The German value of time (VOT) and value of reliability (VOR) study The survey work

Transcription

1 Researc Collection Working Paper Te German value of time (VOT) and value of reliability (VOR) study Te survey work Autor(s): Dubernet, Ilka; Axausen, Kay W. Publication Date: Permanent Link: ttps://doi.org/ /etz-b Rigts / License: In Copyrigt - Non-Commercial Use Permitted Tis page was generated automatically upon download from te ETH Zuric Researc Collection. For more information please consult te Terms of use. ETH Library

2 Te German value of time (VOT) and value of reliability (VOR) study: Te survey work For possible publication in Transportation Ilka Dubernet Kay W. Axausen August 2018

3 Te German value of time (VOT) and value of reliability (VOR) study: Te survey work For possible publication in Transportation Ilka Dubernet IVT ETH Zuric CH-8093 Zuric pone: Kay W. Axausen IVT ETH Zuric CH-8093 Zuric pone: August 2018 Abstract In 2012 Germany s Federal Ministry of Transport and Digital Infrastructure (BMVI) initiated several projects in preparation of te new Federal Transport Investment Plan (BVWP) Tis included an update of te general metodology and in particular of its cost-benefit analysis (CBA) wic is used to evaluate te effects of undreds of German infrastructure projects under study. As part of te work te first official values of time (VOT) and values of reliability (VOR) for personal and business travel for Germany were estimated. From May 2012 until January 2013 nationwide data of more tan 3000 participants was collected in a combined two-stage revealed and stated preference survey. Tis paper discusses te survey design, reports experience of te field pase and analyses te response beaviour of te sample. Te stated coice experiments address mode, route, time of departure, work place and residential/location coice. Te complex multi-attribute experiments of different types cover various aspects of sort and long-term travel coice attributes wic te respondent as to take into consideration during is decision process. Furtermore overlapping variables of te stated and revealed preference experiments are suitable for a joint estimation of te wole data. Additionally numerous socio-demograpic and attitudinal questions plus te large sample size for business and non-business trips make it a unique dataset offering various aspects of travel beaviour and its valuation to explore. i

4 1 Introduction Te German Federal Ministry of Transport and Digital Infrastructure (BMVI) as recently publised te 2030 Federal Transport Investment Plan (Bundesverkerswegeplan, BVWP), its medium- to long-term investment strategy for te countrys transport infrastructure serving longer distance travel (BMVI, 2016). As part of tis, it updated te overall metodology of its central evaluation tool, cost-benefit analysis (CBA). Te effects of undreds of infrastructure projects in transport policies and investments ad to be evaluated wit it. In tis context one project estimated and recommended values of travel time savings (VOT) and reliability (VOR) for personal and business travel (Axausen et al., 2015a). Te new VOTs were estimated to replace existing values wic were based on values from te BVWP92 and ad not been verified independently since ten (BMVBS, 2003). Te VORs were estimated for te first time, altoug tey are (still) not part of te standard appraisal. Te aim of integrating reliability into te new BVWP is, in line wit practice and science, to make transport systems not only faster but also more reliable (BMVI, 2016). To address tis a researc team around te IVT (ETH Zuric) estimated te VOT and VOR for te BMVI (Axausen et al., 2015a). Anoter BMVI-initiated project calculated VOTs and VORs for freigt, but tis is not subject of tis paper (capter 7, page , BVU et al., 2016). Often, travel time savings make up te largest sare of te gains in CBAs (Mackie et al., 2001). Micro-economic models of time allocation ave been used to derive te valuations of tecnologically constrained time use since Becker (1965), Beesley (1965) and DeSerpa (1971), especially on te value of travel time (e.g. Truong and Henser, 1985; Bates, 1987; Jara-Diaz, 1990). Te current state of practice draws largely upon past Britis (Wardman, 1998; Mackie et al., 2003; Department for Transport, 2015; Wardman et al., 2016), Dutc (Significance et al., 2012), and Scandinavian studies (Börjesson and Eliasson, 2014; Ramjerdi et al., 2010; Fosgerau et al., 2007). Time valuation moved from revealed preference (RP) data to a growing reliance on personalized stated preference (SP) experiments to estimate te VOTs and VORs by using suitably formulated discrete coice models of travel beaviour, especially of route and mode coice. Hence, a personalised stated coice survey is te standard approac (e.g. Small, 2012). Swiss studies followed a variant pat, wen compared to international practice by employing more complex SP experiments including multiple modes and multiple elements of te generalized costs of travel in a series of overlapping coice contexts (Axausen et al., 2004, 2008; Weis et al., 2012; Frölic et al., 2013). Wile tese kind of complex coice surveys ave been applied for some years in Switzerland more recently tey were also acknowledged by researcer of oter national VOT studies (e.g. Hess et al., 2017). 1

5 Te design of te German VOT study builds on te experience of tose studies in Switzerland. As described above te features of te survey are complex multi-attribute experiments of different types covering various aspects of sort and long-term travel coice attributes. During is decision process te respondent as to take all tese attributes into consideration. Tis makes te coice situation utmost realistic. Furtermore overlapping variables of te stated and revealed preference experiments are suitable for a joint estimation of te wole data. Additionally numerous socio-demograpic and attitudinal questions plus te large sample size for business and non-business trips make it a unique dataset offering various aspects of travel beaviour and its valuation to explore. Tis paper presents te design of te German VOT and VOR Study in detail. Furter, it will report on te field pase of te study and analyse te response beaviour and te caracter of te attributes of te sample. 2 Te survey and study Design 2.1 Te survey idea and preceding considerations Te first official value of time and reliability estimation for Germany required utmost diligence. As coice experiments wit multiple attributes are igly complex to understand different types of experiments ad to be developed. A combination of mode coice for te estimation of te VOR would even ave been too complex wic is wy some attributes were included in route coice experiments. Also some modes are not relevant for certain groups as not everyone as a car available. Tis information was gatered beforeand in te RP questions and used for te questionnaire assignment. It was not only important for te estimation to include actual decisions on a single trip or route but also long-term decisions to measure te effect of future trips. Tese long-term decisions gave te respondents te opportunity to implement major canges in teir coices but also include a discounted evaluation of te sum of te sort-term trips. As all experiments included joint variables wic made it possible to estimate te required models togeter by pooling te data and even include te collected RP data as a reference. Tis approac was most useful to create a rater realistic tan only a ypotetical coice situation. It was even possible to estimated a pooled sort-term and long-term model (Dubernet et al., 2018). Fig. 1 sows te steps of te study. As business travel is concentrated in a small sare of te population, a complementary sample of suc travellers was recruited in addition to a population-based sample to acieve an adequate sample size. Business travel was defined as all employment-related travel, but excluding emergency services and driving as work (delivery, 2

6 Figure 1: Study metodology Literature review Small travel time savings DEVELOPMENT RP & SP QUESTIONNAIRE Expert interviews on business travel Non business: dual frame population sample (land line 40%, mobile 60%) 1 ST STEP: COLLECTING RP DATA WITH CATI Business: recruitment from existing online access panel ASSIGNING SP GAMES TO RESPONDENT FROM RP DATA Random selection of reference trip, construction of customized SP game sets 2 ND STEP: COLLECTING SP DATA Non business: paper pencil/online Business: online MODEL ESTIMATION Estimation and validation discrete coice models VOT & VOR DETERMINATION bus, coac drivers, etc.). Te category includes various kinds of business travellers from local craftsmen to lawyers and consultants. Te additional sample of business travellers was recruited wit an on-line access panel. On te basis of te revealed preference (RP) data collected, a stated coice (SC) questionnaire was designed in a second step. Te sort-term SP experiments include mode coice, route coice and route coice and reliability experiments. Tey are described in detail in te following capter (Section 2.2). In order to allow te cross-cecking of te results, tis approac was furter expanded to include long term coice contexts, wic also involve travel as an element (residential and work place location coice), wic also ad been trialled in an earlier Swiss study (e.g. Weis et al., 2012). Te long-term SPs include residential and work place coice situations and are described in Section 2.3. At te end of eac stated coice block all respondents ad te opportunity to mark if one or more of te attributes ad no impact on teir decision in te different coice situations or if all attributes were important to te respondents. Te results are described in Section 4.5. All te SP questionnaires included additional attitudinal questions on risk acceptance, environmental protection and variety seeking in daily life. An descriptive analysis of te attitudinal questions can be found in Section 4.6. In addition to te survey itself two secondary subjects of interest were investigated in te first 3

7 pase of te project for furter validation of te survey approac and design. Te first issue was tat business travellers are sometimes not free to coose te mode or even te route of teir travel due to company policy and tereby cannot contribute valid te SP experiments. Tis was cecked before te main survey by conducting a small-scale qualitative study. 24 decision makers ad been recruited to cover te main regions of Germany as well te range of firm sizes. Wile many firms indeed ad policies in place, te sample reported tat teir employees were free to coose teir routes and in te vast majority also te mode of travel. Te small sare allowed us to go aead wit te SC experiments witout aving to fear a major bias in te results (see Capter 3, Axausen et al., 2015a, for a detailed description). Te oter important issue for te BMVI was te treatment of small travel time savings. Te empirical literature on sort-term canges in travel beaviour sows tat small travel time canges (e.g. less tan five minutes) are often ignored or not perceived by te travellers. Still, in te long-term logic of Cost-Benefit Analyses (CBA) tis effect is irrelevant. To account for te effect would be inconsistent wit assumptions of it and would open te cance to manipulate its results troug dividing or aggregating projects into smaller or larger units. After a literature researc on te state of te art on size and sign effect (for example Daly et al., 2011). It was also tested wit te collected data if te size of te travel time differences offered to te respondents in te SP experiments ad an impact on te valuations. After accounting for te oter non-linearities, te models could not identify suc effects. Tus our recommendation for te BMVI was to follow international practice and to value all savings equally (Ereke, 2016). 2.2 Design of te sort-term experiments In te non-business survey RP data on tree trips undertaken by te respondents were collected in a first step. Te purposes of te RP trips were pre-specified: commuting to work and te trips to te most important sopping and leisure (< 50 km) destinations. Also information on te last long-distance trip over 50 km distance was collected, were, if te latter was ground-based, data on te most recent air trip was also collected. On some occasions te purpose of te reported last long-distance trip was business so tat te non-business sample also contains a small number of business trips. Te rationale beind te approac of collecting information on sort and long distance trips is based on te observation tat te bulk of a persons everyday travel is to a very small number of destinations (Aas et al., 2010a,b; Scönfelder and Axausen, 2010). So witin a relatively sort computer assisted telepone interview (CATI) a good range of trips could be obtained. Business travellers reported teir last tree business trips. Te reference trip of a respondent was cosen randomly but aiming for an overall sare of about 4

8 one tird long-distance trips and two-tirds daily trips, so te reference trip was selected wit a bias to longer trips given teir rarity and te interest of te BVWP in intercity travel. Tis selection was corrected in te analysis troug a re-weigting to matc te distance-purpose distribution observed in te most recent German national travel diary survey (Follmer et al., 2010). Te most recent trip became te reference in te business sample. During te CATI te destinations and te route of te reference trip were geocoded using te software Trip Tracer (DDS Digital Data Services GmbH, 2012). Te gatered trip information was complemented wit te usual socio-demograpic information and information about mobility tools as well as attitudinal questions. Te SP experiments were constructed around te reference trip. Information about te non-cosen options were added. Te non-cosen alternatives and teir attributes were based on information from a number of sources. Door-to-door car travel times were computed based on te average travel times reported by Tom-Tom Stats and a NavTeq network for Germany using te MATSim software package (Horni et al., 2016). Te average car travel cost were calculated based on te 2012 ADAC (General German Automobile Club) price-per-kilometer estimate for an average sized car in eac car segment (range from mini to caravan) (ADAC, 2012). Te travel times, eadways, transfers and prices on public transport including air travel were obtained from te relevant websites wit an internet bot programmed by IVT. Table 1 sows te survey design and attribute levels of te different sort term experiments. Bot samples received te SP experiments witin a maximum of two weeks of aving participated in te CATI. Te business trip sample responded via a web-based survey system. Te nonbusiness sample could coose to respond in a paper-and-pencil form or wit a web-based survey. Respondents in te non-business survey received tree different SP experiments. To keep te response burden low for te business sample respondents only received two types of SP experiments - eiter a mode coice, route coice or departure time coice (reliability) experiment but no long-term SP. So in total, respondents were offered between 16 and 24 coice situations. Eac type of SP experiment contained 8 coice situations. Table 2 sows te 18 possible combinations of te different SP experiments for te non-business sample were eac combination represents one type of questionnaire. Te design of te business sample was basically te same only witout te long-term experiments of ome and work place location coice. In te mode coice experiments te respondent ad to coose between tree modal alternatives. Te modes offered depending on te reported reference mode were eiter walking, cycling, car, local public transport (PT) and te various long distance public transport modes: train, air and te newly deregulated coac option. By tat time te lack of familiarity wit te coac as a 5

9 Table 1: Survey design and attribute levels sort term experiments Attribute Attribute levels Alternative Walk Bike Car PT Bus Plane Mode coice (SP 1) Travel time -30%,-10%,+20% of current state x x x x x x Access/egress time 5%,10%,20% of travel time - - x x x x Congestion/waiting time 5%,10%,20% of travel time - - x x x x Travel cost -20%,+10%,+30% of current state - - x x x x Transfers -1,+/-0,+1 time x x x Headway -1,+/-0,+1 step x x x Sare delayed trips 5%, 10%, 20% - - x x x x Route coice (SP 2) Travel time -30%,-10%,+20% of current state - - x x - - Access/egress time 5%,10%,20% of travel time - - x x - - Congestion/waiting time 5%,10%,20% of travel time - - x x - - Travel cost -20%,+10%,+30% of current state - - x x - - Transfers -1,+/-0,+1 time x - - Crowding low, medium, ig x - - Delay every x. trip 5,10, x x - - Departure time and reliability (SP 3) Travel time -30%,-10%,+20% of current state - - x x - - Access/egress time 5%,10%,20% of travel time - - x x - - Congestion/waiting time 5%,10%,20% of travel time - - x x - - Travel cost -20%,+10%,+30% of current state - - x x - - Transfers -1,+/-0,+1 time x - - Sare arriving early 5%, 10%, 20% - - x x - - Sare arriving on time 100%-sare early-sare delayed - - x x - - Sare arriving delayed 10%,20%,40% - - x x - - Time arriving early 5%,15%,25% of travel time - - x x - - Time arriving late 10%,20%,30% of travel time - - x x - - sceduled long-distance alternative resulted in unreliable estimates and no results for te coac option were reported. Belgiawan et al. (2017) faced similar problems wen comparing te mode coice experiments to oter context depending data and deriving values of time for te coac option. Fig. 2 sows an example of a mode coice experiment wit te tree alternatives bike, public transport and car. 6

10 Table 2: Allocation of SP experiments for non-business sample From RP Reference trip Daily trip (2/3 of all trips) Long distance trip (1/3 of all trips) Allocated SP Travel Mode coice Route Reliability mode coice Long-term No Walk Walk / PT / Car Work place 1 Walk Walk / PT / Car Home location 2 Bike Bike / PT / Car Home location 3 Bike Bike / PT / Car Work place 4 PT Bike / PT / Car PT type 1 Work place 5 PT PT PT type 2 Home location 6 Car Walk / PT / Car Car type 1 Home location 7 Car Car Car type 2 Work place 8 PT Bus / PT / Car PT type 3 Work place 9 PT PT PT type 1 Home location 10 Car Bus / PT / Car Car type 3 Home location 11 Car Car Car type 1 Work place 12 PT PT / Car / Airplane PT type 2 Work place 13 PT PT PT type 3 Home location 14 Car PT / Car / Airplane Car type 2 Home location 15 Car Car Car type 3 Work place 16 Airplane PT / Car / Airplane Airplane type 1 Work place 17 Airplane PT / Car / Airplane Airplane type 2 Home location 18 In te route coice experiments respondents were offered two route alternatives for eiter car or public transport. In te long-term experiments te respondents could coose between teir current work or living situation and a constructed alternative. Fig. 3 sows an example of a car route coice experiment. Te departure time and reliability experiment was formulated as route-departure time coice wit an indication of travel time variability. Tree formats of different complexity were tested, but eac allowing to estimate te mean-variance model of sceduling (Li et al., 2010). All tree formats were retained after te pre-test, as it indicated no clear preference between tem in spite of teir growing complexity. Fig. 4 sows te tree different presentation types of reliability using te example of public transport wereas eac column (PT type 1, PT type 2, PT type 3) represents eac type of experiment. 7

11 Figure 2: Example of mode coice task (SP 1) (Translated from German) Situation 1 Public transport Bike Travel time 0:38 Travel time Car 0:27 Travel time 0:19 Tereof Tereof In-veicle time 0:15 In-veicle time Waiting time 0:06 Time in congestion 0:03 Access time 0:06 Access time 0:03 Costs 2,10 Cange(s) Costs 0:13 0 times(s) 1,70 (17 /mont wit 4 trips) (14 /mont wit 4 trips) Every 10 min Sare delayed 20 % Sare delayed 5 % Coice: Situation 2 Figure 3: Example of car route coice task (SP 2) Farrad (Translated from German) Fartzeit 0:38 Route 1 Travel time tereof In-veicle time Öffentlicer Verker Gesamtzeit 0:25 Gesamtzeit 0:21 davon Fartzeit 0:15 davon farend 0:13 davon Wartezeit 0:06 davon im Stau Route 2 0:03 davon Fußweg 0:04 davon Fußweg 4:28 Umsteigen Kosten 3:50 Travel time tereof 1,60 Walking time Travel cost Wal: Delayed every «id» : Coice 52, trip 4:15 (14 /Monat bei 4 Farten) 0:58 Time in congestion 20 min 0:22 Anteil verspätet 1,80 Kosten In-veicle time 0:16 Färt alle 0:05 5:31 1 Mal (13 /Monat bei 4 Farten) Time in congestion Auto Walking time 10 % 0:18 1 % Anteil verspätet Travel cost 47,80 Delayed every «fragebogen» trip Te travel time reliability was varied by providing different congestion probabilities and average congestion times (delay) for automobile travel and by providing te probability of delays (in

12 minutes) from sceduled arrival time for public transport travel (delays were a percentage of te specified tolerance from te RP survey). Furtermore te mode coice experiments included te sare of delayed arrivals and te route coice experiments te sare of trips delayed. Figure 4: Different types of reliability experiments (SP 3) (Translated from German) PT type 1 trip 1 - type 1 PT type 2 trip 1 - type 2 PT type 3 trip 1 - type 3 Departure time 12:00 Departure time 6:06 Departure time 16:55 Expected travel time 0:40 Expected travel time 2:09 Expected travel time 1:15 tereof in-veicle time 0:26 tereof in-veicle time 143 tereof in-veicle time 1:04 tereof waiting time 0:05 tereof waiting time 0:17 tereof waiting time 0:04 tereof access time 0:09 tereof access time 0:09 tereof access time 0:07 Expected arrival time 12:40 Expected arrival time 8:15 Expected arrival time 18:10 sare 25min early 10 % (55 % of te cases) (75 % of te cases) sare on time 80 % 5 % of te cases 8:05 5 % of te cases 17:35 sare 35min delay 10 % 40% of te cases 8:25 20 % of te cases 18:35 Transfer( s) 0 time(s) Transfer( s) 2 time(s) Transfer(s) 1 time(s) Costs 4.80 Costs 4.80 Costs 1.80 Coice Coice Coice Comparison arrival time distribution 100% 80% 60% 40% 20% 0%.. I I 17:35 18:10 18: Design of te long-term experiments Most value of time studies consider sort term decisions framing experiments around a situation were respondents are presented wit variations to travel time and cost of different modes or routes. Te questions arises if te focus on sort term decisions is te most appropriate? Can for example a commuter vary muc of is daily commute in te sort run or is it peraps more reasonable tat canges in commutes occur because of longer term decisions tat people make suc as were to work or were to live? (Beck et al., 2017). Workplace and residential location influence many oter beavioural coices of travellers as tey define te marginal cost of furter travel and te distances involved. Terefore te focus of several more recent empirical studies sifted to understand and explain everyday travel beaviour as a routine activity canging due to key events suc as residential relocation or workplace decisions. A recent article by Müggenburg et al. (2015) reviews te teoretical framework and te most important studies investigating mobility beaviour in a long-term coice context. 9

13 Table 3: Survey design and attribute levels long-term experiments Attribute Unit Attribute levels Alternative (Current altenative (RP)) (New alternative (SP)) Current New Workplace coice (SP 4) Car commute time (min) 30%, 10%, +20% x x Car commute cost ( /mont) 20%, +10%, +30% x x PT commute time (min) 30%, 10%, +20% x x PT commute cost ( /mont) 20%, +10%, +30% x x Salary before tax ( /mont) 10%, +/ 0%, +10% x x Staff managed (number) 50%, +20%, +100% x x Budget managed (million /year) 50%, +20%, +100% x x Cange of industry needed (yes/no) no, yes no x Cange of company needed (yes/no) no, yes no x Residential location coice (SP 5) Type (ouse/apartment) ouse, apartment x x Size (m 2 ) 20%, +10%, +30% x x Standard (new/renovated/old) new, renovated, old x x Exterior (none/garden/balcony) none, garden, balcony x x Rent/mortgage ( /mont) 20%, +10%, +30% x x Area (urban/suburban/rural) urban, suburban, rural x x Car travel time: - Commute (min) 30%, 10%, +20% x x - Sopping (min) 30%, 10%, +20% x x Car travel costs: - Commute ( /mont) 20%, +10%, +30% x x - Sopping ( /mont) 20%, +10%, +30% x x PT travel time: - Commute (min) 30%, 10%, +20% x x - Sopping (min) 30%, 10%, +20% x x PT travel costs: - Commute ( /mont) 20%, +10%, +30% x x - Sopping ( /mont) 20%, +10%, +30% x x (Scirmer et al., 2014) give a compreensive overview of residential location coice literature and sow tat travel time, commuting and employment canges are significant determinants of coices. 10

14 Trading workplace or residential location, owever, represents a long term coice; it is a decision tat is not made easily and cannot be canged quickly. Te alternatives in te coice situations include travel related variables and in addition a description and variation of work and residence attributes of te respondents. Te respondents were asked to make trade-offs between tese transport and workplace or residence related attributes. Figure 5: Example of work place coice task (SP 4) (Translated from German) Situation 1 Car commute time Current 0:13 New 0:09 Car commute cost 58 / mont 34 / mont PT commute time 0:43 0:36 PT commute cost 54 / mont 32 / mont Salary (before tax) 1600 / mont 1760 / mont Staff managed 4 employees 23 employees Budget managed 1,0 Mio. / year 0,7 Mio. / year Cange industry Cange company No No No Yes Coice: In te workplace games we presented coices via a labelled coice experiment were respondents were asked to coose between teir current workplace and an alternative workplace tat varied in commute times, commute costs, salary and oter workplace attributes. Te attributes and teir variation can be found in Table 3. An example of tis coice task is sown in Fig. 5. A respondent received eigt long-term coice tasks in total. Te residential location games were similar to te workplace ones but wit residential attributes. In addition to te travel cost and time for commute trips te alternatives also sow te time and cost for car and public transport to te nearest sopping location. Te residential attributes regard te appearance and location of te dwelling. All attributes and teir variation can be found in Table 3. An example of tis coice task is sown in Fig. 6. «id» 2 «fragebogen» 11

15 Figure 6: Example of residential location coice task (SP 5) (Translated from German) Situation 1 Type of dwelling Size Standard Exterior space Rent / mortgage Area Car travel time: Commute Sopping Car travel costs: Commute Sopping PT travel time: Commute Sopping PT travel costs: Commute Sopping Current Apartment 120 m 2 Old None 540 / mont rural 0:12 0: :36 0: / mont / mont / mont / mont 0:08 0:13 New House 132 m 2 Renovated Garden 600 / mont rural :32 0: / mont / mont / mont / mont Coice: 3 Response beaviour After te pre-test in May 2012 te two-step survey was carried out in six subsequent waves from July to October For teir participation in te wole survey respondents of te non-business sample receive a lottery ticket ("Aktion Mensc") as an incentive. Respondents of te business sample were recruited by an online access panel and received te usual award for teir participation. Te population based non-business sample was drawn from a dual frame of land-line and mobile numbers (60% and 40%) to ensure tat te growing sare of mobile-only persons are included (ADM Arbeitskreis Deutscer Markt- und Sozialforscungsinstitute e.v., 2014). Te sample was incrementally controlled over te survey period so as to ensure spatial quotas in terms of te German federal states. Before sending«id» out te SP game sets of te first wave 2 (pre-test) te expected «fragebogen» response rates for te paper-pencil and on-line non-business and business sample were predicted following and compared to oter surveys already conducted at te IVT to calculate te number of contacts needed for te aimed-for number of participants Axausen et al. (2015b). In te end all tree observed rates settled in te expected range (see Fig. 7). Te response rate was even iger tan 12

16 for te IVT Swiss value of time study (Axausen et al., 2004). A recruitment rate of over 30% for te CATI and 73% completion rate for te first pases of te RP survey and response rates of 68% (non-business sample) and 91% (business sample) for te second pases in spite of te complexity of te instruments indicate a strong interest for te topic. Figure 7: Response burden and response rates 100 Prior recruitment and incentive Prior recruitment, no incentive No prior recruitment, no motivation call 90.7 German VOT business online German VOT non business paper pencil Response rate German VOT non business online ,000 1,200 1,400 1,600 Ex ante assessment of respondent burden Source: Adaptation from Axausen et al. (2015b) In te RP survey over 4,000 persons completed te questionnaire providing socio- demograpic caracteristics and information on recent trips. During te recruitment pase te data was cecked and controlled so tat tere was a sufficiently large sample of responses for all trip purposes. Including te pre-test data over 2400 non-commercial and over 830 commercial respondents completed te questionnaire including te SP games provided to tem. Hence te sample contains almost 64,000 coice situations (Table 5). Fig. 7, and Table 4 sow tat te response rate of te commercial study is overall iger tan in te non-commercial study as participants were recruited in a business market researc on-line panel. Table 5 gives an overview about te distribution of te number of te completed coice tasks by type of experiment and sample. Sufficient data for all five types of SP experiments was collected. Only te reliability experiments for business trips wit te plane do not contain many cases. As some of te long-distance fligts of te non-business sample were also business trips te number 13

17 Table 4: Response rates Non-business sample Business sample Total sample Paper On- Total Paper On- Total Paper On- Total pen- line pen- line pen- line cil cil cil Pretest Contacts RP (CATI) 200 (30%) 77 (30%) 277 (30%) SP survey 126 (72%) 18 (83%) 144 (73%) 53 (71%) 53 (71%) 180 (72%) 18 (83%) 198 (73%) Main survey Contacts 9,491 1,112 1,112 1,112 10,603 RP (CATI) 3,155 (33%) 864 (76%) 4,003 (38%) SP survey 2,162 (69%) 98 (51%) 2,260 (68%) 786 (91%) 786 (91%) 2,162 (69%) 884 (84%) 3,046 (73%) Total Contacts 10,158 1,112 1,372 1,112 11,530 RP (CATI) 3,355 (33%) 925 (67%) 4,280 (37%) SP survey 2,288 (69%) 116 (55%) 2,404 (68%) 53 (71%) 786 (93%) 839 (91%) 2,341 (69%) 902 (84%) 3,243 (73%) increase to 10 person and 80 completed SPs. However any disaggregated modelling for tis trip purpose, mode and SP experiment as to be done carefully. Fig. 8 sows te response rates by waves sample and medium. As mentioned above respondents in te business sample ave an overall iger response rate (except in te pre-test). Te aimed-for number of participants in te business study was already reaced after wave six so tat in te sevent wave only non-business SP game sets were sent out. In te non-business survey respondents were free to coose between completing te questionnaire on-line or as paper-and-pencil. From almost 3200 respondents wo indicated teir willingness to participate in te SP experiments only 5.6% or 186 person in total cose to complete te questionnaire on-line. Hence, te response rate of te on-line non-business sample varies more tan te oter 14

18 Table 5: Number of completed valid SP games by type of experiment Non-business Business Total Experiment SPs Pers,* Response (%) SPs Pers,* Response (%) (%) Mode coice 12,267 1, , ,706 2, Route coice car 3, , , Route coice put 1, , Reliability car 8,141 1, , ,503 1, Reliability put 5, , , Reliability plane Work place 9,504 1, ,504 1, Home location 8,634 1, ,634 1, Total 50,561 2, , ,953 3, * max, 3 different SP experiments per person Figure 8: Response by sample, medium and wave Response rate [%] Business, online SPs Pers,* Response Nonbusiness, paper pencil Wave 1 (Pretest) Wave 2 Wave 3 Wave 4 Wave 5 Wave 6 Wave 7 Wave Nonbusiness, online samples rates as it s sample is muc smaller. In any case, te response rates for tat medium were te lowest. 15

19 To complete te full on-line SP questionnaire respondents in te business sample needed between 1 minute 18 seconds and 43 minutes 48 seconds and on average 9 minutes 24 seconds. Participants in te non-business survey needed more time, taking between 5 minutes 6 seconds and 58 minutes and on average 17 minutes to fill in te survey questionnaire. As te long-term experiments were only given to te respondents in te non-business sample tey ad to answer to an additional block of 8 different coice situations. Neverteless te absolute number of respondents of te non-business on-line SP survey is about ten times smaller tan te absolute number of participants in te business on-line access panel. Witin two weeks after participating in te CATI respondents received te SP games and te overall time it took tem to send back te questionnaires was recorded. Tose wo did not answer witin 21 days after te send-out received a reminder by tat time. Fig. 9 sows tat te reminder ad only little impact in te two on-line-surveys but did so in te paper pencil one. Figure 9: SP response time in days 100 Response rate [%] Non-business, paper pencil Non-business, online reminder Duration after dispatc[days] Business, online Responses to te two online samples were faster tan in te paper pencil survey. Over alf of te respondents of te online business sample answered witin two days. After one week 80% of respondents ad already completed te SP games. Te reminder ad almost no effect as responses did not substantially increase after it was sent out. In te non-business sample alf of te respondents took a maximum of 4 days to answer te SPs. Most of te respondents (80%) answered witin 14 days. Te reminder increased responses by about 2%. Sending questionnaire by post and back takes more time in general tan answering an on-line survey. First completed SP arrived after 5 days and alf of te questionnaires were sent back 16

20 witin two weeks. Te reminder sent after 21 days motivated an increase between 15% and 20% additional responses after an additional time interval of about four days. 80% of te questionnaires arrived witin 28 days. So it took te respondent almost te twice te time to complete te written questionnaire owever not including te additional time by sending it troug post. Te last questionnaire arrived after 151 days. Besides experience from te pre-test te main study confirmed tat all tree types of reliability presentation delivered equally ig response rates (see Fig. 10). Between te presentation types no clear pattern is recognizable. In te written paper pencil non-business survey te reliability presentation type 2 got most responses wereas respondents in te non-business on-line survey responded best to type tree presentation of reliability. Type 1 turned out to gain most responses in te on-line business survey wereas in total te difference between type 3 and type 1 is about 7%. If one as to decide between te different presentation types it seems reasonable to prefer a grapical presentation of reliability as it is easier for respondents to understand te experiment. Figure 10: Response by presentation of reliability Early/delay - type 1 70 Response rate [%] Non-business paper pencil Non-business online Business online Total Arrival time (tabular) - type 2 Arrival time (grapic) - type 3 Sample 17

21 3.1 Non-traders, lexicograpic beaviour and item non-response Non-traders Non-traders in a stated preference survey are respondents, wo always coose te same alternative across teir coice sets regardless of te available alternatives attributes. Tis may ave several reasons, one of wic is te presence of very strong preference in te context of utility maximisation. Oter reasons could be picking te same alternative for every situation in order to reduce boredom or misunderstanding te questions (Hess et al., 2010). Te total sare of non-traders of all te five different coice experiments in te German VOT study is 25 %. Non-trading occurs far less often in unlabelled coice situations because it doesn t reflect a general preference of te respondent for one of te alternatives. But it still can appen for example if te respondents always cooses te left or rigt alternative (Hess et al., 2010). In te German VOT study te route coice (SP2) and reliability experiments (SP3) are unlabelled coice experiments wereas te mode coice (SP1) and bot long-term experiments - workplace (SP4) and residential coice (SP5) are labelled experiments. In te German VOT study in total 34% of te respondents never varied teir coices in te mode coice experiments (see Fig. 11). Differentiated by mode it can be seen tat te sare of non-traders is iger for car user and persons using non-motorised transport wereas public transport user are more willing to vary teir coices. Non-trading does not necessarily imply inconsistent responses. Hence, te relevant variables, suc as trip distance and purpose and te availability of mobility tools were included in te modelling process rater tan excluding non-traders. In te long-term workplace coice experiment te sare of non-traders was at about 43% wit 14% always coosing te new workplace. In te residential location coice experiment te sare of non-trader was a bit iger wit 51% wit only 7% always coosing te new residential alternative. Overall te sare of non-traders was in te expected range. For te reasons described above te unlabelled experiments (SP 2 and SP 3) include far less non-traders. Overall only 26 respondents (0.1 %) always cose te left or rigt alternative. Wit 22 respondents non-trading occurred mostly in te route coice and departure time experiments (SP3). Over te two labelled experiments 12 respondents always cose te left and 14 always te rigt alternative. Te car route coice experiments included overall more non-traders (18 respondents) tan te public transport coice experiments. 18

22 Sare [%] Te German value of time (VOT) and value of reliability (VOR) study: Te survey work August 2018 Figure 11: Sare of non-traders by mode coice experiment 100% 90% 2% 8% 18% 80% 70% 60% 47% 47% 58% 34% Nontrading 50% 40% 30% Trading 20% 10% 0% Coac PT Plane Car Bike Walk Total Mode Lexicograpic beaviour Lexicograpic beaviour occurs wen over te course of te experiment te respondent evaluates te coice alternatives on a basis of a subset of attributes for example by always coosing te ceapest or fastest alternative (Hess et al., 2010). Te autor s state in te same paper tat true lexicograpic beaviour is ard to detect especially in complex coice situations wit multiple attributes as in tis survey. For example in a coice situation were te respondent always cooses te ceapest alternative te reason wy e did not coose te more expensive one in a certain situation could also be due to more canges during te trip. Also it is sometimes ard to distinguis between lexicograpic and non-trading beaviour (Hess et al., 2010). Neverteless it is interesting to see ow often te respondents decided to always coose te fastest or ceapest alternative in te types of experiments were in tis case te five different types of coice experiments can be even more revealing. In te mode coice experiments (SP1) wic contains data of in total 2062 respondents 390 respondents (19%) always cose te fastest option wic was offered to tem. 18% (376) always cose te ceapest option. 13% (264) always cose te alternative wit te smallest sare of delayed trips. As mentioned above especially in mode coice experiments it is extremely ard to detect real lexicograpic beaviour. For example a person wo always cooses te bike alternative could be eiter a non-trader wit a general preference for taking te bike or could really cose te bike because it is te ceapest option wit zero costs. 19

23 In te car route coice experiments 47% (396 from 841) always cose te fastest in-veicle time and 26% (224) of te respondents te fastest overall travel time. 23% (195) always decided for te ceapest alternative. In te public transport games 19% (60 from 316 respondents) only cose te ceapest option and 35% (112) te fastest. For te route coice and reliability experiments (SP3) te sares are lower. As tese experiments contain even more variables (reliabilty related) tis migt be anoter sign tat no true lexicograpic beaviour is detected. Again in SP4 and SP5 it can not be distinguised if a respondent cooses is current situation or a lexicograpic attribute for example always te igest salary (40%) in SP 4 or te lowest commute time (41%) or rent (56%) in SP5. Neverteless even if it is not possible to see true lexicograpic beaviour te results give us a general insigt for te importance and dominance of certain attributes in te coice set. Also te findings matc wit te ones of te variable importance questions (Section 4.5). Furtermore tey validate te trade-offs generated troug experimental design as most of te respondents did not always coose only one certain low or ig attribute of a coice situation Item non-response Anoter important issues for a survey is item non-response, wic means tat respondents do not answer to a particular unit among te questions. In social sciences tese are often sensitive private information like income or education. Te German VOT study sowed only minor problems wit item non-response, generally te sares of missing values were less 2% or occurred for less important variables. Te questions about being an academic, number of jobs, cildren living in te ouseold and te profession ad a sare of missing values iger tan 20%, but were more or less covered by oter questions, for example by education in general, te number of person living in a ouseold of te respondents or te type of employment (all less tan 1% missing values). However, te variable ouseold income wic was essential for modelling and usually is also one of te more sensitive questions sowed an item non-response rate of only 12.9%. A possible solution to discover patterns or groups beind te non-response at a later stage is to estimate a separate coefficient for missing income. All oter variables in te survey not sown in Table 6 ad item non-response rates of less tan 2%. 20

24 Table 6: Item non-response Do not know Do not say Missing Total Sare (in %) Academic yes/no Number of jobs Profession Cildren <14y in ouseold Number of bikes in ouseold Income Number of cars in ouseold Car availability Descriptive analysis In tis section we present an overview of te collected data using basic descriptive analysis. Te same socio-demograpic attributes (Section 4.1) were collected in bot samples wereas te data of te reference trip (Section 4.2) differs sligtly among te samples. For validation bot samples are compared wit oter German nationwide travel beaviour survey data - te Mobilität in Deutscland 2008 (MiD 2008) (Follmer et al., 2010). Te collected SP data (Section 4.3) is again te same for bot samples only differing by trip purpose. 4.1 Socio-demograpic attributes Table 7 sows te categorical distribution (number of cases) and te percentage sare of te socio-demograpic attributes. Table 7: Socio-demograpic variables SP sample German VOT MiD 2008 Attribute Level N % N % Gender Female 1, , Male 1, , Age group < 18 10, To be continued on te next page 21

25 German VOT MiD 2008 Attribute Level N % N % Age group , , , ,002 7 > , Houseold size , Cildren < 14 y. 0 1, , in ouseold , , , Cildren < 18 y. 0 1, , in ouseold , , , ,043 2 Education Hauptscule Realscule , Abitur , None University degree Yes 1, , No , Missing 1, , Employment Full time 1, , Part time , Education , Job seeking ,087 2 Housewife/-man ,470 7 Retired , Else ,934 6 Net ouseold <1000 (<900= ) ,608 3 income ( ) ,233 9 To be continued on te next page 22

26 German VOT MiD 2008 Attribute Level N % N % Net ouseold , income ( ) , (MID class) ( ) , ( ) , ( ) , ( ) , ( ) , ( ) , ( ) ( ) >6500 (>6600 ) ,506 3 Driver s license Yes 3, , (car & motorcycle) No , Number of cars 0 4,302 7 in ouseold 1 1, , , , ,706 8 > ,342 2 Car availability Always 2, , Sometimes ,278 7 Never ,967 5 PT ticket None 2, , Montly ,045 2 Annual ,471 6 Else , Bancard (train) None 2, % reduction % reduction % reduction Number of bikes in ouseold , > To be continued on te next page 23

27 German VOT MiD 2008 Attribute Level N % N % Federal State Scleswig-Holstein ,464 4 Hamburg ,598 3 Lower Saxony , Bremen ,634 3 Nort Rine-Westpalia , Hesse ,525 9 Rineland-Palatinate ,320 6 Baden-Wuerttemberg , Bavaria , Saarland ,810 3 Berlin ,582 4 Brandenburg ,102 4 Mecklenburg-Western Pom ,481 2 Saxony ,772 6 Saxony-Analt ,192 4 Turingia ,358 4 Te total number of cases differs between te variables as not all 3,243 respondents answered all of te socio-demograpic questions. However, only te valid percentage sare of te levels are sown. Te education category Hauptscule represents te lower secondary education wit 8 (Volksscule) to 10 (Hauptscule) scool years. Realscule represents secondary education wit 10 scool years usually followed by an apprenticesip. Te category Abitur includes te German Abitur or Allgemeine Hocsculreife and Facabitur wic allows te pupil to enter iger education eiter at a university or at a Facocscule ( tecnical college) wit a Facabitur. Te lower bound of te income categories sown is always above te printed value and te upper bound vice versa, e.g. income class represents an income above 1,000 and below 1,500 per mont. In te collected sample older iger educated male respondents working full-time and owning a car are over-represented compared to te population average (Statistisces Bundesamt, 2014; Follmer et al., 2010). One reason could be te over-sampling of business trips but often tis socio economic group is also more likley to participate in surveys (Follmer et al., 2010, e.g.). 24

28 Even toug some variables migt not represent te German population in all cases te data set contains tree weigting variables to acieve representativeness if needed. 4.2 Reference trip Te following Table 8 sows te same parameters as described above for te reported reference trip of te two RP sample. Te parameters for te trip purposes commute, sopping and leisure and long-distance are derived from te non-business sample. Te business sample provides te derived parameters for business trips. Te questions differed sligtly for long-distance and business trips wic results in te different or less variables sown in te table. Te parameters sow tat te collected data lies witin te expected, plausible range. Table 8: Trip variables from RP sample Attribute Level N % Mean Std. dev. Min. Max. Commuting Travel time (min) Frequency (days/week) Arrival time (:mm) :05 01:59 03:30 21:45 Noticable delay (min) Frequency delay Never < 1/mont < 1/week /week /week 50 7 > 2/week Don t know/say 4.5 Regularity of mode Always same coice Switcing Sopping Travel time (min) Frequency (days/week) Arrival time (:mm) :00 Noticable delay (min) To be continued on te next page 25

29 Attribute Level N % Mean Std. dev. Min. Max. Frequency delay Never < 1/mont < 1/week /week /week 14 2 > 2/week 8 1 Don t know/say 4.5 Regularity of mode Always same coice Switcing Leisure < 50 km Travel time (min) Frequency (days/week) Frequency (days/mont) Arrival time (:mm) :00 Noticable delay (min) Frequency delay Never < 1/mont < 1/week /week /week 13 2 > 2/week 20 3 Don t know/say 0 0 Regularity of mode Always same coice Switcing Long distance trip Travel time (min) Arrival time (:mm) :59 Number of long None distance trips witin last 12 mont Don t know/say 11.3 To be continued on te next page 26