with this respect, then in year 2030 it will have 400 million
passenger cars on roads, hungry for gasoline (Hu et al., 2009). In city level situation is
showing similar frightening growth potential, based on Kenworthy (2002) in the USA
per capita consumption of energy for private cars is 60,000 MJ (in cities), while in China it
is only 2,500 MJ. Only strict policies and regulations have been able to constrain this
development; in Hong Kong and Singapore amount of private cars is five to ten times
lower compared to cities in Europe and the USA (Cameron et al., 2004). But these only due
to very unfavourable cost implications of owning and driving private car. Among
policies and regulations, careful planning of urban areas and closeness of people living
besides each other as well as services, decreases throughout the world traveling by
private car (Cameron et al., 2003; Karathodorou et al., 2010). However, even using UITP
(2005) database, and analyzing private cars per 1,000 persons in larger cities gives clear
causal message (see Figure 1, including also Singapore and Hong Kong): higher the
economic prosperity, typically higher is the amount of private cars. de Jong and van de Riet
(2008) confirm this with extensive literature survey and statistical analysis; only hope
900
800
y = 0.0053x + 310.62
Number of private cars per 1,000 persons
R2 = 0.1498
700
600
500
400
300
200
100
0 Figure 1.
0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000 45,000 50,000 Economic prosperity
GDP per Capita (City) drives car ownership
Note: n = 52
in larger cities
Source: UITP (2005)
4. BIJ lies in increasing amounts of older population, and their inability to use private cars,
18,1 but this potential positive direction could easily be substituted by smaller sized
households.
Although, being against of private car usage, research has not given that many
answers, how and by what manner city-level short-distance public transportation
systems should be built. Such issues as number of stop points, round-trip time, routes
26 and operating hours need to be carefully planned to have needed utilization for public
transportation system (Lao and Liu, 2009; Karttunen et al., 2010). However, rail-based
systems need considerable investments before they could operate (comprehensive
up-to-date analysis, see Flyvbjerg et al. (2008), especially in situations where they are
operating mostly underground (e.g. Jubilee line extension in London being built during
period of 2002-2007 has total length of 22 kilometer, and total cost of GBP 3,600 million,
resulting in per kilometer cost of GBP 163.6 million). Bus-based public transportation
system is much cheaper to construct, since typically infrastructure has already been
paid by road investments (and in most of the cases by private cars), and, e.g. stations are
much easier and cheaper to add in the system.
Earlier research has shown (Lao and Liu, 2009; Karttunen et al., 2010) that inside city
there exist some routes, which are extremely popular, and even could be profitable to
operate. However, this is not complete solution and package, since numerous other routes
need to be operated too in order to satisfy transportation needs, and their popularity and
profitability could be questionable. Efficiency is not the ultimate answer here, in other
words efficiency of actor level (or route level, like George and Rangaraj (2008) illustrated);
most interesting is how efficient public passenger transportation systems are from
system’s perspective (city level). This efficiency concerns both objectives: public
passenger transportation system services used, and space needed to build this system to
operate. These objectives could have tradeoffs as well, where favouring efficiency of
another one will have decreasing efficiency in other respect.
3. Research methodology and used data
International Association of Public Transport (UITP, 2005) maintains database
concerning public transportation in larger cities. Latest version from year 2001 of this
database contains 52 cities around the world. However, most of these are from Europe
(47 of total). This database has been used in earlier scientific studies (Kenworthy, 2002;
Cameron et al., 2004; Karathodorou et al., 2010; Albalate and Bel, 2010) and could be
assumed to have needed reliability of indicators gathered. From most recent studies
Karathodorou et al. (2010) and Albalate and Bel (2010) both used UITP database as
main source within their large-scale statistical analysis, not only describing the large
cities public transport environment as has been the case in the earlier studies
(Kenworthy, 2002; Cameron et al., 2004).
In Figures 2 and 3 two DEA efficiency measurement models of this research work
are shown; one concerning space utilization efficiency and second one service usage
efficiency. We altered these models in a manner that amounts of inputs were either two
(population and urban population density) or all five. In all of the following efficiency
measurement models, outputs are the same in each case (space and service used per se),
four of these outputs are rail related and one concerns busses. Of course, walking and
biking should be part of public passenger transportation system, but we were forced to
leave them out of measurement due to lack of data.
5. As in larger efficiency measurement models five inputs and five outputs were being Efficiency
used, some cities were not included in the efficiency analysis. Mostly, reason for of public
excluding was the lack of data concerning inputs of “urban population plus job density”
and “proportion of jobs in the central business district”. Therefore, our analysis in the transport
larger models consist 43 cities. However, in smaller models all of the database cities were
involved.
In the following analysis, we have only analyzed constant return on scale (CCR) 27
performance of actors using DEA efficiency analysis method. CCR assumes that scale
economics is linear and constant and does not take into account that larger decision-making
units (in this case cities) have much better probability for higher performance than smaller
ones. However, authors cannot recognize this as a caveat, since outputs in all four models
are scaled with area or population, and therefore probability for scale economics
concerning larger decision making units should not be that present (this is actually
supported by following analysis). However, variable return on scale, Banker, Charnes and
Cooper (BCC), is other alternative to take into account scale economics, but it is not applied
in the following. It should be reminded that CCR is original method for DEA efficiency
analysis (Charnes et al., 1978), and BCC development followed as derivative later on
(Banker et al., 1984). Thus, seminal research manuscript from DEA discussed about
scale efficiency function already five decades ago (Farrell, 1957).
Idea in developed two DEA models is following: identified inputs (two/five) of UITP
(2005) databases drive the public passenger transportation needs. These needs are
fulfilled by transportation system, which could be based on road transportation (e.g. buses
and private cars), mechanical transportation modes (like bikes), rail or by walking. In our
study, we are interested from bus and rail (tram, light rail, metro or sub-urban rail) based
systems, and their ability to fulfil public transportation need. Introduced two different
measurement models are developed with an idea that they could be used in any sized
city – inputs are just enablers of public transportation need, and these are converted
Inputs Outputs
Population Bus kilometers per urban hectare
Urban population density Tramway vehicle kilometers per urban hectare Figure 2.
Public DEA measurement model
Urban population + Light rail vehicle kilometers per urban hectare
job density
transportation of public passenger
process Metro vehicle kilometers per urban hectare transport concerning
Proportion of jobs in the
central business district Suburban railway vehicle kilometers per urban hectare efficiency of space being
used
GDP per inhabitant
Inputs
Population Outputs
Bus kilometers per inhabitant Figure 3.
Urban population density
DEA measurement model
Tramway vehicle kilometers per inhabitant
Urban population + Public of public passenger
job density transportation Light rail vehicle kilometers per inhabitant transport concerning
Proportion of jobs in the process Metro vehicle kilometers per inhabitant
efficiency of
central business district transportation services
Suburban railway vehicle kilometers per inhabitant used per inhabitant
GDP per inhabitant
6. BIJ into transactions with the transportation system. So, basically our model treats smaller
18,1 and more spread around cities similarly to those of having well concentrated and high
amount of urban population. Results in the following mostly concern, how efficient public
transportation system is in the space use and in service use. These both are scaled with
area or population. So, therefore results are not necessarily supporting earlier studies,
which give more significance on volume (absolute, like number of passengers and/or
28 passenger-kilometers).
4. Benchmarking efficiency of public passenger transports in large cities
Small DEA efficiency measurement model
As analyzing results of Tables I and II in overall, it could be concluded that between cities
in both models, there exist quite significant variation, and some cities seem to outperform
whole observation group. This outperforming group of efficiency frontier performance in
¨
both of the models is following: Bern, Munich, Prague and Zurich. Among these four,
Stockholm is nearly frontier in both of the models. All of these cities are medium-sized
among larger cities database, and they are located in central Europe (if Stockholm is not
taken into account). Also, common in all five identified high-performing cities is the
dominance of rail-based public transportation (as all railway transportation modes are
summed together) over busses. This is particularly the case with Bern, Munich and Zurich.¨
Lowest performing cities are not that homogenous group in Tables I and II, but three
Spanish cities (Barcelona, Sevilla and Valencia) as well as Athens could be identified
presenting such group. However, there does not exist any “easy explanation” behind
this low efficiency. It is true that Athens and Sevilla rely on public transport in busses,
but Barcelona and Valencia are having quite significant proportion for both
transportation modes (bus and rail).
Interesting further insight in the analysis of smaller DEA results is the high performance
of mega-cities in space utilization, but very low performance in the services used. Cities
behaving in such manner are following: Hong Kong, Moscow, London, Singapore,
Sao Paulo, Berlin and Madrid. Contrarian behaviour also exists, as like Melbourne, Dublin
and Newcastle illustrate. Only explaining factor in these is that in very large-scale cities
transportation use per inhabitant is just smaller. However, counter argumentation could
also be stated, e.g. Melbourne is behaving oppositely although being mega-city.
Large DEA efficiency measurement model
As further developed DEA model is used, differences between evaluated cities are
becoming smaller (Tables III and IV). In larger models 12 cities make the frontier
simultaneously in both of the models, and London as well as Copenhagen are rather near
of this group too. Basically, mega-cities start to show much better performance as
additional three input items are included in the models. In mega-city performance
appraisal cases, reason is simply that these actors do show much lower performance in
jobs located in central business district (as these cities have several job concentration
areas) as well as in some cases in gross domestic product per capita.
Lowest performance in both of the larger DEA models has now more cities included,
but all lower performing actors, which were included in both smaller and larger
analyses, are still in the worst performance group (Athens, Barcelona and Sevilla).
However, in larger DEA model lowest performance is present in Dubai, but efficiency
of Chicago is not good either. These two mentioned cities were not performing showing
7. City Space used (%)
Efficiency
of public
1 Bern 100.0
2 Gent 100.0 transport
3 Hong Kong 100.0
4 Krakow 100.0
5 London 100.0
6 Moscow 100.0 29
7 Munich 100.0
8 Prague 100.0
9 Vienna 100.0
10 ¨
Zurich 100.0
11 Tallinn 99.1
12 Stockholm 98.4
13 Warsaw 94.8
14 Brussels 94.7
15 Berlin 93.3
16 Singapore 84.2
17 Copenhagen 83.8
18 Amsterdam 79.3
19 Graz 76.5
20 Budapest 75.8
21 Helsinki 75.3
22 Stuttgart 68.1
23 Bologna 67.8
24 Glasgow 67.3
25 Nantes 64.9
26 Clermont-Ferrand 60.5
27 Oslo 59.3
28 Paris 59.2
29 Sao Paulo 58.9
30 Hamburg 57.5
31 Marseilles 57.3
32 Geneva 56.7
33 Milan 53.6
34 Madrid 53.3
35 Newcastle 52.5
36 Manchester 46.9
37 Lille 46.4
38 Rome 43.5
39 Dublin 40.4
40 Lisbon 40.0
41 Bilbao 36.7
42 Barcelona 36.2
43 Tunis 36.0
44 Lyons 35.4
45 Melbourne 34.7
46 Rotterdam 33.1
47 Turin 30.6
48 Athens 25.2
49 Valencia 24.9
50 Chicago 24.5 Table I.
51 Dubai 18.3 Efficiency of analyzed
52 Se villa 17.3 cities with respect of
space used by public
Notes: Small DEA; n ¼ 52 transportation system
8. BIJ City Services used (%)
18,1
1 Bern 100.0
2 Gent 100.0
3 Munich 100.0
4 Prague 100.0
5 ¨
Zurich 100.0
30 6 Tallinn 100.0
7 Stockholm 100.0
8 Copenhagen 100.0
9 Graz 100.0
10 Helsinki 100.0
11 Melbourne 100.0
12 Krakow 97.5
13 Glasgow 89.4
14 Newcastle 83.5
15 Warsaw 81.6
16 Dublin 81.3
17 Stuttgart 81.1
18 Oslo 80.9
19 Nantes 77.4
20 Vienna 75.0
21 Budapest 67.1
22 Amsterdam 66.7
23 Brussels 62.5
24 Clermont-Ferrand 60.8
25 Berlin 56.9
26 Bologna 56.5
27 Marseilles 50.3
28 Manchester 49.6
29 Geneva 49.1
30 Hamburg 47.8
31 Lisbon 46.8
32 London 45.5
33 Lille 43.7
34 Paris 43.6
35 Lyons 41.8
36 Singapore 38.9
37 Bilbao 37.7
38 Rotterdam 37.5
39 Dubai 36.2
40 Turin 35.8
41 Chicago 35.8
42 Milan 30.4
43 Rome 29.1
44 Sao Paulo 26.7
45 Hong Kong 24.8
46 Madrid 22.6
47 Valencia 22.5
48 Sevilla 22.2
Table II. 49 Moscow 19.2
Efficiency of analyzed 50 Tunis 17.6
cities with respect of used 51 Athens 15.8
transportation services of 52 Barcelona 13.8
public transportation
system Notes: Small DEA; n ¼ 52
9. Efficiency
City Space used (%)
of public
1 Bern 100.0 transport
2 Brussels 100.0
3 Budapest 100.0
4 Graz 100.0
5 Hong Kong 100.0 31
6 London 100.0
7 Moscow 100.0
8 Munich 100.0
9 Prague 100.0
10 Stockholm 100.0
11 Stuttgart 100.0
12 Vienna 100.0
13 Warsaw 100.0
14 ¨
Zurich 100.0
15 Copenhagen 90.2
16 Singapore 84.2
17 Bologna 81.9
18 Clermont-Ferrand 81.3
19 Amsterdam 79.8
20 Helsinki 76.4
21 Sao Paulo 76.0
22 Nantes 75.5
23 Glasgow 72.1
24 Geneva 71.2
25 Paris 62.5
26 Lille 62.2
27 Oslo 59.3
28 Marseilles 57.3
29 Madrid 56.8
30 Newcastle 54.6
31 Manchester 50.0
32 Rome 46.9
33 Lisbon 43.6
34 Bilbao 41.2
35 Melbourne 39.0
36 Barcelona 38.3
37 Lyons 37.4
38 Rotterdam 36.2
39 Turin 34.4
40 Athens 27.3
41 Chicago 25.2 Table III.
42 Sevilla 19.0 Efficiency of analyzed
43 Dubai 18.3 cities with respect of
space used by public
Notes: Large DEA; n ¼ 43 transportation system
high performance in smaller DEA, but in enlarged models their performance was clearly
belonging into lowest performing cities of this evaluated group. However, it should be
noted that in current analysis “services used” model shows much better efficiency
in overall; it is clearly on the higher level on the average as compared to “space used”
model, and the lowest performing cities have also significantly better starting level as in
10. BIJ
City Services used (%)
18,1
1 Bern 100.0
2 Budapest 100.0
3 Graz 100.0
4 Hong Kong 100.0
32 5 Moscow 100.0
6 Munich 100.0
7 Prague 100.0
8 Stockholm 100.0
9 Stuttgart 100.0
10 Vienna 100.0
11 Warsaw 100.0
12 ¨
Zurich 100.0
13 Copenhagen 100.0
14 Helsinki 100.0
15 Sao Paulo 100.0
16 Glasgow 100.0
17 Newcastle 100.0
18 Melbourne 100.0
19 Manchester 99.1
20 London 94.8
21 Nantes 93.8
22 Singapore 86.7
23 Clermont-Ferrand 84.1
24 Bologna 81.2
25 Oslo 81.1
26 Amsterdam 75.2
27 Lisbon 73.1
28 Brussels 71.9
29 Lille 69.5
30 Paris 67.6
31 Geneva 67.0
32 Madrid 65.1
33 Bilbao 56.0
34 Marseilles 54.7
35 Turin 51.6
36 Rome 47.9
37 Sevilla 46.1
38 Barcelona 45.6
39 Lyons 45.0
Table IV. 40 Athens 44.0
Efficiency of analyzed 41 Rotterdam 40.4
cities with respect of used 42 Chicago 39.5
transportation services of 43 Dubai 39.3
public transportation
system Notes: Large DEA; n ¼ 43
the “space used” model. So, based on the analysis it could be argued that public
transportation has regarding to service use some more brighter future, but land use is
still challenge in numerous cities.
As number of cities was able to upgrade their performance, the contrasting results
between two models were getting less frequent. However, some still remain to show
11. such behaviour, like Newcastle, Manchester and Melbourne. Also quite number of Efficiency
other cities shows similar, much higher performance in services used model, but do of public
have problems regarding to the use of space in public transportation system. Opposite
situation occurs only in significant terms in Brussels, where space utilization efficiency transport
is high, and service utilization 28.1 percentage points lower.
5. Discussion 33
Used four efficiency measurement models yielded similar findings, and this was verified
by correlation analysis too (Table V and Figure 3). Especially, strong connection was
found between two “space used” DEA models, positive correlation co-efficient was above
0.958. Difference between small and large models is present in correlation analysis; strong
positive correlation exist between “space used” and “service used” larger models (0.806),
but interestingly between smaller counterparts this co-efficient was much lower (0.532).
It should be emphasized that all of the used models were having relationships
between each other, and these were analyzed to be statistically significant too. However,
as it is evident from Figure 4, in some cases higher efficiency does not lead to higher
performance in other measure. This is caused by different variance in error term as we
move on higher values of efficiency – this in turn decreases the predictability of
behaviour in all situations. For cities, e.g. this means that any efficiency measurement
model is good enough to follow with low-efficiency standard, but as performance
improves further, increasing number of efficiency measurement models is needed to
measure performance and implications of decisions correctly.
As Table VI illustrates, only four cities from central Europe (all of these belong to
medium-sized group of large cities) reached the frontier in all four efficiency
¨
measurement models, and these are namely Bern, Munich, Prague and Zurich. Lack of
mega-cities in the frontiers of all four used models was earlier explained with the amount
of public transportation used per inhabitant. Similar central European favouring
findings were reported in the most recent public transportation supply/demand study
(Albalate and Bel, 2010), which used regression analysis with same UITP (2005)
database, and illustrated that supply of public transport mostly occurs in the center
Correlations
UH_large I_large UH_small I small
UH_large Pearson correlation 1 0.806 0.958 * 0.547 *
Sig. (two-tailed) 0.000 0.000 0.000
N 43 43 43 43
I_large Pearson correlation 0.806 * 1 0.768 * 0.681 *
Sig. (two-tailed) 0.000 0.000 0.000
N 43 43 43 43
UH_small Pearson correlation 0.958 * 0.768 * 1 0.532 *
Sig. (two-tailed) 0.000 0.000 0.000
N 43 43 43 43
I_small pearson correlation 0.547 * 0.581 * 0.532 * 1 Table V.
Sig. (two-tailed) 0.000 0.000 0.000 Correlations are positive
N 43 43 43 43 and significant between
all four public
Notes: *Correlation is significant at the 0.01 level (two-tailed); denotation, UH ¼ urban hectare and transportation DEA
I ¼ inhabitant per vehicle kilometre models
12. BIJ
18,1
UH_large
34
I_large
UH_small
I_small
Figure 4.
Correlations in matrix
figure between four
different public
transportation DEA
efficiency measurement UH_large I_large UH_small I_small
models
Notes: Denotation, UH = urban hectare and I = inhabitant per vehicle kilometre
Space used Services used Space used Services used
City (large) (large) (small) (small)
1 Bern 1.000 1.000 1.000 1.000
2 Munich 1.000 1.000 1.000 1.000
3 Prague 1.000 1.000 1.000 1.000
4 ¨
Zurich 1.000 1.000 1.000 1.000
5 Vienna 1.000 1.000 1.000 0.750
6 Hong Kong 1.000 1.000 1.000 0.248
7 Moscow 1.000 1.000 1.000 0.192
8 London 1.000 0.948 1.000 0.455
9 Stockholm 1.000 1.000 0.984 1.000
10 Warsaw 1.000 1.000 0.948 0.816
11 Brussels 1.000 0.719 0.947 0.625
12 Graz 1.000 1.000 0.765 1.000
Table VI. 13 Budapest 1.000 1.000 0.758 0.671
The most efficient 14 Stuttgart 1.000 1.000 0.681 0.811
20 public transportation 15 Copenhagen 0.902 1.000 0.838 1.000
cities sorted first with 16 Singapore 0.842 0.867 0.842 0.389
space used models 17 Bologna 0.819 0.812 0.678 0.565
(small/large) and second 18 Clermont-Ferrand 0.813 0.841 0.605 0.608
by services 19 Amsterdam 0.798 0.752 0.793 0.667
used (small/large) 20 Helsinki 0.764 1.000 0.753 1.000
13. of Europe rather than in its peripheries. Interestingly, our research work confirmed that Efficiency
this supply has also in quite many cities needed frontier efficiency. of public
If smaller DEA concerning services used is not taken into account, then we have cities
from positions 5-10 as potential frontier and exemplary actors of public transports. transport
However, it should be emphasized that cities having positions from 11-20 in Table VI are
not poor performers either. Some of these might have weakness in either service used or
space used side, but overall show high performance, and have possibly reached frontier in 35
one of the two measurement dimensions. This research work also gives some guidance for
these lower performing (but still meeting good standard) cities, how to proceed further with
public transportation system efficiency improvement. For example, lower performance in
space used DEA models indicates that all of the routes in the transportation system are
not necessarily needed and/or overlapping in routes for the regions between public
transportation modes should be carefully examined. In other situation, when “space used”
DEA is indicating frontier performance and “services used” is lacking behind, then new
alternative routes and/or modification of current service structure should be examined
further.
What then could very low-performing cities do with the results of this study? After
taking into account special characteristics of the particular city, it should either be start
to develop public transportation system further from the perspective of “space used” or
“services used”. For example, in new mega-cities of Asia “space used” sounds reasonable
(Kenworthy, 2002) as medium-sized cities in Europe could follow “services used”
strategy. However, in some cities, like in the USA (e.g. Chicago and Phoenix)
and Australia (e.g. Perth), new bold steps to develop public transportation system as
functional, should simply need to be taken (analyzed in details in Cameron et al. (2004)).
Argued development steps are further supported from the environmental perspective
too; we analyzed relationships of different measures of private car use, and found
interesting as well as statistically strong connection between share of private car use
(or motorized vehicle) and measured DEA efficiency. In Figure 5 is shown this
relationship between space used DEA model (small) and share of private car use – linear
regression enjoys R 2-value of 35 per cent (this regression relationship was also found to
be , 0.001 statistically significant in regression analysis, see Appendix). As could
be clearly noted, lower the efficiency of the public transportation DEA model, the
correspondingly higher use of private cars. This relationship holds very nicely until the
level of 0.9 DEA efficiency – interestingly some cities having frontier performance could
have high car use or other way around. Similar statistically significant relationships
were found within both larger DEA models (space and services; see Appendix for
regression analysis); these also repeated similar causal relationship of private car use
and DEA efficiency, which was having strong explanation power until frontier
efficiency. Only DEA model from four used, smaller services used DEA, did not have any
relationship with the share of private car use. However, it should be remembered that in
the earlier correlation analysis it was showed, that all DEA models are in statistically
strong and positive relationship with others, and therefore we cannot neglect larger
services used DEA from the further use. Thus, in the future further analyses should be
completed to reveal whether it has connection with the private car use or not.
Based on this research work, we are in environmental terms slightly in favour of
space used DEA model, since both large and small were in statistically strong and
significant relationship with share of private car use; interesting fact, among the lack
14. BIJ 100
y = –34.888x + 89.254
18,1 R2 = 0.3453
Percentage of daily mechanised and motorised trips
90
80
by private motorised modes
70
36
60
50
40
30
Figure 5.
Scatter plot and linear 20
regression relationship
between space used DEA 10
model (small) and relative
0
share of private road 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
transportation
Space used DEA (small)
of relationship with smaller services used DEA model, is that even larger services used
DEA model was having lower statistical significance with private car use (, 0.01).
6. Conclusions
Problem with economic growth is that larger amounts of people will get an opportunity
to use private car transportation – in larger scale it has also been shown several times
that economic prosperity leads into increased private car sales. However, this is
frightening development for larger cities, as climate change and scarcity of oil are
becoming as the fact in following decade. Therefore, we need to have much more efficient
public passenger transport systems to support the traveling needs of inhabitants.
As argued by Ausubel and Marchetti (2001) even middle-aged people in cities were
having needs to daily trips and size of wall protected cities was Approx. 2.5 kilometer
radius circle (resulted in 1 hour of traveling each day). Currently, this medieval circle is
insufficiently sized, since millions of people are living in one city only. However, we need
to modify our transportation systems in a manner that travel need is fulfilled by the
most environmentally friendly means. This does not necessarily mean that private car
travel would not be existing, but surely not having the magnitude of 600-700 cars
per 1,000 persons living. First, cities need to offer good enough covered public
transportation system, which reaches as many people as possible with the least amount
of space being used. This research work has pointed some frontier cities, which others
could follow in their planning, implementation and enhancement processes. Problematic
part with public transport is that it has costs, and these costs are seldom covered by the
usage fees. Therefore, cities in general have temptation to select cheapest possible
configuration, which favours road transports. However, there does not exist any support
that these sorts of systems could favour nor support the objectives of year 2020 or 2030 in
terms of transportation sustainability.
15. To continue further with our research work, we would like to take into account Efficiency
possible future scenarios and changes of public passenger transport. For example, one of public
scenario could be the higher priced oil (e.g. 150 USD per barrel), while CO2 emission
regulations could be other changing factor (and corresponding into extra cost of emitting transport
too much). So, this would lead us most probably in the development of two-staged DEA
efficiency measurement model, where intermediate level (to our already tested DEAs)
would consist these inputs (and number of others, like investments and labour used). Our 37
research work currently weights similarly rail-based emission free passenger transport
with busses using diesel oil or fuel. However, latter alternative could also be developed
further using gas, electricity or alternative fuels and also being environmentally
sustainable. Therefore, further research is needed to find most sustainable cities in the
world regarding to passenger transports – setting new standards for the future.
References
Albate, D. and Bel, G. (2010), “What shapes local public transportation in Europe? Economics,
mobility, institutions and geography”, Transportation Research Part E, Vol. 46 No. 5, pp. 775-90.
Amos, P. and Thompson, L. (2007), Railways in Development: Global Round-Up 1996-2005,
World Bank Transport Note No. TRN-36, World Bank, Washington, DC.
Amtrak (2008), National Railroad Passenger Corporation and Subsidiaries (Amtrak) Consolidated
Financial Statements for the Years Ended September 30, 2008 and 2007, Amtrak,
Washington, DC.
Ausubel, J.H. and Marchetti, C. (2001), “The evolution of transports”, The Industrial Physicist,
Vol. 7 No. 2, pp. 20-4.
Banker, R.D., Charnes, R.F. and Cooper, W.W. (1984), “Some models for estimating technical and
scale inefficiencies in data envelopment analysis”, Management Science, Vol. 30 No. 9,
pp. 1078-92.
Cameron, I., Kenworthy, J.R. and Lyons, T.J. (2003), “Understanding and predicting private
motorized urban mobility”, Transportation Research: Part D, Vol. 8 No. 4, pp. 267-83.
Cameron, I., Lyons, T.J. and Kenworthy, J.R. (2004), “Trends in vehicle kilometers travel in world
cities, 1960-1990: underlying drivers and policy responses”, Transport Policy, Vol. 11 No. 3,
pp. 287-98.
Charnes, A., Cooper, W.W. and Rhodes, E. (1978), “Measuring the efficiency of decision making
units”, European Journal of Operational Research, Vol. 2 No. 6, pp. 429-44.
Cowie, J. (1999), “The technical efficiency of public and private ownership in the rail industry”,
Journal of Transport Economics and Policy, Vol. 33 No. 3, pp. 241-51.
Cowie, J. and Asenova, D. (1999), “Organisation form, scale effects and efficiency in the British
bus industry”, Transportation, Vol. 26, pp. 231-48.
Debnath, R.M. and Shankar, R. (2008), “Benchmarking telecommunication service in India – an
application of data envelopment analysis”, Benchmarking: An International Journal,
Vol. 15 No. 5, pp. 584-98.
de Jong, G. and van de Riet, O. (2008), “The driving factors of passenger transport”, European
Journal of Transport and Infrastructure Research, Vol. 8 No. 3, pp. 227-50.
EU (2010), Energy – Statistical Pocketbook, available at: http://ec.europa.eu/energy/
publications/statistics/statistics_en.htm (accessed February 2010).
Farrell, M.J. (1957), “Measurement of productive efficiency”, Journal of the Royal Statistical
Society: Series A (General), Vol. 120 No. 3, pp. 253-90.
16. BIJ Flyvbjerg, B., Bruzelius, N. and van Wee, B. (2008), “Comparison of capital costs per
route-kilometre in urban rail”, European Journal of Transport and Infrastructure Research,
18,1 Vol. 8 No. 1, pp. 17-30.
George, S.A. and Rangaraj, N. (2008), “A performance benchmarking of Indian railway zones”,
Benchmarking: An International Journal, Vol. 15 No. 5, pp. 599-617.
Goncharuk, A.G. (2008), “Performance benchmarking in gas distribution industry”,
38 Benchmarking: An International Journal, Vol. 15 No. 5, pp. 548-59.
Hilmola, O.-P. (2007), “European railway freight transportation and adaptation to demand decline –
efficiency and partial productivity analysis from period of 1980-2003”, International Journal of
Productivity and Performance Management, Vol. 56 No. 3, pp. 205-25.
Hilmola, O.-P. (2009), “Global railway passenger transports – efficiency analysis from period of
1980-2004”, International Journal of Logistics Economics and Globalisation, Vol. 2 No. 1,
pp. 23-39.
Hu, X., Chang, S., Li, J. and Qin, Y. (2009), “Energy for sustainable road transports in China:
challenges, initiatives and policy implications”, Energy, Vol. 35 No. 11, pp. 4289-301.
Jain, P., Cullinane, S. and Cullinane, K. (2008), “The impact of governance development models on
urban rail efficiency”, Transportation Research Part A, Vol. 42 No. 10, pp. 1283-94.
Joo, S.-J., Stoeberl, P.A. and Kwon, I.-W.G. (2007), “Benchmarking efficiencies and strategies for
resale operations of a charity organization”, Benchmarking: An International Journal,
Vol. 14 No. 4, pp. 455-64.
Karathodorou, N., Graham, D.J. and Noland, R.B. (2010), “Estimating the effect of urban density
on fuel demand”, Energy Economics, Vol. 32 No. 1, pp. 86-92.
Karttunen, J., Hilmola, O-P. and Saranen, J. (2010), “Evaluating light rail as a short distance
passenger transportation solution in a midsized town”, World Review of Intermodal
Transportation Research, Vol. 3 Nos 1/2, pp. 121-36.
Keh, H.T. and Chu, S. (2003), “Retail productivity and scale economics at the firm level: a DEA
approach”, Omega – The International Journal of Management Science, Vol. 31 No. 2,
pp. 75-82.
Kenworthy, J. (2002), “Traffic 2042 – a more global perspective”, Transport Policy, Vol. 9 No. 1,
pp. 11-15.
Lao, Y. and Liu, L. (2009), “Performance evaluation of bus lines with data envelopment analysis
and geographic information systems”, Computers, Environment and Urban Systems,
Vol. 33 No. 4, pp. 247-55.
Maggio, G. and Cacciola, G. (2009), “A variant of the Hubbert curve for oil production forecasts”,
Energy Policy, Vol. 37 No. 11, pp. 4761-70.
Odek, J. (2008), “The effect of mergers on efficiency and productivity of public transport
services”, Transportation Research: Part A, Vol. 42 No. 4, pp. 696-708.
Qin, Z. (2008), “Improving public transit access to in-city villages”, International Journal of Data
Analysis Techniques and Strategies, Vol. 1 No. 2, pp. 141-52.
Rhoades, D.L., Williams, M.J. and Green, D.J. (2006), “Imperfect substitutes: competitive analysis
failure in US intercity passenger rail”, World Review of Intermodal Transportation
Research, Vol. 1 No. 1, pp. 82-93.
Sandalow, D. (2008), Freedom from Oil, McGraw-Hill, New York, NY.
UITP (2005), International Association of Public Transport – Mobility in Cities Database,
Belgium, Brussels.
17. Veolia (2009), Consolidated Financial Statements for the Half-year Ended June 30, , available at: Efficiency
www.veolia-finance.com/docs/CONSOLIDATES-FINANCIAL-STATEMENTS-FOR-THE-
HALF-YEAR-ENDED-JUNE-0030-2009.pdf (accessed December 2009). of public
Yu, M.-M. (2008), “Assessing the technical efficiency, service effectiveness, and technical transport
effectiveness of the world’s railways through NDEA analysis”, Transportation Research
Part A, Vol. 42 No. 10, pp. 1283-94.
Yu, M.-M. and Lin, E.T.J. (2008), “Efficiency and effectiveness in railway performance using a 39
multi-activity network DEA model”, Omega – International Journal of Management
Science, Vol. 36 No. 6, pp. 1005-17.
Further reading
Goncharuk, A.G. (2009), “Improving of the efficiency through benchmarking: a case of
Ukrainian breweries”, Benchmarking: An International Journal, Vol. 16 No. 1, pp. 70-87.
Appendix. Regression statistics from the causality of used four DEA models of the
study and the share of private car use
Regression statistics
Multiple R 0.5876
R2 0.3453
Adjusted R 2 0.3317
SE 0.2232
Observations 50
ANOVA
Significance
df SS MS F F
Regression 1 1.2609 1.2609 25.3174 0.0000
Residual 48 2.3905 0.0498
Total 49 3.6514
Upper Lower
Coefficients SE t-stat. p-value Lower 95% 95% 95.0%
Intercept 1.3113 0.1345 9.7520 0.0000 1.0410 1.5817 1.0410
Percentage of daily
mechanised and Table AI.
motorised trips by private Summary output – space
motorised modes 2 0.0099 0.0020 25.0316 0.0000 2 0.0139 2 0.0059 2 0.0139 used DEA modal (small)
18. BIJ
Regression statistics
18,1 Multiple R 0.0825
R2 0.0068
Adjusted R 2 2 0.0139
SE 0.2962
Observations 50
40 ANOVA
Significance
df SS MS F F
Regression 1 0.0289 0.0289 0.3291 0.5689
Residual 48 4.2127 0.0878
Total 49 4.2416
Upper
Coefficients SE t-stat. p-value Lower 95% 95% L%
Intercept 0.6978 0.1785 3.9091 0.0003 0.3389 1.0567 0.3389
Percentage of daily
Table AII. mechanised and
Summary output – motorised trips by
services used DEA private motorised
model (small) modes 2 0.0015 0.0026 20.5737 0.5689 20.0067 0.0038 2 00067
Regression statistics
Multiple R 0.5966
R2 0.3559
Adjusted R 2 0.3402
SE 0.2224
Observations 43
ANOVA
Significance
df SS MS F F
Regression 1 1.1209 1.1209 22.6580 0.0000
Residual 41 2.0283 0.0495
Total 42 3.1492
Upper
Coefficients SE t-stat. p-value Lower 95% 95%
Intercept 1.3596 0.1432 9.4927 0.0000 1.0703 16.488
Table AIII. Percentage of daily mechanised and
Summary output – space motorised trips by private motorised
used DEA model (large) modes 2 0.0099 0.0021 24.7600 0.0000 20.0140 2 0.0057
19. Efficiency
Regression statistics
Multiple R 0.4466 of public
R2 0.1994 transport
2
Adjusted R 0.1799
SE 0.2053
Observations 43
ANOVA 41
Significance
df SS MS F F
Regression 1 0.4306 0.4306 10.2127 0.0027
Residual 41 1.7287 0.0422
Total 42 2.1592
Upper Lower Upper
Coefficients SE t-stat. p-value Lower 95% 95% 95.0% 95.0%
Intercept 1.2060 0.1322 9.1205 0.0000 0.9389 1.4730 0.5389 1.4730
Percentage of
daily
mechanised
and motorised Table AIV.
trips by private Summary output –
motorised services used DEA
modes 20.0061 0.0019 23.1957 0.0027 2 0.0100 20.0022 2 0.0100 2 0.0022 model (large)
About the author
Olli-Pekka Hilmola PhD is an Acting Professor of Logistics in Lappeenranta University of
Technology (LUT), in Kouvola, Finland. Concurrently he serves as a Visiting Professor of
¨
Logistics in University of Skovde, Sweden. He is affiliated with numerous international journals
through editorial boards, including Baltic Journal of Management, Industrial Management and
Data Systems, as well as Decision Support Systems. Olli-Pekka Hilmola can be contacted at:
olli-pekka.hilmola@lut.fi
To purchase reprints of this article please e-mail: reprints@emeraldinsight.com
Or visit our web site for further details: www.emeraldinsight.com/reprints