University of Amsterdam
Programme «Quantitative Finance»
Big Data based customer traffic analysis:
The impact of four Key Point of Interests (KPIs) in the retail
sector.
Master Thesis
Student: Roxan T. M. Stevens
Student ID number: 10675248
Supervisor: Dr. T. Yorulmazer
Submission date: 01.07.2018
2018
Acknowledgements
I want to thank and recognize my supervisor mister Dr. Yorulmazer who guided and supported me throughout the whole process of writing my thesis.
Further, I want to thank the company RMC for their trust to allow me to use their database in order to find empirical answers to the posed research questions. Moreover, a word of gratitude goes to all the employees from RMC that provided me with their support and advice during their daily work and for their participation in this research.
Statement of Originality
I, Roxan Stevens, hereby declare to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that all references, to
Abstract
The retail sector is assumed to be the pre-eminent sector to benefit from new marketing strategies based on Big Data analysis. In this respect a thesis has been done in cooperation with the company RMC (KvK 29033906, Amsterdam). This company is tracking foot traffic in shopping areas through WiFi signals in 69 cities in the Netherlands. Based on the data gathered by these sensors, the relationship between four Key Point of Interest (KPIs) has been investigated. First, the relationship was investigated using the entire database, then a separation was made based on cities part of the Randstad or the G4, and on the time seasons. The results show that an increase in the number of unique visitors increases the percentage of new first time visitors, the frequency of visitors return, and the average dwell time. On the contrary, the percentage of new visitors, the average frequency, and the average duration are inversely correlated. Also the dependency of these four KPIs on the characteristics of a city was investigated. The results show that an increase in the amount of jobs in the “Culture, Recreation and other Services”-sector will decrease the unique number of visitors. At the same time the amount of jobs in tourism is positively correlated with this parameter. An increase in the average persons per household results in a decrease in the percentage of new visitors. Finally, an increase in the number of people receiving “WW” social aid decreases the average frequency, while an increase in the percentage of non-western immigrants will increase the frequency. This thesis reveals that Big Data analysis on foot traffic in shopping areas allows retailers to extract valuable information from their store surroundings. Additionally, the individual relevance and influence of the predetermined KPIs are confirmed. Based on these findings, new marketing strategies can be developed for retailers offering them a competitive advantage.
1. Introduction
6
2. Theoretical Framework
9
2.1 RMC 9
2.1.1 Key Performance Indicators versus Key Point of Interest 10
2.2 Customer traffic analysis 11
2.3 The use of Big Data in the retail industry 12
2.4 The use of the Internet of Things in the retail industry 14
3. Hypotheses & Methodology
15
3.1 Hypotheses 15
3.2 Research Method 21
4. Data and Descriptive Statistics
22
4.1 Data 23
4.2 Variables 28
5. Results
31
5.1 The unique number of visitors as dependent variable 31
5.2 The percentage of new visitors as dependent variable 32
5.3 The visitors frequency as dependent variable 33
5.4 The average duration as dependent variable 34
5.5 The dependency of the four KPIs on the city characteristics 35
5.6 VAR-model requirements check 36
6. Conclusion
38
Limitations 39
References
41
List of abbreviations
CBS: Central Bureau of Statistics
CRM: Customer Relationship Management IoT: Internet of Things
JB: Jacque-Bera test KPI: Key Point of Interest
OIS: Research, Information and Statistics VAR: Vector AutoRegressive model
WOMM: Word-Of-Mouth Marketing
List of tables
Table 1: The expected effect of the chosen eight city characteristics on the unique number of visitors, the percentage of new visitors, the average dwell time, and the frequency. Table 2: The expected interaction effect of the chosen four KPIs.
Table 3: The expected effect of the chosen eight city characteristics on the four KPIs. Table 4: Data characteristics of dataset retrieved from RMC on the four KPIs.
Table 5: Correlation table of the first dataset containing the four KPIs.
Table 6: Data characteristics of the second dataset containing the eight city characteristics and the four KPIs
Table A: Characteristics of the 69 cities of the dataset.
Table B: Correlation matrix of the four KPIs using the entire dataset.
Table C: Correlation matrices of the four KPIs concerning cities in and outside of the Randstad area.
Table D: Correlation matrices of the four KPIs concerning cities part of G4 and not. Table E: Correlation matrices of the four KPIs in summer, spring, autumn and winter. Table F: Regression results with the unique number of visitors as dependent variable. Table G: Regression results with the percentage of new visitors as dependent variable. Table H: Regression results with the visitors frequency as dependent variable.
Table I: Regression results with the average duration as dependent variable.
Table J: Regression results with one of the four KPIs as dependent variable, and the eight city characteristics as explanatory variables.
1. Introduction
The retail sector is one of the most rapidly developing industries of the world (Marr, 2015). New selling channels and advanced digital technologies provide new sources of data for retail organizations. From transaction information and sensor data to social media files, such data present great opportunities for retailers to gain a competitive advantage in an expanding and competitive industry. Retailers are constantly trying to find new and innovative ways to extract insights from the ever-increasing amount of information available about the behavior of their customers (Marr, 2015). The convergence of all these data coming from multiple sources is called Big Data. Big Data describes the large volume of data, both structured and unstructured, that deluges a company on a day-to-day basis. The importance of Big Data revolves around analysing it and trying to find solutions for cost and time reductions, new product development and innovation, and ultimately for smart decisions making (SAS, 2018,). According to Hofacker et al. (2016), the use of Big Data analytics can also be beneficial for retailers by studying consumer behavior. Consumer behavior researchers should use and exploit the opportunities that Big Data provides in order to generate new insights. For example, the use of Big Data has the potential to increase our understanding of each stage in the consumer decision-making process and therefore generate and increase the profits (Hofacker, 2016).
One of the ways to increase knowledge about customer behavior by Big Data is through customer traffic analytics. As stated by Nakahara and Yada (2011), shopping traffic data which records customer’s position and time information is increasingly being used as new marketing data. These analyses have enabled accurate measures for customer behavior in shopping areas. The required data can be gathered through sensors, actuators, and microprocessors. New technologies in the smartphone era enable this measurement of retail customer traffic. Location-sharing devices, like smartphones, will likely become a major metric for brick-and-mortar campaign success. When mining this real world data, retailers can determine what marketing strategies are working when comparing their data to the competition. This will ultimately change how retail promotions are evaluated and even how retailers compete (Handly, 2018).
Gathering data by using location-sharing devices is among other things part of the so called Internet of Things (IoT). Wigmore (2014) defines the Internet of Things (IoT) as the all-connected network of electronic physical devices. The term “things” refers to an “inextricable
mixture of hardware, software, data and service that allows for autonomous data exchange and service provision” (Wigmore, 2014). The detailed data collected from such objects can be implemented for optimization of goods, services, and business models.
For my thesis I got in contact with the company RMC, that focuses on customer traffic analysis. By using more than 400 sensors in 93 cities in The Netherlands and Belgium, RMC collects costumer traffic data by phone-tracking through WiFi-signals. Our phones are constantly searching for the strongest WiFi-signal available by pinging modems in its direct surroundings. The sensors of RMC measure 24/7 hours a day whenever a smartphone sends out such a pinging signal to any wireless access points nearby. This IoT solution can provide retailers with detailed data on customer traffic, illustrating the flow of customers throughout their stores surroundings, as well as revealing the time spent within the shopping area covered by the respective sensors of RMC. When store managers utilize this data efficiently, they can be proactive and efficiently organizing staff schedules, creating new strategies for product promotion, and creating an insight on the best placement for a new store within a certain shopping area.
Based on the collected data, RMC develops customer indexes for retailers, and inspect the effect of certain Key Point of Interest (KPI). In their opinion, the four most important KPI their retail customers should focus on are: the number of unique visitors, the percentage of new visitors under these unique number of visitors, the dwell time in the geographic shopping area, and the visitors frequency.
Upon request and in agreement with RMC, the purpose of this research is to get a better understanding of the relationship between the four stated KPIs. Given the opportunity to make use of this unique database, I wanted to elaborate on the variables in order to get a better understanding of the four KPIs. Therefore, their antecedents will also be researched in more detail by investigating their potential dependency on certain city characteristics. Eventually this thesis will try to answer the following research question:
What is the interacting relation between the four stated KPIs, and how do they depend on the characteristics of a city?
In order to answer this research questions, the research consists of two parts. To start, the interacting relationship between the four stated KPIs will be tested. This relationship will be compared between multiple separations of the dataset. The relation will be tested using
the entire dataset. Secondly, cities located within and outside of the “Randstad” are separated. More specific, the Randstad concerns all available cities located within the municipalities Amsterdam, Rotterdam, The Hague, and Utrecht.
Then, the dataset will be split in cities part of the “G4” and the other cities. The G4 are the four biggest cities of the Netherlands: Amsterdam, Rotterdam, The Hague, and Utrecht. Finally, the dataset will be separated according to the four seasons. These separations are made upon discussion and advice from RMC. In chapter 4.2 each separation of the database will be elaborated. For this part a fixed effects regression is used, with fixed effects for both the different cities and for all the weeks.
Given the opportunity to use such an unique database, this thesis will also try to get a more profound understanding of these four KPIs themselves. Therefore, the second part of the research focuses on trying to understand what city variables explain the four used KPIs. In other words, what city characteristics explain their movements? The city characteristics chosen are the average number of people living within a household, an index for the crowdedness of the city, called the “cityness”, the age of the population split in five categories, the percentage of non-western immigrants in the population, the number of citizens receiving two types of social welfare, the number of business establishments in the following two sectors: “Retail, Food & Beverages”, and secondly, “Culture, Recreation, and Other Services”, and finally the amount of jobs in the tourism sector as an index for the number of tourists. For this part will be a Vector AutoRegression model (VAR) used. In chapter 3.2 the rationale of this chosen methodology will be provided. Further, in chapter 4.2 are the respective variables will be explained in more detail.
The entire dataset for this research consists of weekly data on the four KPIs, from August 2016 until May 2018 from 69 cities in the Netherlands. RMC has sensors in more cities in the Netherlands, but due to incomplete data from some of the cities, only 69 cities will be used for this research. Based on the lack of sufficient sensors in Belgium cities, and intending to meet the requests of RMC, this research will focus on the Netherlands, and therefore not including the sensors in Belgium. Since research on the benefits of retail traffic and the use of these particular KPI is non-existent, the conclusion of this thesis will contribute to theory.
This thesis will have the following structure. First, the theoretical framework is presented, in which also the company RMC is introduced. Hereafter, the use of shopping
traffic data in order to get a better understanding of customer behavior will be explained. Finally, without getting in too much detail about information technologies, will this literature based chapter be concluded by elaborating on the applications and benefits of Big Data and IoT in the retail business. In the third chapter are the hypotheses and the selected methodology to test this relationship provided and explained. The fourth chapter will outline the dataset constructed with RMC and are the characteristics discussed. Chapter five presents a discussion on the findings of the regressions and a variables check for the VAR model assumptions. The last chapter concludes the work and considers the limitations of the research.
2. Theoretical Framework
In this chapter is elaborated on the theoretical concepts underlying this thesis. More specific, the theoretical framework will describe: general information about the company RMC, Key Performance Indicators versus Key Point of Interest, customer behavior and traffic analysis, and the use of Big Data and the Internet of Things in the retail sector. The presented theory will be used when analysing the empirical data further on in this thesis, and will be the foundation for the conclusion.
2.1 RMC
This research is written in collaboration with the company RMC. RMC is an Amsterdam based company that provides customer analysis to retailers and municipalities based on the “CityTraffic Method”. This method is based on phone tracking through WiFi signals. Whether a smartphone is connected to a WiFi network or not, it is constantly searching for new WiFi networks to add to the list of available networks close by. Moreover, when a phone’s WiFi radio is active, it will constantly sends out signals to wireless access points. RMC has placed more than 400 sensors in 93 cities in the Netherlands and Belgium that keep track of the number of times a smartphone sends such a signal to one of the networks nearby. These WiFi-based sensors collect data in some of the most crowded shopping areas in these particular cities for 24 hours a day, 7 days a week, and 365 days a year. This database provides an accurate insight into the foot traffic related measurements of the passengers, such as customer frequency, dwell time, and the walking trends of customers.
Based on this accurately collected database RMC develops and provides retail indices, which can be useful for retailers and municipalities. Their main index is the National Shopping Index for the Netherlands and Belgium. This index summarizes and visualizes the collected CityTraffic data, where the impact of the weather is taken into account. Using this index, a retailer is able to analyze and understand the state of the passenger traffic of the past. Based on this data retailers can for example evaluate their performance of a certain period relative to a previous period. Or they can analyze the impact of a newly placed marketing or promotion initiative on the foot traffic around their store. More about the usefulness of this information is provided later on in the literature review.
Next to providing these indices, RMC also focuses on the significance of certain Key Point of Interest (KPI). In their opinion, the four most important KPIs that their retail customers should focus on are: the number of unique visitors, the percentage of new visitors amongst these unique visitors, the average dwell time, and the frequency visitors return. These KPIs will be specified more in detail at the variables descriptive in chapter 4.2.
An example of one of the cases RMC researched is the investigation on the best location for the first store of Kiehl’s in the Netherlands. L’oreal started the luxury cosmetic brand Kiehl’s and wanted to open a monobrand store in the Netherlands. The assignment for RMC was to advise what the best location for their first store would be to introduce their newly developed cosmetic brand. To come up with a well-reasoned proposition, RMC started by making a long list of all the shopping streets in the Netherlands. They collected passengers data and conducted traffic analysis for each of these shopping streets. Additionally, they conducted a location- and competition analysis. After they qualified and evaluated certain establishment criteria for all streets, they criticized all the possible establishment locations by considering the different aspects of each establishment location using one of their own developed models called “The Best Location”. This project resulted in a quantitative and quantitative based and well-argued advice regarding the best place of establishment for the first Kiehl’s store for L’oreal.
2.1.1 Key Performance Indicators versus Key Point of Interest
According to Anand and Grover (2015), the measurement of business performance has become a key focus of management and the source of competitive advantage and value for firms. Improving a firms performance is a continuous method that requires a performance
measurement system to quantify the efficiency and effectiveness of decisions and actions (Anand & Grover, 2015) . Such a system to help a company measure their performance on a daily basis are based on a certain set of measurements, called Key Performance Indicators (KPIs) (Gopal & Thakkar, 2012). These KPIs can be either financial or non-financial performance measures. As Ananad and Grover (2015) states, performance measures are critical to achieve the identified target by the firms. Moreover, it is not only about objectively measuring performance. Each retailer sets his own KPIs that not just become the target for the sales team and the scorecard in marketing department, these KPIs will also be integrated within the politics, emotions and several other behavioral issues within the firm (Gopal & Thakkar, 2012).
The service RMC provides to their customers is focused on some benchmark variables to create value. Instead of Key Performance Indicators, they focus on their own developed Key Point of Interest. However, the idea of both measurements are the same. According to RMC retailers should focus on these four Key Point of Interest to extract as much possible valuable information from their store surroundings. By analysing and trying to positively influence these factors, retailers will be able to create value and in consequence achieve a competitive advantage, just as when focusing on Key Performance Indicators. Since both benchmark variables are abbreviated by ‘KPI’, in this thesis will this abbreviation refer to Key Point of Interest instead of Key Performance Indicators.
2.2 Customer traffic analysis
As stated by Howe (2014), customer relationship management (CRM) has always been one of the most important strategic issues for retailers. The development of sensor network technologies in the recent years is significantly changing the market strategy situation. The development of sensor networks has enabled accurate measuring of customer behavior in shopping areas. The increasing and accumulated knowledge of customer behavior enables retailers to adapt their business strategies by leveraging this intelligence (Howe, 2014).
Measuring the foot traffic in shopping areas provides interesting and valuable information on store activity trends, peak periods, and the assessment of high-impact variables holidays, national feasts, and the weather (Howe, 2014). Measuring this foot traffic in shopping areas can create a significant amount of valuable insights to retailers. Several aspects are marketing and advertising, and staffing (Vend, 2018). For example, foot traffic
enables to research the effects of newly placed advertising and banners. Furthermore, when comparing the foot traffic around the store before and after the new store displays, it can be determine whether it attracts new potential customers. Retailers need to understand what their collected data entails and search for explanations for fluctuations of the data. Possible reasons could be the effect of the weather on the dwell time of customers, or potentially another retailer close by introduced a major sale and therefore lurked customers towards his or hers store. Therefore, it is more than just analysing the collected customer traffic data (Nicasio, 2014).
Another way retailer can create real value using foot traffic analytics, is by combining these insights with operational data, such as workforce management. Information about store peak traffic hours can be useful when making staffing decisions accordingly. Whenever there is a peak, a retailer can decide to put more associates in the store during this time to ensure a positive and beneficial staff to customer ratio. On the other hand, the opposite can be done when there is a drop of customers to be expected (Vend, 2018). In this way costs of labor can be planned efficiently and in consequence return on labor investment be improved.
Based on the KPIs that are taken into consideration, a customer traffic flow map can be made. The routes mostly taken by customers within a shopping area can identified. Radas and Lewis (2009) contributes to the retailing literature by clarifying the respective roles of store traffic and customer traffic flow. Their results show that gross customer traffic flow and realized customer traffic flow are strong drivers of money spending. Therefore, the understanding of these customer flow is of great value of retailers.
Furthermore, also for municipalities shopping area traffic analysis can provides valuable knowledge. For example, it can clarify the geographical layout of the shopping area. Moreover, foot traffic analysis can confirm which parts of the area are getting relatively more or less traffic. In addition, the data can also show if there are any bottlenecks that disrupt the traffic flow. All this knowledge enables retailers and municipalities to improve to optimally organize the shopping area (Vend, 2018).
2.3 The use of Big Data in the retail industry
There is not an overall definition of Big Data itself. Forbes defines Big Data is “a collection of data from traditional and digital sources inside and outside your company that represents a source for ongoing discovery and analysis” (Arthur, 2013). However, a more frequently used
definition states that Big Data concerns three V’s which are: volume, velocity, and variety. Big Data is high-volume, high-velocity, and high-variety information assets that use effective and innovative ways of information processing to create enhancing insights and for decision making. Where volume is the scale of the data, variety is about the many different sources and forms of the data, and velocity is about the speed of the data (Sicular, 2013).
Nowadays Big Data analytics are taking over the retail industry. As seen in the study by Khale (2016), Big Data can help to turn customer analytics into value-increasing information. Furthermore, it allows companies to predict buyer behavior. This enables the respective organizations to improve their sales numbers, market optimizations, inventory planning, and many more applications. McKinsey also highlights the importance and relevance of the use of Big Data analytics for retailers. They estimated that the use of Big Data specific in the retail sector increases the margins by up to 60%. The increasing and accumulated knowledge of customer behavior enables retailers to adapt their business strategies by leveraging this intelligence (Brown et al., 2011).
Elragal and Raslan (2014) state that the use of Big Data analytics has great potential for customer intelligence. These analyses allow for more transparency, and makes relevant data more accessible to managers and stakeholders. Therefore, they provide the ability to make more informed strategic decisions, by developing predictive models for customer behavior and purchase patterns, and therefore raising overall profitability. Additionally, amongst other things Big Data analytics can be used to gain a better and more profound understanding of locations and customer frequency (Elragal & Raslan, 2014).
Not everybody is as positive about the general use of Big Data. The universality of the internet combined with the ability to easy and eternally store endless amounts of data, means that our personal information is spread out entirely through the digital universe. Personal data gets collected, even when we are not online. Data can be collected and traced back to individuals in various forms without the person in question being aware of it. This can be done through internet tracking, location data, and video data amongst other things (Church & Kon, 2007). These possible invasions in our privacy is one of the biggest concerns of many ordinary people and politicians have about real-time analysis of Big Data (Hayes, 2015). Another negative aspect that is easily underrated, is the logistical issue of Big Data. Companies planning on analysing Big Data will need to consider data identification, data security, and a Big Data governance strategy (Ciklum, 2017). It is already a challenge to
compile and store those terabytes of information on consumers, from which much is unstructured data from social media sources. It is another challenge to organize all this data and be able to extract the crucial insights hidden in the data, which enables retailers to understand, and even predict, the behavior of consumers in real life (van Zanten, 2012). Furthermore, the credibility of the data is an important issue when implementing Big Data solutions. The risk of having poor data or a significant amount of noise is a common problem for all types of organizations. However, validating all the available data can costs an enormous amount of resources (Redman, 2013).
2.4 The use of the Internet of Things in the retail industry
The Internet of Things (IoT) is, just as the use of Big Data analytics, a paradigm that aims to bridge the digital world with the real world. According to Whitmore et al. (2015), the IoT is just as Big Data analytics expected to change the way customers experience shopping. While the exact definition of IoT is still in its formation stage, it is generally described as a paradigm where everyday things or objects are embedded with technology that equips them with sensing, identifying, networking and processing capabilities. These capabilities enable them to communicate with other devices and services through the internet (Whitmore, Agarwal, & Xu, 2015). It is stimulating innovation and new opportunities as every device, consumer and activity are integrated into the digital realm. Given the potential of IoT to create new opportunities and business perspectives, it is receiving worldwide attention from researchers, practitioners, and the public (Gregory, 2015).
While the applications of the IoT technology differ across industries, one of the most prominent areas of its application includes the retail industry (Pantano & Timmermans, 2014). The IoT is empowered by a wide range of technologies (Manyika & Chui, 2015). Such as hardware, sensors, devices, mobile apps, telemetry, data, and cloud connection. Furthermore, it is the new answer that drives an innovation in the retail industry and is a solution for the overwhelming number of choices. A retailer is able to create a perfect customized strategy for every individual and boost production frontiers by having enough customer data (Chernev, 2013).
As said before, particularly in the retail industry the use of IoT will leave its mark. The IoT is empowered by a wide range of technologies. Such as hardware, sensors, mobile apps, data, and cloud connection (Manyika & Chui, 2015). Already retailers are experimenting with
ways to use intelligent, connected devices to offer new service and also to reshape the customers experience. The IoT movement offers retailers opportunities in three areas: customer experience, the supply chain, and new channels and revenue streams (Gregory, 2015).
One of the ways IoT can create valuable information for retailers, is by the ability to tap into the GPS location through smartphones. The CityTraffic Method created by RMC allows for WiFi to be the IoT data gathering medium concerning this purpose. Most electronic mobile devices regularly emit a WiFi signal, even though it is not connected to a certain WiFi signal. The mobile device often sends out a signal seeking for surrounding networks to potentially connect with. These signals they send out contain an identification marker, allowing a Wi-Fi sensor point to associate the market as a unique mobile device. However, the sensor cannot identify an individual shopper by any personal information. It can hereby simply determine if it has “seen” a particular marker before or how long this marker has been in the geographic area of the sensors. Wi-Fi can re-identify the same mobile device as long as it moves within the same geographic area, or if it returns another time (RetailNext, 2018). Another aspect in which retailers can benefit from analysing the GPS location of customers, is by creating a digital geofence around their store to alert customers. Whenever a customer enters this geofence, the retailer can send them personalized messages and offers to try to lure them to the store (St. Louis, 2018).
3. Hypotheses & Methodology
In this section the hypotheses will be introduced and the methodology to study these hypotheses will be described. The purpose of this chapter is to describe the selected hypotheses in accordance with the requests of RMC. Hereafter, the methodology that will be used will be outlined. The goal of this study is to gain a profound understanding in the relation between the four stated KPIs, and research how eight pre-selected city characteristics explain these four KPIs.
3.1 Hypotheses
Since this research is done in cooperation with the company RMC, the primary goal is to address RMC’s strategic requests. RMC wants to investigate the relation between four stated KPIs and whether this relation varies when dividing the dataset according to different
benchmarks. See table 2 below for the expected effect of each KPIs as explanatory variable on the other KPIs as dependent variables.
Table 2: The expected effect of the expected effect of the four KPIs as explanatory variables (vertical row) on the other KPIs as independent variables (horizontal row).
Based on this table the hypotheses to be tested are the following:
H0: The percentage of new visitors, the average dwell time, and the frequency will have a positive effect on the unique number of visitors.
H0: The unique number of visitors will positively affect the percentage of new visitors, on the contrary to the negative effect of both the average duration and the frequency on this percentage.
H0: The unique number of visitors, the percentage of new visitors and the frequency will have a positive effect on the average dwell time.
H0: The unique number of visitors and the average duration will have a positive effect on the frequency, on the other hand the percentage of new visitors will have decreasing effect on the average frequency.
The theory behind most of the positive expected relations is based on Word-of-Mout Marketing (WOMM). For example consider the first hypothesis, if the frequency and/or the duration increases, intuitionally is expected that customers appreciate the shopping area. That is why they intend to return to that particular shopping area and/or stay there for a longer period of time. Whenever a customer prefers a certain shopping area, they will tell their friends and others. This will result in new people to be attracted to the shopping area, which in turn will increase the percentage of new visitors and the unique number of visitors. This
Unique # visitors New % visitors Dwell time Frequency
Unique # visitors + + +
New % visitors + +
-Dwell time + + +
phenomena is called Word-Of-Mouth Marketing (WOMM). According to Forbes (2014), WOMM is the the most valuable form of marketing. In a survey done by Nielsen (2012), with more than 28,000 respondents in 56 countries, 92% of consumers believe recommendations made by their friends and family better than all forms of advertising.
The reasoning why the average dwell time and the percentage of new visitors will negatively affect each other, is due to tourism. A large percentage of new visitors can be the result of a lot of tourists as visitors. Additionally, tourists will lower the frequency as they will not return to the same shopping area again.
Due to the opportunity to use such an interesting and unique dataset, I wanted to get a more profound understanding of the four KPIs than just their interacting relationship. Therefore, the first part of the research will elaborate on the dataset and demands of RMC, and the second part focuses on getting a better understanding of the four KPIs itself. Eight city characteristics are chosen that possible impact the four KPIs: (1) the average number of persons living within one household, (2) an index for the crowdedness of the city, called the “cityness”, (3) the age of the population split in five categories, (4) the percentage of non-western immigrants in the population, (5) the number of citizens receiving social welfare, the number of business establishments in the following two sectors: (6) “the Retail and the Food & Beverage Industry” and secondly, (7) “Culture, Recreation, and Other Services”, and (8) the number of jobs in the tourism sector as an index for the amount of tourists. See table 3 for the expected effect of the eight chosen city characteristics on the unique number of visitors, the percentage of new visitors, the dwell time, and the frequency. Here a plus (+) indication means an expected positive effect and a minus (-) indication means an expected negative effect.
Table 3: The expected effect of the chosen eight city characteristics on the unique number of visitors, the percentage of new visitors, the average dwell time, and the frequency.
Unique # visitors New % visitors Dwell time Frequency Number of persons per household + + + +
Cityness index + + +
-Age population +/- +/- +/-
+/-Percentage non-western immigrants - - - -Percentage receiving welfare + + + +
The average number of persons within a household will increase all KPIs. When more persons live within a city, there will be a higher unique number of visitors. As a result of WOMM this will also attract more new visitors. Whenever the unique number of visitors increases, the shopping area will become more attractive for retailers to open a business establishment there. This will cause more shops to locate in this certain shopping area. With the area becoming more attractive, people will stay for a longer period of time and probably come back more frequently. Consequently it will have an increasing effect on all the four KPIs. The level of cityness will have a similar effect as the average number of people in one household. A more crowded city, represented by a higher level of the cityness index, will result in a higher number of unique visitors. As said earlier, WOMM will cause the percentage of new visitors to increase. However, a lower level of the cityness index indicates that the population is more widely spread and therefore the distance to the shopping area may increase. When the travel time to the shopping area therefore increases, people will go less often to the shopping area, resulting in a negative effect on the frequency. However, whenever a visitors did bridge the distance, he or she will stay for a longer period of time and buy all they want to make sure they do not have to travel there again soon. This results in an increasing effect on the average dwell time.
The age of the population is separated into five categories: (1) 0 - 15 years, (2) 15 - 25 years, (3) 25 - 45 years, (4) 45 - 65 years, (5) 65+ years. The first and last categories will probably be the less mobile categories of the five, and will therefore have a negative effect on all the four KPIs. The other three categories will mostly include students and working people. Therefore, these age categories will have the opportunity and especially the resources to go shopping, causing a positive effect on all four KPIs.
Boheim and Mayr (2005) researched the relation between immigration and public pending. Their results showed that an increase of highly educated immigrants will increase the amount of public spending, and an increase of low educated immigrants will decrease the amount of public spending. However, taking into consideration multiple factors, the overall results showed a decrease of public spending due to an increase of immigrants. Based on
Retail, food & drinks industry + + + + Culture, recreation, and others + + + +
-this results, the expected effect on the four KPIs by the percentage of non-western immigrations will be negative.
The amount of people receiving social welfare concerns two types: (1) social aid for whenever someone is temporarily or permanently without a job, this is called “WW” , and (2) for people who have a job but are incapable of earning sufficient to maintain a certain livelihood, this is called “Bijstand” (CBS). Without a job, people have more time to go to a shopping area. Even though they potentially do not have the money to spend in the shopping area, they have the time to travel there. Therefore the expected effect of the number of people receiving social aid will be positive on the four KPIs.
The amount of jobs in the tourism sector is an index for the amount of tourisms. Whenever the number of jobs increases, the amount of tourists is presumed to increase too. More tourists within a city will increase the number of unique visits, the percentage of new visitors and the dwell time. It will however have a negative effect on the frequency, since tourists generally not come back multiple times to a city within the same week.
Based on these expected effects the following two sub-hypotheses are constructed: H0: The average number of persons living within one household, the level of cityness, the percentage of people within the age of (1) 15 - 25 years, (2) 25 - 45, and (3) 45 - 65 years, the number of people receiving social welfare, the number of business establishments in the two relevant sectors and the number of jobs in the tourism sector will have a positive effect on the four KPIs.
H0: The percentage of people within the age of (1) 0 - 15 years, and (2) 65+ years, the percentage of non-western immigrants will have a negative effect on the four KPIs.
The first part of the research, concerning the relation between the four stated KPIs, will be first tested on the entire dataset including all the 69 cities. Hereafter, three different separations of the dataset are tested upon request of RMC. The second part of this research, which is focused on the two sub-hypotheses, which will be tested on the entire dataset only. In chapter 4.1 the characteristics of the exact database will be given in more detail.
In total there is weekly data collected from 69 cities in the Netherlands between the period period of August 2016 until May 2018 considering the four described KPIs. For the eight city characteristics there is yearly data is collected from 2013 until 2018.
Concerning the main hypothesis, the first parting of the main dataset is based on cities that are part of the Randstad area, and cities located outside of the Randstad. The Randstad is not a single city in the Netherlands, but a conglomerate of large to midsize cities in the Netherlands. Nearly half of the country’s population live in this dynamic metropolitan (i.e. 8.1 million citizens). With a gross regional product of 367 million euros, the Randstad is the fourth-largest metropolitan region in Europe. The region geographically offers direct connections to the rest of the world, which makes the region in addition to other factors appealing for international businesses, conferences and tourists (Regio Randstad, 2017). Due to this international economical capacity it is presumed to be an interesting draw out of the entire sample to test the hypotheses on.
According to the CBS (2000), all the cities within the municipalities Amsterdam, Rotterdam, The Hague, and Utrecht are part of the Randstad. From our available database this concerns the following cities: Amstedam, Amstelveen, Rotterdam, The Hague, Delft, Leidschendam-Voorburg, Utrecht, Nieuwegein, and Zeist.
The second separation of the dataset is based on the so called ‘G4’. The G4 represents the partnership of the four biggest cities in the Netherlands. According to CBS, the cities are Amsterdam, The Hague, Rotterdam, and Utrecht. These cities are presumed to have similar dimensions, composition of population, and at least 250.000 citizens (OIS, 2013).
The third separation is based on the weather seasons. It will be tested whether the relation between the four KPIs differ between the different seasons. The exact starting dates of the seasons change each year. Based on the astrophysical information provided by KNMI, the spring is supposed to start on the 20th of March until the 20th of June, summer starts at the 21th of June and ends at the 20th of September, autumn is from the 21th of September until the 20th of December, and lastly winter starts at the 21th of December until the 20th of March (KNMI).
This last separation potentially explains the effect of the weather on the consumer decisions. Moreover, research have shown that an increase of sunlight exposure tends to negatively affect the consumer spendings (Murray, Muro, Finn, & Leszczyc, 2010). According to KNMI the sun hours in in 2017 the Netherlands was 636 hours in the summer, 610 hours in the spring, 330 hours of sunlight in the autumn, and in the winter of 2017 to 2018 there
were 244 hours of sunlight exposure. Based on this different amount of average sunlight hours we can expect a different relation between the KPIs between the four seasons.
3.2 Research Method
The first part of the research will focuses on the main hypothesis to test the relationship between the four KPIs. This will be tested using a fixed effect regression model. A fixed effecs regression model employed in a panel data is an analysis that allows to control for time-invariant individual characteristics that possibly are correlated to the observed independent variables (Imai & Kim, 2017). In this case, fixed effects for both the weeks of the year and for each city are taken in consideration. To test this hypothesis four regressions will be run. Each time another KPI is used as the dependent variable, with the other three KPIs as independent variables. Running these four regressions will not just show the relation to one another, which is the case by a simple correlation matrix, but the exact effect of one KPI on another. These four regressions are conducted on the entire dataset, and on each of the three separations. The regressions to test the main hypothesis will look as follows:
!
Where Unique means the unique number of visitors, New means the percentage of new visitors of this unique number of visitors, Minutes is the average dwell time expressed in minutes, and Frequency is the average weekly frequency. The theta stands for the fixed effects for each city, and the phi represents the fixed effects for each week. The last part of the regressions is the error term.
The second part of the research focuses on the effect of eight city characteristics on the four KPIs. An adequate model to test this is a Vector AutoRegressive model (VAR). One of the assumption of the VAR model is that it only runs on a time series data. Since the dataset originally was panel data, it first needs to be converted into a time series. Due to the unbalanced dataset, only the first lag of each dependent variable can be taken into the
regression. Attached to the use of a VAR model there are some requirements that needs to be followed in order to be able to run a VAR model. In chapter 5.6 the results of testing the dataset and variables on the needed requirements are presented.
The regression to test the two sub-hypotheses will be tested four times, whereas each time one of the four KPIs is the dependent variable. The results of a Lagrange-Multiplier (LM) test showed that a VAR model with only the lag of the dependent variable is a better model for the data than a VAR model with the lagged variables of all the independent variables included. Therefore, the regressions will look as follows:
Where Y is one of the four KPIs, Y(t-1) is the first lag of the dependent variable, Pphouse is the average people in one household, Ctynss is the level of the cityness index, Age1 are people between the age of 0 - 15 years, Age2 are people between the age of 15 - 25 years, Age3 are people between the age of 25 - 45 years, Age4 are people between the age of 45 - 65 years, Age4 are people between the age of 65+ years, Imgrnts is the percentage of the population that are non-western immigrants, WW is the amount of people receiving the so-called “WW” social aid, Bijstnd is the amount of people receiving social aid called “Bijstand”, RFB is the number of business establishments in the “Retail and Food & Beverages” industry, CR is the number of business establishments in the “Culture, Recreation, and Other Services” industry, and final Trsm is the number of jobs in the tourism industry.
4. Data and Descriptive Statistics
In this chapter the dataset used for the research is described. First, the database received from RMC concerning the four KPIs is described. Secondly, the collected database to research the effect of eight city characteristics on the four KPIs is explained in more detail.
4.1 Data
For the first part of the research which investigated the interacting relation between the four KPIs, weekly data is used. As mentioned, this database exists of 69 cities within the Netherlands. See table A of the Appendix for the full list of the cities and there characteristics. As said before, due to the research questions from RMC this research only focuses on the Netherlands, despite having a number of sensors placed in several cities in Belgium. Not all the sensors of RMC are placed at the same time, and in consequence are the starting dates different between cities. For each city there is at least data from week 49 of 2016 until week 21 of 2018 is available, whereas for some cities there is also data available from weeks earlier in the year 2016.
The dataset I received from RMC required some readjustments to eventually become a reliable dataset used for this research. The original dataset existed of (1) the total number of visitors, (2) the unique number of visitors of this total amount, (3) the number of new visitors from this unique number of visitors, (4) the average duration expressed in seconds, and (5) the weekly frequency. Again, this all concerns all weekly values and averages. To retrieve the percentage of new visitors from the unique number of visitors, I divided the number of new visitors by the unique number of visitors. To express the average weekly dwell time in minutes instead of seconds, I divided the average duration by 60. See table 4 below for the characteristics of all the variables in the original dataset and the final dataset used for the research.
Table 4: Data characteristics of dataset retrieved from RMC. These are weekly numbers and averages for total amount of visitors, the unique number of visitors, the new number of visitors from the unique number of visitors, the percentage of new visitors from the unique number of visitors, the average duration expressed in seconds, the dwell time expressed in minutes, and the average weekly frequency.
Obs. Mean Std. Dev. Min Max Total # Visits 5.394 16604.63 22112.84 0 178972 Unique # Visits 5.394 12925.73 16464.21 0 133476 New # Visits 5.394 6385.25 8900.42 0 79198 New % Visits 5.394 0.5166 0.1823 0 1 Duration Avg. 5.394 4336.51 936.69 0 11370 Minutes 5.394 72,275 15,612 0 189.5
First, the entire database has been used to test the relation between the four KPIs. See below for the correlation matrix between the four KPIs. A correlation between |0.5| and | 1.0| can be interpreted as a strong correlation. Any correlation below |0.5| is considered to be a weak correlation. One can see in table 5 below that only the percentage of new visitors and the visitors frequency have a strong, negative, relation. This means that whenever the percentage of new visitors decreases, the relation visitors frequency will increase, and vice versa. A possible explanation for this correlation could be that an increase in frequency suggests many regular visitors, and therefore less new visitors. The other three of the KPIs all have weak correlations.
Table 5: Correlation table of the unique number of visors, the percentage of new visitors from this unique number of visitors, the dwell time, and the frequency visions return. All the variables are on a weekly basis. For this table all the data of the 69 cities is used.
The first separation in the database to test the main hypotheses is based on the Randstad. See table B.1 and B.2 in the Appendix for the correlation matrix for cities part of the Randstad, and cities outside of the Randstad. The correlation between the four KPIs are relatively similar for both group of cities, as most corrections are considered to be weak. For cities outside of the Randstad the percentage of new visitors and the frequency have a strong negative relation equal to -0.6278. For cities within the Randstad this has a value of -0.4863, which is also a negative relation, but smaller than |0.5| and therefore not considered as a “strong correlation”. Another difference is that for cities within the Randstad, the average dwell time and the frequency have a strong positive relation. For cities outside of the Randstad this is a quite week correlation, but still positive.
Freq. Avg. 5.394 1,223 176 0 1.8
Unique # visitors New % visitors Dwell time Frequency Unique # visitors 1,000
New % visitors -0.1127 1,000
Dwell time 0.3699 -0.0562 1,000
The second split of the dataset is based on the G4. The G4 includes Amsterdam, Rotterdam, Utrecht, and The Hague. See table C.1 and C.2 in the Appendix for the correlation matrices of the G4 and cities outside of this area are given. Just as when comparing cities in- and outside of the Randstad, only the correlation between the percentage of new visitors and the frequency is considered to be a strong correlation. With a difference of 0.0001 they have an almost identical negative correlation for cities part of the G4 and for the other cities.
The last separation of the dataset is based on the four seasons. In the Appendix the tables D.1, D.2, D.3 and D.4 give the correlations matrixes between the four KPIs for the summer, spring, autumn, and winter. Again, only the correlation between the visitors frequency and the percentage of new visitors is considered to be strong. During the autumn this correlation is the strongest, namely -0.7611.
The second part of the research is focused on explaining the four stated KPIs. For this part data is collected on the following variables: (1) the average number of people living within a household, (2) an index for the crowdedness of the city, called the “cityness”, (3) the age of the population split in five categories, (4) the percentage of non-western immigrants in the population, (5) the amount of citizens receiving social aid, the number of business establishments in the following two sectors: (6) “the retail and the food & beverage industry” and secondly, (7) “culture, recreation, and other services”, and (8) the number of jobs in the tourism sector as an index for the amount of tourists. In the next chapter the chosen variables will be explained in more detail. For all of these variables yearly data from 2013 and 2017. Except for the amount of jobs in tourism which is collected from LISA, all the data is collected from the Central Bureau of Statistics (CBS). The dataset from the first part of the four KPIs need to be recalculated from weekly into yearly data, as resulted in a unbalanced dataset, since the two datasets only have overlap in 2016 and 2017. See the table 6 below for the variables characteristics for the dataset including both the variables from CBS and LISA, and the four KPI.
Table 6: Data characteristics of the dataset used for the second part. These are yearly numbers and averages for the number of people per household, the index of “cityness”, the number of people between the age of 0 - 15 years, 15 - 25 years, 25 - 45 years, 45 - 65 years, 65+ years, the percentage of the population which are non-western immigrants, the amount of people receiving “WW” or “Bijstand” social aid, the number of jobs in “the retail and the food and beverages industry” and in “culture, recreation, and other services”, the number of jobs in tourism, the unique number of
visitors, the percentage of new visitors from the unique number of visitors, the dwell time expressed in minutes, and the average yearly frequency.
The correlation between the eight city characteristics is presented in table E in the Appendix. There is to be a strong positive correlation between the population between 0 - 15 years and the population of 25 - 45 years. Between the age of 25 until 45 years old people typically will start a family, which will increase the population 0 until 15 years old. The percentage of the population between 45 - 65 years is strongly positive correlated to the percentage of people between the age of 0 -15 years, 15 - 25 years, and 25 - 45 years. More people with the age between 45 and 65 year could be explained by a lot of parents with children or even grandchildren. Therefore, the percentage of people younger will increase as the percentage of people between 45 and 65 increases.
The percentage of the population that are non-western immigrants has a positive strong correlation with the percentage of people in the age of 0 -15 years old, 15 - 25 years, and 25 - 45 years. The majority of immigrants are people searching for work and/or for a
Obs. Mean Std. Dev. Min Max Source Persons per household 320 2,205 145 1,697 2,547 CBS Cityness 320 2194 838 1 4 CBS 0 - 15 years 320 17,023 5,954 5,146 85,060 CBS 15 - 25 years 320 12,794 5,448 4,045 72,506 CBS 25 - 45 years 320 26,341 12,032 6,431 184,345 CBS 45 - 65 years 320 28,622 9,274 10,453 137,580 CBS 65+ years 320 18,355 6,147 7,341 91,067 CBS Immigrants % 320 14,839 8,706 3 91,067 CBS WW 320 2794.844 3476.479 90 22650 CBS Bijstand 320 4217.125 7715.017 50 42710 CBS Jobs RFB 320 2,061.813 2,644.584 170 20,380 CBS Jobs in CR 320 1,389.625 2,557.564 55 22,970 CBS Jobs Tourism 312 7,415.083 14,016.07 80 127,224 LISA Unique # 187 10670.78 15351.55 0 82870.96 RMC
New % 183 0.588 0.220 0.189 1 RMC
Minutes 187 75,562 21,538 0 189.5 RMC Frequency 187 1,163 0.235 0 1,711 RMC
better future for their family. Both of these motives matches these certain age categories. Another positive strong correlation is found between the amount of amount of business establishments of the two relevant sectors and the percentage of people receiving one of the two sorts of social aids. The people receiving social aid do not have a job at all or potentially not a full-time job since they are not capable of earning their livelihood. They spend their free time possible they might spend at these business establishments. The two social aids also have a strong positive correlation with each other. If the amount of people receiving one of the two socials aids increases this may have a negative effect on the local economy. Due to this decrease of the local economy jobs could disappear which in turn will cause an even more increase in the overall amount of people receiving social aid. It will in part turn into a downwards spiral.
Another interesting positive strong correlation that is found, is the relation between the amount of jobs in tourism and the amount of relevant business establishments. Whenever the amount of tourists increases, an increase in the amount of jobs in the tourist sector will be the result. Businesses will anticipate on this increase of tourists and start opening new locations, and vice versa. Whenever there is an increase of the two categories of businesses, tourists will be attracted towards this city and therefore cause an increase of the jobs in the tourism sector. An increase in the tourism sector will probably happen in an blooming and interesting city. A city with these characteristics is not only attractive for tourists, but also for people starting their careers and families. This can be an explanation for the positive correlation between the percentage of people in the age of 25 until 45 years and the jobs in the tourism sector.
The last positive correlation to investigate is the relation between the amount of jobs in the tourism sector and the amount of people receiving one of the two social aids. Whenever there are more tourists coming to a city, represented by a high amount of jobs in the tourism sector, an increase the economy of a city is noticed. The combination of more tourists and a wealthier city will stimulate the prices to rise. With products and services becoming more expensive, it will become more difficult for the poorer part of the population to maintaining their previous livelihood. This will in part cause a growth of people receiving either form of social aid.
4.2 Variables
As said in the introduction, the first part of the research looks at are (1) the number of unique visitors, (2) the percentage of this unique number of visitors which are new visitors, (3) the duration passengers stay within a geographic shopping area, and (4) the visitors frequency. All of the data is collected on a weekly basis. The number of unique visitors, the percentage of new visitors, and the duration are also possible to be collected on a daily basis. However, the dataset the company RMC had available was on weekly basis. I was hoping to retrieve data from the WiFi sensors themselves. Since this was not possible, the research will be done with the available weekly averages instead of raw data from the sensors.
The number of unique visitors
In each city were RMC is active, they have placed multiple sensors on different spots in the city. The number of unique visitors shows the unique number of visitors within the total geographic range covered by these sensors. In general, this is a certain shopping street or a small shopping area consisting of a few streets. When a certain visitor has been spotted by two or more sensors, this visitor still counts as one unique passenger. This variable is expressed in 1000s of people. Meaning that a relative increase of 1 is equal to an increase of 1000 visitors.
The percentage of new visitors
Based on the number of new visitors of this unique number of visitors, the percentage of new visitors is calculated. A visitors is treated as a ‘new’ visitor when it has never been spotted before by any of the sensors of a certain geographic area, or when he or she has not been spotted for at least 2 months. This variables is expressed in percentages.
The average duration of stay
The average time a visitor stays within an area covered by some sensors, is called the duration of stay, or the dwell time. This variables is an weekly average expressed in minutes. To filter out employees and home owners who work or live in the area, there is a minimum stay of 5 minutes and a maximum stay of 6 hours to be treated as a visitor of the shopping area.
This visitors frequency measures the average times a certain individual is seen in that particular shopping area. In general a visitor will not visit a certain shopping area multiple times during one day, therefore this variable is not calculated on a daily basis but on a weekly basis. It is expressed in absolute numbers.
The second part of the research focuses on explaining the four stated KPI by city characteristics. The city variables chosen are (1) the number of people per household, (2) the index of “cityness”, the number of people between the age of (3) 0 - 15 years, (4) 15 - 25 years, (5) 25 - 45 years, (6) 45 - 65 years, (7) 65+ years, (8) the percentage of the population which are non-western immigrants, the amount of people receiving (9) “WW” or (10) “Bijstand” social aid, the number of jobs in (11) “the retail and the food and beverages industry” and in (12) “culture, recreation, and other services”, (13) the number of jobs in tourism. The first twelve variables are retrieved from CBS, the Central Bureau of Statistics, and the amount of jobs in the tourism sector from LISA. All from 2013 until 2017. The data from the four stated KPI needed to be recalculated from weekly data into yearly data to match this dataset. This resulted in a unbalanced dataset, since the two datasets only have overlap in 2016 and 2017.
The number of persons per household
The first city characteristic to explain the four KPI is the average number of people per one household. This is calculated by dividing the total amount of citizens by the total number of houses within a city. Both the number of citizens of a city and the total amount of houses are measured on the first of January of each year.
The index of “cityness”
Based on the number of addresses within a squared kilometer, the environmental density of a city can be measured. CBS expresses this environmental density through the index of “cityness”. Within this index there are five levels, where a higher level of “cityness” means a higher density.
1. At least 2500 addresses within one squared kilometer.
2. Between 1500 and 2500 addresses within one squared kilometer. 3. Between 1000 and 1500 addresses within one squared kilometer. 4. Between 500 and 1000 addresses within one squared kilometer.
5. Less than 500 addresses within one squared kilometer.
The age of the population divided in five segments
The population is divided percentages of the following five age categories: 1. People of 15 years old and younger
2. People between 15 and 25 years old 3. People between 25 and 45 years old 4. People between 45 and 65 years old 5. People of 65 years old and older
The number of people within each category is measured on the first of January of each year. The percentage of the people which are non-western immigrants
This is the percentage of immigrants within a city who are assumed to be ‘non-western’. Non-western means that they have an Africa, South-America, Asia (including Indonesia and Japan) or Turkish nationality.
The amount of the people who receive social aid
The amount of people who receive social aid is separated into two groups:
1. People who receive so called “WW”. One can request “WW” social aid whenever someone is temporarily or permanently without a job.
2. People who receive so called “Bijstand”. Someone can request “Bijstand” social aid whenever they have a job but still are not capable of adequately earning his livelihood. The amount of people for both of the social aid categories is calculated on the first of January of each year.
The number of business establishments in two relevant sectors
RMC mainly has its sensors placed in shopping areas selling non-daily products. Therefore, their clients are mainly retailers of non-daily products. In this research the effect of the amount of shops selling non-daily products are taken in consideration too. The CBS provided seven categories representing all the sectors of business establishments:
1. Agriculture, Forestry, and Fishery. 2. Industrial and Energy
3. Retail and the Food & Beverages industry 4. Transport, Logistics and Communication
5. Financial Services and Real Estate 6. Business Facilities
7. Culture, Recreation, and Other Services
To represent the number of business establishments selling mainly non-daily products, the following two of the categories above are chosen: “Retail and Food & Beverage industry” and “Culture, Recreation, and Others Services”. The number of these business establishments are calculated on the first of January of each year.
The amount of jobs in tourism
Since there is no data available on the tourists flows in cities, the amount of jobs in the tourism sector are taken as an index for the amount of tourists in this research. If there are more jobs focused on tourists in an city, there may be presumed that there will be more tourists.
5. Results
This section presents the results of the quantitative analysis which tests the hypotheses of the research. First, the results are given concerning the first part of the research. In table F, G, H, and I of the Appendix, the results are given with each KPI as dependent variable separately. Then, the result of testing the dependency of the four KPIs on certain city characteristics is given. In table J of the Appendix on page, the results are presented using the city
characteristics as explanatory variables for each of the four KPIs.
5.1 The unique number of visitors as dependent variable
The results of the regressions run with the unique number of visitors taken as dependent variable are given in table F in the Appendix. For each of the tables F, G, H, and I the first column shows the result for the regression on the entire dataset, column 2 and 3 for cities inside and outside of the Randstad, column 4 and 5 for cities part and not part of the G4, and lastly the last four columns for the results during the summer, spring, autumn, and winter. Looking at table F in the Appendix, the percentage of new visitors is only significant as an explanatory variable in the separation based on spring. In this case an increase of 1% new visitors, ceteris paribus, will cause an increase of 17816 visitors at a 1% significance level.
