Sie sind auf Seite 1von 15

Transportation Research Part E 41 (2005) 201215

www.elsevier.com/locate/tre

Logistics managers stated preferences for freight service attributes


Romeo Danielis
a

a,*

, Edoardo Marcucci b, Lucia Rotaris

Dipartimento di Scienze Economiche e Statistiche, Universita degli Studi di Trieste, P.le Europa 1, 34127 Trieste, Italy b Istituto di Scienze Economiche, Universita di Urbino, via Sa 42, 61029 Urbino PU, Italy Received 6 July 2003; received in revised form 27 February 2004; accepted 23 April 2004

Abstract This paper reports the results of an adaptive conjoint analysis (ACA) experiment performed in two Italian regions that estimates logistics managers preferences for freight service attributes. Two sets of estimates have been obtained: (a) the utilities associated to each attribute level for each experiment, and (b) the attribute utility via an ordered probit model. Both estimates indicate, on average, a strong preference for attributes of quality (time, reliability, and safety) as opposed to cost. 2004 Elsevier Ltd. All rights reserved.
Keywords: Freight transport; Transport demand; Stated preference; Logistics; Conjoint analysis

1. Introduction In a globalised and competitive environment, characterised by complex logistics and supply chain structures, assessing rms value of service for freight transport is important for many different actors. Carriers, for example, could take advantage of knowledge of a rms willingness to pay for specic service characteristics in order to customize their services, dierentiating them properly and strengthening their own competitive position. Public agencies, sectoral authorities

Corresponding author. Tel.: +39-040-5587076; fax: +39-040-567543. E-mail address: danielis@univ.trieste.it (R. Danielis).

1366-5545/$ - see front matter 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.tre.2004.04.003

202

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

or local governments could acquire valuable information and make better investments decisions or implement regulations to increase the eciency of the transport system and improve the competitiveness of a region. Finally, researchers could use estimates of freight service attributes to feed their transport demand models. Several methodologies can be used to analyse how shippers evaluate and select freight transport services. Mangan et al. (2002) propose the following classication: input-oriented, output-oriented and process-oriented models. This paper implements and tests the potential of an adaptive conjoint analysis (ACA) methodologyan output-oriented type of model in Mangans classication. Conjoint analysis is a stated preference-based technique, largely used and discussed in transport studies on freight and passenger transport (see, for example, Fowkes and Shinghal, 2002a,b; Maier et al., 2002; Maier and Bergman, 2001; Bolis and Maggi, 2002; Fridstrm and Madslien, 2002). Stated preferences are collected via choice experiments, often in face-to-face interviews, and analysed within the framework of discrete choice models. Though widely used in transport studies, applying these to freight transport is still fraught with diculties. Bergkvist (2001) argues that the evaluation of freight transport attributes is still in its infancy compared to passenger transport. Important issues concerning theory, data and sampling, experimental design, modelling, and interview and software still need to be tackled. An important theoretical issue is which framework best accounts for the fact that freight service evaluation and selection happens within a rm, possibly with the involvement of more than one person. Winston (1979) suggests a utility maximising framework in which the logistic managers seeks to maximise his personal utility by performing outstandingly and by devising ecient transport solutions for his company. Should he fail, he may ultimately lose his job. A potential principal-agent problem may arise if the transport manager has dierent information and goals than the company owners. Rose and Hensher (2004) propose an interactive agency conjoint experiment methodology to take into account that choices are made in a context of interdependence both with the shipper and with the carriers. The theoretical framework inuences how the experiment is designed and conducted, which attributes to include, if to specify the mode of transport as an attribute or not, and so on. Various modelling approaches could be used to analyse the data collected in the experiment, ranging from the frequently used logit models to the more recent mixed logit or latent class models. Experiments can be conducted by mail, paper-and-pencil questionnaires, face-to-face interviews, computer-administered software or, possibly, by making use of the world-wide-web. Various computer software existssuch as LASP, MINT, and ACAspecically designed for conjoint experiments. Notwithstanding the large variety of approaches adopted, freight transport stated preference studies recently conducted in Europe found some interesting empirical regularities. For instance, there is a consensus on the importance of reliability in determining mode choice. Estimates of value of time performed so far fall within a reasonably small range (see de Jong, 2000; Bergkvist, 2001). On the other hand, some authors nd relevant dierences among shippers preferences according to the specic sector of activity (Maier and Bergman, 2001), while others see a more homogenous picture (Bolis and Maggi, 2002). Building on this literature, this paper investigates logistics managers preferences for service attributes in two Italian regions, Friuli Venezia Giulia and Marche, located in the North-East and in

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

203

the Centre of Italy, respectively. Experiments are conducted with the ACA software developed by Sawtooth Software Inc. for marketing applications (Sawtooth Software, 19912002). This software is characterised by the following features: (a) It provides utility estimates for each attribute level at the end of each interview. This feature allows us to perform segmentation analyses of the relationship between stated preferences and shipments and rms characteristics. (b) It collects data on respondents choices among hypothetical options to perform standard econometric estimates of discrete choice models. (c) It is adaptive since questions are customised from previous answers, thus making interviews more time-ecient, 1 though aecting design orthogonality. Consequently, two sets of results will be presented and comparedindividual estimates for each shipper, and aggregate estimates for the whole and segments of the sample. The hypothetical alternatives presented in the experiments are consistent with the theoretical principles of the abstract-mode inventory model developed by, among others, Baumol and Vinod (1970). In the abstract-mode inventory model, rms decisions are based on the total logistics cost, which is the sum of the direct shipping cost, in-transit carrying cost, ordering cost, recipients inventory carrying cost, safety stock cost, and loss and damage cost. These costs are product and rm specic and depend on travel time, reliability, and loss and damage but do not require mode specication (Danielis and Rotaris, 2002). Hypothetical options are, therefore, characterised by service attributes only. This model specication has the advantage of avoiding a potential correlation bias between service attributes and the mode variable and is consistent with any unimodal or multimodal shipment type. The disadvantage is that it does not allow us to estimate mode-specic constants to be used in unimodal choice forecasting. Four attributes characterise each hypothetical option: cost, time, reliability and damage. The rst three attributes appear quite frequently in the literature, whereas the fourth one is seldom included, though it resulted very important in our experiments. Methodological and experimental issues are discussed in Section 2. The sample is presented in Section 3, and ACA estimates are reported in Section 4. Econometric estimates are presented in Section 5. Section 6 contains some comments and conclusions. 2. Methodology Face-to-face interviews with logistics managers were administered via a laptop computer equipped with ACA version 4. Each interview consisted of two parts. In the rst, basic information about the rm was collected and managers asked to describe the characteristics of the typical
The ACA manual describes adaptive as follows: The term adaptive refers to the fact that the computeradministered interview is customised for each respondent; at each step, previous answers are used to decide which question to ask next, to obtain the most information about the respondents preferences . . . Questioning is done in an intelligent way; the respondents utilities are continually re-estimated as the interview progresses, and each question is chosen to provide the most additional information, given what is already known about the respondents values. Respondent utilities are available upon completion of the interview.
1

204

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

transport relation on the input and output sides of the company. In the second, with reference to the typical shipment, managers were asked to answer the questions generated by the ACA software. Two independent ACA experiments were generally performed: one regarding the input side and one the output side of the rm. In a number of cases only one experiment could be completed. Each interview lasted about an hour. An ACA experiment is made up of four sets of questions: (1) unacceptable levels; (2) importance of attribute levels; (3) paired-comparison trade-os; and (4) calibrating concepts. Detailed description of each section can be found on the ACA technical paper (Sawtooth Software, 1991 2002). Few theoretical points are discussed in this section. An important characteristic of ACA software is that it estimates the utility associated with each level of each attribute, both for qualitative and quantitative variables. In order to compare preferences across respondents it is therefore necessary to keep attribute levels xed. On the other hand, it is important to customise hypothetical scenarios as much as possible since freight shipments are highly heterogeneous. To meet both requirements, we expressed attribute levels as reported in Table 1. The base scenario is the following: transport cost is equal to the current cost of the typical shipment; travel time is equal to current travel time of the typical shipment; there is no risk of delay and no risk of damage and loss. Variations from the base scenario are expressed in absolute terms for the time attributes, and in percentage terms for transport cost and risk of damage. In the rst step, the ACA software allows the respondent to dene which attribute levels are considered acceptable and which are not. Attribute levels judged unacceptable are not used in the hypothetical proles. In Swaits terminology (Swait, 2001), unacceptable levels in the ACA procedure are assumed to be hard cut-os, that is compensatory behaviour takes place only within limits dened by the respondent. Given the heterogeneity among shipments, we initially made use of this feature to enhance the realism of the hypothetical proles and to improve the informational content of the choice experiments. Excluding unacceptable levels has the further positive eect of reducing the number of attribute levels, therefore reducing the number of questions asked, shortening the length of interviews and avoiding possible fatigue distortions. Respondents made large use of the unacceptability feature (Table 2). A 10% increase in transport cost is considered unacceptable in 49 out of 64 cases, a risk of 3 days delay in 56 cases, a 3 days of more travel time is frequently viewed as unacceptable, as it is a risk of damage or loss equal to 10% of the shipment value.
Table 1 Attributes and attribute levels used in the ACA experiment Attribute # 1 Cost 10% below current cost 5% below current cost Equal to the current cost 5% above current cost 10% above current cost Attribute # 2 Travel time Equal to current travel time 1 more day than current travel time 3 days more than current travel time 5 days more than current travel time Attribute # 3 Punctuality No risk of delay Risk of a 1/2-day delay Risk of a 1-day delay Risk of a 3-day delay Attribute # 4 Damage and loss No risk of damage and loss Risk of damage and loss equal to 5% of shipment value Risk of damage and loss equal to 10% of shipment value

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215 Table 2 Number of times a level was judged unacceptable over 64 experiments Attributes and levels Cost 10% below 5% below Equal 5% above 10% above Travel time Equal 1 more day 3 more days 5 more days Risk of delay Zero risk Risk of a 1/2-day Risk of a 1-day Risk of a 3-day Risk of damage and loss Zero risk Risk of 5% Risk of 10% No. 1 0 0 30 49

205

0 0 54 59

0 0 27 56

0 1 56

The unacceptability feature, however, has two negative side eects. Firstly, a mis-specication error might occur by not showing levels outside the respondents cut-os. Swait (2001, p. 906) states that in fact, any method that does not account for cut-os and their potential violation, if cut-os are in fact being used and violated, will cause parameter bias. Secondly, since respondents dene their individual choice set, individual utility estimates are not homogeneous across interviews. More precisely, the software distributes 400 utility points among attribute levels. The number of attribute levels might dier among rms as respondents may exclude some of them, such that individual utilities are not easily comparable. For this reason, we decided to perform 29 experiments without the unacceptability feature. The central part of the ACA questionnaire is the choice section, called the paired-comparison trade-o questions section. Two proles appear on the computer screen as in Fig. 1. Proles can be described using all attributes at once (full prole), or just some of them (partial prole). We decided to use the full-prole setting for the rst 64 experiments, and partial-prole for the remaining ones, since the former produced unsatisfactory econometric results (as discussed in Section 4). Respondents were asked to compare the two proles and express their preferences using a 9-point scale. As explained in the ACA technical paper, these values are priors that are updated after each graded paired-comparison via a quasi-ordinary least square estimating procedure. Finally, utilities are calibrated with the information obtained in the calibration concept section.

206

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

Which transport service would you prefer? 10% above current cost Zero risk of delay Zero risk of damage and loss 1 day more than the current time 5% below current cost Risk of a 1-day delay Risk of damage and loss equal to 10% of shipment value 3 days more than the current time

Strongly Prefer Left 1 2 3 4 5 6 7 8 Indifferent

Strongly Prefer Right 9

Fig. 1. An example of the a graded paired-comparison question.

3. The sample The sample consists of 65 manufacturing rms, specialised in a variety of products. 35 rms are located in Friuli Venezia Giulia, a region in the north-east of Italy bordering with Austria and Slovenia, while the remaining 30 rms are located in Marche, a region in the centre of Italy. Firms are mostly small or medium sized: 21 have less than 100 employees, 36 have between 100 and 500 employees, and 8 rms have more than 500 employees. All rms buy transport services from third-party providers, while only 15 outsource transport, inventory and/or packaging services. Whenever possible, two separate experiments were performed: one for the shipment of outputs and one for that of inputs. A total of 93 experiments were collected, 65 regarding outputs and 28 regarding inputs. Managers were asked to dene how they organise logistics on the input and output side. On the input side, 38 managers responded that they buy inputs according to production forecasts, 21 according to the actual demand, and only 5 try to follow just-in-time principles. On the output side, 35 rms produce according to sale forecasts, 14 on demand and 16 rms adopt a just-in-time approach. Shipments classied by transit time (less than 6 h, 612 h, 1224 h, and more than 24 h) appear to be almost equally represented, both for inputs and outputs. In 83 out of 93 experiments, road was identied as the mode of transport used for the typical shipment. This information will be used to perform segmentation analysis.

4. ACA estimates For each experiment, the ACA software provides an estimate of the utility associated with attribute levels. Estimates can be analysed separately, aggregated over sample segments, or over the

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215 Table 3 Average utility by attribute level in the with unacceptable and without unacceptable sub-samples Attribute and levels With unacceptables sub-sample Average utility (absolute value) (1) Cost 10% below 5% below Equal to current cost 5% above 10% above Transit time Equal to current transit time 1 day more 3 days more 5 days more Risk of delay No risk of delay Risk of a 1/2-day Risk of a 1-day Risk of a 3-day Risk of damage and loss No risk of damage and loss Risk of 5% Risk of 10% 79.1 49.8 19.5 8.6 4.3 Average utility (dierence) (2) 59.7 30.4 0.0 10.9 15.2

207

Without unacceptables sub-sample Average utility (absolute value) (3) 40.0 32.6 22.9 11.0 0.2 Average utility (dierence) (4) 17.1 9.6 0.0 11.9 22.8

60.2 4.1 3.8 0.0

0.0 56.1 56.4 60.2

53.3 38.6 18.4 1.0

0.0 14.7 34.9 52.2

56.5 15.1 4.3 2.1

0.0 41.3 52.2 54.4

50.8 34.5 19.2 0.5

0.0 16.3 31.6 50.3

106.1 2.2 1.3

0.0 103.9 104.9

52.7 24.2 0.0

0.0 28.5 52.7

entire sample. Table 3 presents average utility estimates over the with unacceptable and without unacceptable sub-samples. Table 3 is to be read as follows. The rst column presents average utility values for each attribute level for the with unacceptables sub-sample. It indicates that, given a total of 400 points, on average 79.1 is the utility associated to a 10% transport cost discount. The utility decreases by almost 20 points (to 49.8) when the 5% reduction is applied. The second column presents the differences in utility associated with each level minus the utility of the base scenario levels. This allows us to make comparisons across attributes. For instance, a transit time increase of one day decreases utility by an amount equal to 1.85 times a transport cost decrease of 5% with respect to the current cost. Due to the unacceptability feature, utility changes rapidly as one deviates from the status quo but then stays almost constant. Without the unacceptability feature (column 3 and 4), however, there is an almost linear relationship between utility and attribute level change. Let us dene the following compensation index (CI) CIXC DU X DU C

208

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

where DUX is the variation of utility associated with either a 1-day increase in travel time, a 1/2 day increase in the risk of delay, or a risk of damage and loss increase equal to 5% of the shipment value, and DUC is the variation of utility associated with a 5% transport cost reduction. Consequently, assuming a linear transport cost function:  CITC (travel time compensation index) represents how many times a 5% transport cost reduction should be applied in order to compensate the disutility of a 1-day increase in travel time.  CIRC (reliability compensation index) represents how many times a 5% transport cost reduction should be applied in order to compensate the disutility of a 1/2 day increase in the risk of delay.  CIDC (damage and loss compensation index) is how many times a 5% transport cost reduction should be applied in order to compensate the disutility of a 5% increase of damage and loss risk. The index is constructed using the levels considered acceptable by all respondents (with only one exception for 5% risk of loss and damage, see Table 2). It therefore represents a reasonable indicator for comparing results obtained with the two data sets. Table 4 presents some aggregate results (rows 2 and 3 can be derived from Table 3). Compensation costs are calculated based on compensation indices, on information on the average transport cost, and on the average value of the commodities transported. The result obtained for the average value of freight travel time is within the range (on the low side) reported in the paper by de Jong (2000) summarising the European value-of-time estimates. Results denote a high aversion to risk of delay: an hour of unexpected delay is valued about 50% more than an hour of expected travel time. Caution should be used in interpreting this results since we did not explicitly specify the exact amount of the risk involved (e.g., 90% or 50%). Our estimates, therefore, measure generic aversion to risk of loss and damage. A more precise specication of risk attributes will be tested in future research projects. Thanks to the availability of utility estimates at the rm level, we used the above indices to compare managers preferences in dierent production sectors. The results reect, among other factors, the inuence of shipments heterogeneity with regard to travel times and to the value of the transported goods.

Table 4 Average compensation indices Compensation index type CITC CIRC 1.41 1.36 1.69 CIDC 3.36 3.42 2.96 Aggregate results 1.80 With-unacceptables sub-sample 1.85 Without-unacceptables sub-sample 1.52 Compensating costs (Euros)

Compensating cost of Compensating cost for Compensating cost of a an hour of travel time the risk of a 1 h delay risk of loss and damage equal to 50 Euros per 1000 Euros shipment value Aggregate results 6.1 9.5 4.1 With-unacceptables sub-sample 6.2 9.2 4.1 Without-unacceptables sub-sample 5.1 11.4 3.6

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

209

Tables 5 and 6 present results for input and output ows, respectively. Let us name cost sensitive those sectors with low indices and quality sensitive those with high indices. Since the sample size within each sector is quite small the results are not statistically robust. However, it is interesting to note that the variability of the compensation index is very high, especially for risk of damage and loss. These ndings are in line with Maier and Bergman (2001). Chemicals and bres, and machines are the sectors which appear to have consistently high quality indices. Furniture scores high on CITC and CIRC, whereas electric and electronic equipment
Table 5 Sectoral results for shipments of inputs Travel time compensation index Chemicals and bres Machines Furniture Wood Textiles and clothing Building Paper and paper products Food and tobacco Electric and electronic equipment Metal products 3.8 2.2 2.1 2.0 1.7 1.7 1.5 1.4 1.3 1.3 Reliability compensation index Furniture Chemicals and bres Paper and paper products Machines Electric and electronic equipment Wood Textiles and clothing Building Food and tobacco Metal products 3.4 2.5 2.1 2.1 1.8 1.8 1.2 1.0 1.0 0.9 Damage and loss compensation index Paper and paper products Chemicals and bres Machines Building Electric and electronic equipment Food and tobacco Wood Furniture Metal products Textiles and clothing 6.5 5.0 4.5 4.1 3.9 3.6 3.3 2.8 2.1 1.4

Table 6 Sectoral results for shipments of outputs Travel time compensation index Food and tobacco Chemicals and bres Electric and electronic equipment Machines Textiles and clothing Building Paper and paper products Rubber and plastic products Minerals Furniture Metal products Wood Leather products Non-metallic minerals 5.2 4.3 3.1 2.6 2.0 2.0 2.0 2.0 1.9 1.7 1.4 1.4 1.0 0.2 Reliability compensation index Food and tobacco Chemicals and bres Leather products Electric and electronic equipment Building Metal products Textiles and clothing Rubber and plastic products Wood Furniture Machines Paper and paper products Minerals Non-metallic minerals 3.0 2.8 2.6 2.3 1.9 1.9 1.8 1.7 1.4 1.3 1.3 1.2 1.0 0.3 Damage and loss compensation index Wood Machines Textiles and clothing Food and tobacco Chemicals and bres Electric and electronic equipment Building Furniture Non-metallic minerals Leather products Paper and paper products Metal products Rubber and plastic products Minerals 6.2 6.1 5.7 5.0 4.5 4.5 4.3 4.1 3.7 3.0 2.9 2.5 2.1 1.2

210

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

presents intermediate values. Among all sectors, metal products shows the lowest quality requirements. With regards to the shipment of outputs, chemicals and bres and food of tobacco are quality sensitive sectors. Electric and electronic equipment ranks higher than in the case of input ows. Among the cost sensitive sectors, non-metallic mineral has the lowest time requirements. Other cost sensitive sectors appear to be minerals, metal products, and paper and paper products which typically involve bulk shipments. Further information on logistics managers preferences can be derived by comparing average compensation indices by type of shipments and rms characteristics. To verify if the dierence in the mean values between two segments of the sample is statistically signicant, a t-test of the dierences in the means is calculated (Table 7).
Table 7 Average compensation indices and the p-value of the t-test of the dierences in the means Input ows CITC CIRC CIDC 1.76 1.71 3.70 Road only 2.22 1.90 3.84 Within 12 h door-to-door travel time 2.61 2.04 3.92 Less than 500 employees 2.20 1.84 4.18 Other than JIT input ows 1.58 1.12 3.66 Other than JIT output ows 2.09 2.05 3.70 Outsourcing transportation only 2.21 1.76 3.84 Output ows 2.60 1.89 4.05 Not road only 1.55 0.91 4.86 More than 12 h door-to-door travel time 1.59 1.49 3.98 More than 500 employees 1.83 1.47 2.51 JIT input ows 2.26 2.68 3.74 JIT output ows 2.41 1.66 4.30 Outsourcing transportation and inventory 1.84 1.94 4.47 p-Value 0.15 0.73 0.62

CITC CIRC CIDC

0.31 0.07 0.24

CITC CIRC CIDC

0.01 0.11 0.93

CITC CIRC CIDC

0.54 0.45 0.03

CITC CIRC CIDC

0.23 0.02 0.94

CITC CIRC CIDC

0.57 0.34 0.38

CITC CIRC CIDC

0.48 0.68 0.36

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

211

The null hypothesis of equal means can be rejected with condence in only few cases. However, we provide hereafter a general comment of the results.  Input vs. output ows: All compensation indices are larger for output than for input ows. Time compensation indices are signicantly dierent. A possible explanation is that customer satisfaction prevails over internal production logistics concerns.  Road-only vs. other-than-road-only shipments (e.g. road and rail): Average time and reliability compensation indices are higher for shipments using road transportation only. The opposite is true for safety. The absolute dierence among means is considerable, although the t-test is statistically signicant only for reliability. The result conrms that road has mainly a timerelated advantage over other modes.  Less than 1-day door-to-door travel time vs. 2-days or more door-to-door travel time: As expected, the perceived importance of travel time and reliability declines the longer the travel time.  Less than 500 employees vs. more than 500 employees: Larger rms are characterised by smaller time reliability and safety compensation indices, that is, cost is more important than the quality. A possible explanation is that larger rms have high bargaining power allowing them to impose higher quality standards on their third-party providers.  Other than JIT input ows vs. JIT input ows: Firms organising their input ows on JIT principles are more sensitive to reliability (with a 0.02 p-value) than those who do not. They are also slightly more sensitive to travel time, whereas there is no dierence with reference to safety.  Other than JIT output ows vs. JIT output ows: Similarly to the case of input ows, JIT output ows are more sensitive to reliability, but the absolute value and statistical signicance is much lower than for inputs. Firms operating on JIT principles are also slightly more sensitive to travel time. There is no dierence with regard to safety.  Outsourcing transportation only vs. outsourcing transportation and inventory: The two types of rms do not show statistically signicant mean values. Outsourcing does not seem to alter quality requirements. So far we have presented results based on individual ACA estimates at the aggregate, sectoral and sample segment level. Although informative, they are ACA point estimates with no information on their statistical signicance. A standard statistical estimate for freight service attributes is presented in the next section.

5. Ordered probit estimate With the data collected during the paired-comparison trade-o section of the ACA test one can estimate an ordered probit model of the choice probabilities of alternative freight services. A logit model was rst tested using data collected with the with unacceptables, full-prole set-up. As reported in Danielis and Rotaris (2002), such data did not produce satisfactory results. Cost, time and reliability were not statistically signicant and showed the wrong sign. Only damage was statistically signicant and showed the expected sign. This result was attributed to a lexicographic bias due to the overwhelming importance of damage overshadowing the role of the other three variables. In other words, choice was determined by damage only since respondents

212

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

were not willing to trade it o with other attributes. Therefore, we decided to turn to a simpler (without unacceptables) partial prole set-up. Interviewees were prompted with a choice between two proles characterized by two attributes at a time. Partial-prole data were collected in 29 experiments providing 239 observations on choices among alternative proles and information on preference intensity. A xed-eect 2 ordered probit model could be estimated. As Greene (2003, p. 736) explains, the model is built around a latent class regression such as y x0 b e where y* is unobserved. What we observe is: y1 y2 y3 . . . y9 if l7 6 y if y 6 0 if 0 < y 6 l1 if l1 < y 6 l2

By assuming e to be normally distributed and normalising its mean and variance respectively to zero and one, the following probabilities are obtained: Pr oby 1jx Ux0 b Pr oby 2jx Ul1 x0 b Ux0 b Pr oby 3jx Ul2 x0 b Ul1 x0 b . . . Pr oby 9jx 1 Ul7 x0 b The resulting log-likelihood function is then optimised. Table 8 presents the results. All variables have the expected sign and are statistically signicant. One can derive (Table 9) the marginal rate of substitution between quality attributes and cost, as well as the implicit compensating cost using the values reported in Table 8, information on the average transport cost and on average value of the commodities transported. Notice that the results are similar to the ACA estimates. Values derived from the ordered probit model show that logistics managers require a higher compensation for accepting an increase in the risk of delay of 1 h than for accepting an increase of 1 h in travel time.

Since each experiment provides a set of repeated observations leading to an auto-correlation bias, following Maier et al. (2002) a corrected model was estimated through a xed-eect variable implemented through a vector of individual specic constants equal to the number of experiments performed.

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215 Table 8 Ordered probit estimates Variables Constant Cost Time Reliability Damage l1 l2 l3 l4 l5 l6 l7 Log-L Log-L(c) Chi-square (32 d.f.) Coecients 1.4079 3.2709 0.3480 0.4767 13.4873 0.3442 0.6893 0.8798 1.0571 1.2783 1.7774 2.2545 t-Statistic 4.25 2.95 6.89 6.54 6.44 4.09 6.70 7.83 8.88 10.45 13.18 14.41 444.00 500.99 113.97

213

Table 9 Marginal rates of substitution and implicit compensating costs in Euros Ordered probit Value Marginal rate of substitution between travel time and cost Marginal rate of substitution between reliability and cost Marginal rate of substitution between loss and damage and cost Compensating cost of an hour of travel time Compensating cost of 1 h deterioration in reliability Compensating cost of a risk of loss and damage equal to 50 Euros per 1000 Euro shipment value 0.1064 0.1457 4.1234 7.1 9.7 4.2 t-Statisticsa 3.41 3.39 3.55

a t-Statistics for the ratio of the coecients are constructed based on a rst-order expansion of a Taylor series for the MRS around the point estimate (see Ortuzar et al., 1994, p. 416).

6. Conclusions An adaptive conjoint analysis was performed, within the theoretical framework of the abstract mode choice model, using a stated-preference database collected by interviewing 65 logistics managers. 93 dierent experiments were conducted with the aid of the ACA 4.0 software. This study estimates logistic managers preference for freight transport service attributes identied as: freight cost, travel time, risk of delay and risk of loss and damage. Two sets of results have been derived: ACA estimates of the utilities associated to each attribute level for each experiment, and econometric estimates of attribute utility within the discrete choice modelling framework. Both estimates indicate, on average, a strong preference for quality attributes over cost. They indicate a high willingness to pay for quality in freight transport services, especially for reliability and safety. Such results indicate a high demand for quality that, given the actual characteristics of

214

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

the Italian freight transport system, will not be easily met due to increasing congestion on the roads and to the inecient intermodal transport system. This result conrms that successful modal shift policies should focus mainly on the quality aspects of the mode to be promoted. The need to improve freight transport quality, in general and of intermodal transport in particular, is well established in political debate. The EU White Paper argues in favour of strategies such as interoperability among railway operators, telematics, ecc. aimed at improving freight transport quality. Political, infrastructural and regulatory decisions are taken at various levels to improve the quality performance of the transport system and the balance among dierent modes of transport. The results presented in this paper can quantify the tradeos implicit in such choices from the point of view of the shippers. But a larger sample size and a more sophisticated econometric technique is needed to improved the predictive power of the model. Moreover the transferability of the results to other geographical areas needs to be tested. The segmentation analysis performed on the ACA estimates and, to a lesser extent, the econometric analysis allowed us to infer some tentative conclusions on the determinants of shippers preferences. The type of good shipped, either as input or output, has a strong inuence on preferences. The shorter the travel time the more important time and reliability become relative to cost. Finally, rm size is negatively related to the intensity of preferences for quality attributes. The adoption of JIT inventory principles increases the preference for reliability, whereas outsourcing arrangements have no inuence on preference structure. Because of recent trends towards an increase in shipment distances, a decrease in rm size, and an increase in the number of rms adopting JIT inventory principles and outsourcing transport services, the resulting trends in freight service attributes valuation are not clear cut, though an increase in the preference for quality relative to cost seems likely. Given the small sample size, these conclusions should be considered tentative, providing hints for successive research eorts, more than for informed decision making at the rm or policy level. However, it is our opinion that estimates of the monetary value of freight transport quality attributes could allow freight transport providers to tailor their services to specic demand needs. The methodology presented in this paper can help us understand how freight service demand is segmented relative to manufacturing sectors, position in the supply chain, and the transport routes used. The estimated model can help simulate and predict market share variations relative to changes in the quality of service supply. Acknowledgement The research has been funded by the Friuli Venezia Giulia Region, by the Ministry of University and Research, and by the National Research Council. Previous versions of the paper beneted from the comments of the participants to the Swiss Transport Research Conference held at Ascona (March 2002) and the Stella meeting held at Siena (June 2002). The help of Jacopo Zotti for the econometric analysis is gratefully acknowledged. All usual caveats apply. References
Baumol, W.J., Vinod, H.D., 1970. An inventory theoretic model of freight transport demand. Management Science 10 (7), 413421.

R. Danielis et al. / Transportation Research Part E 41 (2005) 201215

215

Bergkvist, E., 2001. Freight transportation: Valuation of time and forecasting of ows. Umea Economic Studies 549, Umea University, Sweden. Bolis, S., Maggi, R., 2002. Stated preferenceevidence on shippers transport and logistics choice. In: Danielis, R. (Ed.), Domanda di trasporto merci e preferenze dichiarateFreight Transport Demand and Stated Preference Experiments (bilingual). F. Angeli, Milan. Danielis, R., Rotaris, L., 2002. Shippers preferences for freight transport services: a conjoint analysis experiment for an Italian region. Trasporti Europei 22, 2735. de Jong, G., 2000. Value of freight travel-time savings. In: Hensher, D.A., Button, K.J. (Eds.), Handbook of Transport Modelling. Pergamon. Fowkes, T., Shinghal, N., 2002a. Freight mode choice and adaptive stated preferences. Transportation Research Part E (38), 367378. Fowkes, T., Shinghal, N., 2002b. The leeds adaptive stated preference methodology. In: Danielis, R. (Ed.), Freight Transport Demand and Stated Preference Experiments. F. Angeli, Milan. Fridstrm, L., Madslien, A., 2002. A stated preference analysis of wholesalers freight choice. In: Danielis, R. (Ed.), Domanda di trasporto merci e preferenze dichiarateFreight Transport Demand and Stated Preference Experiments (bilingual). F. Angeli, Milan. Greene, W.H., 2003. Econometric Analysis, Fifth ed. Prentice Hall. Maier, G., Bergman, E.M., 2001. Conjoint analysis of transport options in Austrian regions and industrial clusters. European Research in Regional Science, vol. 11. Reprinted. In: Danielis, R. (Ed.) Freight Transport Demand and Stated Preference Experiments, F. Angeli, Milan. Maier, G., Bergman, E.M., Lehner, P., 2002. Modelling preferences and stability among transport alternatives. Transportation Research Part E 38, 319334. Mangan, J., Lalwani, C., Gardner, B., 2002. Modelling port/ferry choice in RoRo freight transportation. International Journal of Transportation Management 1, 1528. Ortuzar, J. de D., Willumsen, L.G., 1994. Modelling Transport, Second ed. Wiley, Chichester. Rose, J., Hensher, D.A., 2004. Modelling agent interdependency in group decision making. Transportation Research Part E 40 (1), 6379. Sawtooth Software, 19912002. ACA 5.0 Technical Paper. Technical Paper Series. Available from <http:// www.sawtoothsoftware.com/download/techpap/acatech.pdf>. Swait, J.D., 2001. A non-compensatory choice model incorporating attribute cutos. Transportation Research Part B 35 (10), 903928. Winston, C.M., 1979. A Disaggregated Qualitative Mode Choice Model for Intercity Freight Transportation. Dissertation-thesis, MIT Press, Massachusetts.

Das könnte Ihnen auch gefallen