Satisfaction with crowding and other attributes in public transport

Transport Policy 79 (2019) 213–222

Contents lists available at ScienceDirect

Transport Policy journal homepage: www.elsevier.com/locate/tranpol

Satisfaction with crowding and other attributes in public transport a

b,∗

Maria Börjesson , Isak Rubensson a b

T

The Swedish National Road and Transport Research Institute (VTI), Sweden Department for Transport Science, KTH Royal Institute of Technology, Teknikringen 10A, S- 114 28, Stockholm, Sweden

ARTICLE INFO

ABSTRACT

JEL classification: R41 R42 R48

We analyse customer satisfaction surveys conducted among public transport passengers over 15 years in Stockholm. We analyze satisfaction and importance of many attributes and their temporal trends, focusing on attributes that stand out from the rest in some way, which is primarily crowding. Crowding is the attribute with the lowest satisfaction and the only attribute for which satisfaction declines over time. However, in spite of the low satisfaction, crowding is still less important for the total satisfaction than the cognitive attributes reliability and frequency (the most important attributes). Only when crowding levels reach high levels, like that of the most crowded bus services in central Stockholm, does crowding become as important as the cognitive attributes. Also the attribute reliability stands out – it is the most important attribute. For the attributes reliability and crowding, data allow us to compare satisfaction and importance with performance. We find that that satisfaction and importance are influenced by the performance level for both attributes.

Keywords: Customer satisfaction Service quality Public transport Preferences Crowding Reliability

1. Introduction Customer satisfaction surveys (CSS) are frequently used to support decision makers. Since 2001, Stockholm Public Transport Administration has collected monthly waves of customer satisfaction surveys (CSS). Eight European cities collaborates on a benchmarking CSS (best2005.net), train passengers in UK are annually surveyed (Greeno and Joyner, 2016) and The European Union has recently commissioned the CSS Eurobarometer. There are also an increasing number of studies analyzing public transport using CSS (Fujii and Kitamura, 2003; Friman and Fellesson, 2009; Cats et al., 2015; Abenoza et al., 2017). However, the interplay between the satisfaction, the performance, and the importance of the attributes are still unclear. Understanding such interplay is key for policy design: does the performance of an attribute improve satisfaction? Which attributes are most important for policy makers to focus on? In this paper, we therefore analyze the interplay between satisfaction and importance for nine quality attributes in the Stockholm CSS data. For the two attributes for which it is possible, we explore the interplay between the satisfaction and the performance. We study temporal trends and differences between socio-economic groups and differences between public transport modes. Our primary focus is on attributes that stand out from the rest in some way, primarily crowding and reliability. Linking importance and satisfaction of different quality attributes

∗

expressed in CSS are commonplace in the literature (Martilla and James, 1977; de Oña and de Oña, 2014; Oliver, 2014; Cats et al., 2015; Abenoza et al., 2017) either directly requesting an assessment of attribute importance in the CSS or, indirectly, inferring importance in some way from the relation between general satisfaction and satisfaction with specific attributes. We adopted the latter approach modeling trip satisfaction as a function of attribute satisfaction in an ordered logit model (olm). We then interpret attribute satisfaction impact on trip satisfaction as attribute importance. Our most striking finding is that crowding is among the attributes with the lowest satisfaction, and the only attribute for which satisfaction declines over time. Still, crowding is less important than the cognitive attributes reliability and frequency (the most important attributes). There are also indications that crowding is less important at average crowding levels in other countries. For instance, analyses of CSS conducted in London and in other parts of the UK (Greeno and Joyner, 2016) show that satisfaction with crowding has a limited impact on trip satisfaction. However, looking at the most crowded bus service in Stockholm, we find that crowding becomes the most important attribute. Given this finding, a potential explanation to the low average importance of crowding is that crowding needs to surpass a certain threshold before seeing an uptake in importance. Indeed, Tirachini et al. (2013), using multinominal logit (MNL) and error components (EC) models, finds that crowding has this character of

Corresponding author. E-mail address: [email protected] (I. Rubensson).

https://doi.org/10.1016/j.tranpol.2019.05.010 Received 24 August 2017; Received in revised form 24 April 2019; Accepted 8 May 2019 Available online 09 May 2019 0967-070X/ © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).

Transport Policy 79 (2019) 213–222

M. Börjesson and I. Rubensson

being salient only when levels are high. We also find indications of the reversed, that better performance sometimes can increase importance, through increased expectations. Examples are the attributes “information of delays” and “headway”, for which importance has increased over time in tandem with increased level of performance. The role of expectations for importance is stressed in earlier literature (Oliver 1980, 2014; Morfoulaki et al., 2007; Chen and Schwartz, 2008). The importance of the attribute “information of delays” also increase when reliability is low, so there are also interactions between the attributes’ level of performance. We find further that for both reliability and crowding does the performance influence the satisfaction. The better reliability and the lower crowding, the more satisfied are the passengers with these attributes. Segmenting the CSS by public transport mode, we also find indications on how performance impacts satisfaction for the attributes where performance level differs across modes. For instance, passengers on commuter trains, having markedly lower reliability, have significantly lower satisfaction with reliability. Together our results demonstrate that both importance and satisfaction are influenced by the performance of the attributes. However, the relationships are different for different attributes. This is important for interpreting trends over time, but also for understanding differences in satisfaction and importance across different public transport systems. The previous literature concerning the impact on public transport satisfaction by performance has given mixed results. Barabino et al. (2012) find that performance has a large influence on satisfaction, whereas (Mackett and Edwards, 1998; Fujii and Kitamura, 2003; Friman and Fellesson, 2009) find evidence of the opposite. Friman and Fellesson (2009) find a weak correlation between the measured crowding and the satisfaction with crowding (although the measure of crowding they use is fairly imprecise). However, they find no correlation between measured crowding and trip satisfaction when comparing different cities. We are, in this study, able to use automatic passenger count data (APC) and automatic vehicle location data (AVL) to construct more disaggregated performance measures for the attributes of crowding and reliability. Our findings are to some extent consistent with (Friman and Fellesson, 2009): we also find that performance influence satisfaction. PT authorities in several countries have introduced quality-based incentive payments to public transport operators in the past decade (Hensher and Houghton, 2004; Van de Velde et al., 2008; Trafikanalys, 2013) to improve service quality. A better understanding of the interplay between the satisfaction, importance and the performance is then key for designing the contract-incentives and motoring the outcome. Section 2 reviews the theory on consumer satisfaction. Section 3 describes our data. Section 4 explores the relationship between satisfaction and importance for different attributes, how this relationship has evolved and how it differs across modes. In Section 5 we compare the customer satisfaction with the performance by attribute. Section 6 discusses the findings, and section 7 presents the policy implications of our findings.

(Abenoza et al., 2017), using for instance bivariate Pearson correlation, regression analysis, structural equations, or path analysis and neural networks (de Oña and de Oña, 2014). Oliver (2014) advocates inferred importance measures for two reasons. First, adding questions on both satisfaction and importance for all survey attributes might increase the number of questions substantially and increase the risk of respondent fatigue. Second, in surveys where the respondents are directly asked to state the importance of an attribute, it is unclear how the question is interpreted. In this paper, we infer the importance using an ordered logit model, we will also note, comparing to a previous Stockholm survey, similar results using stated levels of importance - which is reassuring. Oliver (2014) suggests that the relationship between performance and importance depends on the type of attribute. Similar ideas can be found in Maslow (1943), Kano et al. (1984) and Matzler et al. (2003). The latter paper divides the attributes into three categories: (1) Attributes having a performance threshold level, which are important primarily when performance drops below the threshold and have a small impact on the importance when above. (2) Attributes for which importance increases when the attribute is delivered but that cause no dissatisfaction if absent. (3) Attributes for which the importance is independent of performance. In the expectancy-disconfirmation framework, customer satisfaction also depends on the performance relative to expectations (Oliver, 1980; Morfoulaki et al., 2007; Chen and Schwartz, 2008). Hence, the literature suggests that satisfaction with an attribute is formed based on performance as well as the consumer's expectation of the performance level. Since the performance of the attribute varies across modes, over time, and over different socioeconomic groups, section 4 explores how the satisfaction with several public transport attributes varies in these dimensions. This indicates whether satisfaction with the attributes is influenced by the performance level and the expectation.

2. Theory

Stockholm Public Transport has 2.9 million passenger boardings on an average winter weekday, and includes metro (1.3 million boardings), bus (1.1), commuter train (0.3) and light rail (0.2). The system spans the Stockholm county of 6.5 thousand square kilometers, with commuter trains providing long distance stem corridors, metro serving a dense inner city and the inner suburbs. There are six different light rail services: two within the city and four operating on longer distances, between outer suburbs and the inner-city. Buses cover the remainder of the county. Public transport is competitive, with modal shares of passengers toward the inner city in the morning peak of almost 80 percent and average weekday modal share of 43 percent(See. Fig. 1). Crowding reaches high levels in certain places during the morning and afternoon peaks, such as the central station, the inner legs of metro services and the commuter train services, inner-city bus services and

3. Data We use the Stockholm Customer satisfaction Survey 2008–2016 and automatic location and passenger counts for 2014 from vehicles operating in Stockholm. To explore the relationship between the performance level, satisfaction and importance of the attribute, we also compare the satisfaction data from 2014 with the gathered vehicle location and passenger count data for the same year. Since Tirachini et al. (2013) convincingly shows that crowding must pass a threshold level to cause discomfort, section 3.1 describes the general crowding levels in the Stockholm Public Transport system. In 3.2, we describe the Stockholm customer satisfaction survey, including survey method and questionnaire. To measure the crowding levels and travel time reliability facing the respondents, we use data from automatic passenger count (APC) and automatic vehicle location (AVL), described in 3.3. 3.1. Stockholm Public Transport network

Consumer satisfaction theory was developed in the 1970s by firms to increase the understanding of customers’ satisfaction with and loyalty to brands and products. Martilla and James (1977) constructed the Importance-Performance (IP) – chart to extract information from quality and satisfaction surveys, relating satisfaction with perceived importance of an attribute. The literature has used two main categories of methods for measuring the importance of each attribute: (a) explicitly asking respondents in the customer satisfaction survey to indicate the importance they attach to each quality attribute (Tyrinopoulos and Aifadopoulou, 2008; Eboli and Mazzulla, 2009; Guirao et al., 2016); and (b) inferring the importance by modeling the explanatory power of each attribute on the total trip satisfaction 214

Transport Policy 79 (2019) 213–222

M. Börjesson and I. Rubensson

trips on the line and state their satisfaction. After the nine statements, the respondents are asked to grade their trip satisfaction with the line, where one is the lowest grade and seven the highest grade. All statements/questions include the escape-alternative “I have no experience”. 3.3. Automatic vehicle location and automatic passenger counts All vehicles in the Stockholm Public Transport system are equipped with Automatic Vehicle Location (AVL). It measures the travel time and reliability for all services on all lines and modes. Approximately 10 percent of all buses and trains are also equipped with Automatic Passenger Counts (APC), measuring the number of boarding and alighting passengers and load factors between stops (APC is not used on the Metro). Operators are instructed to operate the APC vehicles in proportion to observed number of passengers by line and service, and at least twice a month for each service.

Fig. 1. Crowding on an ordinary work day, Metro platform in the central station (left) and commuter train car (right).

Table 1 Average seat occupancy by transport mode in the Stockholm Public Transport, 2004–2016, for the morning peak (6–9 am) and peak hour (7.30–8.30 am). year

2004 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Rush period (6–9 am)

Rush hour (7.30–8.30 am)

Bus

Metro

Commuter train

Light Rail

Bus

Metro

Commuter train

Light Rail

25–40 30–45 30–45 25–45 25–50 35 35 35 35 35 40

45 45 50 55 55 55 55 55 55 55 60

35 35 30 35 35 35 35 30 30 35 35

25–45 30–60 25–55 25–55 20–55 30 35 35 35 35 35

25–60 40 40 40 40 40 40

70 70 75 75 75 75 80

35 35 35 30 35 35 35

20–80 45 45 45 40 45 45

bus services and light rail operating to and from the inner city. Occasionally, crowding reaches levels forcing passengers to wait for the next train/bus but this is unusual and happens only for services with very short headway (such that the waiting time in that case is low). The crowding increases over time as the population of Stockholm increases. This is indicated by the trend increase in average seat occupancy (measured as person kilometers divided by seat kilometers). Table 1 shows how the seat occupancy, by transport mode, has evolved from 2004 to 2016. The table also indicates that the metro has the highest average occupancy. This is largely because the metro operates primarily in the center of the city. The average occupancy levels naturally hide a substantial variation in time and space, and for the other modes the variation is even larger, where the more central services have a higher occupancy level and the more peripheral services often have lower occupation levels.

We use observations for buses where both automatic passenger count (APC) data and automatic vehicle location (AVL) data has been collected. The bus network in Stockholm is divided into 16 tender areas which are the geographical blocks used when preparing calls for tender and tender agreements. Section 5 compares the satisfaction with crowding and reliability with the APC/AVL data on crowding and delays across the 16 tender areas. From this data, we can then derive the total travel time that the passengers on different public transport services spend in vehicles with different load factors. We can thereby compare the satisfaction with crowding and its impact on trip satisfaction between passengers on the most crowded inner-city bus line and passengers on the other bus lines. We can also compare the passengers’ satisfaction with reliability with the actual performance of travel time reliability (also computed from AVL/APC data).

3.2. Stockholm Customer satisfaction survey

4. Satisfaction and importance across time, modes and socioeconomic groups

The Stockholm Customer Satisfaction Survey has been conducted since 2001, with ten survey waves per year.1 Before 2014, each survey wave included 9,000 interviews, but since 2014 each wave has included 10,000 interviews. Paper questionnaires are distributed and collected on board. They include statements about satisfaction with nine quality attributes (see Table 2). The responses are given on a seven-point Likert scale, ranging from ‘7- agree completely’ to ‘1- disagree completely’. The respondents are instructed to have in mind the usual conditions of the line they are using when stating their satisfaction. Hence, the respondents are tacitly asked to recall the travel conditions from previous

Fig. 2 shows the share of satisfied respondents (giving the relevant attribute a grade above four) for crowding, reliability and trip satisfaction throughout the 16 years in which the CSS has been conducted. The trip satisfaction shows a steady trend increase. In fact, most of the quality attributes display a trend increase in satisfaction. The trend of satisfaction with reliability closely follows the trip satisfaction trend. The satisfaction with crowding, on the other hand, displays a trend decline until 2013, which is consistent with the increase in occupancy level shown in Section 3.1. To explore the importance of each attribute for the total satisfaction, we model trip satisfaction as a function of the satisfaction with each attribute applying an ordered logit model (olm). We use the estimated parameters as measures of importance, defining importance as the

1 Each wave is conducted during two weeks. One wave per month is conducted January–May and August–December.

215

Transport Policy 79 (2019) 213–222

M. Börjesson and I. Rubensson

Table 2 Statements concerning the satisfaction with nine quality-attributes labels, and the question on the trip satisfaction.2 Factor

Wording in survey

Response alternatives

Headway Reliability Information delays Clean vehicle Clean platform/stop Crowding Information Personnel Attitude personnel Smooth Driving Trip satisfaction

I am satisfied with the service frequency I am satisfied with the travel time reliability I am satisfied with the provision of information about train delays and timetable changes information on changes and delays The vehicle is clean The bus stops or platforms are clean I am rarely troubled by crowding when I travel with this line Staff are able to answer questions regarding Stockholm Public Transport Staff are well mannered and service-minded The driving style is smooth and comfortable What is your overall grade of this line?

Likert scale Likert scale Likert scale Likert scale Likert scale Likert scale Likert scale Likert scale Likert scale Grade 1-7

1-7 1-7 1-7 1-7 1-7 1-7 1-7 1-7 1-7

satisfaction for 2008/2009 and 2013/2014 separately. These two models are reported in Table 3, including a statistical comparison of the importance and average satisfaction by attribute. All attributes except crowding see a statistically significant increase in satisfaction. Headway, smooth driving, information personnel and information delays combine increased satisfaction with increases in importance. Respondents’ satisfaction with crowding has decreased, while its importance has increased. Thus, three different patterns emerge, first, the attribute with growing dissatisfaction and importance (crowding), second, attributes with increasing satisfaction and importance (information delays, smooth driving, information personnel and headway), and, third, attributes with increasing satisfaction without changes in importance (all the other attributes). Regarding the first pattern, if the decreasing general satisfaction with crowding is an effect of a growing share of the passengers experiencing crowding below the threshold that triggers importance, then the increased importance is expected. Regarding the second pattern, increased importance could be a sign of an excitement attribute, important when present but not important when absent. This could be the case with information delays; traffic information has become increasingly available over the past ten years due to technical developments and the advent of the smartphone. Or it could be shifts in other preferences, such as changed preferences caused by, for example, higher levels of stress, or higher demand for comfort. Regarding the third pattern, if an attribute is above the lower acceptable threshold, or no threshold exists, we would expect increased performance to deliver increased satisfaction, but the importance to remain stable. To further evaluate the responsiveness to performance of the attributes satisfaction and importance, we estimate four models using data for 2008–2014 but segmented into the mode of transport, bus, commuter train, light rail, and, metro. The full table with t-statistics is placed in Table 7 in the appendix. Fig. 3 compares the estimated parameters, i.e. the importance of all modes. For comparison, Fig. 4 shows the average satisfaction by attribute and mode. The most striking result is the high importance of, and low satisfaction with, reliability among commuter train passengers. Commuter trains are also burdened with many more delays than other modes. Also, the attribute information delays has high importance and low satisfaction among commuter train passengers, probably because passengers that experience frequent delays are more dependent on information of these delays. Satisfaction with information is also low for bus, but for bus it is not so important. Hence, it seems that low reliability increases the importance of information delays. Headway has clearly lower importance for commuter trains than for other modes, as well as the attitude of personnel and smooth driving. This might be due to expectations: these passengers are not used to personal contacts or short headways. The most crowded mode is metro, and metro passengers are also least satisfied with crowding. Crowding is also most important for metro passengers. Metro passengers also place higher importance on clean platforms than other modes. Light rail passengers have on average

Fig. 2. The share of respondents satisfied by crowding, reliability and giving their line a grade above four (trip satisfaction) over time in the survey conducted since 2001.

explanatory power that the satisfaction with each attribute has for trip satisfaction. The software used for our modeling is BisonBiogeme (Bierlaire, 2003) The dependent variable in the olm is the response to the trip satisfaction, (indicated on a 7-grade scale, see above). The explanatory variables are the satisfaction with each attribute (indicated on a 7-grade Likert scale). Table 3 shows the olm model estimated on all data 2008–2014 as well as segmented in an earlier and a later model using 2 years of data from the beginning and the end of the data set. For comparison, the table also includes the mean of the satisfaction with the attribute in the table (in italics). We are aware that averages on ordinate scales can be misleading, because the distance between each step in the scale might not have the same meaning. For instance, it is not certain that the distance between “disagree completely’” and “disagree” equals the distance between “neutral” and “disagree”. Still, we make this simplified assumption here because it is easy to make statistical comparisons between two means. Table 3 shows that crowding has the lowest satisfaction, which is consistent with Fig. 3. Table 3 also shows that reliability and headway are more important than the softer attributes describing comfort. Interestingly, reliability and headway are also more important than crowding in spite of the low satisfaction with crowding. As reported in section 3.1, crowding levels have been increasing over time, which, looking at Fig. 2, seem to have taken its toll on satisfaction with crowding. To investigate how the importance of and satisfaction with the attributes have evolved over time, we model trip

2

A Likert scale expresses the level of agreement or disagreement on a symmetrical scale between disagree completely (1) and agree completely (7). The trip satisfaction is not measured on a Likert scale but the respondents are asked to give the service and overall grade, where 1 is the lowest grade and 7 is the highest. 216

Transport Policy 79 (2019) 213–222

M. Börjesson and I. Rubensson

Table 3 Base model of trip satisfaction as a function of quality attribute satisfaction as well as two models to assess changes over time, one for the years 2008/2009 and one for the years 2013/2014. Model parameter is interpreted as the importance of the attribute, in italics the average satisfaction with the attribute is reported. T-tests comparing values between 2008/2009 and 2013/2014 are presented in the last column. Base model (2008–2014)

Early model (2008–2009)

Late model (2013–2014)

t-test: Late - Early data

0.40 170.92 5.00

0.38 84.39 4.95

0.42 104.29 5.04

7.13

0.54 201.58 4.90

0.54 102.68 4.87

0.52 111.38 4.97

0.17 74.39 4.50

0.15 34.18 4.37

0.21 49.81 4.67

10.09

0.20 69.14 5.08

0.20 36.50 4.99

0.20 39.54 5.17

−0.27

0.14 47.35 4.91

0.15 26.39 4.79

0.14 27.16 5.00

0.27 134.92 4.27

0.25 63.76 4.32

0.28 83.12 4.23

0.11 32.23 4.95

0.09 14.18 4.91

0.11 19.32 5.02

Headway model parameter model t-test satisfaction mean Reliability model parameter model t-test satisfaction mean Information delays model parameter model t-test satisfaction mean Clean vehicle model parameter model t-test satisfaction mean Clean platform/stop model parameter model t-test satisfaction mean Crowding model parameter model t-test satisfaction mean Information personnel model parameter model t-test satisfaction mean Attitude personnel model parameter model t-test satisfaction mean Smooth driving model parameter model t-test satisfaction mean

0.31 98.00 5.02

0.32 52.54 4.91

0.30 54.82 5.11

−1.72

0.32 119.04 5.29

0.32 61.10 5.26

0.34 73.71 5.30

2.90

n Likelihood ratio test: Adjusted rho-square:

407 858 706 895.53 0.446

104 537 178 007.17 0.44

141 263 250 569.05 0.455

Fig. 3. Importance by mode and attribute. Importance is defined as the parameter estimates of four models using data for 2008–2014 but segmented into the mode of transport, bus, commuter train, light rail, and, metro.

13.04 −1.57 16.37

38.35

31.27 −1.41 33.85 5.74 −12.92 2.28 14.04

30.98

6.32

Fig. 4. Average satisfaction by mode and attribute.

Does the importance of and the satisfaction with the attributes depend on gender and age? To investigate this, we segmented the models by gender and by age in. Table 4 (one model for children, teenagers and young adults below 30 and one model for older grownups above 30).

higher satisfaction with almost all attributes and with trip satisfaction. The performance of the quality attributes is not in general higher for light rail, but the passengers are richer because the light rail connects more affluent parts of Stockholm.

217

Transport Policy 79 (2019) 213–222

M. Börjesson and I. Rubensson

Table 4 Four models of trip satisfaction as a function of quality attribute satisfaction, two for men and women, one for young (< 30 years old) and one for old passengers (above 30). Model parameter is interpreted as the importance of the attribute, in italics, the average satisfaction with the attribute is reported. The last column presents the t-statistics for pairwise comparisons between the genders/age groups. Women

Men

t-test: women - men

Young

Old

t-test: young - old

0.40 132.45 5.01

0.40 106.82 4.99

0.42

0.39 109.07 4.86

0.39 126.52 5.10

0.63

0.55 154.85 4.92

0.54 127.59 4.86

−2.00

0.48 121.45 4.78

0.61 159.97 4.98

0.17 57.24 4.52

0.17 48.21 4.47

0.21

0.18 49.62 4.57

0.17 55.28 4.45

−1.07

0.19 51.79 5.09

0.21 45.58 5.06

2.91

0.17 38.26 5.10

0.22 56.81 5.08

8.06

0.14 34.98 4.92

0.15 31.26 4.91

0.15 30.94 4.96

0.15 37.41 4.89

0.28 107.10 4.23

0.26 80.55 4.33

−3.62

0.25 80.32 4.15

0.29 107.19 4.35

0.10 23.19 4.97

0.11 22.05 4.92

2.07

0.09 18.90 4.88

0.12 26.25 5.01

0.26 56.66 4.79

0.34 76.07 5.17

12.13

0.28 70.22 5.24

0.35 94.76 5.34

12.58

164 329 260 719.75 0.423

235 500 438 023.06 0.468

Headway model parameter model t-test satisfaction mean Reliability model parameter model t-test satisfaction mean Information delays model parameter model t-test satisfaction mean Clean vehicle model parameter model t-test satisfaction mean Clean platform/stop model parameter model t-test satisfaction mean Crowding model parameter model t-test satisfaction mean Information personnel model parameter model t-test satisfaction mean Attitude personnel model parameter model t-test satisfaction mean Smooth driving model parameter model t-test satisfaction mean

0.32 78.10 5.05

0.29 59.06 4.97

−4.03

0.33 95.25 5.34

0.30 71.49 5.22

−6.81

n Likelihood ratio test: Adjusted rho-square:

241 252 421 788.38 0.452

162 955 280 699.58 0.441

−4.42

−11.52

−8.03

−7.69 2.10 −0.57

18.33

−7.27

−16.09

−26.83

In our gender models we see that satisfaction and importance is very similar with one exception. Women are less satisfied with crowding and deem it more important than men. Women's lower satisfaction and higher importance might be a result of a more negative experience of crowding, due to security. The risk of harassments and unwanted touching probably increases with crowding. Moreover, women are on average shorter than men, and shortness make crowding more of a nuisance, making it hard to reach poles and grab handles, not having free sight lines when standing among taller people. Next, we compare young (< 30 years) and old (> 30 years) passengers. Older passengers have higher satisfaction with, but also place higher importance on, reliability than younger passengers. The higher importance of reliability is probably an effect of more scheduling constraints; adults above 30 years have more responsibilities at work, for children, family, older relatives, in the civil community making their vulnerability to unreliability higher. Older passengers have higher satisfaction and place more importance on the attitude of personnel. In a survey conducted in Stockholm 15 years ago (Olsson et al., 2001) commuters were asked to rank quality attributes in order of importance. The resulting average rankings are shown in Table 5. It shows that public transport users ranked the importance of the quality attributes very similarly to what we derived from the CSS data: frequency and reliability are both at the top and crowding in the bottom half among affective attributes. This supports the evidence from the satisfaction studies, and that impact on trip satisfaction seems to

45.44 23.48 38.88

−19.48

−4.28 0.81 −13.23 8.52 34.26 3.49 18.86

72.02

21.66

Table 5 Comparison between importance rankings of quality attributes made by Stockholm PT users in 2000 (Olsson et al., 2001) and in 2014 (using ordered logit model on Stockholm CSS). Olsson et al., (2001)

Stockholm CSS 2014

1 1 3 4 4 4 7 7 9 10 11 12 13 14 14

1 2 3 4 5 6 7 8 9

Frequency On station signs with time to next departure Reliability Short travel time Clean vehicle Agreeable temperature Smooth driving Clean station/stop Crowding “find a free seat" No noice and bumpiness Comfortable seats Good lighting Access to seats Good lighting Onboard signs with information about next stop

Reliability Frequency Smooth driving Attitude personnel Crowding Clean vehicle Information delays Clean station/stop Information personnel

measure importance. To summarize, over time, crowding is the only attribute for which satisfaction decreases and importance increases. Importance also increases for headway and information delays, the latter probably due to 218

Transport Policy 79 (2019) 213–222

M. Börjesson and I. Rubensson

a change in expectations. Commuter train passengers place the highest importance on reliability, and are least satisfied with reliability, congruent with the lower regularity performance during the modelled period. There are no gender differences in satisfaction and importance of any attributes, except for women deeming crowding less satisfactory and more important than men. Younger passengers are less satisfied with but place lower importance on reliability and personnel attitude. Crowding is consistently, over time and across modes, ranked as less important than reliability and frequency and, at the same time, among the attributes that passengers are least satisfied with.

Table 6 Comparison between all bus services (during all times and all days) and Bus 4 during peak hour on work days. Upper part of the table describes the distribution of crowding between the services and the lower part shows selected variables from the ordered logit models. All bus services

Bus 4 peak hour

share of travel time under different Load factors

%

%

5. Does satisfaction depend on performaNce?

< 0.5 0.5–0.75 0.75–1 1–3.3

61.1 22.7 10.1 6.1

22.5 23.8 18.9 34.8

5.1. Crowding and reliability on bus tender areas

Model and satisfaction data

All bus

Bus 4 Peak

t-test

In this section we analyze how performance of crowding and delays influences the satisfaction and importance of these attributes. We begin by exploring how satisfaction with crowding and reliability depends on the performance. We compare the average satisfaction with actual performance from public transport vehicles by bus-tender area3 (n = 16) using data collected in 2014 only. The actual performance is measured by automatic passenger counts (APC) and/or automatic vehicle location (AVL). The AVL data records the actual travel time by line segment (between two stops). If there are delays, this time might be different from the travel time according to the timetable. The APC data records the number of passengers in the vehicle on each line segment. Total travel time on a line segment is computed as the vehicle travel time times the number of passengers. The occupancy rate is computed as the number of passengers per seat. The level of crowding by tender area is computed as the share of the total travel time that passengers spend in vehicles where occupancy is above 1 (all seats occupied). With a basic linear regression model where the dependent variable is share of satisfied passengers in tender areas and the explanatory variable is level of crowding, as described above, we find a downward slope of −1.2 (satisfaction decreases when crowding increases) with t-stat = −3.9 and R2 = 0.5287. So, in the case of crowding, passengers’ satisfaction does depend on performance. Delays are not as straight forward to measure as crowding. . We use AVL and APC data to assess the total delay time experienced by the passengers by bus-tender area. The AVL data includes information on the delay of each bus departing from the bus stops within the bus route. All stops are included for all buses. However, since the total delay can only be computed at the end of the trip, when the passenger alights, the total delay time cannot be calculated exactly from only the number of passengers per service and line segment. We therefore approximate the total delay time experienced by the passengers on bus service j, T jd , as

Headway model parameter model t-test satisfaction mean Reliability model parameter model t-test satisfaction mean Crowding model parameter model t-test satisfaction mean

0.42 120 4.92

0.38 7.26 5.47

−0.72

0.50 125.23 4.77

0.40 8.3 4.86

−2.03

0.24 77.15 4.50

0.43 11.13 3.73

4.92

223 958 276875.29 0.438

877 2470.96 0.459

T jd

xi

j

max i, i j

n Likelihood ratio test: Adjusted rho-square:

13.57

1.93

−16.39

5.2. Crowding satisfaction on the busiest line The inner-city bus line No. 4 is the busiest bus line in Stockholm. It cuts through the city center of Stockholm and has a low headway (every 4–6 min in the peak). The upper part of Table 6 shows the share of the total in-vehicle time the passengers spend in different occupancy rates, calculated from the ATR data. In the peak, passengers on bus No. 4 spend on average 34 percent of the in-vehicle time at an occupancy rate over 100 percent. The corresponding share for all bus lines, services and times-of-day in the county is 6 percent. The crowding is thus substantially higher for peak passengers on bus No. 4 than for the average bus passengers. The headway is also substantially higher for the former passenger group. The lower part of Table 6 shows selected variables4 and satisfaction averages, estimated separately for peak passengers on bus No. 4 and all bus passengers in the county. As expected, bus No. 4 passengers are more satisfied with the headway and less satisfied with crowding than the latter group. Hence, the performance level facing the respondents does influence the satisfaction with these two attributes. Moreover, for bus no. 4 crowding is the most important attribute, even more important than headway and reliability. This is consistent with the hypothesis that crowding must reach a certain threshold level before it impacts passenger's discomfort (Tirachini et al., 2013). Passengers on bus No. 4 and all bus passengers are equally satisfied with all other attributes than headway and crowding, which is expected since they exhibit no large differences in performance. Hence, we can conclude that although satisfaction with crowding is low among all passengers, it is only important for trip satisfaction when it reaches a certain, high, threshold level. The high threshold level is also indicated by satisfaction studies from other countries. Greeno and Joyner (2016) find a low importance of crowding when analyzing

(1)

where xi is the number of passengers on line segment i and service j, δi is the delay of the service for the line segment i. Hence, we assume that all passengers experience the delay of the line segment with the largest delay. This might overestimate the total delay time, but since delay and passenger volumes are correlated, this construct should at least indicate how delay varies across regions. We estimate a linear regression with the ratio of total delay time over the total travel time of all lines and services in the tender area ( j T jd/ j Tj ) as the explanatory variable. The resulting slope is −0.5 indicate decreasing satisfaction with increasing delays (t-statistic = −1.9, R2 = 0.2428). 3 Administrative geographic units used when defining the extent of a bustender agreement. An agreement can include one or more bus-tender areas.

4 The three variables from the olm model on bus no 4 in peak period that have the strongest impact on trip satisfaction.

219

Transport Policy 79 (2019) 213–222

M. Börjesson and I. Rubensson

National Rail passenger CSS data from the UK, where the average occupancy level is higher than in Stockholm.5 The UK National Rail passenger CSS is conducted on all regional and national rail services and multivariate analyses show that dissatisfaction with crowding explains only 6 percent of the total trip dissatisfaction. In summary, we have shown that satisfaction depends on performance and expectations. Satisfaction with reliability and crowding seems to mirror actual performance in these attributes. We have also shown that when crowding reaches really high levels, as in the case with bus line no. 4, it becomes the most important attribute.

possible product failure in their total satisfaction with the service. Naturally, an underlying necessary assumption in satisfaction studies is that the consumers have the impression that the service provider can influence the performance of the attribute, and that the attribute is part of the service they are asked to evaluate. All attributes but crowding included in the survey used in this paper are clearly part of the service in the sense that the operator is responsible for their performance. However, the finding that crowding is increasing in importance for high levels of crowding speaks against this hypothesis. 7. Policy implications

6. Discussion of findings

Customer satisfaction studies are increasingly used in the planning and design process of the public transport system. Customer satisfaction studies are increasingly used to monitor and steer public transport production in Europe, whether they are studies constructed in a single public transport system (as the Stockholm CSS reported in this article) or nationwide (as the Transport focus studies from UK) or as a multilateral benchmarking project in the form of BEST (best2005.net) between Berlin, Copenhagen, Geneva, Helsinki, Oslo, Stockholm and Vienna. The objectives of these surveys include improving the understanding of consumers’ perceptions and views of the public transport systems, and monitoring the performance of the systems. They are also used to craft strategies to achieve political objectives for the public transport system (related to increased market share for public transport, revenues and customer satisfaction). One of the main applications of the satisfaction studies is to monitor the service level in procured public transport, and to give the operators incentives to maintain or increase the level of service. Here, customer satisfaction surveys are an attractive tool for conducting regular checkups to gauge tender companies' performance as well as the influence of exogenous factors such as weather, strikes, disruptions in infrastructure and other events outside the tendering companies’ responsibility. Results from the customer satisfaction surveys are often linked to bonuses and penalties in the contract with tendering companies. This paper underscores the importance of, and possible pitfalls in interpreting and understanding the interplay between satisfaction(S), importance(I), and, performance(P). Further, how S-I-P interplays is, as have been shown, different for different attributes, which need to be understood to craft efficient policies. Finally, an attractive public transport system is a cornerstone for a vital and green city. In the efforts to maintain or build such a system, the customer satisfaction survey could be a useful tool, provided that the policy makers understand its strengths and weaknesses, which has been the main objective of the paper.

We find a stronger relationship between performance level and satisfaction for crowding and reliability than Friman and Fellesson (2009), who use more aggregate data than we do. Moreover, we find that the satisfaction with different attributes varies across modes in the way that is expected from the performance level and that satisfaction with all attributes except crowding has increased over time for all attributes. We infer attribute importance by statistically testing the strength of the relationship of individual attributes and trip satisfaction. This is common practice in the literature (de Oña and de Oña, 2014; Oliver, 2014), although it must be noted that the direction of these relationships are only assumed plausible and not known. The literature on the Importance-Performance relationship describes why the relationship is expected to vary among different types of attributes (see Section 2). Our results seem to comply with these theories in many ways. Crowding and reliability are the types of attributes that have a threshold under which decreases in satisfaction are not so critical. However, above this threshold, importance rises. Information delay seems to be an excitement attribute, for which increased possibility to deliver performance (smart phones, real time data) increases both satisfaction and importance. The results for headway in our data seem to suggest that this attribute is an excitement attribute: importance goes up with increasing satisfaction with headway. One puzzling result of this study is that, although the importance of crowding is low unless it reaches very high levels, crowding is still the attribute with the lowest satisfaction levels. The low importance attributed by most passengers is not consistent with the high valuation of crowding found in many valuation studies, or the public debate where crowding in the public transport system is seen as a major and increasing problem, justifying mega investments. But why is satisfaction so low when importance is low: could it be due to policy bias (respondents trying to impact changes in policies) or affective bias? Another possible reason for the low satisfaction in combination with the low importance could be that crowding is an affective attribute, and thereby processed differently in the brain. Stradling et al. (2007) show that unwanted arousal from affective attributes (feeling unsafe, crowding etcetera) has a prominent role in deterring public transport trips. Another potential explanation for the low importance of crowding may be that the passengers do not perceive crowding as part of the service provided by the operator, because they do not believe that the operation can or should influence the demand. Folkes (1984) suggests that customers do consider who they believe are responsible for a

Declaration of interest statement Isak Rubensson is an employee of Stockholm County Public Transport Administration and his Phd studies are financed by Stockholm County. Acknowledgements This research is funded by The Stockholm County Administration.

5 Comparing average seat occupancy (crude measure of crowding) all UK rail operators have higher occupancy than the Stockholm light rail and commuter rail systems. The Stockholm metro is placed third among UK rail operators with London Overground and First Transpennine Express having higher occupancy rates. (ORR (2012). Costs and Revenues of Franchised Passenger Train Operators in the UK, Office of Rail Regulation, SLL (2015). Fakta om SL och länet 2014. Fakta om SL och länet: 74.).

220

Transport Policy 79 (2019) 213–222

M. Börjesson and I. Rubensson

Appendix. Importance and satisfaction per mode and pairwise T-statistics Table 7

Models per each transport mode of trip satisfaction as a function of quality attribute satisfaction. Model parameter is interpreted as the importance of the attribute; in italics the average satisfaction with the attribute is reported. The six last columns present the t-statistics for pairwise comparisons between modes.

Headway model parameter model t-test satisfaction mean Reliability model parameter model t-test satisfaction mean Information delays model parameter model t-test satisfaction mean Clean vehicle model parameter model t-test satisfaction mean Clean platform/stop model parameter model t-test satisfaction mean Crowding model parameter model t-test satisfaction mean Information personnel model parameter model t-test satisfaction mean Attitude personnel model parameter model t-test satisfaction mean Smooth driving model parameter model t-test satisfaction mean

Bus (1)

Commuter train (2)

Light rail (3)

Metro (4)

t-test (1)–(2)

t-test (1)–(3)

t-test (1)–(4)

t-test (2)–(3)

t-test (2)–(4)

t-test (3)–(4)

0.42 120.00 4.92

0.32 51.67 4.74

0.42 43.53 4.79

0.43 107.93 5.16

−14.58

0.29

0.95

9.22

14.81

0.19

−19.26

−11.16

40.77

−2.92

−45.49

−30.36

0.50 125.23 4.77

0.77 95.08 4.02

0.52 42.37 5.22

0.47 99.23 5.19

30.07

1.56

−5.37

−17.12

−32.52

−4.05

−84.19

40.77

78.25

−93.14

−137.31

2.66

0.15 43.69 4.03

0.23 32.87 4.34

0.19 17.59 4.72

0.19 49.68 4.88

10.64

3.35

8.90

−3.64

−4.60

0.80

32.46

53.73

132.87

−26.19

−59.55

−13.17

0.18 42.36 5.19

0.18 23.19 4.88

0.26 18.68 5.51

0.19 40.66 4.97

−0.11

5.36

0.32

4.96

0.33

−5.20

−37.61

32.13

−44.68

−53.27

−10.49

55.28

0.11 25.26 4.89

0.12 12.91 4.75

0.10 7.11 5.38

0.20 40.86 4.91

0.50

−0.53

13.87

−0.76

8.47

6.53

−18.52

48.11

4.02

−54.13

−21.61

46.81

0.24 77.15 4.50

0.23 36.43 4.26

0.27 29.32 4.70

0.31 98.09 4.00

−1.57

3.45

14.98

4.00

10.97

3.25

−26.08

15.41

−82.10

−30.56

28.34

55.95

0.11 20.77 5.03

0.13 13.78 4.76

0.08 4.76 5.54

0.10 19.48 4.88

1.70

−1.88

−1.57

−2.69

−2.80

1.22

−24.44

37.62

−20.81

−49.65

−11.21

49.55

0.34 67.34 5.13

0.21 22.33 4.84

0.35 23.73 5.66

0.29 58.91 4.85

−12.60

0.32

−7.71

8.00

7.55

−3.85

−33.53

53.79

−52.60

−68.08

−1.38

82.43

0.35 83.16 5.16

0.21 23.24 5.70

0.36 24.69 5.72

0.34 83.30 5.26

−14.86

0.46

−2.39

8.98

13.54

−1.39

76.08

60.51

18.33

−1.36

64.46

51.07

65 005 73 099.61 0.417

73 341 40 888.36 0.483

45 554 320 834.74 0.463

N 223 958 Likelihood ratio test: 276 875.29 Adjusted rho-square: 0.438

quality attributes in public transportation: narrowing the gap between scientific research and practitioners' needs. Transport Pol. 49, 68–77. Hensher, D.A., Houghton, E., 2004. Performance-based quality contracts for the bus sector: delivering social and commercial value for money. Transp. Res. Part B Methodol. 38 (2), 123–146. Kano, N., Seraku, N., Takahashi, F., Tsuji, S., 1984. Attractive quality and must-Be quality. J. Jpn. Soc. Qual. Contr. 14 (Number 2). Mackett, R.L., Edwards, M., 1998. The impact of new urban public transport systems: will the expectations be met? Transport. Res. Pol. Pract. 32 (4), 231–245. Martilla, J.A., James, J.C., 1977. Importance-performance analysis. J. Mark. 77–79. Maslow, A.H., 1943. A theory of human motivation. Psychol. Rev. 50 (4), 370. Matzler, K., Sauerwein, E., Heischmidt, K., 2003. Importance-performance analysis revisited: the role of the factor structure of customer satisfaction. Serv. Ind. J. 23 (2), 112–129. Morfoulaki, M., Tyrinopoulos, Y., Aifadopoulou, G., 2007. Estimation of satisfied customers in public transport systems: a new methodological approach. J. Transp. Res. Forum 46. Oliver, R.L., 1980. A cognitive model of the antecedents and consequences of satisfaction decisions. J. Mark. Res. 460–469. Oliver, R.L., 2014. Satisfaction: A Behavioral Perspective on the Consumer: A Behavioral Perspective on the Consumer. Routledge. Olsson, C., Widell, J., Algers, S., 2001. Comfort factors and their influence on bus and rail demand. Traveller values, models and forecasts for local work trips. Vinnova Rapport 2001 (8). ORR, 2012. Costs and Revenues of Franchised Passenger Train Operators in the UK. Office of Rail Regulation. SLL, 2015. Fakta Om SL Och Länet 2014. Fakta Om SL Och Länet. pp. 74. Stradling, S., Carreno, M., Rye, T., Noble, A., 2007. Passenger perceptions and the ideal urban bus journey experience. Transport Pol. 14 (4), 283–292. Tirachini, A., Hensher, D.A., Rose, J.M., 2013. Crowding in public transport systems:

References Abenoza, R.F., Cats, O., Susilo, Y.O., 2017. Travel satisfaction with public transport: determinants, user classes, regional disparities and their evolution. Transport. Res. Pol. Pract. 95, 64–84. Barabino, B., Deiana, E., Tilocca, P., 2012. Measuring service quality in urban bus transport: a modified SERVQUAL approach. Int. J. Qual. Serv. Sci. 4 (3), 238–252. Bierlaire, M., 2003. BIOGEME: a free package for the estimation of discrete choice models. In: Swiss Transport Research Conference. Cats, O., Abenoza, R.F., Liu, C., Susilo, Y.O., 2015. Evolution of satisfaction with public transport and its determinants in Sweden: identifying priority areas. Transport. Res. Rec.: J. Transp. Res. Board (2538), 86–95. Chen, C.-C., Schwartz, Z., 2008. Timing matters: travelers' advanced-booking expectations and decisions. J. Travel Res. 47 (1), 35–42. de Oña, J., de Oña, R., 2014. Quality of service in public transport based on customer satisfaction surveys: a review and assessment of methodological approaches. Transport. Sci. 49 (3), 605–622. Eboli, L., Mazzulla, G., 2009. A new customer satisfaction index for evaluating transit service quality. J. Public Transport. 12 (3), 2. Folkes, V.S., 1984. Consumer reactions to product failure: an attributional approach. J. Consum. Res. 10 (4), 398–409. Friman, M., Fellesson, M., 2009. Service supply and customer satisfaction in public transportation: the quality paradox. J. Public Transport. 12 (4), 4. Fujii, S., Kitamura, R., 2003. What does a one-month free bus ticket do to habitual drivers? An experimental analysis of habit and attitude change. Transportation 30 (1), 81–95. Greeno, D., Joyner, R., 2016. National Rail Passenger Survey, vol. 2016 Spring, London, Transportfocus. Guirao, B., García-Pastor, A., López-Lambas, M.E., 2016. The importance of service

221

Transport Policy 79 (2019) 213–222

M. Börjesson and I. Rubensson effects on users, operation and implications for the estimation of demand. Transport. Res. Pol. Pract. 53, 36–52. Trafikanalys, 2013. Utvärdering Av Marknadsöppningar I Kollektivtrafiken, vol. 2013 TRAFA - Transport Analysis, Stockholm, Trafikanalys 13. Tyrinopoulos, Y., Aifadopoulou, G., 2008. A Complete Methodology for the Quality

Control of Passenger Services in the Public Transport Business. EUT Edizioni Università di Trieste. Van de Velde, D., Beck, A., Norheim, B., Longva, F., Dombi, T., Rudolf, N., Mellor, A., Carvalho, D., Macário, R., Terschüren, K.-H., 2008. Contracting in Urban Public Transport. European Commission – DG TREN.

222