The impact of extreme weather conditions on long distance travel behaviour

Transportation Research Part A 77 (2015) 305–319

Contents lists available at ScienceDirect

Transportation Research Part A journal homepage: www.elsevier.com/locate/tra

The impact of extreme weather conditions on long distance travel behaviour Alberto M. Zanni a,⇑, Tim J. Ryley b a b

Transport Studies Group, School of Civil and Building Engineering, Loughborough University, Loughborough LE11 3TU, UK Griffith Aviation, School of Natural Sciences, Nathan Campus, Griffith University, 170 Kessels Road, Brisbane, QLD 4111, Australia

a r t i c l e

i n f o

Article history: Received 15 April 2014 Received in revised form 5 February 2015 Accepted 28 April 2015 Available online 19 May 2015 Keywords: Extreme weather Long distance travel Uncertainty Disruption Climate

a b s t r a c t This paper examines traveller attitudes and responses towards disruption from weather and natural events. An internet-based travel behaviour survey was conducted with more than 2000 respondents in London and Glasgow. Of these respondents, 740 reported information on over 1000 long distance trips affected by extreme weather and natural events over the previous three years. Results show respondents are generally cautious towards travelling during extreme weather events. For a slight majority in the case of air and public transport, and a greater one in the case of car, travellers did not considerably alter their travel plan following the disruption. This was explained not only by less disruptive weather conditions (with heavy snow and volcanic ash being the most disruptive) and impact, but also by the relative importance of their trips. Differences between transport modes were not substantial. Business trips sometimes appeared to give travellers more flexibility, some other times not. Origin and destination did have an impact on reaction, as well as the presence of children whilst travelling. Mixed results were obtained about socio-economic and attitudinal variables. Age in particular did not appear to have a significant effect. Whilst most respondents did acknowledge no external influence in their decision, results showed an important contribution of transport organisation staff, as well as home and mobile internet technology. A limited but still considerable number of respondents indicated their closest friends/relatives as the main influence of their decisions. The results will help planners deploy strategies to mitigate the negative effects of weather related disruptions. Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction, context and objectives of this research The impact of weather and climate on human behaviour and economic activity, from life satisfaction (Maddison and Rehdanz, 2011) to college enrolment (Simonsohn, 2010), passing through migration (Rehdanz and Maddison, 2009) and consumer spending (Murray et al., 2010) has been demonstrated by several empirical studies. The impact of weather and climate on various transport variables such as road safety (Andersson and Chapman, 2011), traffic and/or congestion levels (Hooper et al., 2013; Lam et al., 2008), as well as transport systems resilience and maintenance issues (Jaroszweski et al., 2010; Koetse and Rietveld, 2009) has also been analysed. However, fewer studies have concentrated on the impact of weather on individual travel behaviour. In those cases, including in a number of very recent

⇑ Corresponding author. E-mail addresses: [email protected] (A.M. Zanni), t.ryley@griffith.edu.au (T.J. Ryley). http://dx.doi.org/10.1016/j.tra.2015.04.025 0965-8564/Ó 2015 Elsevier Ltd. All rights reserved.

306

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

papers published on this journal, historical traffic or travel data (the latter from either electronic surveys or diaries) were generally used to analyse and predict the effect of weather on mode and destination choices, public transport ridership or propensity to cycle or walk (Arana et al., 2014; Nosal and Miranda-Moreno, 2014; Sabir et al., 2013; Saneinejad et al., 2012; Singhal et al., 2014). In this paper we focus on the impact of extreme weather (and natural) events on long-distance travel behaviour using data collected from a primary survey carried out in the UK in late 2011/early 2012. Recent studies have provided evidence that extreme weather events will become more frequent as a consequence of climate change (see for example Bronnimann et al., 2012). Whilst last winter (2013/2014) was relatively mild in the UK in temperature terms, it was the wettest since 1910 (Press Association, 2014), and a number of storms hit various regions causing considerable disruptions to buildings and travel infrastructures, with the cost for the railway system estimated at £170 million (Odell, 2014). The winters of 2009/2010 (the coldest for 31 years) and 2010/2011, which were covered in our survey, were also particularly severe in the UK, as well as across Europe (Vajda et al., 2013), causing a number of disruptions not only to travel, but also to electricity and water supplies.1 Travel disruptions, in particular, were estimated to cost £280 million per day to the UK Economy during the 2010 and 2011 severe spells (Prior and Kendon, 2011a,b). In the same period, the eruption of the Eyjafjallajökull volcano in Iceland, and the consequent closure of part of the European airspace, also caused considerable disruptions to UK travellers (Budd et al., 2010; Oxford Economics, 2010). Various passengers’ satisfaction surveys conducted during or before the most disrupted periods, showed a considerable low percentage of travellers satisfied with the way the relevant transport operators had handled the disruptions and their consequences. For example, a survey and focus groups about winter disruptions conducted by the UK Civil Aviation Authority (CAA) in 2011 showed that 74% of air travellers were dissatisfied with the amount of information they received during disruptions, and 81% with the assistance they received (CAA, 2011). Dissatisfaction level were lower for rail passengers in the same period but still considerable at 65% (Passenger Focus, 2010); and whilst the UK Parliament has recently recognised that some progress has been made in the years following the major disruptions of 2009/2010/2011, also as a result of milder weather, the Transport Committee has also recommended an increased attention to passenger welfare (House of Commons, 2013). Finally, although long-distance travel accounts for only 2% of trips in the UK, it clearly covers a much larger proportion of total distance travelled (Dargay and Clark, 2012), and the disruptions and resulting uncertainty are likely to be more severe for travellers, who may find themselves much farther away from home and, possibly, in unfamiliar locations. In this paper we aim to better understand the way travellers change or adjust their plans when facing uncertainty due to extreme weather conditions prior and during travelling for long-distance trips. In particular, we seek to provide an answer to the following research questions: What trip and/or travellers’ characteristics affect the reaction to the disruption the most? What are the main sources of information before or during disruptions? What is their influence on the travellers’ final decisions? In particular, concerning the last point, we also want to see whether referring to a closest friend/relative is an important tool when individuals take decisions under uncertain conditions. Understanding the way travellers react to disruptions has a particular importance to inform transport operators and local and national policy makers in their efforts to avoid or mitigate both the disruptions and their consequences. Disruptions, and the way the system (providers, infrastructure and users) reacts and adapts to them, can also be seen as sort of experiments and opportunities to plan more radical changes in transport policy (Marsden and Docherty, 2013), and their analysis is therefore of particular policy and practical relevance. In our survey, respondents were asked to report information about recent past long distance trips (in the previous three years) that were disrupted by extreme weather events (from heavy snow to extreme heat) or other natural causes (like volcanic ash or landslides). The remainder of this paper is organised as follows. In Section 2 we review the existing literature on the impact of weather on travel behaviour. Section 3 introduces our travel behaviour survey and methodology. Section 4 contains the results of our analysis, whilst Section 5 provides a discussion and conclusions.

2. Literature review on the impact of weather on travel behaviour As noted in the introduction, studies have demonstrated that changes in weather conditions have indeed a strong impact on various transport and travel dimensions. Here we briefly review those studies focusing on the impact on travel behaviour that are more relevant to our research (for a full review see Böcker et al., 2013a; Koetse and Rietveld, 2009). Overall, travel behaviour studies have mainly looked at the relation between weather and propensity to cycle (Böcker et al., 2013b), walk (Clark et al., 2013) and use public transport, either for commuters or leisure travellers (Arana et al., 2014; Sabir et al., 2013). Whilst various studies have demonstrated that, with different magnitudes, public transport ridership generally increases during adverse weather conditions (Aaheim and Hauge, 2005), others have observed the contrary (de Palma and Rochat, 1999; Guo et al., 2007). In Saneinejad et al. (2012), for example, the utility of public transport seems to be positively affected by cloudy and rainy conditions, but only for male travellers, whilst Cools et al. (2010) revealed that snow has the highest impact on commuting trips, whereas extreme temperature, both warm and cold, have the least impact. Kalkstein et al. (2009) analyse the effect of air masses (which include all weather variables such as precipitation levels, temperature, wind speed, cloud cover and humidity) on demand for rail travel in three US cities, and identify a significant 1 For an excellent account of weather- (and volcanic ash) related delays and costs for the EU transport system (also in comparison with the US, China and Australia), as well as passengers protection regulations please see the very recent report by the Mowe-It Project (Mowe-It, 2014).

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

307

relationship between air mass type and ridership, with demand for rail generally higher on days with dry moderate, dry tropical and tropical air mass types, whilst lower with polar and moist polar air mass types. Evidence of the impact on travel behaviour apart from mode choice or trip generation is instead more limited. For example, Böcker et al. (2013a) also looked at the distance travelled and long rather than short term weather trends, Hjorthol (2013) focused on the impact of weather on elderly people activities, Cools and Creemers (2013) analysed the impact of weather forecast on activity–travel behaviour. Of particular relevance for our research is the study by Guiver (2012), that reports the results of a survey looking at the reaction of UK travellers to three different case study extreme weather and nature related disruption events (including the ash crisis and the severe UK 2010 winter). The author identifies a number of possible adjustments to disruption, namely spatial changes (like a change in origin, destination or routes), time changes (frequency, duration, and departure times), modal switches, and other changes like in companionship or activities. And although we focus on a larger range of trip purposes, the impact of weather on the wider travel behaviour can also be drawn from the tourism literature. For example, using a small scale survey among international tourists in New Zealand, Becken and Wilson (2013) show that weather conditions caused ‘substantial’ or ‘some’ travel changes (to timings, route, and recreational activities) to almost 65% of tourists, with a higher percentage of changes for those who more regularly look for weather information. In some cases, however, the weather did not cause disruption to the travel experience (in terms of delays or cancellation), but rather to the recreational activity one (for example, no snow for a ski trip). In another paper targeting international tourists in New Zealand (Jeuring and Becken, 2013), the focus is on the information seeking activity during extreme weather conditions. Their results show that tourism authorities should provide individuals with the adequate information to protect themselves from hazardous conditions (with information tailored in particular to the most vulnerable segments of the populations), but also increase the sense of self-responsibility and protection. Finally, it is useful to observe that when travellers face uncertainty, irrespective of its main cause, suitable resources are needed to predict the likely conditions of the transport system prior to a trip. In this case, personal experience and exchange of information with other trusted individuals can become an effective way of reducing its occurrence and consequences (Barton, 2011; Bonsall, 2004; Guiver and Jain, 2010; Schwanen, 2008). This paper contributes to the existing literature on weather and transport in a number of ways. Firstly, it considers a rich primary dataset on long-distance travel with detailed information about the disrupted trips, the travellers, and their decisions, rather than existing traffic or travel data. As noted by Böcker et al. (2013a: 82) in a recent review, socio-economic characteristics of the travellers, as well as trip (apart from purpose of) and geographical characteristics are ‘‘poorly covered’’ in the available literature. Secondly, the analysis is not limited to mode choice but to a number of other components of travellers’ behaviour under disrupted conditions, namely choices about departure time, itinerary, mode and destinations. Thirdly, a number of extreme weather and natural events are also taken into account. Fourthly and finally, particular attention is given to the travellers’ decision process.

3. Methodology 3.1. Travel behaviour survey In order to look in more details to the disrupted trips, the travellers and their decisions, we had to go beyond the usage of secondary datasets and link traffic or ridership data with weather conditions. Therefore, we organised a primary data collection, where we asked respondents to report information about weather conditions, consequent disruptions, and their reaction to them. We also wanted to cover a relatively large period, including two particularly harsh and disruptive winters in the UK, and also wanted to examine both winter and summer disruptions, so travel diaries data would have been difficult to organise; and given the number of information we wanted to collect and analyse, a relatively large sample was needed, and only a major survey would have had potential to achieve that. The survey also served to collect other pieces of information, which are however not used in this particular paper. There are certainly issues in using self-reported data, namely respondents not remembering exactly, or mis-reporting, the severity of the extreme weather event (and sometime overor under-estimating its impact), as well as other characteristics of the trip. On the other hand, self-reported information was necessary in this instance to precisely understand travellers’ perception of uncertainty and the way they reacted and adjusted to it. This would have not been possible using existing data on weather, demand, road closures, delays and number of cancelled services, for example. Furthermore, it is often complex to link historic weather variables to long-distance travel, as for particularly long trips weather conditions at origin and destination, and along the trip, are often different. Self-reported information provided a solution to this issue. We therefore report in this paper the results of an internet-based travel behaviour survey which was carried out over the summer 2011 and winter 2012 with more than 2000 respondents from the UK cities of London and Glasgow.2 First of all, respondents were asked to indicate the number of times in the previous three years any long distance journey of theirs had been affected by extreme weather or other natural events (these were: heavy snow, freezing temperatures, heavy rain/thunderstorm, flooding – river, flooding – coastal, heat wave, extreme winds/hurricane, landslide, volcanic ash, 2 This survey was carried out by IPSOS Mori using a number of online panels. For information about the sections of the survey not covered in this paper please see Ryley and Zanni (2013), and Zanni and Ryley (2013).

308

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

dense fog). Long distance journeys were defined in the survey as those lasting for more than three hours each way. The 2027 respondents were first asked to indicate the number of disrupted trips, and then faced a number of detailed questions for up to the most recent three of them (if applicable). In total, respondents reported information for over 1400 trips. After thorough checks, it was realised that a number of respondents had reported information for trips which would have normally been below the three-hour threshold, but lasted more than three hours as a consequence of the disruption. For matter of consistency, trips that were less than 50 miles long each way, generally defined by the UK government as being long distance ones for statistical purposes (Dargay and Clark, 2012), were removed from the dataset.3 As a result of this, as well as other checks to remove trips for which respondents had not given complete or precise information, the final dataset was composed of 1125 trips. Information about these trips included origin and destination, reason, method, weather and impact, reaction, importance score and reason of importance, and was reported by a total of 740 respondents (444 respondents reported information about one disrupted trip, 207 about two trips, and 89 respondents about three trips), whilst the remaining respondents did not report any trip. Information about cost of the affected trips was not collected as it was thought that respondents may have had difficulties in recalling the exact total price paid for some (often) complicated multi-modal trips. Also, although cost may have an important impact on travellers’ adaptation strategies, it is often more important to know whether the price paid can be reimbursed by either the operator or a travel insurance, and this is particularly the case for long distance travel. The collected information about the importance of the trips and reason for their importance were able to shed light on this issue. Other collected information included socio-demographic and travel behaviour data, as well as a number of attitudinal questions whose purpose was to profile respondents with respect to their perception of risk whilst travelling under uncertain conditions in general, and weather related uncertainty in particular. The trips dataset contained a number of trips from abroad to the UK. Obviously trips home have special characteristics, the most important of which being that they have to be carried out at some point. Analysing them separately would have however meant having two separate subsamples not easily comparable between each other. Trips home were therefore included, and a variable was considered in the econometric analysis to control for the difference between these and other trips.

3.2. Methods of analysis The large amount of collected information was analysed in various ways. The initial analysis considers respondents’ attitudes and, given the number of answers, these were examined using factorial analysis. Then, particular attention is given to the impact that the particular weather (or natural) extreme event had on the trip, and the way the travellers adapted to the disruptions and their consequences. This is undertaken first through a wide range of descriptive statistics (frequencies and averages), providing a detailed picture of both the disrupted trips and the respondents, and then exploratory econometric analysis in order to detect statistical relationship between respondents’ reaction and both trip (and disruption) characteristics, as well as socio-economic and attitudinal variables. These variables were included in our analysis as they were expected to have an impact on the respondents’ reaction. Importantly, when analysing reaction, we look at trips by air or public transport separately from those by car, as the disruptions and possible reactions can be very different between the two modal groups. This was also reflected in the different types of questions faced by respondents depending on whether their disrupted trips were carried out by car, or by air or public transport. For organised transport (air and public transport) trips, choices may be very limited and travellers may be forced to travel nonetheless on a different day, say if a service has been cancelled, especially if they need to return home. On the other hand, depending on the type of disruption, car travellers may have more choices at their disposal in terms of departure time, routes and destinations. Subsequently, in order to give more insights into respondents’ decisions, our attention turns to the motivations. This was to generate a greater understanding of the main drivers of respondents’ decisions, sources of information and, importantly, whether the travellers actually had any choice given the situations they were involved. This could have been particularly the case for some of those using public transport services, perhaps already on board, which clearly had no alternative options than waiting for the service to resume and eventually reach a destination.

4. Analysis and results 4.1. Sample and disrupted trips The main characteristics of the 740 respondent composing the sample are as follows. 315 respondents were from London, 425 from Glasgow. 310 were male, 430 female, with an average age of 41. 2.5 was the average household size, 188 respondents had children, 1.6 each on average. 428 respondents were employed full-time, 73 part-time, whilst 55 were 3 Respondents indicated both origin and destination of the trips in the questionnaire. Approximate distances between the two points were then calculated using Google Maps. Other European countries, and the US, have slightly different definitions of what constitutes long distance travel for statistical purposes. Please see Dargay and Clark (2012) for more information.

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

309

self-employed. Of the remainder, 37 were unemployed, 55 retired, 51 in education, 13 disabled, and 28 looking after the home. Table 1 illustrates instead the characteristics of the 1125 trips in our database.

4.2. Respondents attitudes to travelling under extreme weather conditions In this section of the questionnaire, respondents were shown 21 attitudinal statements about travelling under uncertain conditions in general, and during extreme weather events in particular, and they were asked to indicate their agreement on a scale from 1 to 5 (1 = Strongly disagree, 2 = Tend to disagree, 3 = Neither agree nor disagree, 4 = Tend to agree, and 5 = Strongly Agree). These are shown in Table 2. Given the large number of statements, it was decided to perform factor analysis in order to make them more manageable and usable for further statistical analysis. First, a correlation matrix was run and a number of positive relationships among the statements were identified. Then, the Kaiser–Meyer–Olkin measure of sampling adequacy (KMO) was calculated, and at 0.755 it showed that factor analysis was appropriate in this case. There were six principal components generated that satisfied the criteria (eigenvalues > 1). The last one is just over 1 and there could be an argument for taking it out for further analysis (it is the most unclear). The last column in Table 2 shows the principal component to which the relevant statement can be associated. Statements are re-ordered in accordance with the component to facilitate understanding. Numbers in italics mean that the relation is negative for that statement. Principal components are then explained after the table. A table showing the Rotated Component Matrix for the 21 statements according to the 6 principal components is available on request. Table 2 shows, among other things, that respondents do no particularly like driving when the weather conditions are not good. Interestingly, the statement to which they disagreed the most was the one about driving when it is very hot, implying that hot temperatures decrease their willingness to drive. On the other hand, hot temperatures seem to make car more appealing than public transport nonetheless. Respondents also appear to be fairly cautious, as they generally start journeys earlier than required in order not to be late, look for information about travel condition before embarking on a journey (but would not be willing to pay for extra information), tend to purchase travel insurance whenever they travel, and do not tend to travel when official warnings are in place. Table 2 also shows that most respondents take decisions on their own, and tend to exchange information with other travellers in uncertain situations more than contacting their friends/relatives for suggestions. Table 2 also shows to which of the six principal components the statements can be referred to. The principal components can be summarised as follows: 1. 2. 3. 4. 5. 6.

Do not mind about uncertain or difficult weather conditions (positive relationship between statements 1–3 and 12). Prefer not travelling, cautious travellers (positive for 5, 7, 13, 15 and 18). Planning and looking up information (positive for 8–10). Prefer travelling by car over public transport due to weather (positive for 4, 6 and 19). Will keep travelling regardless of others or official warnings (positive for 14 and 21). Contact others and willing to pay for extra information/flexible tickets (positive for 11, 17 and 20, and negative for 16).

4.3. Reaction to the disruption: descriptive statistics We now look at the way respondents adjusted to the disruption. In particular we were interested in understanding whether respondents decided to travel nonetheless, and what other adjustments they decided (or somehow were forced) to make, such as postponing their departure to a better moment (in the same day or another day), a change in route, departure point, destination or companionship. As indicated earlier, reaction is analysed and discussed separately depending on whether the disrupted trip was carried out by car or organised transport. In detail, respondents were asked to choose among the options listed in Table 3 in a multi-code answer. The only exception to this was option 4 (I cancelled the trip) which was made as an exclusive code. Table 3 presents descriptive information about the most frequent reactions stated by respondents. For most public transport and air as well as car trips, travellers did not considerably alter their travel plans. In about 30% of case for public transport and 13% for cars, respondents travelled on a different day. In about 9% of cases for both groups of methods respondents cancelled their trips. In the vast majority of cases (1010 trips out of 1125), respondents selected only one option from the eleven listed in Table 3. The information discussed above gives an interesting picture of respondents’ reactions. These reactions need to be examined in more depth as they are likely to have been affected by a number of factors. First of all, they are affected by the type of extreme weather conditions and the consequent type of disruption on the particular trip. Second, they are affected by some other trips characteristics like origin and destination, reason, companionship and importance. Finally, it is also possible that reaction depended on the respondents’ socio-economic characteristics. In order to shed more light on respondents’ behaviour we ran exploratory econometric analysis on the responses illustrated in Table 3.

310

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

Table 1 Disrupted trips main characteristics. Characteristics

No. trips

%

Category

Distance

118 258 225 327 100 26 46 25

10.5 22.9 20.0 29.1 8.9 2.3 4.1 2.2

Within the UK – 50–250 miles Within the UK – 250–500 miles Within the UK – More than 500 miles Either to or within Europe (excl. Russia)a Either to or within North, Central and South America Either to or within Africa Either to or within Far East and Australia/New Zealand Either to or within Russia and the Middle East

Type

336 739 49

30.0 65.7 4.3

Returning home (if within the UK) and/or to the UK from abroad Starting from home (if within the UK) and/or from UK to abroad Other

Purpose

148 319 22 21 382 209 24

13.2 28.4 2.0 1.9 34.0 18.6 2.1

Holiday – weekend Holiday – longer stay Sightseeing Going to a sport/cultural event Visiting friends and relatives Business Other

Companionship

486 299 117 44 14 91 41 17 16

43.2 26.6 10.4 3.9 1.2 8.1 3.6 1.5 1.4

On my own With husband/wife/partner With husband/wife/partner and children With my children With brother/sister With friends With colleagues With parents Other

Trip importance Average score – min 0, max 10

a

6.2 7.5 5.5 6.3 7.1 7.4 7.6 7.1

Holiday – weekend Holiday – longer stay Sightseeing Going to a sport/cultural event Visiting friends and relatives Business Other All trips

Main reason why trips was important and difficult to postpone (collected for 827 trips where score was >5):

165 164 161 137 116 28 24 20 14

20.0 19.8 19.5 19.6 14.0 3.4 2.9 2.4 1.7

I had to return home Important family matter that could not be postponed Important period of the year that I wanted to spend with friends/family Important business matter that could not be postponed I could have not taken leave from work or study in another period Other reasons Important sporting/cultural event I did not want to miss Travel insurance but not sure of reimbursement No travel insurance

Weather or natural event

479 103 119 19 3 12 137 8 211 31 3

42.6 9.2 10.6 1.7 0.3 1.1 12.2 0.7 18.8 2.8 0.3

Heavy snow Big freeze (low temperatures, ice, but no snow)b Heavy rain/thunderstorm Flooding – river Flooding – coastal Heat wave Extreme winds/hurricane Landslide Volcanic ash Dense fog Other

Method of travel

543 19 228 36 216 79 2 2

48.3 1.7 20.3 3.2 19.2 7.0 0.2 0.2

Air Boat Train Coach Car (driver) Car (passenger) Motorcycle Other

Trips that originated and terminated abroad were a significantly low percentage of total trips for all categories (<1%). ‘The Big Freeze’ was the way the UK media and Meteorological authority (Met Office) described the harsh winter weather in 2009/2010 (http://www. metoffice.gov.uk/about-us/who/how/case-studies/big-freeze). b

311

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319 Table 2 The attitudinal statements that relate to travelling in difficult weather conditions.a Statements

Mean

Std. deviation

Principal component

1. I do not mind driving during heavy rain 2. I do not mind driving in snowy conditions 3. I do not mind driving in icy conditions 12. I do not mind uncertain situations (for example when a service is late, or I have to find a new way to get to my destination) when travelling 5. When I find the weather very hot I prefer not to travel at all 7. When I find the weather very cold I prefer not to travel at all 13. As a consequence of previous trips in bad weather conditions, I have become more cautious when travelling 15. When facing travel uncertainty, my reaction is similar to people I know 18. When facing travel uncertainty, I often try to exchange information with strangers in the same situation 8. I tend to start a journey earlier than required in order not to be late 9. I tend to look at a lot of information about travel and weather conditions before starting my journey 10. I tend to look at a lot of information about travel and weather conditions whilst on my journey using portable devices (like satnav, mobile phone, laptop, radio) 4. When I find the weather very hot I prefer travelling by car than using public transport 6. When I find the weather very cold I prefer travelling by car than using public transport 19. I always take travel insurance whenever I travel 14. In bad weather conditions, when other people avoid travelling, I tend to travel nevertheless 21. During bad weather I normally attempt to travel even when an official warning of ‘not to travel unless absolutely necessary’ is in place (for example from the Police, AA, Met Office, Highways Agency or Local Council) 11. I would be willing to pay a fee to have extra information on travel conditions whilst on my journey 16. When facing travel uncertainty, I tend to make decisions about my journey on my own 17. When facing travel uncertainty, I tend to contact my friends or family for suggestions on what to do 20. I tend to buy flexible tickets that would allow me to reschedule/cancel if needed

3.10 2.74 2.52 2.71

1.126 1.186 1.175 1.076

1 1 1 1

2.53 2.72 3.41 3.26 3.24 3.92 3.62 3.22

1.107 1.074 0.987 0.838 1.012 0.942 0.967 1.156

2 2 2 2 2 3 3 3

3.37 3.31 3.53 2.99 2.57

1.133 1.144 1.314 1.015 1.167

4 4 4 5 5

2.31 3.57 2.83 2.68

1.056 0.996 1.061 1.093

6 6 6 6

a The text of the question was: ‘‘To what extent do you agree or disagree with each of the following statements regarding the effects of weather conditions on your general travel habits?’’.

Table 3 Reaction by method of trip. Reaction options

1. I travelled as planned without any major changes to my original plan, apart from those caused by the disruption 2. I delayed the departure time but travelled on the same day 3. I travelled on a different day 4. I cancelled the trip 5. I changed the route 6. I changed the destination 7. I changed the departure point 8. I changed the method of travel 9. I decided to travel on my own rather than with other people 10. I decided to travel with other people rather than on my own 11. Other a

Air/public transport 828 trips

Car 297 trips

Trip no.a

%

Trip no.a

%

271

32.7

142

47.8

116 232 76 69 19 36 69 15 5 34

14.0 28.0 9.2 8.3 2.3 4.3 8.3 1.8 0.6 4.1

32 39 28 56 5 8 14 4 5 16

10.8 13.1 9.4 18.9 1.7 2.7 4.7 1.3 1.7 5.4

Given the multi-code nature of the answer these figures do not always sum to the total of trips and 100%.

4.4. Reaction to the disruption: econometric analysis Given the (large) number of options in the reaction questions, their ‘tick all apply’ nature, and the relatively low number of cases for some of them, we have carried out exploratory analysis on the reactions with the highest frequency only. These were: Reaction 1 ‘‘I travelled as planned. . .’’, which was selected as a solo answer in 254 cases of the 858 air and public transport (organised transport) trips and in combination with others, mainly reaction 2 ‘‘I delayed the departure time but travelled the same day’’, in 17 cases; Reaction 2, selected solo in 95 cases, and in combination with others, mainly reaction 5, in 21 cases; and Reaction 3 ‘‘I travelled on a different day’’, which appeared in 192 cases as a solo response, and 40 times in combination with others, mainly Reaction 5 ‘‘I changed the route’’ or 8 ‘‘I changed method of travel’’, for public transport trips. For car trips we have looked at Reaction 1 (chosen solo in 125 out of 297 cases for car trips, and in 17 cases in combination with other reactions, mainly Reaction 5). Given the yes/no nature of the responses, binomial logit was applied to the choice data. Before that, correlation was expectedly detected between weather conditions and type of disruption; and for this reason, two models were estimated

312

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

for each considered reaction, one with weather (to detect the presence of a weather effect) and trip characteristics (these included reason, companionship, method, location, and importance), and one with the type of disruption for air and public transport (short or long-delay, cancellation of the service, rerouting or other impacts) or car (closed roads, difficult driving conditions, congestion, necessity to call breakdown service, need to abandon car and other) and the same trip characteristics. Socio-economic variables (gender, age, city, number of affected trips in the past, attitudes, household size and working status) were also considered. A simple binomial logit would have been sufficient in the case of the trips being carried out by different individuals. However, some of the trips in our dataset were reported by the same person, possibly displaying some similarity between the reaction to one or another trip and therefore taking the form of an unbalanced panel (repeated choice), with some of the variables, i.e. the socio-economic characteristics, being the same across some of the observations. When data are clustered in this way, it is advisable to correct the estimator in order to take into account possible correlation across observations in the same group. This can be done in different ways. Panel data approaches for binomial observations may consider fixed or random effects, which directly take into account the correlation issue. These were attempted on our data but did not satisfactorily perform, with convergence problems experienced. Another simpler option was to apply a cluster correction to the estimated asymptotic covariance matrix, leaving the maximum likelihood estimator unmodified. Taking into account possible correlation or other forms of connection between observations in the same pre-defined cluster (in our case responses from the same respondents) is part of those methods that are normally referred to as the ‘robust’ covariance matrix estimation, as they attempt to correct for hypothetical or unobserved limitations of the model. Because the correction is applied to the covariance matrix only, and not the estimator itself, correlation is less explicitly taken into account than in more sophisticated panel data approaches (Greene, 2012). The clustered correction, despite its simplicity, did outperform the more complex methods and was therefore deemed to be the most appropriate tool for our exploratory analysis. Finally, given the sometimes lower number of observations for certain variables, some of them were recoded, and therefore the variables used here do not always reflect the details in information showed in the descriptive statistics illustrated earlier. The following three tables show the results of our analysis (carried out with Limdep 10). The first two tables refer to air and public transport trips, the third one to private cars trips. Model 1 in Table 4 considers variables describing the trip, socio-economic characteristics, as well as variables describing the type of disruption caused by the extreme weather or natural event. First of all, both the variables linked to leisure and visiting friends and relatives are significant showing that respondents undertaking trips with these purposes were more likely not to change their plans than those travelling for business reasons, possibly holders of a more flexible ticket, as well as more likely to be insured. Results also show that travellers were more likely not to change their plans in the case of an intercontinental trip (with respect to a UK one), as they probably did not have much flexibility. Unexpectedly, cases of long delays were more likely, with respect to short delays, to determine travellers’ likelihood not to change their plans. On the other hand, in accordance with expectations, respondents had to change their travel plans if instead the disruption caused ‘other impacts’. This included inability to access the starting points and other impacts, like missed connections or broken down trains, likely to cause a much greater disruption than short delays. Also expectedly, respondents who did not experience a cancellation in their service were more likely to be able and willing to stick to their original travel plans. The same applied to those respondents for whom the affected trip was more important, who therefore were more in need to complete it without any major change. In terms of travellers’ characteristics, results show that women were more likely to have been able/willing to not significantly alter their travel plans. The same applied to those with the highest number of trips affected by weather in the past, possibly more experienced than other. More likely to have not altered their plans were also those respondents with the lowest score for the identified factor 6 ‘‘contact others and willing to pay for extra information/flexible tickets’’ (Section 4.2). It is very important to note that by inserting the different factors’ scores among the set of explanatory variables we are not attempting to assess a causal relationship but rather a simple correlation. This is because respondents’ attitudes were detected at the moment they were surveyed, which clearly followed chronologically their disrupted trips. The variables linked to the factors score are therefore likely to be endogenous (being influenced by past disruption rather than affecting them). We have however kept a simpler specification of the model, given the exploratory nature of our analysis, the relatively low number of observations (but high number of collected information), and the difficulties in detecting and treating endogeneity in discrete choice models (Walker et al., 2011). Model 2 contains weather rather than impact variables. The estimated coefficients show that respondents were more likely not to change their plans if their trip was affected by extreme wind, rather than snow, and by snow rather than by the volcanic ash crisis or ‘other’ weather events. This is reasonable as the extreme wind and snow were likely to (or perceived to) be more disruptive than their counterparts. The same applied to those travelling for business rather than other reasons (probably reflecting the difference in importance), and to those travelling by air rather than by coach (possibly reflecting the difference in cost) as well as, intuitively, for those whose trips were more important. Results also show that employed people were more likely to have changed their plans than those not working (long or short terms unemployed, as well as disabled people not working). However, retired people were more likely not to change their travel plans. The same applied to those with the highest scores for Factor 1 and Factor 5, and those not really affected by weather conditions in general (and even by official warnings). These were, therefore, fairly intuitive correlations. The same analysis above was applied to responses to the reaction 2 ‘‘I delayed the departure time but travelled on the same day’’. However, a really limited number of variables were detected as significant determinants of choice, and therefore results are not showed here. In what follows we look at the responses to reaction 3: ‘‘I travelled on a different day’’.

313

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319 Table 4 Air and Public Transport – Reaction 1: Travelled as planned – Cluster adjusted binomial logit – N = 828. Variable (and base category where applicable)

Constant Return home (away from) Leisure (business) Visiting friends and relatives (business) Other reason (business) Train (air) Coach (air) Boat (air) With others (solo) With children (solo) Europe (UK) Intercontinental (UK) Long delay (Short delays) Cancellation (Short delays) Rerouting (Short delays) Other impact (Short delays) Freeze (Snow) Rain/thunder (Snow) Wind (Snow) Ash (Snow) Other weather (Snow) Importance Gender (Women) Age City (Glasgow) Total trips Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6 Household size Employed (Unemployed) Retired (Unemployed) Education (Unemployed) Log-likelihood McFadden R2 * ** ***

1. Impact, trip characteristics and socio-economic

2. Weather, trip characteristics and socio-economic

Beta coeff.

Beta coeff.

1.062 0.149 0.518 0.415 0.182 0.005 0.088 0.878 0.215 0.215 0.293 0.872 0.945 2.544 0.043 1.555

0.059 0.522 0.006 0.070 0.080 0.114 0.073 0.012 0.080 0.051 0.311 0.090 0.334 0.499 0.383

T-value 1.37 0.71 1.68* 1.71* 0.22 0.01 0.18 1.15 0.79 0.78 1.2 2.44** 6.06*** 9.35*** 0.15 7.43***

1.51 2.17** 0.62 0.31 2.17** 1.08 0.60 0.12 0.71 0.47 2.81*** 0.88 1.31 1.05 0.87 354.250 0.323

T-value

0.212 0.013 0.438 0.275 1.338 0.232 0.716 0.024 0.067 0.227 0.166 0.824

0.29 0.07 1.55 1.29 1.95* 0.54 1.64* 0.03 0.29 0.99 0.79 2.74***

0.501 0.198 0.859 1.067 1.354 0.088 0.439 0.001 0.034 0.041 0.204 0.027 0.078 0.010 0.226 0.139 0.127 0.414 0.686 0.340

1.76* 0.61 3.10*** 3.99*** 1.81* 2.47** 2.14** 0.06 0.17 1.05 2.24** 0.3 0.91 0.11 2.43** 1.49 1.47 1.82* 1.69* 0.85 455.705 0.130

Significant at 90% level. Significant at 95% level. Significant at 99% level.

Table 5 shows that the trip being the return leg back home was a significant determinant of travelling on a different day (in the models considering weather). This appears counterintuitive. The same applied, in Model 4, to trips undertaken for other purposes rather than business (possibly because of more flexibility), in both specifications to trips with children and in Europe (with respect to solo trips, possibly to wait for a better day in terms of weather and general conditions to travel with children), and with respect to trips in the UK (where other travel arrangements could perhaps have been made to travel on the same day), again in all specifications, and with respect to intercontinental trips (in the variants of the model considering impact rather than weather). As expected, travelling the following day was sometimes necessary or chosen, in case of unavailability of a similar service in the same day, for those who experienced a cancellation, with respect to those who experienced a relatively short delay. Unexpectedly, however, the same applied to those experiencing short delays rather than long ones and other impacts. In terms of weather, snow rather than wind and rain/thunder, and ash rather than snow, were determinants of travelling on a different day. Those results do reflect the more likely disruptive power of both snow and ash with respect to the other weather conditions. No other variables were detected as significant determinant of the travellers’ reaction apart from the number of affected trips in the past (and only for the weather specification). The same applied with those with the highest score for Factor 3 and Factor 4 (those more likely to plan and look up information, and those who prefer travelling by car over public transport due to weather), possibly both as a result of their previous experience with weather disruptions. Table 6 reveals that those travelling by car for other reasons than business, and passengers rather than car drivers, were more likely to have not considerably changed their travel plans. The same applied, intuitively, to those who did not

314

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

Table 5 Air and Public Transport – Reaction 3 – Travelled on a different day – Cluster adjusted binomial logit – N = 828. Variable (and base category where applicable)

3. Impact, trip characteristics and socio-economic Beta coeff.

Constant Return home (away from) Leisure (business) Visiting friends and relatives (business) Other reason (business) Train (air) Coach (air) Boat (air) With others (solo) With children (solo) Europe (UK) Intercontinental (UK) Long delay (Short delays) Cancellation (Short delays) Rerouting (Short delays) Other impact (Short delays) Freeze (Snow) Rain/thunder (Snow) Wind (Snow) Ash (Snow) Other weather (Snow) Importance Gender (Women) Age City (Glasgow) Total trips Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6 Household size Employed (Unemployed) Retired (Unemployed) Education (Unemployed) Log-likelihood McFadden R2 * ** ***

1.645 0.284 0.150 0.065 0.820 0.601 0.065 1.107 0.275 0.418 0.563 0.613 0.694 1.872 0.327 0.843

0.019 0.125 0.015 0.117 0.057 0.109 0.127 0.228 0.226 0.045 0.119 0.007 0.014 0.637 0.719

4. Weather, trip characteristics and socio-economic

T-value

Beta coeff. *

1.74 1.40 0.41 0.26 1.11 0.86 0.09 1.12 0.93 1.44 2.28** 1.96** 2.91*** 9.70*** 0.96 2.53**

0.52 0.53 1.49 0.57 1.67 0.99 1.15 2.28** 2.20** 0.40 1.13 0.09 0.05 1.37 1.32 361.952 0.273

T-value

1.783 0.373 0.131 0.066 1.546 0.215 0.670 0.314 0.257 0.515 0.487 0.396

2.35** 2.01** 0.42 0.30 2.21** 0.37 1.12 0.29 1.00 2.02** 2.28** 1.35

0.465 0.732 0.702 0.562 0.462 0.042 0.163 0.009 0.054 0.057 0.005 0.119 0.247 0.150 0.073 0.024 0.063 0.131 0.266 0.665

1.25 1.86* 1.93* 2.65** 0.69 1.33 0.81 0.94 0.28 1.66* 0.06 1.27 2.81** 1.67* 0.75 0.25 0.89 0.54 0.61 1.48 491.117 0.112

Significant at 90% level. Significant at 95% level. Significant at 99% level.

experience road closures, but only difficult driving conditions, very low temperatures and rain/thunderstorm, rather than snow, and for whom the trip was more important. No socio-economic characteristics were found to be statistically significant apart from the not working status. There was also a positive relationship between the number of past affected trips and reaction 1 (‘‘Travelled as planned’’), denoting perhaps an experience effect. Finally, a high score for Factor 4 ‘‘prefer travelling by car over public transport due to weather’’ was also correlated with reaction 1, as expected.

4.5. Influence on the decision In this section we attempt to gain a better understanding of travellers’ reaction to disruption by looking at what influenced their decisions. This is done in order to understand if and in what way respondents’ choices were limited by the conditions they experienced, as simply crossing disruption with reaction does not always give a complete picture. Whilst if following a short delay a traveller has said that they travelled on the following day, it is reasonable to say that the travellers had some alternative options, we cannot be so certain for a traveller who decided to travel the following day due to a cancellation, as we do not know whether any other equivalent service was available on the same day. Moreover, analysing influence also enabled us to test the hypothesis that travellers, when facing uncertainty often exchange information with others experiencing the same disruption and, especially refer to their closest friends and relatives (not travelling with them) for advice and suggestions. Table 7 presents some descriptive statistics concerning the question ‘‘which of the following best describes what influenced the decision you took about this journey as a result of the weather or natural event?’’.

315

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319 Table 6 Car – Reaction 1 – Travelled as planned – Cluster adjusted binomial logit – N = 297. Variable

1. Impact, trip characteristics and socio-economic Beta coeff.

Constant Return home (away from) Leisure (business) Visiting friends and relatives (business) Other reasons (business) Car passenger (driver) With others (solo) With children (solo) Abroad (UK) Roads closed (yes/no) Driving difficult (yes/no) Congestion (yes/no) Breakdown (yes/no) Abandon (yes/no) Other impact (yes/no) Freeze (snow) Rain/thunder (snow) Wind (snow) Other weather (snow) Importance Gender (Female) Age City (Glasgow) Total trips Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6 Household size Employed (unemployed) Retired (unemployed) Education (unemployed)

2.267 0.330 0.175 0.012 1.549 0.695 0.456 0.092 0.317 1.012 0.862 0.442 1.429 0.764 0.956

0.230 0.473 0.016 0.216 0.116 0.004 0.073 0.259 0.335 0.085 0.188 0.035 0.773 0.510 0.079

Log-likelihood McFadden R2 * ** ***

2. Weather, trip characteristics and socio-economic

T-value

Beta coeff. **

2.16 1.13 0.34 0.03 2.21** 1.84* 1.13 0.17 0.4 2.54*** 1.86* 1.2 1.34 0.68 1.52

3.26*** 1.12 0.97 0.67 1.92* 0.03 0.49 1.57 1.69* 0.48 1.01 0.26 1.81* 0.58 0.11

T-value

2.273 0.209 0.075 0.176 1.698 0.716 0.499 0.020 0.379

2.33** 0.71 0.16 0.46 2.13** 1.9 1.29 0.04 0.55

0.883 1.095 0.523 0.356 0.230 0.386 0.008 0.158 0.096 0.087 0.082 0.220 0.339 0.132 0.116 0.047 0.662 0.276 0.012

1.81* 2.62*** 1.17 0.21 3.24 1.01 0.55 0.58 1.55 0.63 0.53 1.38 1.80* 0.81 0.7 0.36 1.72* 0.33 0.02

205.580 0.196

205.580 0.159

Significant at 90% level. Significant at 95% level. Significant at 99% level.

Table 7 Main influence behind decision. Reaction options

1. Whilst travelling, I phoned people I know for suggestions on what to do 2. Whilst travelling, I asked other travellers (I did not know beforehand) 3. Whilst travelling, I asked transport staff 4. Whilst travelling, I looked for info on mobile phone 5. I listened out for info on my car radio 6. I looked for info using satnav 7. I looked for info on the internet at home 8. I looked for info on television 9. I exchanged info on travel discussion websites 10. I took decision with the people travelling with me 11. I took the decision on my own 12. Somebody else decided for me 13. Could not decide as I had no alternatives 14. Other reasons

Air/public transport 828 trips

Car 297 trips

Trip no.

%

Trip no.

%

19 19 137 86 15 6 114 42 9 75 189 75 24 18

2.3 2.3 16.5 10.4 1.8 0.7 13.8 5.1 1.1 9.1 22.8 9.1 2.9 2.2

7 2 1 15 81 15 37 13 1 53 59 10 0 3

2.4 0.7 0.3 5.1 27.3 5.1 12.5 4.4 0.3 17.8 19.9 3.4 0.0 1.0

316

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

Table 7 shows that in most instances (23%) for air and public transport users, respondents did not acknowledge any particular influence on their decision as they said they took their decision on their own. This was followed by information obtained by transport staff whilst travelling (16.5%), from internet at home (14%) and from mobile phone whilst travelling (10.4%). 2% did cite car radio as the main source of influence, likely to be on the way to the airport or railway station. For car travel, in most cases respondents said that their decision was mainly affected by information they heard on their car radio (27%). This was followed by ‘‘I took the decision on my own’’ (20%) and ‘‘I took the decision with the people travelling with me’’ (18%). Internet and mobile phone represented an important source of information for a considerable proportion of both types of travellers (mobile phone less so for car drivers, fortunately so), reiterating the importance of ICT (Information and Communications Technology) devices as means to reduce uncertainty whilst travelling, as already demonstrated in other studies (Barton, 2011). For both cases internet at home had a much higher impact than television. In a relatively low number of cases (about 5% in total for both types of trips) respondents stated that referring to their social circle, or other travellers, was the main determinant of their decision. However, in 9% of cases for air and public transport and 18% for cars respondents took their decisions together with the people travelling with them, likely to be among their closest contacts. Evidence gathered from a separate section of the survey (Ryley and Zanni, 2013) shows that in the vast majority of uncertain situations respondents do get in touch with their closest contacts for suggestions and advice, or simply to inform them of the disruption (and the eventual consequences like delays). Here we provide further evidence that discussion and exchange of information with closest social contacts were an important determinant of decisions in a considerable number of trips.

Table 8 Air + Public Transport – Motivation 11 – I took the decision on my own – Cluster adjusted binomial logit – N = 828. Variable (and base category where applicable)

Constant Return home (away from) Leisure (business) Visiting friends and relatives (business) Other reason (business) Train (air) Coach (air) Boat (air) With others (solo) With children (solo) Europe (UK) Intercontinental (UK) Long delay (Short delays) Cancellation (Short delays) Rerouting (Short delays) Other impact (Short delays) Freeze (Snow) Rain/thunder (Snow) Wind (Snow) Ash (Snow) Other weather (Snow) Importance Gender (Women) Age City (Glasgow) Total trips Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6 Household size Employed (Unemployed) Retired (Unemployed) Education (Unemployed) Log-likelihood McFadden R2 * ** ***

Significant at 90% level. Significant at 95% level. Significant at 99% level.

1. Impact, trip characteristics and socio-economic

2. Weather, trip characteristics and socio-economic

Beta coeff.

Beta coeff.

1.277 0.173 0.150 0.217 0.375 0.923 1.175 2.826 0.687 1.422 0.014 0.041 0.018 0.198 0.626 0.408

0.034 0.461 0.002 0.116 0.007 0.017 0.013 0.147 0.116 0.215 0.009 0.105 0.414 0.302 0.537

T-value 1.54 0.78 0.49 0.87 0.49 1.97** 2.47** 2.13** 2.21** 5.07*** 0.05 0.12 0.11 1.09 1.84* 1.71

0.88 2.02** 0.17 0.53 0.16 0.16 0.11 1.45 1.1 1.94* 0.09 1.11 1.63 0.58 1.09 397.415 0.106

T-value

1.221 0.203 0.157 0.178 0.343 0.941 1.116 2.982 0.642 1.339 0.071 0.015

1.54 0.91 0.53 0.71 0.39 2.08** 2.39** 2.33** 2.11* 4.85*** 0.26 0.05

0.561 0.001 0.547 0.036 0.613 0.024 0.438 0.003 0.133 0.001 0.008 0.008 0.169 0.096 0.222 0.003 0.088 0.413 0.297 0.472

1.75* 0.00 1.92* 0.14 0.96 0.64 1.90* 0.24 0.61 0.02 0.08 0.07 1.67* 0.93 1.98** 0.03 0.95 1.63 0.56 0.95 397.451 0.106

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

317

In 3% of air and public transport trips respondents said they did not have alternatives. This answer code was however not showed to respondents as an option but was the product of our recoding of ‘other’ answers. No car respondents directly declared they had no freedom in deciding, which confirms the typical higher flexibility generally enjoyed by car drivers even during disruptions. However, it must be observed that a display of lack of freedom can also be detected among those respondents who selected the option ‘someone else decided for me’ (9% for air and public transport and 3.5% for cars) which, on reflection, was not properly phrased. And although inserting a clear option indicating lack of freedom in the decision would have certainly improved this particular section of our questionnaire, respondents were given the possibility of selecting ‘other’ and better explain the situation, so we do believe answers to this question give a more than satisfactory picture of what influenced respondents’ reaction. We have also performed exploratory econometric analysis on the influence question. Given the ‘tick one only’ nature of the answer the ideal model would have been a multinomial logit. However, given the number of options, and the relatively low number of observations, as well as the necessity of the cluster correction,4 we have run a set of cluster corrected binomial logit models on the motivations most frequently chosen by respondents for both group of methods. Table 8 shows that (in both specifications) travellers using air were more likely to have mainly decided on their own rather than those using all of the other organised modes, who were more reliant on other resources. Surprisingly, those travelling with others were more likely to have decided on their own rather than those travelling solo. The same applied, expectedly, to those travelling with children. In terms of impact, those who suffered rerouting (rather than a relatively short delay) were less likely to indicate they mainly decided on their own. This was expected as rerouting certainly is a more disruptive impact than a short delay and often requires extra information. In terms of weather, solo decisions were more typical of those whose trips were affected by low temperatures and wind, rather than snow, reflecting perhaps the intensity of the event and/or the consequent disruption. Of the socio-economic variables, only gender had a significant influence, indicating that men were more likely than women to have decided (or believed they had) on their own. Two of the attitudinal scores were also significant, indicating an intuitive correlation between the independent decisions and high scores for factor 5 ‘‘keep travelling regardless of others or official warning’’, and slightly less intuitively with the lowest scores for Factor 3 ‘‘Planning and looking up information’’, although the latter could indicate that people planning beforehand did not need extra help whilst travelling, or that people who used to decide by themselves now rely more on planning and other source of information before travelling. The same analysis was also performed on other answers as well as on car trips. However, the estimations produced a limited number of significant variables and are therefore not reported.5 Nonetheless, for air and public transport users, the fact of being in a return leg, in the UK and with impact other than a cancelled serviced were positively linked with information from transport staff as the main determinant of decisions. Given the importance of reaching home, and the absence of language barriers, this is an intuitive result. Understandably, internet was the main source for those suffering a cancellation, due to snow and the ash crisis in particular. Finally for car travel, being younger, male and been travelling with children, were the main determinants of independent decisions. The same applied to the variable indicating that the trip was affected by rain rather than snow. No other variables were significant for car trips.

5. Discussion and conclusions In this paper we have analysed detailed information over more than 1000 trips affected by extreme weather or natural events reported by 740 survey respondents. Through descriptive statistics, multivariate analysis and exploratory econometric models we have examined the trip characteristics, respondents’ attitudes, their reaction the disruption and what influenced their decisions. This has provided planners with elements to inform the deployment of strategies (such as maintenance, information, assistance and forecast systems) likely to mitigate the negative effects of these conditions. The main limitation of our study possibly lies in the reported nature of the information we analysed and the necessity to trust respondents’ perceptions. Reported information has, however, enabled us to provide a good picture of disrupted trips and their reaction. On reflection, more qualitative information would have helped to better understand the nature of the disruption (their timing and location along the trip, for example), as well as the reaction to it. There remains a number of unanswered questions, however, this papers does contribute to the advance in the understanding of the impact of weather on general transport dimensions and travel behaviour in particular. It is useful to remind first that the trips in our dataset were generally judged to be of high importance by respondents (with an average importance score of 7.1 out of 10). The cost of the trip was not among the main reasons for its importance, but the most frequent reason was the trip being the return leg to home, very closely followed by an important family matter that could not be postponed, an important period of the year that the respondent wanted to spend with friends/family, or an important business matter that could not be postponed. More than three-quarters of trips were made by air or public transport services, with the remainder by car, about 20% of the trips were for business reasons. Heavy snow was by far the most disruptive extreme event, with about 30% of trips experiencing long delay and more than 25% being cancelled for air and 4 We experienced a number of convergence issues when the number of variables exceeded a certain magnitude when applying the cluster correction. We believe that the pros of the clustered versions exceed the cons of not using a multinomial specification for the purpose of our analysis. 5 Detailed results are available from the authors upon request.

318

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

public transport, and closed roads and difficult driving conditions as the main impact for car driving, followed by volcanic ash (75% cancellations) and extreme winds/hurricane, with long delays experienced in more than 40% of the affected trips. Before discussing our results, we note that these are not always directly comparable with those from the available literature as no other studies looked at the impact of a wide range of weather variables on long-distance, and with different purposes, trips but rather concentrated on specific types of trips like commuting and/or shopping or leisure one, with obviously different implications. Last but not least, no other studies match the amount of information at our disposal on trips, travellers and their decision process. The attitudinal analysis showed travellers in our sample as generally cautious towards travelling during extreme weather events. However, we cannot precisely say whether the same attitudes were present during their disrupted trips or are actually a product of them. A slight majority in the case of air and public transport, and a greater one in the case of car, did not considerably alter their travel plans as a result of the extreme weather conditions. This was explained not only by less disruptive conditions and impact, but also by the relative importance of their trips, that possibly needed to be completed nonetheless. Some respondents did decide to travel on another day, sometimes despite a relatively minor impact on their journeys, other times, like following a cancellation, most likely as a necessity rather than a choice. Our first research question concerned the trip and travellers’ characteristics more likely to have an impact on the reaction. Exploratory econometric analysis showed that importance of the trips, weather (with snow generally having a greater impact) and type of disruption (cancellations and long delays especially) had, expectedly, a significant impact on the reaction choices. Differences between different transport modes (air, rail and coach) were not substantial, perhaps simply indicating similar conditions and reactions in organised transport. Similarly to earlier studies (for example Cools et al., 2010) the purpose of the trip had indeed an impact on the reaction to the weather disruptions, with business trips sometimes appearing to give travellers more flexibility, some other times not. Origins and destinations did have an impact on reaction, possibly revealing the important impact of familiarity on travel behaviour, as well as the presence of children whilst travelling. Our results show that not considerably altering travel plans did not depend on whether the disrupted trip was a returning leg home or not. The type of disruption was a more important factor. However, returning home was a determinant for those deciding to travel on the following day nonetheless, in general following a cancellation. Mixed results were also obtained about socio-economic and attitudinal variables. Gender did have a little impact, as well as working status and past exposure to disrupted travel, whilst the remaining characteristics did not. And despite previous evidence of a considerable effect of weather (winter in particular) on activities undertaken by the elderly and travel behaviour (Hjorthol, 2013), age did not appear to have a significant effect, although the status of being retired had. Analysis of the main influences behind the decision helped form a better understanding of respondents’ reactions by giving insight into their sources of information and decision process, the objects of our second and third research questions. It also helped assessing the degree of flexibility respondents had following the disruption. Whilst most respondents acknowledged no external influence on their decision, results showed an importance contribution of transport organisation staff as well as both home and mobile internet technology. Whilst precise conclusions on the flexibility enjoyed by respondent could not be made, it does appear most respondents had some alternatives when facing the disruption. Although from the attitudinal analysis people appeared to be more used to exchange information and support with strangers in the same situation, rather than contacting friends and family for suggestions, more people indicated the latter rather than the former as the main influence behind their decision. There was a limited but still considerable number of respondents who indicated their closest friends/relatives as the main influence of their travel adjustments decisions, confirming therefore evidence gathered elsewhere (Barton, 2011; Guiver and Jain, 2010). Transport operators and authorities certainly have plans in force to set operational changes and adjustments during disruption events. However, recent history in the UK has shown that operators and the relevant authorities need to make additional efforts to better understand travellers’ possible reaction to disruption in order to more effectively plan service recovery activities. A considerable understanding of the socio-economic characteristics of typical travellers, the reasons and nature of their trips (without invading privacy) can therefore provide useful insights. The results of our study give further support to the already recognised necessity of supplying timely and precise information, using a variety of methods, to travellers affected by weather related disruptions and general uncertainly whilst travelling, especially for those travellers finding themselves in unfamiliar locations. Travellers do not appear to be keen to pay for further information, and therefore a good provision of information during service failure and recovery should be based on the operators’ investments only and be available to everyone. A larger diffusion of affordable flexible tickets would probably help in mitigating the effect of disruption, by reducing travellers’ stress and anxiety, and therefore also making service recovery operations easier and more efficient. Dealing with business travellers can be complex, given the often high importance of their trips but also the relatively flexibility, in terms of ticketing and costs, that business travellers often enjoy. Finally, although the impact of age was not confirmed by our empirical analysis, the status of being in retirement did so to some extent, and we therefore offer further evidence of the need for devising specific mitigation plans for more vulnerable travellers, who sometimes do not receive any special treatment during disruptions. Acknowledgements The authors would like to acknowledge the UK based Engineering and Physical Sciences Research Council (EPSRC) for funding the ‘FUTURENET’ (Future Resilient Transport Networks) project. The travel behaviour survey was conducted as

A.M. Zanni, T.J. Ryley / Transportation Research Part A 77 (2015) 305–319

319

one component of the FUTURENET project. The authors would like to thank the FUTURENET partners for some useful discussions, and also acknowledge the market research company Ipsos MORI that conducted the data collection. References Aaheim, H.A., Hauge, K.E., 2005. Impacts of Climate Change on Travel Habits: A National Assessment based on Individual Choices. Centre for International Climate and Environmental Research, Oslo. Andersson, A.K., Chapman, L., 2011. The impact of climate change on winter road maintenance and traffic accidents in West Midlands, UK. Accid. Anal. Prev. 43, 284–289. Arana, P., Cabezudo, S., Peñalba, M., 2014. Influence of weather conditions on transit ridership: a statistical study using data from Smartcards. Transport. Res. Part A: Policy Pract. 59, 1–12. Barton, D., 2011. People and technologies as resources in times of uncertainty. Mobilities 6, 57–65. Becken, S., Wilson, J., 2013. The impacts of weather on tourist travel. Tourism Geogr. 15, 620–639. Böcker, L., Dijst, M., Prillwitz, J., 2013a. Impact of everyday weather on individual daily travel behaviours in perspective: a literature review. Transp. Rev. 33, 71–91. Böcker, L., Prillwitz, J., Dijst, M., 2013b. Climate change impacts on mode choices and travelled distances: a comparison of present with 2050 weather conditions for the Randstad Holland. J. Transp. Geogr. 28, 176–185. Bonsall, P., 2004. Traveller behaviour: decision-making in an unpredictable world. Intell. Transport. Syst. 8, 45–60. Bronnimann, S., Martius, O., Von Waldow, H., Welker, C., Luterbacher, J., Compo, G.P., Sardeshmukh, P.D., Usbeck, T., 2012. Extreme winds at northern midlatitudes since 1871. Meteorol. Z. 21, 13–27. Budd, L., Griggs, S., Howarth, D., Ison, S., 2010. A fiasco of volcanic proportions? Eyjafjallajökull and the closure of European airspace. Mobilities 6, 31–40. CAA, 2011. Aviation’s Response to Major Disruption. Civil Aviation Authority (CAA), London. Clark, A., Scott, D., Yiannakoulias, N., 2013. Examining the relationship between active travel, weather, and the built environment: a multilevel approach using a GPS-enhanced dataset. Transportation, 1–14. Cools, M., Creemers, L., 2013. The dual role of weather forecasts on changes in activity-travel behavior. J. Transp. Geogr. 28, 167–175. Cools, M., Moons, E., Creemers, L., Wets, G., 2010. Changes in travel behavior in response to weather conditions: whether type of weather and trip purpose matter. Transport. Res. Rec.: J. Transport. Res. Board 2157, 22–28. Dargay, J., Clark, S., 2012. The determinants of long distance travel in Great Britain. Transport. Res. Part A: Policy Pract. 46, 576–587. de Palma, A., Rochat, D., 1999. Understanding individual travel decisions: results from a commuters survey in Geneva. Transportation 26, 263–281. Greene, W.H., 2012. LIMDEP – Version 10 – Econometric Modeling Guide. Econometric Software, Inc., New York. Guiver, J., 2012. Travel Disruption: Three Case Studies. Report to Disruption Project – University of Central Lancashire. Guiver, J., Jain, J., 2010. Grounded: impacts of and insights from the volcanic ash cloud disruption. Mobilities 6, 41–55. Guo, Z., Wilson, N.H.M., Rahbee, A., 2007. Impact of weather on transit ridership in Chicago, Illinois. Transport. Res. Rec., 3–10 Hjorthol, R., 2013. Winter weather – an obstacle to older people’s activities? J. Transp. Geogr. 28, 186–191. Hooper, E., Chapman, L., Quinn, A., 2013. The impact of precipitation on speed–flow relationships along a UK motorway corridor. Theor. Appl. Climatol., 1– 14 House of Commons, 2013. Ready and Waiting? Transport Preparations for Winter Weather – Tenth Report of Session 2013–14. Transport Committee, House of Commons, London. Jaroszweski, D., Chapman, L., Petts, J., 2010. Assessing the potential impact of climate change on transportation: the need for an interdisciplinary approach. J. Transp. Geogr. 18, 331–335. Jeuring, J., Becken, S., 2013. Tourists and severe weather – an exploration of the role of ‘Locus of Responsibility’ in protective behaviour decisions. Tourism Manage. 37, 193–202. Kalkstein, A.J., Kuby, M., Gerrity, D., Clancy, J.J., 2009. An analysis of air mass effects on rail ridership in three US cities. J. Transp. Geogr. 17, 198–207. Koetse, M.J., Rietveld, P., 2009. The impact of climate change and weather on transport. Transport. Res. Part D: Transp. Environ. 14, 205–221. Lam, W.H.K., Shao, H., Sumalee, A., 2008. Modeling impacts of adverse weather conditions on a road network with uncertainties in demand and supply. Transport. Res. Part B: Methodol. 42, 890–910. Maddison, D., Rehdanz, K., 2011. The impact of climate on life satisfaction. Ecol. Econ. 70, 2437–2445. Marsden, G., Docherty, I., 2013. Insights on disruptions as opportunities for transport policy change. Transport. Res. Part A: Policy Pract. 51, 46–55. Mowe-It, 2014. Policy Guidelines for User Protection, Long Term Operational Resilience and Cross-Modal Transferability between Air and Surface-based Transport Sectors. Management of Weather Events in Transport System (Mowe-It) Project. Deliverable 8.2 – 27/10/2014. Murray, K.B., Di Muro, F., Finn, A., Popkowski Leszczyc, P., 2010. The effect of weather on consumer spending. J. Retail. Consum. Serv. 17, 512–520. Nosal, T., Miranda-Moreno, L.F., 2014. The effect of weather on the use of North American bicycle facilities: a multi-city analysis using automatic counts. Transport. Res. Part A: Policy Pract. 66, 213–225. Odell, M., 2014. Network Rail Boss puts Storms Cost at £170m. The Financial Times (25.02.14). Oxford Economics, 2010. The Economic Impacts of Air Travel Restrictions Due to Volcanic Ash. A Report Prepared for Airbus, Oxford Economics, London. Passenger Focus, 2010. Delays and Disruptions – Rail Passengers have their Say. Passenger Focus, London. Press Association, 2014. UK Suffers Wettest Winter on Record. The Guardian (20.02.14). Prior, J., Kendon, M., 2011a. The disruptive snowfalls and very low temperatures of late 2010. Weather 66, 315–321. Prior, J., Kendon, M., 2011b. The UK winter of 2009/2010 compared with severe winters of the last 100 years. Weather 66, 4–10. Rehdanz, K., Maddison, D., 2009. The amenity value of climate to households in Germany. Oxford Econ. Pap. 61, 150–167. Ryley, T.J., Zanni, A.M., 2013. An examination of the relationship between social interactions and travel uncertainty. J. Transp. Geogr. 31, 249–257. Sabir, M., van Ommeren, J., Rietveld, P., 2013. Weather to travel to the beach. Transport. Res. Part A: Policy Pract. 58, 79–86. Saneinejad, S., Roorda, M.J., Kennedy, C., 2012. Modelling the impact of weather conditions on active transportation travel behaviour. Transport. Res. Part D: Transp. Environ. 17, 129–137. Schwanen, T., 2008. Managing uncertain arrival times through sociomaterial associations. Environ. Plann. B 35, 997–1011. Simonsohn, U., 2010. Weather to go to college. Econ. J. 120, 270–280. Singhal, A., Kamga, C., Yazici, A., 2014. Impact of weather on urban transit ridership. Transport. Res. Part A: Policy Pract. 69, 379–391. Vajda, A., Tuomenvirta, H., Juga, I., Nurmi, P., Jokinen, P., Rauhala, J., 2013. Severe weather affecting European transport systems: the identification, classification and frequencies of events. Nat. Hazards, 1–20. Walker, J.L., Ehlers, E., Banerjee, I., Dugundji, E.R., 2011. Correcting for endogeneity in behavioral choice models with social influence variables. Transport. Res. Part A: Policy Pract. 45, 362–374. Zanni, A.M., Ryley, T., 2013. Discrete choice and social networks: an analysis of extreme weather uncertainty in travel behaviour. In: ENVECON Conference, London. The Royal Society (15.03.13).