Assessing preferences for a mega shopping centre in the
Netherlands: A conjoint measurement approach
Citation for published version (APA):
Borgers, A. W. J., & Vosters, C. (2010). Assessing preferences for a mega shopping centre in the Netherlands: A conjoint measurement approach. In Proceedings of the European Institute of Retailing and Services Studies conference (RASS) (pp. 21-). Technische Universiteit Eindhoven / EIRASS.
Document status and date: Published: 01/01/2010
Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page numbers.
Link to publication
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:
www.tue.nl/taverne
Take down policy
If you believe that this document breaches copyright please contact us at:
openaccess@tue.nl
providing details and we will investigate your claim.
Assessing Preferences for a Mega Shopping Centre in the Netherlands:
A Conjoint Measurement Approach
Aloys Borgers
Eindhoven University of Technology Urban Planning Group
Eindhoven, The Netherlands
a.w.j.borgers@tue.nl Cindy Vosters
Advin BV Oss, The Netherlands
Cindy.vosters@advin.nl
Abstract
In 2004, the Dutch central government decided to liberalise her restricted retail policy. This stimulated some retail developers to prepare plans for mega shopping centres. As mega shopping centres do not exist in the Netherlands, this study aims at eliciting consumers’ preferences for this kind of new developments. Consumers visiting a down town shopping centre and one of the largest out-of-town shopping centres in the Netherlands were presented descriptions of different hypothetical mega shopping centres, systematically varying on 10 attributes. The consumers were asked to select the centre they preferred most from sets of two centres. The following attributes were used to define the mega shopping centres: accessibility by car, accessibility by public transport, parking tariff, length of the main shopping streets, type of shopping supply, type of anchor stores, type of traffic allowed in the shopping centre, design style, scale of the shopping streets, and type of activities in the shopping centre.
Over 300 respondents completed the online questionnaire. Discrete choice models (both multinomial and mixed logit) were estimated to assess the importance of each attribute. Overall, the estimation results confirm expectations. Shoppers prefer well accessible shopping centres and free parking. The preferred time needed to walk through the main streets of the shopping centre is 45 minutes; 30 minutes is still acceptable, but 15 minutes is not preferred at all. Shoppers do not prefer a shopping supply existing of small and medium sized (local) shops, and specialised/exclusive shops are preferred over the well known national chains. Regarding anchor stores, shoppers seem to dislike the very large electronics stores and traditional department stores are preferred over flagship stores. Only pedestrians should be allowed to enter the shopping centre. Other traffic modes like bicycles and especially motorized modes are not preferred. The design style should be historically while a Disney style is detested. A modern design style is somewhere in between. The preferred scale of the shopping streets is a mixture of short/narrow and long/wide streets. Only long/wide scale shopping streets are not preferred. Finally, the type of activities offered by the shopping centre should be a mixture of passive and active activities. Shoppers seem to be less happy with active activities only. Although all attributes have a significant impact on the preference for a shopping centre, parking fee and design style appear to be the most important attributes. In addition to the overall effects, significant differences between females and males, between younger and older respondents, and between respondents recruited in the down town shopping centre and respondents recruited in the large out-of-town shopping centre were found. Some interactions between attributes were significant as well. The models perform very satisfactory.
1. Introduction
Until 2004, the Dutch central government pursued a restrictive retail policy. Shopping centres had to be hierarchically organised with the down town shopping centre on top of a city’s hierarchy. Out of town or peripheral shopping developments were restricted to particular types of shops in selected cities (see also Gorter et al. 2003). In the 2004 policy document on spatial planning, the Dutch government delegated retail decisions to the municipalities. Provinces were assigned to supervise and coordinate municipal plans. This reversal in policy liberalised the Dutch retail system. Although some earlier plans for mega shopping centres in the Netherlands failed, in September 2007, a plan for a mega shopping mall in the medium sized city of Tilburg was announced. The 100.000 m2 mega shopping mall was planned at a former military area in the northern periphery of the city. As may be expected from experiences in other countries (see e.g. Howard & Davies, 1993; Marjanen, 1995; Williams, 1995), many objections rose against this plan. Neighbouring municipalities worried about environmental aspects and the viability of their retail facilities and established retailers expected decreasing turnover figures. The municipality of Tilburg as well as other municipalities hired consultants to assess likely economic and environmental impacts. After some years of political discussions, the plan was cancelled by referendum in 2009. In the mean time, it was decided to investigate potential customers’ preferences regarding a peripheral mega shopping centre. This paper reports the approach and results of this investigation.
Because mega shopping centres like the one planned in Tilburg do not exist in the Netherlands, preferences regarding mega shopping centres cannot be derived from observed shopping centre patronage. Therefore, it was decided to use a conjoint choice model to measure customers’ preferences. Conjoint preference or choice models (Louviere et al. 2000) have been applied many times in the context of retailing. For example, Oppewal et al. (1997) developed a conjoint choice model to measure the effects of shopping centre size and marketing mix on customers’ choice behaviour; Oppewal & Timmermans (1999) applied a stated preference model to measure the effect of physical aspects of shopping centres on consumer perceptions; Borgers et al. (2006) used a stated choice model to assess the impact of peripheral retail centres on traditional urban shopping centres in a Dutch city; and Kim et al. (2009) used conjoint analysis to design a novel suburban luxury brand outlet mall in S Korea.
Conjoint choice analysis involves a number of steps. First, attributes (or characteristics) of the alternatives that are assumed relevant have to be identified, along with their so-called attribute levels (section 2). Each combination of attribute levels defines an alternative (a mega shopping centre). As the number of alternatives may grow huge if the number of attributes and/or the number of attribute levels increases, some experimental design is used to select a representative fraction from the complete set of alternatives. Given the hypothetical mega shopping centres, experimentally controlled choice situations must be created and presented to respondents (section 3). To assess the preferences, data must be collected by asking respondents (potential shoppers) to choose the mega shopping centre they prefer from the choice situations created in section 3. This will be explained in section 4. Next, discrete choice models have to be specified to estimate the effect of the attributes (and respondents’ characteristics) on the respondents’ preferences regarding mega shopping centres (section 5). The results of the model estimation will be presented in section 6. Finally, in section 7, conclusions will be drawn and implications for future development of mega shopping centres will be discussed.
2. Selection of attributes
Developing large shopping centres involves many decisions. Location and accessibility are very important decision variables to assure a sufficient number of potential customers. In addition, the shopping centre should attract many shoppers. Supply of retail and entertainment outlets, but also aspects related to design, layout, atmospherics, et cetera are important. Although the plans for a mega shopping mall in Tilburg induced this research project, the purpose is to investigate variables of special interest in the first stages of designing a mega shopping centre somewhere in the Netherlands. Based on the literature and opinions of industry experts, the selected variables are listed in Table 1. For each variable (attribute), three levels were defined. The effects of these attribute levels on shoppers’ preferences for mega shopping centres will be investigated.
As mega shopping centres are likely to be located in the periphery of urbanised areas, accessibility by car and public transport should be guaranteed. Although accessibility can be improved by means of infrastructural measures, it is of interest to assess the importance of the positioning of the mega centre relative to the highway exit and public transport stop. Accessibility by car represents the ease to reach the shopping centre after leaving the highway. This is expressed by the number of obstacles between the highway exit and the shopping centre. Examples of obstacles are traffic lights and busy intersections. Accessibility by public transport is expressed by the time to walk from the nearest public transport stop to the shopping centre and vice versa. Although
parking tariff is not a main decision variable in the beginning of the design process, it
was included in the list of variables as a kind of reference (or benchmark) variable. In the Netherlands, many cities introduced paid parking at the parking facilities of the main shopping centres (Van der Waerden et al. 2009). By including this attribute, the importance of the other attributes can be related to the importance of parking costs. The levels represent the range of commonly used tariffs at large, non down town, shopping centres in the Netherlands. Note that the accessibility variables do not take into consideration the time or distance to travel from home to the shopping centre. It is assumed that before starting the design process, a suitable location for the mega shopping centre already has been selected.
According to Reimers & Clolow (2004) consumers may be reluctant to walk excessive distances in a shopping centre. Therefore, they advice creating compact shopping environments. However, as shopping trips to mega shopping centres mainly can be considered as recreative or hedonic shopping trips, consumers may be less sensitive to walking distances. The length of the main shopping streets expresses the time needed to traverse the main streets in the shopping centre. This does not include the time to visit shops, window shopping or take a rest. Also related to the layout of a shopping centre is the scale of the shopping streets. The shopping centre may consist of a network of short and narrow shopping streets with narrow shop fronts. On the other hand, the streets may be long and wide with wide shop fronts. The third level of this attributes is defined by a mixture of both short/narrow and long/wide streets. Although most large shopping centres in the Netherlands allow pedestrians only, it was questioned whether shoppers would prefer (limited) access by bicycles or other transportation modes as well in the case of extremely large shopping centres. Therefore, the attribute type of traffic allowed was taken into consideration as well.
One of the long-run decisions regarding new shopping centres concerns the selection of anchor stores. Finn & Louviere (1996) concluded from their research in the Edmonton region that anchor stores have a dominant role on shopping centre image.
The types of anchor stores selected in this study are department stores, very large electronics shops, or very large fashion shops. The latter type is also known as flagship stores (Kozinets et al., 2002; Kent, 2002). In addition to anchor stores, the type of
shopping supply is considered relevant as well. As national (and international) chains
dominate many shopping centres, it was questioned whether consumers would prefer other types of shopping supply in the mega shopping centre. Therefore, small to medium sized and specialized/exclusive shops were considered as the main type of shopping supply as well.
Table 1: Attributes and attribute levels
Attribute level description
Accessibility by car 1 2 3
1 obstacle between highway exit and shopping centre 3 obstacles between highway exit and shopping centre 5 obstacles between highway exit and shopping centre Accessibility by public
transport
1 2 3
First PT-stop at 3 minutes walking from shopping centre First PT-stop at 6 minutes walking from shopping centre First PT-stop at 9 minutes walking from shopping centre
Parking tariff 1 2 3 Free parking €1,00 per hour €2.00 per hour Length of main shopping
streets 1 2 3 15 minutes walking 30 minutes walking 45 minutes walking Type of shopping supply 1
2 3
Well known national chains Small to medium sized shops Specialized and exclusive shops Type of anchor stores 1
2 3
Department store Mega electronics store Flagship store (fashion) Type of traffic allowed 1
2 3
Pedestrians only
Pedestrians and bicyclists All transport modes
Design style 1 2 3 Historical Modern Disney style Scale of shopping streets 1
2 3
Many short and narrow shopping streets Some long and wide shopping streets Mixture of both types
Type of activities 1 2 3
Passive, like a restaurant or a cinema Active, like a fun-fair or a bowling alley Mixture of both types of activities
Wakefield & Baker (1998) conclude that, amongst other things, overall architectural design of the mall and entertainment outlets like a theatre or family recreation centre may generate excitement and improve a mall’s competitive position. Also Sit et al. (2003) conclude that entertainment is essential. However, Haynes & Talpade (1996) warn that mall developers should use caution in developing a mall with an entertainment centre. Teller & Reuttener (2008) found that entertainment does not impact the evaluation of the attractiveness of a shopping centre. Although architectural design and entertainment may be important, it is not clear which type of architectural design is preferred and what kind of entertainment outlets should provide. Therefore,
attributes. Regarding the type of activities, a distinction is made between passive and active activities. A mixture of both is considered as well. Regarding the architectural design style, a historical, modern, and Disney style were chosen. In the case of a historical style, the shopping centre consists of historical look-alike buildings. The modern style represents a more present-day and technological character, while the Disney style refers to a specific theme in a picturesque architecture.
By selecting one level for each attribute, a description of a hypothetical shopping centre is generated. In total, 310 different hypothetical shopping centres can be generated, which is an impractical high number of alternatives. However by taking an orthogonal fraction of the full set of 310 alternatives, preferences can still be estimated. Therefore, a fraction of 81 alternative shopping centres was selected. This selection allows for the estimation of all main effects and the interaction effects between the first five attributes listed in Table 1.
3. Choice tasks
One way to assess preferences regarding shopping centre attributes is to present respondents choice sets and ask them to identify the most preferred alternative in each choice set. In this study, the choice set is composed of different hypothetical shopping centres. To keep the choice task simple, each choice set was composed of two hypothetical shopping centres and a ‘no preference’ alternative which can be chosen if the respondent has no preference regarding one of the two hypothetical centres in the set. An example of a choice task is presented in Figure 1.
Information about the characteristics:IINNFFOO
Characteristic Shopping centre 1 Shopping centre 2
Accessibility by car 3 obstacles to the shopping centre
5 obstacles to the shopping centre
Accessibility by public transport First PT-stop at 6 minutes walking
First PT-stop at 9 minutes walking
Parking tariff €2.00 per hour €2.00 per hour Length of main shopping streets 30 minutes walking 45 minutes walking Type of shopping supply Specialized and exclusive Specialized and exclusive Type of anchor stores Mega electronics store Department store
Type of traffic allowed All transport modes Pedestrians only Architectural design style Disney style Disney style Scale of shopping streets Many short and narrow shopping
streets
Mixture of short/narrow and long/wide streets
Type of activities Mixture of both passive and active activities
Mixture of both passive and active activities
Which alternative do you prefer?
O shopping centre 1 O shopping centre 2 O no preference Figure 1: Example of a choice task
Each respondent was presented 14 choice tasks. For each respondent, the 14 choice sets were generated by randomly selecting two alternatives from the set of 81 alternatives. The on-line questionnaire started with an introduction emphasizing that the context was recreational shopping. So, a respondent had to imagine that the main purpose of visiting the shopping centre is to enjoy his/her leisure time. It was explained that the respondent would be presented 14 choice situations. The task was explained by an example choice set which was shown prior to the 14 choice sets. The example was used to explain how the two shopping centres are defined by the ten attributes and how the respondent can identify his/her preference for one of the two shopping centres, or, if applicable, neither shopping centre. Furthermore, it was explained that any time the respondent could ask for an extensive description of the attributes by clicking the INFO-button on top of the screen. These extensive descriptions showed some pictures for the attribute Type of
anchor stores, Design style, and Scale of the shopping streets. The respondents were
instructed to assume that the two mega shopping centres presented in each choice situation only differ in terms of the listed characteristics. No information about the location of the mega shopping centre was provided.
4. Data collection
Data was collected by means of an internet based questionnaire. After a short introduction, the respondent was presented the 14 choice tasks as described in the previous section. At the end of the questionnaire, the respondent was asked to provide information regarding personal characteristics (gender and age), postal code, in which shopping centre he/she was invited to participate in the research project, and the preferred type of shopping centres for recreational shopping (down town shopping centres, district shopping centres or other types of shopping centres). As five gift coupons were raffled among the participants, the respondent was asked to provide his/her email-address to notify whether a gift coupon was won.
It was decided to recruit respondents among customers in large shopping centres such as down town shopping centres of Dutch cities and large peripheral shopping centres. The down town shopping centre of Den Bosch and the peripheral shopping centre Alexandrium, located in Rotterdam, were chosen to recruit respondents. Both shopping centres attract customers from a wide region. The down town shopping centre of Den Bosch is one of the most popular down town shopping centres in the Netherlands. Alexandrium is one of the largest peripheral shopping centres in the Netherlands (see also Gorter et al. 2003). In each shopping centre, respondents were recruited during three days at the end of June and the beginning of July 2008. Weather conditions were fine. As the Alexandrium is an indoor shopping centre and the down town of Den Bosch is an open air shopping centre, rainy days might have reduced the number of customers in Den Bosch. Customers were personally asked whether they were willing to participate. If yes, their email-addresses were registered. Next, these respondents were sent an email inviting them to visit the website containing the questionnaire. Respondents not responding to the first invitation within two to three weeks were sent a recall mail. To encourage participation, 5 gift coupons of €10,00 each were raffled among the respondents.
In total, 667 usable email-addresses were collected in the two shopping centres. Eventually, 312 (47%) respondents completed the online questionnaire. Table 2 lists some characteristics of the respondents. Compared with national statistics regarding recreational shopping in 2006/2007 (CBS / Statistics Netherlands), the male-female
ratio is approximately representative, but the age category of 15-24 is overrepresented and the oldest category (over 65 years of age) is underrepresented. The number of respondents recruited in Den Bosch is higher than in Rotterdam. This was expected as relatively more shoppers in Den Bosch were willing to provide their email-address. All pairs of characteristics (gender × age, gender × location, age × location) are independent of each other according to the Chi2-test (if the 41-65 and >65 age categories are merged). However, there is a significant difference between these subsamples in terms of preferred type of shopping centre. For the Alexandrium-sample, the ratio down town centre – district centre is approximately 50-50, while this ratio is about 85-15 for the Den Bosch-sample. This may be attributed to the lack of attractive district or out of town centres in the Den Bosch region.
Table 2: Respondents’ characteristics
# % CBS %1 Gender male 88 29 32 female 212 71 68 unknown 12 -- Age 15-24 years 101 34 17 25-40 years 74 25 34 41-65 years 115 38 33
older than 65 years 9 3 15
unknown 13 --
Location of Alexandrium Rotterdam 117 41
recruitment down town Den Bosch 167 59
unknown 28 --
1) Note that the age category 0-15 was excluded as children are usually accompanied by adults
5. Model specification
The data collected from the choice situations were used to estimate a random utility choice model. Each choice situation consisted of two hypothetical mega shopping centres and a ‘no preference’ option. Thus, one of three choice alternatives has been chosen. According to random utility theory (e.g. Train, 2003), each alternative i has a utility (Ui). This utility consists of a structural (Vi) and a random (εi) component:
Ui = Vi + εi (1)
The structural component is assumed to be an additive function of the characteristics of the alternative:
Vi = Σk βk Xik (2)
where Xik represents characteristic k of alternative i and βk is the parameter for characteristic k. Note that the mega shopping centres are characterized by 10 attributes. However, as each attribute consists of three levels (which can be considered as characteristics), effect coding (see Table 3) was used to estimate the part-worth utility of each characteristic. This means that 20 variables are needed to estimate all part-worth
utilities. The part-worth utility of the first level of the first attribute is equal to β1, of the second level to β2, and of the third level to –(β1+β2), and so on. The utility of the ‘no preference’ option is measured by a constant: β0.
Table 3: Effect coding
Attribute level Coding
1 1 0
2 0 1
3 -1 -1
If it is assumed that the random utility components are identically and independently distributed, the multinomial logit model can be used to calculate the probability pi that alternative i will be chosen. This model is defined as:
pi = exp(Vi) / Σj exp(Vj) (3)
The parameters are estimated by maximum likelihood estimation, which maximizes the predicted probabilities of the chosen alternatives. Using the null-model (all parameters are equal to 0.0) as a reference model, a goodness-of-fit measure Rho2 can be computed. This measure ranges between 0.0 (no improvement compared with the null-model) to 1.0 (a perfect prediction of each observed choice). According to Hensher et al. (2005), a
Rho2 of 0.3 or higher represents a decent fit for a discrete choice model. However, according to Louviere et al (2000) values between 0.2 and 0.4 can be considered to be indicative of extremely good model fits.
The parameters β1 … β20 represent the main effects of the attributes. In fact, they represent the preferences for the attribute levels. However, preferences may vary across individuals’ characteristics. For example, Dholakia (1999) found that more married women seem to enjoy going to the mall than married men and that the recreational and expressive nature of shopping at the mall seems to appeal to the female shopper more than to the male shopper. Ruiz et al. (2004) revealed four segments of shoppers: recreational shoppers, full experience mall shoppers, traditional shoppers, and mission shoppers. The first segment includes far more elderly people while the last segment includes a higher proportion of young adults. Thus, it may be of interest to investigate whether these personal characteristics affect the main effects of the attributes. In addition to gender and age, the location of recruitment was taken into consideration as well, because the respondents recruited in Rotterdam prefer other types of shopping centres than the respondents recruited in Den Bosch.
By creating contrast variables, additional parameters can be estimated to test for differences between subsamples (e.g. males and females). For the first subsample, all X-variables should be copied into Z-X-variables, while for the second subsample, the negative of the X-variables must be copied into the Z-variables. If values are estimated for the β-parameters (related to the X-variables) and δ-parameters (related to the Z-variables), the part-worth utility for variable k is equal to βk Xik + δk Zik, which is equal to (βk +δk)Xik in the case of the first subsample and to (βk -δk)Xik in the case of the second subsample. If δk is not significantly different from zero, the part-worth utility for both subsamples is βkXik, meaning there are no differences between the subsamples. For each X-variable (and also for the constant measuring the utility of the ‘no preference’ option) a contrast variable is created for gender (males: Zik,gender = -Xik; females: Zik,gender = +Xik), age (15-24 years: Zik,age = -Xik; 25-40 years: Zik,age = 0; over 40 years: Zik,age = +Xik), and
location of recruitment (Rotterdam Zik,location = -Xik; Den Bosch: Zik,location = +Xik). Note that three subsamples were specified for age by joining the group aged 41-65 and the small group aged over 65 years. The contrast effect for the middle age group is set to zero, implying that a linear age effect is assumed. Contrast effects may also be referred to as interaction effects between attributes and respondents’ characteristics, see e.g. Alberini et al. (2003).
The experimental design that was used to generate the shopping centres allows for the estimation of interaction effects between the first five attributes of Table 1. As each attribute has three levels and consequently two indicator variables (see Table 3), four variables define the interaction between the two attributes. For example, the first two attributes are specified by variables X1, X2, X3, and X4. The interactions between these attributes are equal to Ii1 = Xi1×Xi3, Ii2 = Xi1×Xi4, Ii3 = Xi2×Xi3, and Ii4 = Xi2×Xi4. In total, 40 interaction variables (Ii1 … Ii40) must be specified to measure all possible first order interaction effects between the first five attributes. Now, equation 2 can be extended to:
Vi = β0 + Σk=1,20 βk Xik + Σk=1,20 δk,gender Zik,gender + Σk=1,20 δk,age Zik,age + Σk=1,20 δk,location Zik,location + Σk=1,40 θk Iik (4)
In this equation, β0 represents the utility of the ‘no preference’ option, the βk-parameters measure the main effect of the attributes, the δk-parameters measure the contrast effects between subsamples regarding gender, age, and location of recruitment, and the θk -parameters represent the interaction effects between attributes.
The multinomial logit model assumes homogeneity (no taste variation among respondents). To test for heterogeneity among the respondents, a mixed (or random parameter) logit model (see e.g. Train, 2003) was estimated as well. Random parameter models assume that respondents share the same kind of preference function, but vary in terms of the weights they attach to the attributes. Such taste differentiation is captured by estimating a distribution for the parameters of the utility function. For each βk -parameter, a Normally distributed random component υk was added with mean 0.0 and standard deviation σk. The equation for the structural utility then becomes:
Vi = (β0+υ0) + Σk=1,20 (βk+υk) Xik + Σk=1,20 δk,gender Zik,gender + Σk=1,20 δk,age Zik,age + Σk=1,20 δk,location Zik,location + Σk=1,40 θk Iik (5)
The standard deviation (σk) was estimated for each variable, in addition to the mean value (βk). According to a mixed logit model, the choice probabilities are calculated by repeatedly applying the multinomial logit. For each individual, random numbers are drawn for the random variables and individual choice probabilities are calculated. This is repeated R times for each individual and the probabilities for each alternative are averaged across the R drawings. For a good performance, very large numbers of draws are required. However, instead of a large number of random draws, a Halton sequence of draws can be used (Bhat, 2001). Halton draws give a fairly even coverage over the domain of the distributions and the draws for one observation tend to fill in the spaces that were left empty by the previous observations. A Halton sequence of draws with only one tenth the number of random draws is often equally effective. As per respondent fourteen choices were observed, the random draws per variable were kept
constant for each respondent. If some of the standard deviations are significantly different from zero, the assumption of homogeneity underlying the standard MNL model is not valid.
Suárez et al. (2004) specified a random effects model as well. However, in their model, heterogeneity was taken into consideration by two market segments differentiating on the effects of the attributes. As in our model the influence of respondents’ characteristics is already measured by means of contrast effects, an additional random heterogeneity component for each main effect was considered appropriate.
6. Estimation
Both the multinomial and the mixed logit model have been estimated. In the case of the MNL model, the structural utility is defined by eq. 4, while this utility for the mixed logit model is defined by eq. 5. The parameters of both choice models have been estimated by Nlogit 4.0 (Greene, 2007). This was done stepwise. After the first run, all variables with significance P[|Z|>z]| > 0.50 were removed from the model. This criterion was gradually decreased until 0.10. Thus only parameters that are significant at the significance level of 10% are included in the models. The estimated parameters according to the models are presented in Tables 4 and 5. The significance of each parameter is displayed between brackets. The column labelled ‘Overall’ represents the estimates for all respondents (the β’s and the σ’s in the case of the mixed logit model), regardless gender, age, and location of recruitment. The other columns show the significant contrast effects for gender, age, and location (the δ’s). Remember that for males, young respondents (aged 15-24) and respondents recruited in Rotterdam, the contrast effects should be subtracted from the main effect, and for females, older respondents (aged over 40 years) and respondents recruited in Den Bosch, the contrast effects should be added to the main effects. As the part-worth utility of the third level of an attribute has to be inferred from the corresponding parameters, it has been italicised in the tables. For ease of interpretation, Figures 2 and 4 represent the part-worth utilities for all attribute levels and all respondents in general. In addition, these figures represent the effects for the gender, age, and location segments.
The interaction effects represent utility adjustments in the case of specific combinations of attribute levels. The interaction effects (the θ‘s) between the attributes are displayed in Figures 3 and 5 and will be discussed separately. The variable labelled ‘Con’ is equal to zero for the two shopping centre alternatives in each choice set, and equal to one for the ‘no preference’ option. In fact, the parameter for this variable measures the utility for the ‘no preference’ option.
The multinomial logit model
The multinomial logit model performs relatively well, rho2 is equal to 0.23. According to Table 4, the utility of the ‘no preference’ option is on average -1.68. This rather strong negative utility implies that in most cases, respondents made a choice between one of the two shopping centres presented in the choice sets, supporting the estimation of the attribute effects. Note that according to the contrast effects, younger respondents have a higher tendency to choose the ‘no preference’ option than the older respondents. Also respondents recruited at the Alexandrium shopping centre in Rotterdam have a
higher tendency to choose this option than respondents recruited in the down town shopping centre of Den Bosch.
The effects of the first three attributes (accessibility by car, accessibility by public transport, and parking tariff) are linear. The first level of each of these attributes (one obstacle by car, 3 minutes walking to the public transport stop, free parking) is positive, the last level (five obstacles by car, 9 minutes walking to the public transport stop, €2,00 parking fee per hour) is negative, and the level in between has a zero utility. Note that the parking tariff has a strong effect on the preference for an alternative. Figure 2 shows that, according to males, the overall part-worth utility of the first level of the first attribute increases and decreases for the third level. For females, the opposite occurs. This means that males attach more weight to the accessibility by car than women do. Males appreciate only one obstacle significantly more than women. Also younger respondents appreciate only one obstacle significantly more than older respondents. The gender effect on accessibility by public transport violates the overall linearity of this attribute. Females still attach a positive value to 6 minutes walking, but they are more negative about 9 minutes walking to the public transport stop. Males, however, do not really differentiate between 6 and 9 minutes, they attach a small negative value to both levels. Again, young respondents attach more weight to the accessibility by public transport than older respondents. For the older respondents, the difference between 3 minutes walking or 9 minutes walking to the public transport stop is rather small. Regarding the parking fee, respondents recruited in Rotterdam attach a negative utility to one euro per hour, while the respondents recruited in Den Bosch still attach a positive utility to this parking fee. However, the latter group of respondents is more discontent with the two euro per hour tariff than the respondents recruited in Rotterdam. A possible explanation may be that respondents from the Rotterdam region are used to the higher parking tariffs commonly applied in the denser urban areas of the Dutch Randstad region.
Regarding the length of the main shopping streets in the shopping centre, respondents do not appreciate a shopping centre that can be traversed in about a quarter of an hour. Remember that the respondents were asked to choose the shopping centre they prefer most in the context of a recreational shopping trip. In this context, a larger shopping centre is preferred. The walking distances of half an hour and three quarters of an hour are appreciated almost equally. This means that if we assume a rather slow walking speed of 2 to 3 kilometres per hour, the total length of the main shopping streets should be 1 to 2 kilometres. There are no significant contrast effects related to gender, age, or location of recruitment regarding this attribute.
Overall, the respondents do not like a shopping centre with small or medium sized (and possibly local) shops as the main type of shops. A shopping centre with specialised and/or exclusive shops appears to be preferred. However, young respondents prefer the national chains over the specialised and exclusive shops while the older respondents appreciate the specialised and exclusive shops much more than average. A similar effect appears for the location of recruitment: the respondents recruited in Rotterdam prefer the national chains and the specialised/exclusive shops approximately equally, while the respondents recruited in Den Bosch prefer the special and exclusive shops more than the other types. The differences between the age groups however are larger than between the location groups.
The overall preference for anchor stores is department stores, followed by flagship stores. Mega electronics stores appear to be disliked. However, there are strong differences between males and females. Females prefer department stores more and
mega electronics stores much less than the average respondent. In contrast, males do not like the flagship stores and attach a positive utility to mega electronics stores.
Regarding traffic in the shopping centre, only pedestrians are preferred. In general, allowing all transport modes (pedestrians, bicyclists, and motorized transport modes) in the shopping centre is not preferred. The option of allowing pedestrians and bicyclists is positioned in between. Young respondents attach less weight to this attribute, the difference in utility between the first (pedestrians only) and third level (all modes) is much smaller than on average. For older respondents, however, this difference is much bigger, meaning that they attach more weight to this attribute. There are also significant differences between respondents recruited in Rotterdam and respondents recruited in Den Bosch. Those recruited in Rotterdam do not really differentiate between the second (pedestrians and bicyclists) and third level (all modes), while those recruited in Den Bosch attach an enlarged negative utility to the third level.
A historical design style is clearly preferred over the other design styles. In general, respondents attach a negative utility to both the modern and Disney style, with the latter being disliked most. However, there are a few exceptions. Females, young respondents and (to a lesser extent) respondents recruited in Rotterdam are less distinct than males, older respondents and respondents recruited in Den Bosch. In some cases (females and young respondents), the difference between a modern style and a Disney style disappears.
Another attribute related to design concerns the scale of the shopping streets. Overall, a mixture of short/narrow and long/wide streets is preferred, with only short/narrow streets in second position. Only long/wide streets are not preferred. For young respondents, the utilities for only short/narrow and only long/wide are almost equal and if respondents get older, the utility of short/narrow streets increases while the utility for long/wide streets decreases. Older people attach more weight to this attribute than young people. Something similar holds for the location of recruitment. Respondents recruited in Rotterdam do not really differentiate between the short/narrow and long/wide streets, while respondents recruited in Den Bosch dislike the long/wide streets.
In general, respondents prefer a mixture of passive and active activities in the shopping centre; only active activities are not preferred. Males care less about this attribute than women. Respondents recruited in Den Bosch slightly prefer the passive activities over the mixture of both passive and active activities.
The range between the highest and lowest utility of an attribute can be considered as a measure of the impact of the attribute on shoppers’ preferences. This range is largest for the parking fee attribute. This also holds for each subsample of shoppers. Thus, it can be concluded that, according to the MNL model, parking fee is the most important attribute from the list of attributes in Table 1. Remember that this attribute was included as a kind of benchmark attribute. The next most important attributes are design style and, at some distance, type of anchor stores. Regarding design style, especially males, older shoppers, and shoppers recruited in Den Bosch like the historical style and dislike the Disney style. Regarding type of anchors, especially the female shoppers dislike the mega electronics shops. Accessibility by public transport has the smallest range in utilities and thus can be considered the least important attribute taken into consideration in this study. Probably, most people will travel by car to a mega shopping centre. Type of shopping supply is the next least important attribute. However, compared to the middle age segment, both the segment of young respondents and the segment of older respondents attach more weight to this attribute, e.g. more than to the type of activities attribute.
Table 4: Estimated parameters multinomial logit model (significance between brackets)
Var. Attribute level Overall Gender Age Location
Con ‘no preference’ -1.683 (.000) -0.207 (.002) -0.148 (.011) Accessibility by car
A1 1 obstacle 0.160 (.000) -0.054 (.097) -0.064 (.061) A2 3 obstacles
A3 5 obstacles -0.160 0.054 0.064
Accessibility by public transport
B1 3 min. walking 0.093 (.001) -0.069 (.040) B2 6 min. walking 0.057 (.056) B3 9 min. walking -0.093 -0.057 0.069 Parking tariff C1 Free 0.405 (.000) C2 €1,00 per hour 0.076 (.012) C3 €2,00 per hour -0.405 -0.076
Length of main shopping streets
D1 15 min. walking -0.169 (.000) D2 30 min. walking 0.087 (.005) D3 45 min. walking 0.082
Type of shopping supply
E1 National chains -0.127 (.000) -0.052 (.081)
E2 Small/medium -0.098 (.001)
E3 Special/exclusive 0.098 0.127 0.052
Type of anchor stores
F1 Dept stores 0.170 (.000) 0.105 (.003) F2 Mega electro stores -0.183 (.000) -0.215 (.000) F3 Flagship stores 0.013 0.110
Type of traffic allowed
G1 Pedestrians 0.151 (.000) 0.073 (.027) G2 Peds + bicyclists 0.082 (.004) G3 All modes -0.151 -0.073 -0.082 Design style H1 Historical 0.317 (.000) -0.106 (.001) 0.122 (.000) 0.050 (.091) H2 Modern -0.098 (.003) H3 Disney style -0.219 0.106 -0.122 -0.050
Scale of shopping streets
I1 Short and narrow 0.062 (.100) 0.057 (.088)
I2 Long and wide -0.154 (.000) -0.082 (.039) -0.079 (.028)
I3 Mixed 0.154 0.020 0.022 Type of activities J1 Passive 0.061 (.042) J2 Active -0.108 (.000) -0.063 (.040) J3 Mixed 0.108 0.063 -0.061 Interaction effects B2×C2 -0.067 (.053) B1×E2 -0.060 (.083)
Log-likelihood = -3650.161; Rho2 = 0.230; Rho2adj
A1 A2 A3 B1 B2 B3 C1 C2 C3 D1 D2 D3 E1 E2 E3 F1 F2 F3 G1 G2 G3 H1 H2 H3 I1 I2 I3 J1 J2 J3 Mean -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 Ma le ( bl ac k ) F em a le (gr ey ) -0.4 -0.2 0.0 0.2 0.4 Y ou ng ( b lac k ) O ld (gr ey ) -0.4 -0.2 0.0 0.2 0.4 R’ da m ( b lac k ) D. B os c h (gr ey ) -0.4 -0.2 0.0 0.2 0.4
Figure 2: Part-worth utilities and contrast effects; MNL model
The interaction effect B2×C2 indicates that the combinations of the second and third levels of the corresponding attributes generate special effects. The multiplication of the B2- and C2-variable is different from zero in four cases. In the case of 6 minutes walking to the public transport stop and the shopping centre (B2) and €2,00 parking costs per hour (C2), the utility derived from both attributes decrease from 0.0 (main effects only) to 0.067 (interaction effect). This also occurs in the case of 9 minutes walking (B3) and €2,00 (C3). In the other cases (the combination of B2 and C3 or the combination of B3 and C2), the interaction effect increases the utility by 0.067. The interaction effect B1×E2 generates special effects for the combinations of 3 or 9 minutes walking from/to public transport stop and small/medium or special/exclusive shopping supply. The effect is equal to a decrease (B1×E2, B3×E3) of 0.06 or an increase (B1×E3, B3×E2) of 0.06. Note that in the case of 9 minutes walking time (B3), both interaction effects (with parking costs and with type of shopping supply) have to be
taken into consideration. In Figure 3, the interaction effects are displayed. Although the two interaction effects are significant at the 10% level, the effects appear to be rather limited. Therefore, the multinomial logit model was re-estimated without the interaction effects. The log-likelihood decreased from -3650.161 to -3653.398. According to the likelihood ratio test, this difference is significant at the 5% level. Therefore, the interaction effects should not be removed from the model.
E1 E2 E3 E1 E2 E3 E1 E2 E3 E1 E2 E3 E1 E2 E3 E1 E2 E3 E1 E2 E3 E1 E2 E3 E1 E2 E3 C1 C1 C1 C2 C2 C2 C3 C3 C3 C1 C1 C1 C2 C2 C2 C3 C3 C3 C1 C1 C1 C2 C2 C2 C3 C3 C3 B1 B1 B1 B1 B1 B1 B1 B1 B1 B2 B2 B2 B2 B2 B2 B2 B2 B2 B3 B3 B3 B3 B3 B3 B3 B3 B3 W it ho u t i nte rac ti on eff ec ts ( bl ac k ) W it h i nte rac ti on eff ec ts ( grey ) -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Figure 3 Attributes B, C, and E, with and without interaction effects; MNL model
The mixed logit model
The mixed logit or random parameter model allows the weights attached to the attributes to vary across individuals. For each β-parameter, the standard deviation of a Normally distributed random component was estimated. The results of the estimation are listed in Table 5. The number of Halton draws (R) was set to 1000, however, 500 draws produced almost the same parameter values. The random parameter (mixed) logit model outperforms the multinomial logit model, rho2 is equal to 0.27. If applicable, estimated significant standard deviations are printed below the corresponding mean parameter values.
Table 5: Estimated parameters mixed logit model (significance between brackets)
Var. Attribute level Overall Gender Age Location
Con ‘no preference’ -2.447 (.000) -0.311 (.028) -0.234 (.059) (st.dev.) 1.519 (.000)
Accessibility by car
A1 1 obstacle 0.153 (.000) -0.087 (.038)
A2 3 obstacles
Table 5: Estimated parameters mixed logit model (continued)
Var. Attribute level Overall Gender Age Location
Accessibility by public transport
B1 3 min. walking 0.109 (.002) -0.085 (.039) B2 6 min. walking 0.063 (.083) B3 9 min. walking -0.109 -0.063 0.085 Parking tariff C1 Free 0.505 (.000) (st.dev.) 0.404 (.000) C2 €1,00 per hour 0.097 (.011) C3 €2,00 per hour -0.505 -0.097
Length of main shopping streets
D1 15 min. walking -0.227 (.000) (st.dev.) 0.531 (.000) D2 30 min. walking 0.104 (.007) D3 45 min. walking 0.123
Type of shopping supply
E1 National chains 0.000 -0.156 (.004) -0.082 (.085) (st.dev.) 0.515 (.000)
E2 Small/medium -0.113 (.002)
E3 Special/exclusive 0.113 0.156 0.082
Type of anchor stores
F1 Dept stores 0.206 (.000) 0.139 (.001) F2 Mega electro stores -0.236 (.000) -0.284 (.000)
(st.dev.) 0.444 (.000)
F3 Flagship stores 0.030 0.145
Type of traffic allowed
G1 Pedestrians 0.202 (.000) 0.095 (.026) (st.dev.) 0.184 (.038) G2 Peds + bicyclists 0.099 (.005) G3 All modes -0.202 -0.095 -0.099 Design style H1 Historical 0.404 (.000) -0.119 (.023) 0.204 (.000) 0.086 (.078) (st.dev.) 0.514 (.000) H2 Modern -0.134 (.003) (st.dev.) 0.361 (.000) H3 Disney style -0.270 0.119 -0.204 -0.086
Scale of shopping streets
I1 Short and narrow 0.103 (.026) 0.074 (.072)
I2 Long and wide -0.194 (.000) -0.113 (.018) -0.091 (.037)
I3 Mixed 0.194 0.010 0.017 Type of activities J1 Passive 0.000 0.109 (.012) (st.dev.) 0.357 (.000) J2 Active -0.126 (.002) -0.091 (.026) -0.084 (.056) (st.dev.) 0.176 (.070) J3 Mixed 0.126 0.091 0.084 -0.109 Interaction effects A1×C1 -0.088 (.072) A1×C2 0.089 (.072) B2×C2 -0.073 (.088)
The results are also shown in Figure 4. For three attributes (accessibility by car and public transport and the scale of the shopping streets) standard deviations were not significantly different from zero. This means that there is not much random variation across the respondents regarding these items. For the remaining attributes at least one part-worth utility is represented by a random parameter with a standard deviation significantly different from zero. Also the constant for the ‘no preference’ option has a random component. For two attributes levels (shopping supply by national chains (E1) and passive activities in the shopping centre (J1)), the standard deviation is significantly different from zero, while the corresponding mean value is not. This suggests that preferences regarding these attribute levels fluctuate around zero, cancelling out to neutral mean values.
A1 A2 A3 B1 B2 B3 C1 C2 C3 D1 D2 D3 E1 E2 E3 F1 F2 F3 G1 G2 G3 H1 H2 H3 I1 I2 I3 J1 J2 J3 Me an ( bl ac k ) S tan da rd de v iati on ( grey ) -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 Ma le ( bl ac k ) F em a le (gr ey ) -0.4 -0.2 0.0 0.2 0.4 Y ou ng ( b lac k ) O ld (gr ey ) -0.4 -0.2 0.0 0.2 0.4 R’ da m ( b lac k ) D. B os c h (gr ey ) -0.4 -0.2 0.0 0.2 0.4
In the upper part of Figure 4, it can be seen that the standard deviations are rather large compared to the mean values. Although standard deviations are non-negative by definition, the standard deviations in this figure were given the same direction as the corresponding mean values to ease interpretation. Note that if both the first and the second level of an attribute have significant standard deviations (Design
style and Type of activities), the standard deviation of the third level is equal to the root
of the sum of the squared standard deviations for the first and second level because the mixed logit model specified in the study assumes uncorrelated random parameters. Compared with the estimation results for the multinomial logit model, the estimated parameters in general have (as expected) a higher (positive or negative) value. Overall, however, the pattern of main effects and contrast effects is similar, apart from a few exceptions. The gender contrast effect for accessibility by car is no longer significant. On the other hand, the age effect on active activities has become significant according to the random parameter model. Furthermore the B1×E2 interaction effect in MNL model has been replaced by the A1×C1 and A1×C2 interaction effects, meaning that all interactions are related to the accessibility variables. The interaction effects are illustrated in Figure 5. If the interaction effects are omitted, the likelihood ratio statistic is significant at the p=0.068 level. Thus, if one sticks to the 5% significance level, the interaction effect may be deleted from the mixed logit model.
C1 C2 C3 C1 C2 C3 C1 C2 C3 C1 C2 C3 C1 C2 C3 C1 C2 C3 C1 C2 C3 C1 C2 C3 C1 C2 C3 B1 B1 B1 B2 B2 B2 B3 B3 B3 B1 B1 B1 B2 B2 B2 B3 B3 B3 B1 B1 B1 B2 B2 B2 B3 B3 B3 A1 A1 A1 A1 A1 A1 A1 A1 A1 A2 A2 A2 A2 A2 A2 A2 A2 A2 A3 A3 A3 A3 A3 A3 A3 A3 A3 W it ho u t i nte rac ti on eff ec ts ( bl ac k ) W it h i nte rac ti on eff ec ts ( grey ) -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
Figure 5: Attributes A, B, and C, with and without interaction effects; ML model
7. Conclusions and recommendations
Since the Dutch policy regarding mega shopping centre has become more liberal, some retail developers have prepared plans to develop such a centre. However, as mega shopping centres do not exist in the Netherlands yet, it is hard to assess customers’ preferences regarding these very large shopping facilities. Therefore, the purpose of this paper was to investigate customers’ preferences regarding shopping centre attributes that are considered relevant in the first stages of the design process of a mega shopping centre. A stated choice approach was used to measure customers’ preferences regarding
accessibility, scale and length of shopping streets and traffic allowed in the streets, the architectural design style, type of anchor stores, shopping supply, and type of activities in the shopping centre. In addition, parking tariff was included as a benchmark. Contrast effects for gender and age categories were included, as well as for the locations where respondents were recruited. Furthermore, interaction effects between attributes were also considered. Two types of discrete choice models were estimated, the standard multinomial logit model and the random parameter model, a mixed logit model allowing for taste heterogeneity. Respondents were asked to choose the most preferred shopping centre from choice sets containing two shopping centres (and a ‘no preference’ option) in the context of a recreational shopping trip.
According to both models, the parking tariff is the most important attribute followed by the design attribute. Next, although at some distance, type of anchor stores, accessibility by car, scale and total length of the shopping streets, type of traffic allowed in the shopping centre, are the most important attributes. Finally, type of activities in the shopping centre, type of shopping supply and accessibility by public transport appear to be the least important attributes.
There are however some noteworthy exceptions regarding the segments of respondents considered. For males, the anchor stores seem to be considerably less important than for females, while males put much more weight on the design attribute. Young respondents (15-24 years of age) attach relatively less weight to shopping supply and architectural design than the older respondents. The differences between respondents recruited in the Rotterdam Alexandrium shopping centre and the respondents recruited in the down town shopping centre of Den Bosch are less distinct. However, there are still significant differences between the two groups. It is unclear whether these differences originate from the differences in residential areas (the Rotterdam region versus the Den Bosch region) or from the difference in type of shopping centre used to recruit respondents (a peripheral indoor shopping centre versus an outdoor down town shopping centre).
Although structural differences between subgroups of customers have been taken into consideration by means of contrast effects, there is still considerable random variation in the main effect parameters. The random parameter model estimated significant variation in the utilities regarding all attributes, except accessibility by car and public transport and the scale of the streets in the shopping area. This means that some customers prefer a particular attribute level much more and others much less than average. The random parameter model also shows that in the case of some attribute levels (shopping supply by national chains and passive activities) significant, but opposing individual preferences exist, resulting in insignificant mean utility values. Taking into consideration this heterogeneity improves the performance of the model considerably. Rho2adjusted for the multinomial logit model is equal to 0.228, and for the random parameter equal to 0.269.
The number of mega shopping centres that can be realized in the Netherlands is limited. If a developer is planning to build one, thorough investigation regarding consumer preferences and shopping behaviour is important. This study provides some insights in consumer preferences regarding a mega shopping centre in the Netherlands. According to the main findings, it should be advised to implement a historical architectural design, contract department stores as the main anchors, find a location near a highway, create both long/wide and short/narrow shopping streets only allowing pedestrians and providing one to two kilometres of walking distance, and offering a mixture of both active and passive entertainment activities. Special and exclusive shops, as well as
shops from national chains should be provided. Finally, a good accessibility by public transport may be advisable, although this is the least important attribute. The preference for such a mega shopping centre can be considerably affected by manipulating (some of) the attributes investigated in this study. However, it should be noted these attributes are less important than the parking fee. A relatively high parking tariff may considerably reduce the utility of a well designed shopping centre. Furthermore, it should be stressed that preferences may vary extensively across consumers.
The findings from this study can also be used to determine the best selection of attributes levels for a specific segment of shoppers. If a developer wants to develop a shopping centre that is especially appreciated by young males or females, less special/exclusive shops and more national chains should be supplied. Also the young shoppers appreciate the historical architectural design style considerably less than the older customers. For the young females, the difference between a historical design style and a Disney design style is rather small. According to the mixed logit model, young males slightly prefer active activities in the shopping centre while the other shoppers prefer a mixture of both passive and active activities. Although a short walking distance between public transport stop and shopping centre is hardly relevant for older shopper, it may help attracting young shoppers.
References
Alberini, A., Riganti, P., Longo, A., 2003. Can people value the aesthetic and use services of urban sites? Evidence from a survey of Belfast residents. Journal of Cultural Economics 27, 193-213.
Bhat, C.R., 2001. Quasi-random maximum simulated likelihood estimation of the mixed multinomial logit model. Transportation Research 35B, 677-695.
Borgers, A., Van Swaay, S., Janssen, I., 2006. Assessing the impact of peripheral mega retail centres on traditional urban shopping centres. Belgeo 44, 53-66.
Dholakia, R.R., 1999. Going shopping: key determinants of shopping behaviors and motivations. International Journal of Retail & Distribution Management 27, 154-165.
Finn, A., Louviere, J.J., 1996. Shopping center image, consideration, and choice: anchor store contribution. Journal of Business Research 35, 241-251.
Gorter, C., Nijkamp, P., Klamer, P., 2003. The attraction force of out-of-town shopping malls: a case study on run-fun shopping in the Netherlands. Tijdschrift voor Economische en Sociale Geografie 94, 219-229.
Greene, W.H., 2007. NLOGIT Version 4.0: Reference guide, Econometric Software, Inc., Plainview USA and Castle Hill, Australia.
Haynes, J.B., Talpade, S., 1996. Does entertainment draw shoppers?: the effects of entertainment centers on shopping behavior in malls. Journal of Shopping Center Research 3, 29-48.
Hensher, D., Rose, J., Greene, W., 2005. Applied choice analysis: a primer. Cambridge University Press, Cambridge.
Howard, E.B., Davies, R.L., 1993. The impact of regional out-of-town retail centres: the case of the Metro Centre. Progress in Planning 40, 89-165.
Kent, T., 2007. Creative space: design and the retail environment. International Journal of Retail & Distribution Management 35, 734-745.
Kim, G., Kim, A., Sohn, S.Y., 2009. Conjoint analysis for luxury brand outlet in Korea with consideration of customer lifetime value. Expert Systems with Applications 36, 922-932.
Kozinets, R.V., Sherry, J.F., DeBerry-Spence, B., Duhachek, A., Nuttavuthisit, K., Storm, D., 2002. Themed flagship brand stores in the new millennium: theory, practice, prospects. Journal of Retailing 78, 17-29.
Louviere, J.J., Hensher, D.A. Swait, J., 2000. Stated choice methods: analysis and application. Cambridge University Press, Cambridge.
Marjanen, H., 1995. Longitudinal study on consumer spatial shopping behaviour with special reference to out-of-town shopping: Experiences from Turku, Finland. Journal of Retailing and Consumer Services 2, 163-174.
Oppewal, H., Timmermans, H., 1999. Modeling consumer perception of public space in shopping centers. Environment and Behavior 31, 45-65.
Oppewal. H., Timmermans, H., Louviere, L., 1997. Modeling the effects of shopping centre size and store variety on consumer choice behaviour. Environment and Planning A 29, 1073-1090.
Reimers, V., Clulow, V., 2004. Retail concentration: a comparison of spatial convenience in shopping strips and shopping centres. Journal of Retailing and Consumer Services 11, 207-221.
Ruiz, J.-P., Chebat, J.-C., Hansen, P., 2004. Another trip to the mall: a segmentation study of customers based on their activities. Journal of Retailing and Consumer Services 11, 333-350.
Sit, J., Merrilees, B., Birch, D., 2003. Entertainment-seeking shopping centre patrons: the missing segments. International Journal of Retail & Distribution Management 31, 80-94.
Suárez, A., Rodríguez del Bosque, I., Rodríguez-Poo, J.M., Moral, I., 2004. Accounting for heterogeneity in shopping centre choice models. Journal of Retailing and Consumer Services 11, 119-129.
Teller, C., Reutterer, T., 2008. The evolving concept of retail attractiveness: what makes retail agglomerations attractive when customers shop at them? Journal of Retailing and Consumers Services 15, 127-143.
Train, K.E., 2003. Discrete Choice Models with Simulation. Cambridge University Press, Cambridge.
Van der Waerden, P., Borgers, A., Timmermans, H., 2009. Consumer response to the introduction of paid parking in a regional shopping center. Transportation Research Record 2118, 16-23.
Williams, C.C., 1995. Opposition to regional shopping centres in Great Britain: a clash of cultures? Journal of Retailing and Consumer Services 2, 241-249.
Wakefield, K.L. Baker, J., 1998. Excitement at the Mall: determinants and effects on shopping response. Journal of Retailing 74, 515-539.
