Value of travel time savings for shopping trips in Switzerland

Alexander Erath, IVT, ETH Zürich

Conference paper STRC 2006

-1-

Value of travel time savings for shopping trips in Switzerland Alex Erath ETH Zürich Switzerland Phone: +41 44 633 3092 Fax: +41 44 633 1057 email: [email protected] March 2006

Abstract The Swiss study of the value of travel time savings (VTTS), conducted by König, Axhausen und Abay (2004) reported unexpectedly high values for shopping trips from early experimental stated preference experiments, which were finally not included in the main study. Therefore a study focusing only shopping travel was conducted to see, if these results could be replicated. This study estimates VTTS for shopping trips which are significantly higher than those in other studies. Values between 53 and 110 CHF/h are observed depending on the preferred supermarket. The effect of the preference for shopping in low price supermarkets on the VTTS is much stronger than income or the travel distance. In contradiction to the findings of the study by Koenig et al. the value of travel time savings decreases with longer distances. Further work which considers the travel and shopping expenditures separately besides shopping location preferences could probably give an even better understanding of VTTS of shopping travel.

Keywords Value of travel time savings, shopping trips, Discrete Choice Modelling, Swiss Transport Research Conference, STRC 2006, IVT, ETH Zürich

-2-

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

1. Estimating the value of travel time savings 1.1

Approach and the resent studies

The 2005 market entry of the German discount supermarket Aldi in Switzerland led to uncertainties about possible demand shifts from established supermarket chains such as Migros or Coop. It is obvious that not only the price of goods effects consumer behaviour as the location choice is certainly also affected by quality and accessibility of supermarkets. While price and quality differences are obvious (Vock 2004) it is difficult to give a reliable value of the driving costs as part of the generalised cost of a shopping trip since the estimation of value of travel time saving is a complex task. As the Swiss study of value of travel time savings (VTTS), conducted by König, Axhausen und Abay (2004) reported unexpectedly high values for shopping trips from early experimental stated preference (SP) experiments, which were finally not included in the main study. Therefore a study focusing only shopping travel and extending the experimental statepreference approach was conducted, as part of an MSc thesis.

3

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

2. Survey design and descriptive analysis 2.1

Survey area

The survey was conducted at supermarkets in the tri-national region of Basel, one in each country. To obtain an interesting sample locations with different clientele and price level were chosen; however the supermarkets had to provide a wide range of products which covers the demand of typical weekly shopping tour. On two Saturdays 386 Swiss costumers of M-Park Dreispitz (CH) (Migros), Marktkauf Weil am Rhein (Germany) and Géant St. Louis (France) were interviewed when they were queuing for the cash desk. Since answering stated preference scenarios needs some time and concentration the interviewee received a prepaid envelope with a more detailed questionnaire containing the stated preference experiments. 110 persons returned complete questionnaires.

2.2

Survey design

The SP-scenarios were defined by the variables travel time, price of goods and quality of supermarket in four grades (Figure 1), which in the literature were regularly significant for shopping destination choices.(Baltas et al. (2003), Petrevu et al.(2000), Simma et al. (2004), Thang et al. (2003), Davies et al. (2001)) Figure 1

Design of the stated-choice experiments

Supermarket A

Supermarket B

Trave time to A: 10 minutes

Trave time B: 45 minutes

Price of goods: 400 CHF

Price of goods: 200 CHF

Quality of shop: 4 ‚Top Quality’

Quality of shop: 1 ‚Discounter’s paradise

‡

← Your Choice →

‡

The experimental design aims to create situations whose attribute values are not or only marginally correlated and which are on the other hand intuitive and realistic for the

4

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

respondent. Table 1 presents the range of values employed. The offered trade-offs range from -270 CHF/h to 400 CHF/h.

Supermarket B

Supermarket A

Table 1

Experimental Design

Variable

Values

Travel time

5, 10, 15, 20 [min]

Quality

’Top Quality’ (4) [Globus, Jelmoli] ‚Good Range – Good quality’(3) [Migors Coop, Géant]

Price

60, 130, 200, 300, 400 [CHF]

Travel time

for revealed TT< 15 minutes: 10, 20 [min] for revealed TT 15 < x 30 Minuten 30, 45 [Min] (Zone 3)

Quality

‚Discounter’s Paradise’ (1) [Denner, Aldi, Lidl] ‚Wide range - small prices’ (2) [Marktkauf, Kaufland] ‚Good Range - Good quality’(3) [Migors Coop, Géant]

Price (factor)

for Quality 1 0,5/0,66 for Quality 2 or 3: 0,66/0,8/0,9

To uncouple the quality variable from a specific supermarket chain, it was precoded with four generic categories. To make it not too complicated for the respondents each quality category itself was described by mentioning comparable well-known supermarkets. Besides this the interviewees had to indicate relevant socio-demographic information such as sex, household size and income and personal assessments of quality factors of shopping location choice like friendliness of staff or range of local products.

2.3

Descriptive analysis

Figure 2 shows the spatial dispersion of the home residences of the respondents depending on the chosen supermarket. It is obvious that Migros and Géant attract people from a more local area than Marktkauf, which acts as a supra-regional player with costumers who drive from 150km far away places such as Brienz or. Figure 2

Places of residence separated by visited supermarket

5

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

Less surprising are the figures of mode choice ranging from 81% car use for M-Park and 90% for Marktkauf. The analysis of the socio-demographic data showed that costumers of Migros and Géant tend to have higher incomes and to be older. Moreover costumers of Marktkauf live in larger households whereas the largest share of economically active persons can be found in households which prefer Migros. Table 2

Socio-demograohic data of the interviewed costumers Migros ∅

[σ]

Géant ∅

6

[σ]

Marktkauf ∅

[σ]

all ∅

[σ]

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

Income

8425

3836

7731

3994

6161

3078

7545

3782

Age

49.7

12.3

43.5

13.8

38

11.2

44.2

13.3

Income

49.7

38.0

43.5

44.2

49.7 3078

7545 3782

Travel time

12.9

7.1

37.0

23.0

19.1

17.3

24.5

22.0

Distance

10.3

12.8

42.5

38.6

15.1

24.1

25.7

35.4

229.9

137.4

138.7

159.2

122.0

Preis Güter [CHF]1

D/F

87.2

79.4

80.4

The analysis of the personal preferences of shopping centre attributes provided instructive results. One would assume that the choice of a shopping centre can be explained by variables like availability of car-parking, level of prices, origin of goods, product quality or availability of organic food, but no strong differences between the set of costumers are obvious Figure 3 shows of the importance of shopping centre attributes by the stated preference for a particular supermarket.

1

1€ =1.50 CHF

7

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

Figure 3

Importance of shopping centre attributes separated by preferred supermarket

Product quality Price level Car park Short queuing Supermarket qualtiy Product origin Distance to residence Friendliness Local Products Image Organic food Familiar product range

Migros Marktkauf

other stores available

Géant Protest against high prices in CH

Coop

Support Switzerland 0

1

2

3

4

No strong differences between costumers who prefer Migros and Marktkauf can be found (except the protest against the high Swiss prices). However, Coop costumers tend to have higher quality demands, which could reflect the market positioning of Coop. It is also surprising that Géant costumers seem to have the lowest quality demands although an outside view would appraise Géant higher on a quality scale than Marktkauf. To assess the importance of the attributes the respondents were also asked to rank their four most important attributes. The score is calculated as the sum of rank points (Rank 1: 4 points, rank 2: 3 points etc.) divided by the number of observations for each category. Figure 4 shows that price level and product quality are the most important attributes, for costumers who like to shop at Migros or Géant. Coop costumers appreciate mostly the choice of organic food, whereas costumers of Marktkauf (which has the widest product range) value overall quality of 8

5

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

the supermarket most. Distance to residence seems to be less important. It has to be noted that the sample includes only 103 Persons. Figure 4

Most important shopping attributes, separated by preferred supermarket

Price level

Product quality

Car park

Organic Food

Supermarket quality Migros Marktkauf Géant Coop

Distance to residence

Familiar product range

0

5

10

15

20

25 Score

9

30

35

40

45

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

3. Modelling 3.1

Theory

The use of discrete choice models is established in all aspects of transport planning since their initial development (Domencich and McFadden, 1975 and for an early synthesis Ben-Akiva and Lerman, 1985). The root of these econometric models is the idea of utility maximisation, i.e. the assumption that a decider (traveller) will select the alternative with the highest subjective utility. The modelling task is to identify those aspects of the alternative, of the choice situation and of the person, which influence this utility. It is clear, that these measurable elements will not capture the utility fully, so that there is a need for a stochastic element which varies from person to person. These stochastic – error - terms will correlate between alternatives capturing their similarity in the eyes of the traveller. It is equally clear that the evaluation of the objective elements will also vary from person to person. A complete model would therefore allow for taste differences between persons and complex patterns of correlation between alternative specific error terms. Adding the assumption of perfect information of the traveller about the objective characteristics of the alternatives one arrives at the following two part model (see Ortuzar and Willumsen, 1994 or Maier and Weiss, 1990): •

The measurable, systematic part Vjq, representing the value of the objective utility of an alternative j for a person q

•

The stochastic part respectively error εjq, of Ujq considering unobserved evaluation by each user

The total utility Ujq is then:

U jq = V jq + ε jq with Vjq considering the attributes of the alternatives, the specific choice situation of the user and the characteristics of the user. Travel time, or its component parts access time, waiting time, in-vehicle-time, transfer time etc., and travel costs are central elements of the description of the choice alternatives. It can be shown that the ratio of the parameters of cost and time in the utility function specifies the marginal value of one unit of time (saved; expended additionally) (Abay and Axhausen, 2000; Bates, 1987 or Jara-Diaz, 2000). For non-linear utility functions the values of travel time savings are a function of the travel time differences. 10

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

3.2

Modelling

The initial model contains only the travel time, price of goods and the quality of the supermarket. Afterwards, by including further variables as well as using non linear functional forms it is attempted to improve the model. Especially the hypotheses of major influences on individuals’ VTTS of Mackie et al. (2003) are investigated as far as variables are available and an application to shopping travel is reasonable. Table 3 summarises the investigated hypotheses. Table 3

Applicable hypotheses of Mackie et al. (2003)

Hypothesis about major influence on VTTS I Income and time budget

Model formulation Income elasticity, employment

II. Personal/ Household characteristics

Age, favoured supermarket

III Other activities during the travel

Social interaction: Number of accompanying persons

VIII Size of the time saving

Time saving elasticity, Distance elasticity

In the following only the models formulations with significant parameters and improved ρ2 are discussed. 3.2.1

Basic Model

Table 4 shows the utility function of the basic model. Usually also inertia terms are included but appropriate variables were not collected as this study focussed on motorised shopping travel. However, it can be regretted that no information about car/season ticket availability or mileage per year is available. Table 4 Basic model

Utility Function basic model U=

βP* Price βD* Distance βQ,i * Qualityi

The stated preference situations present one-way travel times. As it can be assumed that typical weekly shopping trips are roundtrips the travel times were doubled for the modelling.

11

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

Table5 Results ‘basic model’

Parameter

Model charactristics

N

857

L (0)

-594.0

L (β)

-529.0

LL – Ratio Test

129.8

ρ2

0.109 Unit

coeff.

t-Test

sign.2

Travel time

Min

-0.03

-8.83

y

Price of goods

CHF

-0.01

-6.12

y

Quality of shop class1 (DiscountParadise)

-0.22

-0.74

n

Quality of shop class 2 (many items – cheap prices)

0.05

0.25

n

Quality of shop class 3(quality and assortment)

0.17

0.86

n

VTT

CHF

128.85

Interestingly the quality levels 2 and 3 are valued more highly than the reference quality 4, but none of the quality parameters are significant, which is surprising and suggests that description of shop quality needs improvement. The variables of travel time and price of goods are significant with expected signs and lead to a VTTS of 129.85 CHF/h. 3.2.2

Modelling with RPL, scaling parameters and non linear functions

Random Parameter Logit (RPL) It can be assumed that the estimated parameters are not fixed but rather randomly distributed between the respondents. Therefore the use of Random parameter Logit (RPL) models was tested and effected an increase of ρ2 from 0.109 to 0.119 with a significant parameter variance

σw of the travel time parameter, but a non-significant variance σp of the price parameter Scaling Parameters

2

Significance at a 95% level

12

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

In the descriptive analysis it was hypothesised that decision behaviour could depend on the preference of the supermarket. Therefore scaling parameters for preferred supermarket, separated by country, were introduced. Technically, three utility functions were estimated, with differed only in the multiplicative scaling parameter attached to each. These parameters are significant and led to a slightly better model fit (ρ2=0.123). Non linear functions To investigate if VTTS depends on income and distance non-linear elasticity formulations were introduced. Table 6 shows these model formulations. Table 6

Utility function of the model including elasticity of income and distance, with and without RPL-Parameters and scaling parameters for preferred supermarkets

Model 1 (RPL)

U=

βp [σp] * (Income/8586)εInc * (Distance/34.09)εDist * Price βw [σw] * Distance + βQ,ji * Quality,i

Model 2

U=

βp * (Income/8586)εInc * (Distance/34.09)εDist * Price βw * Distance + βQ,ji * Quality,i

The model was estimated once with and once without RPL parameters. In both models the elasticity parameters were significant in contrast to the RPL parameters. As Table 7 shows the model fit increased. The sign of the income parameter is negative as expected which means that persons with a higher income tend to have a higher VTTS. In contrast to the findings of König et al. (2004) the distance elasticity is positive: the longer the trip distance the lower the VTTS. This can be explained by the fact that it makes in relative terms a bigger difference if two shops are 5 and 10 minutes away from residence as 30 and 35 minutes.

13

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

Table 7

Model results including elasticity of income and distance, without RPLParameters and scaling parameters for preferred supermarkets

scaling paramters

Parameter

Model charactristics

N

839

L (0)

-581.55

L (β)

-507.83

LL – Ratio Test

147.44

ρ2

0.127 Unit

Coeff.

t-Test

Sign

Travel time

Min

-0.02

-2.91

y

Price of goods

CHF

-0.02

-5.38

y

Quality of shop class1 (DiscountParadise)

-0.24

-0.77

n

Quality of shop class 2 (many items – cheap prices)

0.06

0.25

n

Quality of shop class 3(quality and assortment)

0.35

1.5

n

εInc

-

-0.63

-3.56

y

εDist

-

0.20

3.88

y

Reference Germany

-

1.00

0.00

-

France

-

1.01

0.02

n

Switzerland

-

0.94

-0.22

n

VTTS

CHF

14

56.85

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

3.2.3

Socio demographic variables

Employment Mackie et al. (2003) argue that people with stronger time budget constraints should have higher VTTS. Therefore a dummy variable with an assigned value of one if the interviewee was working was introduced and included as interaction term combined with the variable of distance assuming that persons with a higher income value time more. The positive parameter indicates that this assumption can be confirmed but as it was not significant further models were estimated without such a formulation. Preferred supermarket In the descriptive analyses it was assumed that the supermarket preference could be associated with people’s estimation of the VTTS. Although the model results in Table 7 shows insignificant scaling parameters an additional model which includes supermarket preference was estimated. For this purpose the data was divided by country of the preferred supermarket, accepting the resulting relatively small data sets. This caused numerical problems estimating RPL models. Nevertheless non-RPL models were successfully estimated with the result of increasing model fits and strongly different VTT as Table 8 shows.

15

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

Table 8

Model results including elasticity of income and distance and data separation after preferred supermarket preferred supermarket in

Switzerland

Model charactristics

N

France

344

360

135

L (0)

-238.44

-249.53

-93.57

L (β)

-202.31

-197.45

-93.57

LL – Ratio Test

72.26

104.16

-70.41

ρ2

0.152

0.209

0.172

Coeff.

Parameter

Germany

t-Test

Coeff.

t-Test

Coeff.

t-Test

Travel time

-0.03

-3.10

-0.01

-0.64

-0.003

-0.28

Price of goods

-0.02

-3.77

-0.02

-3.96

-0.03

-3.28

Quality of shop class1 (Discount-Paradise)

-0.77

-1.66

0.76

1.46

-1.84

-2.09

Quality of shop class 2 (many items –cheap prices)

-0.56

-1.74

1.37

3.49

-1.78

-2.73

Quality of shop class 3 (quality and assortment)

-0.11

-0.33

1.09

2.78

0.22

0.38

εInc

-0.49

-2.01

-0.46

-1.68

-1.13

-3.12

εDist

0.14

1.57

0.26

2.93

0.23

4.91

VTTS

CHF

100.2

CHF

19.90

CHF

28.97

3.2.4

Interaction term: matching preferred and offered supermarket

Although the supermarket quality was defined as a key variable none of the models estimates the associated parameters as significant. Therefore new dummy variables were introduced which contain the preferred shopping country of the interviewee. Modelled as an interaction term, it shows if people with difference location choice preferences valuate the price differently (Table 9). Supporters of German shops act as reference, DF stands for France, DS for Switzerland. Table 9 Model 1

Utility function model with elasticities and country preference Ui =

βp * (Income./8586)εInc * (Distance/34.09)εDist * (1 + βF* DF + βS* DS) * Price + βw * Distance + βQ,Group,i * Qualgroupi

16

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

The newly introduced parameters are highly significant. Furthermore the model fit is better as for the model presented in Table 6. Except the quality all other parameters show strong t-test values. Therefore this model was chosen as best available model. Table 11

Utility function model with elasticities and country preference

Parameter

Model charactersitics

N

3.3

839

L (0)

-581.55

L (β)

-494.35

LL – Ratio Test

174.4

ρ2

0.150 Unit

Koeff.

t-Test

Sign

Travel time

Min

-0.02

-4.88

y

Price of goods

CHF

-0.03

-7.10

y

Quality of shop class1 (DiscountParadise)

-0.50

-1.69

n

Quality of shop class 2 (many items – cheap prices)

-0.09

-0.43

n

Quality of shop class 3 (quality and assortment)

0.19

0.91

n

βDF

-

-0.51

-4.99

y

βDS

-

-0.52

-6.50

y

εInc

-

-0.40

-3.24

y

εDist

-

0.11

2.41

y

VTTS

CHF

52.9

Results

As shown in Table 11 the VTTS for shopping trips strongly depends on personal preferences in shopping behaviour. Whereas people who like to shop in Switzerland have a VTTS around 110CHF/h typical shopping tourists have a significant lower VTTS. In addition to this also income and distance from residence plays a role quantifying the VTTS as Figure 1 shows. Figure 4

VTT for shopping trips, costumers who prefer Migros

17

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

200 180 160 140

100 80 60 40 20 5

0 00 10

00 30

20 00 50

00 70

00 90 0 Inco 00 0 11 me 00 [ CH 0 F/ m 13 00 o. ] 15

18

35 50 0 00 17

[k m ]

VTTS [CHF/h]

120

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

4. Evaluation and further work For definite conclusions the data pool is too small and not representative. However, interesting findings were made, showing that personal preferences have a significant effect on the estimated VTTS. Furthermore the findings of König et. al. are confirmed. For a better understanding of VTTS of shopping travel two major improvements can be suggested after this work: The main application of VTTS estimates is the cost benefit analysis of road projects. It can be assumed that people do not value travel time savings equal to savings which results from choosing low cost supermarket. Therefore a stated preference experiment which separates cost of travel (example road pricing) and cost of goods as well as supermarket qualities would be desirable. Everybody would expect that the quality or type of the supermarket has a significant influence on the choices. More sophisticated specifications of supermarket quality than used in this work could lead to significant parameters and interesting conclusion about the valuation of quality.

19

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

5. Acknowledgements For the support during the master thesis: Professor K.W. Axhausen and Konrad Meister For the kind permission of interviews: Mr. Nötzold (M-Park, Basel), Mr. Mestrowitsch (Marktkauf, Weil am Rhein) and Mr. Groelly (Géant St. Louis)

20

Swiss Transport Research Conference _______________________________________________________________________________ March 15-17, 2006

6. References Abay, G. and K.W. Axhausen (2000) Zeitkostenansätze im Personenverkehr: Vorstudie, SVI Forschungsberichte 42/00, Bundesamt für Strassen, Bern. Baltas, G. und P. Papastathopoulou (2003) Shopper characteristics, product and store choice criteria: A survey in the Greek grocery sector, International Journal of Retail & Distribution Management, 31 (10) 498-507. Davies, F., M. Goode, L. Moutinho and E. Ogbaonna (2001) Critical factors in consumer supermarket shopping behaviour: A neural network approach, Journal of Consumer Behaviour, 1 (1) 35-39. Hensher, D. A. and T.K. Truong (1985) Valuation of travel time savings - a direct experimental approach. Journal of Transport Economics and Policy, 19 (2) 237-261. Jara-Diaz, S. (2000) Allocation and valuation of travel time savings, in D.A. Hensher and K. Button (eds.) Handbook of Transportation, 303-319, Pergamon Press, Oxford.König, A. and K.W. Axhausen (2004) Zeitkostenansätze im Personenverkehr, final report for SVI 2001/534, Schriftenreihe, 1065, Bundesamt für Strassen, UVEK, Bern. Mackie, P.J., S. Jara-Diaz and A.S. Fowkes (2001) The value of travel time savings in evaluation, Transportation Research, 37 E (1) 91-106. Papatla, P. and A. Bhatnaga (2002) Shopping Style Segmentation, Marketing Letters, 12 (2) 91-106. Putrevu, S. and K. Lord (2001) Search dimensions, patterns and segment profiles of grocery shoppers, Journal of Retailing and Consumer Services, 8 (3) 127-137. Simma, A., P Cattaneo, M. Baumeler and K. Axhausen (2004) Factors influencing the individual shopping behaviour: The case of Switzerland, Arbeitsberichte Verkehrs- und Raumplanung, 247, Institut für Verkehrsplanung und Transportsysteme (IVT), ETH Zürich, Zürich. Thang, D and B. Tan (2003) Linking consumer perception to preference of retail stores: an empirical assesment of the multi-attributes of store image, Journal of Retailing and Consumer Services, 10 (4) 193-200. Vock, I. (2004) Preisunterschiede zwischen Deutschland und der Schweiz – Eine empirische Untersuchung im Billigmarktsegment des Detailhandels, Masterarbeit, Lehrstuhl für Finanzwissenschaften, unveröffentlicht, Universität Freiburg, Freiburg

21