Revision_PGD_April2016

NEIGHBORHOOD DEPRIVATION

IN ENSCHEDE Application of Geo-tagged Social Media Indicator

YASMIN AFSHARGHOTLI

June, 2015

SUPERVISORS:

Dr. J. A. Martinez Martin

Ing. F.H.M.van den Bosch

Individual final assignment report submitted to the Faculty of Geo-Information

Science and Earth Observation of the University of Twente in partial fulfilment of

the requirements for the Post Graduate Diploma in Geo-information Science and

Earth Observation.

Specialization: Urban Planning and Management

SUPERVISORS:



IFA ASSESSMENT BOARD:

Dr. S. Amer (Chair)



[etc]

NEIGHBORHOOD DEPRIVATION

IN ENSCHEDE Application of Geo-tagged Social Media Indicator

YASMIN AFSHARGHOTLI

Enschede, The Netherlands, June, 2015

DISCLAIMER

This document describes work undertaken as part of a programme of study at the Faculty of Geo-Information Science and

Earth Observation of the University of Twente. All views and opinions expressed therein remain the sole responsibility of the

author, and do not necessarily represent those of the Faculty.

i

ABSTRACT

The importance of Multiple Deprivation has raised significantly in Europe since 2010. ‘Europe 20120

Strategy’ is a framework that points the social inclusion as a target and states that social inclusion should be

promoted through poverty reduction by people who are exposed to the risk of poverty (The European

Union, 2012).

The Index of Multiple Deprivation tries to describe the socioeconomic composition of resident’s condition

in a particular spatial unit such as census divisions (Meijer, Engholm, Gritter, & Bloomfield, 2013).

Therefore, different indices of deprivation such as ‘The English Indices of Deprivation’ have executed for

the purpose of poverty measurement in Europe.

This research develops a social economic and physical deprivation index, which was adapted from previous

frameworks to quantify deprivation in Enschede City at Neighbourhood scale. Based on Mpata Wekisa

(2014) a geo-tagged social media indicator is built and further studies are conducted to seek the relationship

between social media indicator and deprivation level in the Enschede’s neighbourhoods.

To this, a quantitative geo-tagged social media indicator based on Instagram data was constructed. The

Factor Analysis categorizes this indicator as a physical indicator. Further, the content of the photo’s captions

extracted from Instagram was examined with Word Clouds analysis method. In this method, the frequency

of positive and negative words in the captions have been discovered.

The geo-visualization for the social economic dimension of Enschede deprivation shows that rural areas of

Enschede have lower social economic deprivation. Also, analysing the individual indices shows that in

deprived neighbourhoods, the percentage of inactive people and people with non-western origin are usually

higher. Moreover, the spatial pattern revealed from geo-visualization of physical deprivation shows that

neighbourhoods located within built-up areas have a lower deprivation. The more distance from the City

Centre neighbourhood located causes the more physical deprivation degree.

Final results showed that at 5% significance level, the final index of social economic deprivation is associated

reversely with the final index of physical deprivation, however, the correlation is not strong.

Keywords: Social economic deprivation, physical deprivation, geo-tagged, social media indicator, Instagram

ii

ACKNOWLEDGEMENTS

The following research could not be done without the help of my dedicated supervisors Dr. Martinez and

Mr Ing. F.H.M.van den Bosch.

Moreover, I would like to acknowledge my best friend from Geoinformatic department, Mr. Felipe Diniz,

for his countless effort on Instagram’s data extraction needed for execution of this project.

Also, I would like to express my appreciation to my dedicated family for their unlimited passion and strong

support during my academic life especially in the Netherlands.

iii

TABLE OF CONTENTS

1. Introduction ........................................................................................................................................................... 1

1.1. Urban Poverty and Deprivation ...............................................................................................................................1 1.2. Index of Multiple Deprivation ..................................................................................................................................1 1.3. Geotagging social media .............................................................................................................................................2 1.4. Brief Overview on Enschede ....................................................................................................................................2 1.5. Research Justification ..................................................................................................................................................3 1.6. Objective and questions .............................................................................................................................................4

1.6.1. Objectives ................................................................................................................................................... 4

1.6.2. Questions ................................................................................................................................................... 5

1.7. Report structure ...........................................................................................................................................................5

2. Methodology .......................................................................................................................................................... 7

2.1. Introduction .................................................................................................................................................................7 2.2. Conceptual Framework ..............................................................................................................................................8

2.2.1. Indicators .................................................................................................................................................... 9

2.3. Data and Resources .................................................................................................................................................. 11

2.3.1. Instagram data extraction ..................................................................................................................... 12

2.4. Factor analysis ........................................................................................................................................................... 14 2.5. Standardize and weighting ...................................................................................................................................... 14 2.6. Content analysis ........................................................................................................................................................ 16 2.7. Limitations ................................................................................................................................................................. 17

3. Results and discussion ....................................................................................................................................... 19

3.1. Analysis of the conceptual framework ................................................................................................................. 19

3.1.1. Correlation analysis ................................................................................................................................ 19

3.1.2. Factor Analysis ....................................................................................................................................... 25

3.2. Mapping the indicators ............................................................................................................................................ 26

3.2.1. Social Economic Indicators.................................................................................................................. 26

3.2.2. Physical indicators .................................................................................................................................. 29

3.3. Neighbourhood Deprivation in Enschede .......................................................................................................... 32

3.3.1. Social Economic Deprivation Status in Enschede ........................................................................... 32

3.3.2. Physical Deprivation Status in Enschede ........................................................................................... 33

3.4. Content analysis ........................................................................................................................................................ 34

3.4.1. Word Clouds for Social Economic Deprivation............................................................................... 34

3.4.2. Word Clouds for physical Deprivation .............................................................................................. 35

3.5. Illustrations ................................................................................................................................................................ 36

3.5.1. Social Economic Deprivation status ................................................................................................... 36

3.5.2. Physical Deprivation status .................................................................................................................. 37

4. Conclusion and recomendation ....................................................................................................................... 39

4.1. Conclusion ................................................................................................................................................................. 39 4.2. Research limitations ................................................................................................................................................. 40 4.3. Recommendation ..................................................................................................................................................... 40

Annex 1 ........................................................................................................................................................................ 43

Annex 2 ........................................................................................................................................................................ 44

Annex 3 ........................................................................................................................................................................ 45

iv

LIST OF FIGURES

Figure 1-1: Trend of unemployment rate in Enschede 2004-2014 ...........................................................................................3

Figure 1-2 : Enschede municipality and Neighbourhood division ............................................................................................4

Figure 2-1: Methodological framework ..........................................................................................................................................7

Figure 2-2: Conceptual framework for Social economic and Physical Deprivation, .............................................................8

Figure 2-3: Instagram data extraction process ........................................................................................................................... 12

Figure 2-4: Spatial distribution of the extracted Instagram photos ........................................................................................ 13

Figure 2-5: Neighbourhoods with unavailable data .................................................................................................................. 17

Figure 3-1: Scatter plot between population density and Instagram photo density ............................................................ 19

Figure 3-2: Correlation between geo-tagged social media indicator ....................................................................................... 21

Figure 3-3: Correlation between geo-tagged social media indicator ....................................................................................... 22

Figure 3-4: Scatter plot between % households with low income and final index of social economic deprivation ..... 22

Figure 3-5: Scatter plot between % of inactive people and final index of social economic deprivation ......................... 22

Figure 3-6: Scatter plot between % of people with non-western origin and final index of social economic deprivation............... 23

Figure 3-7: Scatter plot between Average Distance to secondary eduction and final index of physical deprivation .... 23

Figure 3-8: Scatter plot between geo-tagged social media indicator and average distance to daily necessities .............. 23

Figure 3-9: Correlation between geo-tagged social media indicator and House Affordability .......................................... 23

Figure 3-10: Result of KMO and Bartlett test ............................................................................................................................ 25

Figure 3-11: Factor loadings .......................................................................................................................................................... 25

Figure 3-12: Communalities .......................................................................................................................................................... 25

Figure 3-13: Percentage of people with non-western origin at neighbourhood level in Enschede .................................. 26

Figure 3-14: Percentage of inactive people at neighbourhood level in Enschede ............................................................... 27

Figure 3-15: Percentage of households with low income at neighbourhood level in Enschede ....................................... 28

Figure 3-16: Number of new houses at neighbourhood level in Enschede ......................................................................... 29

Figure 3-17: Average Distance to café teria at neighbourhood level in Enschede .............................................................. 30

Figure 3-18: Instagram photo density at neighbourhood level in Enschede ........................................................................ 31

Figure 3-19: Social Economic Deprivation at neighbourhood level in Ensche ................................................................... 32

Figure 3-20: Physical Deprivation at neighbourhood level in Ensche ................................................................................... 33

Figure 3-21: Scatter plot between Social economic deprivation and Physical deprivation ................................................ 34

Figure 3-22: Word cloud for Cromhoffsbleek-Kotman Neighbourhood ............................................................................. 34

Figure 3-23: Word cloud for Wesselerbrink Zuid-Oost Neighbourhood ............................................................................. 34

Figure 3-24: Word cloud for Ruwenbos Neighbourhood ....................................................................................................... 35

Figure 3-25: Word cloud for Oikos Neighbourhood ............................................................................................................... 35

Figure 3-26: Word cloud for Boekelverd Neighbourhood ...................................................................................................... 35

Figure 3-27: Word cloud for Buurtschap Twekkelo ................................................................................................................. 35

Figure 3-28: Word cloud for Marssteden Neighbourhood ...................................................................................................... 35

Figure 3-29: Word cloud for Boddenkamp Neighbourhood .................................................................................................. 35

Figure 0-1: Percentage of persons 65 years and older at neighbourhood level in Enschede............................................. 45

Figure 0-2: Relative Mortality Rate at neighbourhood level in Enschede ............................................................................. 46

Figure 0-3: Household size at neighbourhood level in Enschede .......................................................................................... 46

Figure 0-4: House affordability at neighbourhood level in Enschede ................................................................................... 46

Figure 0-5: Average Distance to bus stops at neighbourhood level in Enschede ............................................................... 46

Figure 0-6: Average Distance to daily necessities at neighbourhood level in Enschede .................................................... 46

Figure 0-7: Population Density at neighbourhood level in Enschede ................................................................................... 46

Figure 0-8: Average Distance to General Practitioners at neighbourhood level in Enschede .......................................... 46

Figure 0-9: Average Distance to secondary education at neighbourhood level in Enschede ............................................ 46

v

LIST OF TABLES

Table 2-1 – Structure of CBS associated data with framework ......................................................................... 11

Table 2-2 - Cost and Benefit .................................................................................................................................... 15

Table 2-3 – Weighting .............................................................................................................................................. 15

Table 3-1: Correlation Matrix ................................................................................................................................... 20

Table 0-1: Meta Data .................................................................................................................................................. 43

NEIGHBOURHOOD DEPRIVATION IN ENSCHEDE APPLICATION OF GEO-TAGGED SOCIAL MEDIA INDICATOR

1

1. INTRODUCTION

1.1. Urban Poverty and Deprivation

Urban poverty and deprivation are often used interchangeably (Noble, Wright, Smith, & Dibben, 2006).

Although there is not clear distinction between these concepts, the most dominant definition for urban

deprivation is presented by Peter Townsend: ‘‘People can be said to be deprived if they lack the types of

diet, clothing, housing, household facilities and fuel and environmental, educational, working and social

conditions, activities and facilities which are customary, or at least widely encouraged and approved, in the

societies to which they belong.’’ (Townsend, 1987, p.125-126).

Followed by Townsend’s definition, European Union Council (EUC) in 1975 adopt a definition of poverty

and social exclusion and extended to more concrete one in 1985: “The persons whose resources(material,

cultural and social) are so limited as to exclude them from the minimum acceptable way of life in the Member

State to which they belong”(EEC, 1985).

According to above definitions, deprivation is a wide range concept that need to be formulated in several

aspects. This concept encompasses not only broad way of different dimensions, but also it includes every

individual who is experiencing deprivation (Noble et al., 2006).

Pacione (1995), discusses the anatomy of deprivation. According to his idea, problems of poverty-related

issues are multidimensional in a way that they interact with each other. He mentions that crime, delinquency,

poor housing, unemployment, increased mortality and mental illnesses are among the poverty effects that

tend to adopt a spatial pattern as well.

Although deprivation is a multidimensional phenomenon, it is a relative concept which can be characterized

by its context (Martínez, 2015). ‘The English Indices of Deprivation’ is an example of governmental report

in United Kingdom which tries to define deprivation in seven domains such as income, employment, health,

education, housing and services, living environment and crime (Noble et al., 2011).

To battle social exclusion, the importance of Multiple Deprivation has raised significantly in Europe since

2010. ‘Europe 20120 Strategy’ is a policy framework that points the social inclusion as one of the most

important objectives. This target articulates that social inclusion should promote through poverty reduction

by people who are exposed to the risk of poverty (European Union, 2012).

1.2. Index of Multiple Deprivation

‘The index of multiple deprivation (IMD)’ is a method that provides a better insight into urban poverty issue

(Baud, Sridharan, & Pfeffer, 2008). In addition, this tool tries to measure the level of disadvantages at small

spatial scale in the context of the western economy (Yuan & Wu, 2014). The Index of Multiple Deprivation

also tries to describe the socioeconomic composition of resident’s condition in a particular spatial unit such

as census divisions (Meijer et al., 2013).

The concept of IMD will reflect various aspects of poverty into discrete domains. Besides, particular groups

of indicators in each domain will address the problem of poverty by means of measuring observables (Yuan

& Wu, 2014).

As it mentioned earlier, ‘The English Indices of Deprivation’, benefits from a combination of seven different

domains that form the index of multiple deprivation. This particular index along with ‘Welsh Index of

Multiple Deprivation’, ‘Northern Ireland Measures of Multiple Deprivation’, and ‘Scottish Indices of

Deprivation’(Noble et al., 2011) are executed for purpose of poverty measurement in British context.


2

In order to map the urban poverty, Baud et al. (2008), develop a ‘livelihood assets framework’ to examine

deprivation level in Delhi city, India. This framework benefits from multiple aspects of poverty such as

social capital, human capital, financial capital, and physical capital.

The spatial scale for the index of multiple deprivation will quantify deprivation at a lower level, and assist

policy-makers to identify specific areas which should benefit from allocation of public resources (Noble et

al., 2006).

1.3. Geotagging social media

According to Wikipedia (2015a) geotagging is a metadata information that adds geospatial identification to

different types of media such images, videos, websites and others. The coordinates, usually expressed as

latitude and longitude, can be derived from the smartphone’s GPS, or proximity to places of known

coordinates, or even manual input by the user. On the other hand, Twitter, Instagram, and Flickr are among

the popular social networking services that enable users to share their posts along with geo-referencing that

provides locational information for each tweet or photo.

Instagram is an online photo service, and it gained a drastic popularity among users from 1 million in

December 2010 to over 300 million in December 2014 (Wikipedia, 2015b). This social media also benefits

from different features and tools implemented such as hashtag usage.

The data which is generated by the social media like Foursquare, Flickr, and Twitter can be considered as a

rich data source for spatio-temporal data, since they cover a long time range of different types of data. These

data sources can be used not only to consider social networking and people’s behaviour, but also they have

a spatio-temporal pattern that is observable in particular area (Li, Goodchild, & Xu, 2013).

The research that has done by Quercia & Saez (2014), uses foursquare data as a reliable land use data source

for the city of London. In this research, the possible relationship between certain types of venues and

Neighbourhoods’ socio-economic deprivation have been examined. The results prove that the correlation

between Foursquare categories and Neighbourhood’s index of multiple deprivation is statistically significant.

Moreover, certain types of land use categories are associated with certain deprived or well-off

Neighbourhoods.

In the research initiated by Li et al., (2013) explore the correlation between the density of tweets and photos,

and social economic specifications using Partial least square regression (PLSR) at county level in California,

United States. Moreover, this research identifies the spatial and temporal pattern of geo-referenced tweet

and photos in California. Results show the relationship between tweet densities and percentage of well-

educated people with an advance degree and a good salary, and the correlation between photo densities and

a high percentage of white and Asian people. Spatial distribution of tweets and images, also shows the

concentration of tweets and photo densities along major highways and roads.

1.4. Brief Overview on Enschede

The city of Enschede is located in eastern part of The Netherlands in the urban region of Twente. With the

population of 158000, Enschede is considered as the largest city in both Twente Region and Province

Overijssel (Provincie Overijssel, 2015).


3

1Currently, Enschede is facing to problem of stable unemployment. The trend of the unemployment rate is

at its highest point-8 percent- during the last ten years (Fig 1.1).

The municipality of Enschede is divided into 70 Neighbourhood area as the smallest spatial unit, and rural

areas dominant. The share of built-up area in this municipality is only 23.5 percent, and 19 Neighbourhoods

are located in the areas which are more than 50 % rural. Considering the fact that most inhabitants reside in

the built-up area, the average population per Neighbourhood is 2528 in built-up area. Figure 1-2 shows

Enschede built-up area along with Neighbourhood divisions.

Neighbourhoods namely, Buurtschap Zuid-Esmarke, Buurtschap Lonneker-West, Buurtschap

Broekheurne, Noord Esmarke, Boekelerveld, Buurtschap Twekkelo, and Drienerveld-U.T- home of the

University of Twente- are totally located in rural areas.

1.5. Research Justification

Mpata Wekisa (2014), defines the relationship between number of geo-tagged Flickr photo and Multiple

Deprivation Index for the city of Enschede at a Neighbourhood level. The author uses Flickr image hosting

and concludes that social media can be used as a tool to measure urban deprivation.

Therefore, to develop an Index of Multiple Deprivation, this research aims to seek the relationship of the

geo-tagged social media indicator and its contribution to the deprivation level of Enschede’s

Neighbourhood. To this end, Instagram geotagged photos have been adopted.

The spatial distribution of geo-referenced Instagram’s photo illustrate the concentration pattern of social

media indicator and will define the hotspots of photos in city’s Neighbourhood level. Moreover, the current

research will perform a qualitative analysis on the geo-tagged photos in Instagram for a most recent year.

This qualitative analysis not only reveals the contents in Neighbourhood level in Enschede but also is a

strength part of the research since it tries to organize the sample photos into visual word clouds.

1 Data source: CBS data for labor force, active population, and unemployed labor for Enschede municipality 2004-2014.

0

2

4

6

8

10

2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

per

cen

tage

year

Unemployment Rate

Figure 1-1: Trend of unemployment rate in Enschede 2004-2014


4

Figure 1-2 : Enschede municipality and Neighbourhood division

1.6. Objective and questions

The main objectives of this research implementation are described as follow. Further, research questions

are designed in accordance of each objective.

1.6.1. Objectives

To explore geo-tagged social media as an indicator for identifying deprived areas at the

Neighbourhood level in the city of Enschede

Mapping social economic deprivation as well as physical deprivation at Neighbourhood level

Understand the relationship between social economic deprivation and physical deprivation


5

1.6.2. Questions

How the geo-tagged social media indicator incorporates with final indices of deprivation?

The answer to this question will construct a social media indicator to assess the deprivation at

Neighbourhood level.

Is there a relationship between geo-tagged social media indicator and final deprivation indices?

This question will test the possible relationships between geo-tagged social media indicator and deprivation

indices.

o What is the content of geo-tagged social media indicator?

The output of this question will present the content of geotagged photos at Neighbourhoods of Enschede

city.

o What is the social economic deprivation status at the Neighbourhood level?

The result of this question will geo-visualize deprivation based on a set of indicators in social economic

dimension.

o What is the physical deprivation status at the Neighbourhood level?

The output of this question will map the deprivation status at Neighbourhood level of Enschede based on

physical dimension.

o To what extent the social economic deprivation is correlated with physical deprivation?

The outcome will investigate the relationship between social economic deprivation and physical deprivation

at the Neighbourhood level for city of Enschede.

1.7. Report structure

This project will be organized in chapters as follow:

1. Introduction

This chapter frames the research problems “Deprivation in Enschede at Neighbourhood level”, and review

fundamental concept about Multiple Deprivation Index (MDI), different types of deprivation, and the role

of social media as a quantitative indicator. Moreover, it presents a brief overview on Enschede Municipality

as a case study area.

2. Methodology

This chapter will present the designated conceptual framework, and discuss methodological steps that

consist of factor analysis, standardize-weighting, and calculating the final index.

3. Results and Discussion

Findings of this research will be set of different maps corresponding to deprivation status in relation to

indicators. Besides, critical discussion on maps will clarify the relationship between social economic

deprivation and physical deprivation. This chapter will carry the most part of this research.

4. Conclusion and recommendation

This chapter tries to answer to main research questions and presenting related recommendations as future

works.


6


7

2. METHODOLOGY

2.1. Introduction

Figure 2.1 presents methodological steps for the construction of Multiple Deprivation Index (MDI) with

the application of geo-tagged social media indicator.

Figure 2-1: Methodological framework


8

The work starts with a literature review in previous constructions of Multiple Deprivation Index, focusing in a European context of deprivation. A framework is constructed based on the context of the city of Enschede and data availability. The data sources used are discussed in section 2.3, and its limitations in section 2.7. The proposed conceptual framework and the indicators used are presented in section 2.2.

In section 2.4 the choice of the indicators is evaluated by means of a Factor Analysis. The process is followed

by standardization of the indices and weighting (section 2.5) and construction of the final MDI indices for

Social economic and Physical deprivation.

In addition, content analysis is performed on the captions of the Instagram data. The process of sampling

and word frequency analysis are further explained in Section 2.6. At the end of this chapter are presented

the limitations of the methodology, especially regarding data availability.

2.2. Conceptual Framework

The following conceptual framework tries to measure deprivation at Neighbourhood scale for Enschede.

Therefore, this proposed framework is both theoretical and data driven which consists of three levels.

At first level, the Index of Multiple Deprivation is broken into two major dimensions: social economic, and

physical deprivation. To this end, the final index calculation will result into two different scores for each of

these dimensions. As stated by European Union (2012), social economic dimension of deprivation refers to

insufficient resources and affordability issues while physical deprivation mentions about the availability of

facility and amenities. In addition, Material deprivation includes four major components such as ‘Housing

condition, Basic amenities, Local environment, and Accessibility’.

The second step of this framework, articulates concept of social economic and physical deprivation in

various discrete domains. Population composition, employment, and income are the domains which try to

measure social economic concept. Furthermore, physical deprivation is described by accessibility, housing,

health and living condition, education, and place popularity separate domains.

At the third level, indicators are presented as a measurement tool. According to Martínez & Dopheide

(2014), indicator is defined as: “…qualitative or quantitative data that describe features of a certain

phenomenon and communicate an assessment of the phenomenon involved.” Selection of each indicator

for this research is based on relevancy and data availability regarding to each specific domain.

Figure 2-2: Conceptual framework for Social economic and Physical Deprivation, Adapted(Noble et al., 2011)


9

2.2.1. Indicators

This section, presents a rationale for a number of indicators and a brief description on each of the

indicators.

Population composition

In European context there is an association between deprivation and level of social class (Yuan & Wu,

2014). This is defined by economical features such as labour force and social feature such as people and

group composition. Therefore, in this study focus will be on certain groups of people are more vulnerable

to deprivation and poverty.

Percentage of persons 65 years or above

Number of inhabitants of age 65 or older expressed as a percentage of total population.

Relative Mortality

Total number of deaths per 1000 inhabitants. The relative number of deaths is calculated based on

the number of inhabitants on January 1.

Household size

Average household size calculated as the number of persons dwelling in private houses divided by

the total number of private houses.

Percentage of people with non-western origin

Number of immigrants with a non-western origin expressed as a percentage of total population.

Percentage of Inactive

The number of income recipients aged from 15 to 65 with 52 weeks receiving benefit as the main

source of income in the previous year, expressed as a percentage of the total number of income

recipients aged 15 to 65 years. Individuals with unemployment, disabled, and pensioners are among

inactive.

Percentage of households with low income

Total number of private houses classified as low income expressed as a percentage of total private

houses. The definition of low income is based on the 40% lower income households which have a

total annual income lower than 25100 euros.

House affordability

This indicator takes into consideration the fields Average house value (WOZ) and Average income

per year (INK_INW) from CBS data:

WOZ: Average value, in thousands of euros, of residential buildings based on the real

estate determined on the Valuation of Immovable Property Act (Wet Waardering

Onroerende Zaken)

INK_INW: Average value of total annual income based on total population. The indicator is calculated using the following formula:

𝐻𝑜𝑢𝑠𝑒 𝑎𝑓𝑓𝑜𝑟𝑑𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 𝑊𝑂𝑍𝐼𝑁𝐾_𝐼𝑁𝑊⁄


10

Average Distance to bus stops

Average distance from each point inside the Neighbourhood to the closest bus stop. This indicator

was calculated using the attribute Bus Stops from NGR. First Euclidean distance of the point layer

of bus stops was calculated using Euclidean Distance function from Arc Toolbox, then the average

value for the distance inside each Neighbourhood was calculated using the function Zonal Statistics,

also available in Arc Toolbox.

Average distance to daily necessities The average distance of all residents to the nearest groceries shop, bakery, or supermarket, liquor

store, tobacco shop, etc. This distance is calculated following the roads (network distance, not

Euclidean distance)

Population density

Total population divided by the area of the Neighbourhood in km2.

Growth of new housing Total number of dwellings in 2011 minus the total number of dwellings in 2007. The buildings

considered in this indicator are the ones which were built with the purpose of permanent residence.

𝐺𝑟𝑜𝑤𝑡ℎ 𝑜𝑓 𝑛𝑒𝑤 ℎ𝑜𝑢𝑠𝑖𝑛𝑔 = 𝑊𝑂𝑁𝐼𝑁𝐺𝐸𝑁2011 − 𝑊𝑂𝑁𝐼𝑁𝐺𝐸𝑁2007

Assumption: The less number of new houses, the more deprivation.

Average distance to GP

Average distance of all residents to the nearest general practitioner. This distance is calculated

following the roads (network distance, not Euclidean distance).

Average distance to cafeteria Average distance to cafeterias, fast food restaurants, kebab shop, coffee shop, etc. This distance is

calculated following the roads (network distance, not Euclidean distance).

Assumption: The less distance to cafeteria , the more deprivation.

Average distance to secondary education

Average distance of all residents in the Neighbourhood to the nearest government-funded full-time

secondary education. This distance is calculated following the roads (network distance, not

Euclidean distance).

Geo-tagged social media

To eliminate the influence of population in different neighbourhoods, the geo-tagged social

media indicator consider the number of geo-tagged Instagram photos divided by population for

each neighbourhood. Therefore, the geo-tagged social media indicator in this research can be

interpret as a measure for photo density in each neighbourhood.

This indicator consider the amount of geo-tagged Instagram posts between 31may2014 and

31may2015 inside each Neighbourhood divided by the population of that Neighbourhood

(AANT_INW), as presented in the formula below:

𝐺𝑒𝑜 − 𝑡𝑎𝑔𝑔𝑒𝑑 𝑠𝑜𝑐𝑖𝑎𝑙 𝑚𝑒𝑑𝑖𝑎 =#𝐼𝑛𝑠𝑡𝑎𝑔𝑟𝑎𝑚 𝑝ℎ𝑜𝑡𝑜𝑠

𝐴𝐴𝑁𝑇_𝐼𝑁𝑊⁄


11

The following assumption for this indicator has been made: Photos have been posted just by

residents of each neighbourhood.

This indicator consider as a cost to deprivation status, so the less Instagram photo density for each

neighbourhood, the more deprived is the neighbourhood.

2.3. Data and Resources

The main data source for this research was the Centraal Bureau Voor de Statistiek (CBS, 2015) for the year

of 2011. This year was chosen due to having a complete range of attributes comparing to the years of 2012,

2013 and 2014 which were available on CBS website. In addition, the attribute WONINGEN was obtained

for the year of 2007 for a comparative analysis. Table 2.1 presents the associated fields of CBS spatial data

and indicators2.

Table 2-1 – Structure of CBS associated data with framework

Complementing CBS data, the coordinates of bus stops for 2011 were obtained in the Nationaal Georegister

website (NGR, 2015).

Also, geo-tagged social media data (from Instagram) was used in addition to the statistical data. The

Instagram API was used for the extraction of geo-tagged photos in the municipality of Enschede between

31 may 2014 and 31 may 2015. The Instagram data extraction is further explained in Section 2.3.1.

2 The full metadata for the data used is presented at Annex 1.

Domain Indicator Data Field Data Field_English

Percentage of persons 65 years or above P_65_EO_JR P_OVER65

Mortality relative P_STERFT MORTALITY

Household size GEM_HH_GR HH_SIZE

Percentage of people with non-western origin P_N_W_AL P_NONWESTERN

Percentage of Inactive P_NIETACT P_INACTIVE

Percentage of households with low income P_LAAGINKH P_LowIncome

House affordability (Houseaff)WOZ/INK_INW Not available in CBS

Average Distance to bus stops AvgDist_bus Not available in CBS

Average distance to daily necessities AF_DAGLMD AvgDist_Daily

Population density BEV_DICHTH POP_DENSITY

Growth of new housing (newhouse)WONINGEN Not available in CBS

Average distance to GP AF_ARTSPR AvgDis_GP

Average distance to cafeteria AF_CAFTAR AvgDist_Cafeteria

Education Average distance to secondary education AF_ONDVRT AvgDist_S_EDU

Place popularity Geo-tagged social media Instagram Not available in CBS

Housing

Health and living condition

Social-economic


Employment

Physical

Accessibility


12

2.3.1. Instagram data extraction

This section will address more specifically the procedures for obtaining the geo-tagged Instagram data.

Figure 2.3 presents the procedures for the extraction of the geo-tagged Instagram images.

In order to have access to Instagram data, an Instagram account and the creation of a client ID is needed.

The client ID is the authentication for the use of the Instagram Application Program Interface (API). The

Instagram API is an open architecture based on Hypertext Transfer Protocol (HTTP) requests for sharing

data between users and applications (Wikipedia, 2015c).

The Instagram API allows the retrieval of different types of data through endpoints. The Media endpoint

allowed the access of public feed of Instagram photos and was used in this research. This endpoint has a

search method with the following parameters:

o LAT: Latitude of the centre point coordinate of the search area in WGS84.

o LNG: Longitude of the centre point coordinate of the search area in WGS84.

o DISTANCE: Radius of search with a maximum distance of 5km.

o MIN_TIMESTAMP: Minimum time for the search.

o MAX_TIMESTAMP: Maximum time for the search

The response to an API request is a JSON file (Wikipedia, 2015d). Further information on Instagram API

can be obtained at Instagram (2015). From the JSON file response, relevant attributes were selected and

saved to a CSV file. The following attributes were found relevant:

Figure 2-3: Instagram data extraction process


13

link of the Instagram post

latitude

longitude

Place: This is a manual location name which is written by users in case of check-in in a particular

location.

Time: Exact time and date for the photo

Photo: An individual URL link for the photo

User caption: Additional image description wrote by user which contain simple text or hashtags.

The function Add XY Data from the software ArcGIS uses CSV file to convert the text into the points.

Each point is a representative for a posted Instagram photo. Further, the CSV file was converted to a GDB

file and, a reprojection from WGS84 3 to RD_New coordinate system was performed to match the

coordinate system with the Neighbourhood boundaries. The data also have improved by removing

duplicate points and removing photos outside Enschede boundary. The duplicates occur due to overlaps of

buffers in photo retrieval, and the ID was used to find duplicates. Also for the same reason exists photos

outside Enschede that were eliminated using ArcGIS Clip function.

In total 42362 Instagram photos were extracted from Enschede Municipality boundary for the period of 31

may 2014 to 31 may 2015. Figure 2.4 illustrates the spatial distribution of geo-tagged photos, and represent

heat locations based on photo densities.

3 WGS84 is a default projection for Instagram.

Figure 2-4: Spatial distribution of the extracted Instagram photos


14

In order to generate the final layer with geo-tagged social media indicator, the final step performs a spatial

join between the Neighbourhood layer and the Instagram photos point layer. This is desirable since the

deprivation in this research will be evaluated in Neighbourhood level, using the density of Instagram posts

as an indicator. As a result, an attribute with the number of Instagram photos was created in the

Neighbourhood layer.

2.4. Factor analysis

Factor Analysis (FA) is one of the useful techniques which look into the latent structure of data, and tries

to describe it. Based on conceptual framework discussed earlier, the final index of deprivation is divided

into two dimensions in this research: Social economic and physical deprivation. Factor analysis will perform

to explore the contribution of geo-tagged social media indicator either to social economic deprivation or

physical deprivation. The sample size for this analysis consist of Enschede’s neighbourhoods, and variables

are 15 indicators for measurement. In addition, KMO and Bartlett test will perform within the analysis to

confirm the adequacy of sample size. The result of this analysis will be discussed in chapter 3 respectively.

2.5. Standardization and weighting

In order to perform comparison and computation between indicators, a standardization is needed. This

occurs due the fact that the indicators benefit from various units of measurement. The standardization will

be undertaken in order to determine the direction of each indicator. The standardized indicators (SI) were

calculated with the following formulas whether they are benefit or cost to the concept of deprivation:

𝑆𝐼𝐵𝑒𝑛𝑒𝑓𝑖𝑡 =𝐼 − 𝐼𝑚𝑖𝑛

𝐼𝑚𝑎𝑥 − 𝐼𝑚𝑖𝑛

𝑆𝐼𝐶𝑜𝑠𝑡 = 1 −𝐼 − 𝐼𝑚𝑖𝑛

𝐼𝑚𝑎𝑥 − 𝐼𝑚𝑖𝑛

Where 𝐼 is the indicator value, 𝐼𝑚𝑖𝑛 and 𝐼𝑚𝑎𝑥 are the minimum and maximum values of the indicator

respectively. As a result of standardization, all indices will range between 0 to 1, where 0 value represents

the least deprived and 1 represents the most deprived neighbourhood.

In order to define the direction of each indicator’s contribution to deprivation, the following question has

been asked: ‘What will be the deprivation’s results when the value of each indicator increases?’. In case of

an increase of deprivation, the direction will be “benefit”, and in the event of decrease the direction will be

“cost”. Table 2.2 presents the direction of each indicator as benefit or cost to the final index of deprivation.


15

Table 2-2 - Cost and Benefit

For the calculation of the final values for Social economic and Physical final index, a weighting procedure

was applied over the standardized indicators. The equal weighting distribution has been chosen, where all

the indicators within the same domain have received the same weight, and every domain within the same

dimension received the same weight. The final weight per indicator is calculated by multiplying the domain

weight by the indicator weight. Table 2.3 presents the weight for each indicator.

Table 2-3 – Weighting

In conclusion, using the indices final weights the final values, the Social economic and Physical indicator

can be calculated as following:

Domain Indicator Data Field Direction

Percentage of persons 65 years or above P_65_EO_JR benefit

Mortality relative P_STERFT benefit

Household size GEM_HH_GR benefit

Percentage of people with non-western origin P_N_W_AL benefit

Percentage of Inactive P_NIETACT benefit

Percentage of households with low income P_LAAGINKH benefit

House affordability (Houseaff)WOZ/INK_INW benefit

Average Distance to bus stops AvgDist_bus benefit

Average distance to daily necessities AF_DAGLMD benefit

Population density BEV_DICHTH benefit

Growth of new housing (newhouse)WONINGEN cost

Average distance to GP AF_ARTSPR benefit

Average distance to cafeteria AF_CAFTAR cost

Education Average distance to secondary education AF_ONDVRT benefit

Place popularity Geo-tagged social media Instagram cost

Housing

Health and living condition

Employment

Social-economic

Physical


Accessibility

Dimension Final weight

%persons 65 years and over 0.25 0.0825

Mortality relative 0.25 0.0825

householdsize 0.25 0.0825

%people with nonwestern origin 0.25 0.0825

Employment 0.33 % of inactive 1 0.33

% household with low income 0.5 0.165

House affordability 0.5 0.165

Average Distance to bus stops 0.5 0.1

Average distance to daily necessities 0.5 0.1

Population density 0.5 0.1

Growth of new housing 0.5 0.1

Average distance to GP 0.5 0.1

Average distance to cafeteria 0.5 0.1

Education 0.2 Average distance to secondary education 1 0.2

Place Popularity 0.2 Geotagged social media 1 0.2

Physical

0.2

0.2

0.2

Housing

Health and living

condition

Accessibility

Socialeconomic

0.33

0.33

Domain Indicator

Population

Composition

Income


16

𝐼𝑆𝑜𝑐𝑖𝑜𝑒𝑐𝑜𝑛𝑜𝑚𝑖𝑐 = 0.0825(𝑖𝑎𝑔𝑒65 + 𝑖𝑚𝑜𝑟𝑡 + 𝑖ℎ𝑜𝑢𝑠𝑒𝑠𝑖𝑧 + 𝑖𝑛𝑜𝑛𝑤𝑒𝑠𝑡) + 0.33 𝑖𝑖𝑛𝑎𝑐𝑡𝑖𝑣𝑒 +

+0.165(𝑖𝑙𝑜𝑤𝐼𝑛𝑐 + 𝑖ℎ𝑜𝑢𝑠𝑒𝑎𝑓𝑓)

𝐼𝑝ℎ𝑦𝑠𝑖𝑐𝑎𝑙 = 0.1(𝑖𝑏𝑢𝑠 + 𝑖𝑑𝑎𝑖𝑙𝑦𝑁𝑒𝑐 + 𝑖𝑑𝑒𝑛𝑠𝑖𝑡𝑦 + 𝑖𝑛𝑒𝑤ℎ𝑜𝑢𝑠𝑒 + 𝑖𝑔𝑝 + 𝑖𝑐𝑎𝑓𝑒) +

+0.2( 𝑖𝑠𝑒𝑐𝐸𝑑𝑢 + 𝑖𝑖𝑛𝑠𝑡𝑎𝑔𝑟𝑎𝑚)

The formula used the standardized values for the indicators and the variable 𝑖𝑎𝑔𝑒65 is the % of persons 65

years and older, 𝑖𝑚𝑜𝑟𝑡 is the mortality rate, 𝑖ℎ𝑜𝑢𝑠𝑒𝑠𝑖𝑧 is the household size, 𝑖𝑛𝑜𝑛𝑤𝑒𝑠𝑡 is the % of people with

nonwestern origin, 𝑖𝑖𝑛𝑎𝑐𝑡𝑖𝑣𝑒is the % of inactive population, 𝑖𝑙𝑜𝑤𝐼𝑛𝑐 is the % of people with low income,

𝑖ℎ𝑜𝑢𝑠𝑒𝑎𝑓𝑓 is the house affordability, 𝑖𝑏𝑢𝑠 is the average distance to bus stops, 𝑖𝑑𝑎𝑖𝑙𝑦𝑁𝑒𝑐 is the average

distance to daily necessities, 𝑖𝑑𝑒𝑛𝑠𝑖𝑡𝑦 is the population density, 𝑖𝑛𝑒𝑤ℎ𝑜𝑢𝑠𝑒 is the growth of new housing, 𝑖𝑔𝑝

is the average distance to general practitioner, 𝑖𝑐𝑎𝑓𝑒 is the average distance to cafeteria, 𝑖𝑠𝑒𝑐𝐸𝑑𝑢 is the average

distance to secondary education, and 𝑖𝑖𝑛𝑠𝑡𝑎𝑔𝑟𝑎𝑚 is the geo-tagged social media indicator.

2.6. Content analysis

Content analysis through the word cloud visualization, provides a general overview on the people’s

insight about each of the neighbourhoods. The textual description in Instagram’s caption can be a

representative and provide an insight into the activities of the residents and visitors of each neighbourhoods.

For desired analysis, the least deprived and the most deprived neighbourhoods were sampled in both social

economic and physical derivation. Word clouds were performed based on meaningful words exploring the

presence of negative or positive words.

As mentioned in Section 2.3.1 the caption of the Instagram photos was also retrieved. Instagram’s caption

can provide better information than other social media such as Twitter, since the caption limitation is up to

2200 characters.

Due to the high amount of text descriptions which need to be analysed, not all Instagram captions will be

considered. Based on the final result of the social economic and physical index, neighbourhoods with the

lowest and the highest values will be chosen and a word frequency analysis will be performed individually.

The website Text is Beautiful (Kapiche, 2014), will be used for the word frequency analysis and the result

will visualize as a word cloud (Wikipedia, 2015e). The mentioned website accepts 100000 input characters.

If total amount of characters in each photo caption exceeds the limit value, characters will be removed by

chance until the number of characters reach the specified limit. Another limitation of this website consist

of case sensitivity4 and disability in removing irrelevant words such as conjunction vocabulary in both

English and Dutch languages. Moreover, this website cannot handle special characters5.

4 Word cloud is case sensitive. For example ‘Enschede’ and ‘enschede’ are considered as separate entities. 5 Special characters consist of series of symbols such as @#?!*-_


17

2.7. Limitations

The limitations of the adopted methodology are all associated with data in terms of availability and quality.

Neighbourhoods “De Broeid”,’tweldink, and “Koekoekbeekhoek” are among the three neighbourhoods

with lack of data availability. The field values are shown with -99999999 in CBS file. Moreover, the CBS

data associate some fields of some Neighbourhoods with the value -99999997, which are data classified as

secret.

Therefore, 13 other neighbourhoods are excluded from analysis due to the fact that at least had one field

with value classified as secret. These fields consists of indicators for relative mortality rate, percentage of

inactive people, percentage of households with low income, and house affordability 6 . In total, 16

Neighbourhoods (approximate 25% of the Neighbourhoods) are excluded from analysis of this research.

On the other hand, the limitations for Instagram data refers to both geo-tagging accuracy and account

privacy. Moreover, photos with a non-geotagged option cannot be extracted since they do not associate

with geographical coordinates. The extractable Instagram photos are the ones that have the good accuracy

in terms of geo-tagging related to places, and have the public access for API.

Figure 2.5 presents the Neighbourhoods that presents some kind of unavailable data. These neighbourhoods

are not considered in calculation of final index and further analysis such as factor analysis and correlation

analysis. Therefore, among 70 neighbourhoods in Enschede municipality 54 take part in analysis.

6 This indicator is associated with two fields: average house price (WOZ) and average income per year (INKW).

Figure 2-5: Neighbourhoods with unavailable data


18


19

3. RESULTS AND DISCUSSION

3.1. Analysis of the conceptual framework

As it is stated in the first research question, incorporation of social media indicator with the final index of

deprivation-social economic or physical- is desirable. Therefore, the analysis of the conceptual framework

will take place in two steps. Firstly, the bivariate correlation analysis will be conducted to consider the

possible relationships between geo-tagged social media indicator and Final indices of deprivation. Second,

the relevance of geo-tagged social media to final deprivation indices will be determined using Factor analysis

technique.

3.1.1. Correlation analysis

The correlation analysis will be done for following objectives:

1. To explore highly associated variables in analysis.

2. To examine the possible relationship of geo-tagged social media indicator and final indices of

deprivation

3. To examine possible association of all other indicators and final indices of deprivation

4. To examine possible relationship between geo-tagged social media indicator and all other indicators

For analysing mentioned objectives, bivariate correlation has been performed. Table 0-2 presents the

Pearson correlation at 2-tailed significance test generated at SPSS. First the correlation matrix will generate

for 54 neighbourhoods which they cover data availability7. Figure 3-1 demonstrates the scatterplot between

population density and geo-tagged social media indicator. As the figure indicates, the neighbourhood of city

Center shows a different behavior from rest of the neighbourhoods.

Despite the low amount of population for city center neighbourhood, the geo-tagged social media indicator

value is quite high. This shows that this particular neighbourhoods benefits from relatively high number of

Instagram photos from different groups of people rather than residents. The population density for this

particular neighbourhood is 4934 people per square kilometer, and the geo-tagged social media indicator is

1.3.

For the purpose of performing analysis in a more effective way, the City Center neighbourhood will be

excluded as an outlier from further analysis.

7 Refer to Annex 2 for correlation matrix of 54 neighbourhoods including City Center neighbourhood.

Figure 3-1: Scatter plot between population density and Instagram photo density


20

At the second step the correlation matrix has been derived based on 53 neighbourhood (Table 3-1).

Table 3-1: Correlation Matrix


21

Explore highly associated variables in analysis

According to Field (2005), the acceptable correlation between indicators should be less than 0.9 in bivariate

table. Based on correlation matrix (Table 3-1), two indicators of ‘Percentage of inactive people’ and

‘Percentage of households with low income’ have a high correlation (0.946), which is not acceptable. Besides,

the 2-tailed significance test shows that correlation is acceptable with 1% significance level. Although the

correlation between several indicator pairs are significant, but the correlation value is out of the defined

range by Field.

This correlation can be explained due to the fact that inactive people receives a benefit considered as low

income, so the higher the number of inactive people the higher will be the number of households with low

income.

To examine the relationship between geo-tagged social media indicator and final indices of

deprivation, following assumption has been made:

𝐻0: 𝑇ℎ𝑒𝑟𝑒 𝑖𝑠 𝑛𝑜 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛

𝐻1: 𝑇ℎ𝑒 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛 𝑖𝑠 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡

Based on correlation matrix (Table 3-1) the value of 2-tailed Pearson correlation is significant between geo-

tagged social media indicator and Final Physical deprivation index. The value of this correlation is 0.41.

Figure 3-2 illustrates the scatter plot between geo-tagged social media indicator and physical deprivation

final index at 1% significance level. Therefore, null hypothesis would be rejected, and the correlation is

significant. We can conclude that there is a relationship between geo-tagged social media indicator and final

index of physical deprivation, although the relationship might not be strong.

On the other hand, the correlation value between geo-tagged social media indicator and final index of social economic deprivation is low (-0.24) and the correlation is not significant. Therefore, null hypothesis cannot be rejected, and we can conclude that relationship between these particular indicators is not significant (Fig 3-3).

Figure 3-2: Correlation between geo-tagged social media indicator

and Physical deprivation


22

To examine possible association of all other indicators and final indices of deprivation

Moreover, the correlation matrix reveals that there are significant and strong correlation between other

indicators such as percentage of household with low income (0.92), percentage of inactive people (0.95) and

percentage of people with non-western origin (0.77) and final index of Social economic Deprivation. Scatter

plots have been made to illustrate this fact in figure 3-4 to 3-6 as follow.

Figure 3-3: Correlation between geo-tagged social media indicator

and Social Economic deprivation

Figure 3-4: Scatter plot between % households with low income and final index of social economic deprivation

Figure 3-5: Scatter plot between % of inactive people and final index of social economic deprivation


23

Also, the final index of physical Deprivation has a significant and strong correlation with indicator of

Average distance to secondary education with the value of 0.77 (Figure3-7).

To examine possible relationship between geo-tagged social media indicator and all other indicators

It is important to notice that table 3-1 also shows significant correlations at 99% level between geo-tagged

social media indicator and average distance to daily necessity (0.528), average distance to General

Practitioner (0.542), house affordability (0.561), and average distance to bus stops (0.634). Since the value is

between 0.5-0.7, these correlations can be described as a moderate correlation.

To find out the possible relationship between geo-tagged social media and these four indicators, figures 3-

8 to 3-11 are presented. Although the correlation matrix reveals the positive and significant linear relation

between geo-tagged social media indicator and indicators of average distance to daily necessities, house

affordability, average distance to General Practitioner, and average distance to bus stops, but the scatter

plots reveals that distribution of the neighbourhood does not follow the linear pattern. Therefore, the fact

of positive linear relation between geo-tagged social media indicator and these four indicators cannot be

justified due to the scatter plots.

Figure 3-6: Scatter plot between % of people with non-western origin and final index of social economic deprivation

Figure 3-7: Scatter plot between Average Distance to secondary eduction and final index of physical deprivation


24


25

3.1.2. Factor Analysis

To prove the suitability of Factor analysis for the research sample size which in this case is the number of

neighbourhoods, KMO test has performed. Values of KMO sampling adequacy for this analysis is 0.774

which according to Field (2005), considered as good (more than 0.7) and Factor analysis is acceptable (Figure

-12). Moreover, to maintain analysis consistency with correlation matrix, neighbourhood of City Center is

not considered in FA analysis as well.

At the second step, table of communalities reveals that value of common variance for indicator of growth of

new houses is 0.269. This value shows that common extraction for this indicator is too low, therefore, the

indicator of growth of new houses should be removed for future analysis from the indicator list (Figure 3-

14).

At the end, the factor analysis for the proposed conceptual framework returned to 3 components.

Factor loadings for factor 1 shows the highly associated of variables Average distance to bus stops, Average

distance to daily necessities, Average distance to general practitioner, and house affordability. Based on

associated variables this components can be named provision of services.

Factor 2 is highly associated with percentage of inactive people and percentage of low income households.

Moreover it has a moderate association with household size and percentage of non-western origin. Based

on mentioned factor loading this component will be called livelihood and ethnicity. Factor 3 has moderate

association with percentage of people over age of 65 and relative mortality rate and will be called factor of

age composition.

Due to observable factor loadings, the geo-tagged social media indicator with value of 0.714 has the

moderate association with factor of provision of services (Figure 3-13).

Therefore, based on above results we can conclude that geo-tagged social media indicator contributes to

physical deprivation and factor analysis prove this fact statistically.

Figure 3-10: Result of KMO and Bartlett test

Figure 3-12: Communalities Figure 3-11: Factor loadings


26

3.2. Mapping the indicators

To explore the spatial pattern that each of indicators follow, this section will present the map of each indicator result at neighbourhood level in Enschede.8

3.2.1. Social Economic Indicators

% of people with non-western origin

As the figure demonstrate, the concentration of people with non-western origin in Enschede mainly is

in the southern part of the city including neighbourhoods of Cromhoffsbleek-Kotman, and

Wesselerbrinks.

8 This section address to more critical maps and rest of the maps are presented in Annex 3.

Figure 3-13: Percentage of people with non-western origin at neighbourhood level in Enschede


27

%inactive people AS the figure shows, the most inactive people are accumulated at neighbourhood of Drienerveld-U.T. (in the area of University of Twente), and the city Center. Also neighbourhood Cromhoffsbleek-Kotman and Wesselerbrink Noord oost also have the high percentage of inactive people.

Figure 3-14: Percentage of inactive people at neighbourhood level in Enschede


28

%households with low income As it can be observed from the figure, the percentage of households that receive low income who quite a high in the built-up area of Enschede. This pattern is at its highest in neighbourhood of City center, Drienerveld-U.T, and south west of Enschede.

Figure 3-15: Percentage of households with low income at neighbourhood level in Enschede


29

3.2.2. Physical indicators

Growth of new housing

As Figure 3-16 shows, some few neighbourhoods in Enschede such as Drienerveld-U.T, Roombeek-

Roomveldje, and Mekkelholt had a growth in number of new houses in a period of 2007-2011. Since this

indicator considered as a cost to deprivation, therefore, the neighbourhoods with lower growth of new

houses are desirable. These neighbourhoods consists of Velve-Lindenhof and Eekmaat are shown with

red in following map.

Figure 3-16: Number of new houses at neighbourhood level in Enschede


30

Distance to cafeteria

The indicator of average distance to cafeteria is also considered as a cost to deprivation. This means that

neighbourhoods which are closer to cafeterias are more deprived, and better neighbourhoods are the ones

which have a distance with the location of Cafeterias. Since most of the cafeteria are concentrated in built-

up area especially the neighbourhoods near city center of Enschede , these neighbourhoods have the least

distance to cafeterias and fast food restaurants.

Figure 3-17: Average Distance to café teria at neighbourhood level in Enschede


31

Instagram Photo density Mapping the indicator of Instagram photo density has shown in figure 3-18. Since the Instagram density involves with population for each neighbourhood, the spatial pattern that has emerged from the map can be interpret as follow: The fringe areas such as Buurtschap Twekkelo, Buurtschap Broekheurne, Buurtschap Usselo, and Buurtschap Lonneker-West in Enschede have a very few population. The fact that they benefit from high value of photo density indicator can be interpret because of number of high number of geo-tagged photos in Instagram since these areas are recreational.

Figure 3-18: Instagram photo density at neighbourhood level in Enschede


32

3.3. Neighbourhood Deprivation in Enschede

The final result of this study, tries to map both social economic and physical Deprivation status at

neighbourhood level for city of Enschede. Figure 3-19 and 3-20 present different deprivation status

respectively. High deprivation status at each neighbourhood associates with higher value in each map.

3.3.1. Social Economic Deprivation Status in Enschede

The correlation analysis in section 3.1.1 shows that social economic deprivation is highly associated with

percentage of low income households and percentage of inactive people, and has a moderate association

with percentage of people with non-western origin. . The overlay map follow the same pattern excluding

neighbourhoods with non-available data.

Neighbourhoods of City Center, De Bothoven, Wesselerbrink Noord-West, Wesselerbrink Zuid-Oost,

Wesselerbrink Noord-Oost, Cromhoffsbleek-Kotman, Mekkelholt, Dolphia, and Twekkelerveld are

among the most deprived neighbourhoods in Enschede socioeconomically.

In addition, neighbourhoods of Eekmaat West, Oikos, Eilermarke, Hogeland-Zuid, 't Zwering, Bolhaar,

Helmerhoek-Noord, Stroinkslanden Noord-West, Stroinkslanden Noord-Oost, and Ruwenbos are the

least social economic deprived neighbourhoods in Enschede.

Figure 3-19: Social Economic Deprivation at neighbourhood level in Ensche


33

3.3.2. Physical Deprivation Status in Enschede

The correlation analysis in section 3.1.1 shows that physical deprivation is highly associated with distance

to secondary education. The overlay map behave in a same way. If secondary education locations

considered as the focal points, the more distance from them causes the most physical deprivation.

Generaly, the neighbourhoods which are located within the built-up area boundary are less deprived since

they have a better accessibility to different facilities.

Neighbourhoods of Buurtschap Twekkelo, Buurtschap Broekheurne, Buurtschap Lonneker-West, Noord

Esmarke, and Boekelerveld are the most physically deprived neighbourhoods since they have a quit high

distance to bus stops, daily necessities, General practitioners, and secondary educations. In addition, neighbourhoods of City Center, De Leuriks, Boddenkamp, Horstlanden-Stadsweide,

Roombeek-Roomveldje, Varvik-Diekman, Walhof-Roessingh, 't Zwering, Cromhoffsbleek-Kotman,

Mekkelholt, and Marssteden are the least physically deprived neighbourhoods in Enschede.

Figure 3-20: Physical Deprivation at neighbourhood level in Ensche


34

The value of correlation between final indices is also calculated in Table 3-1. At the 5% significance level

the final index of social economic deprivation is associated to final index of physical deprivation. The

strength of this relationship is (-2.90) which means the relationship between final indices of deprivation is

reverse and not that high. Figure 3-21 has added to clarify the behaviour of each neighbourhoods in both

social economic and physical deprivation. Moreover, this correlation analysis has been done for all the

neighbourhoods with data available in every indicators which consists of 54 neighbourhood in total.

In addition neighbourhoods of Bolhaar and 't Zwering are the least deprived neighbourhoods in both social

economic and physical dimension.

3.4. Content analysis

Content analysis for extracted Instagram photos consists of two parts. At first, a word cloud analysis will

perform on the captions of every photos in least and most deprived areas and try to present the captions

frequency in terms of positive or negative words. On the other hand, second part demonstrate deprived

areas by showing some representative images for both social economic and physical deprivation status.

3.4.1. Word Clouds for Social Economic Deprivation

Figure 3-22 and 3-23 present the word clouds for the two highest values social economic deprivation,

Cromhoffsbleek-Kotman (𝐼𝑆𝑜𝑐𝑖𝑜𝑒𝑐𝑜𝑛𝑜𝑚𝑖𝑐 = 0.59) and Wesselerbrink Zuid-Oost (𝐼𝑆𝑜𝑐𝑖𝑜𝑒𝑐𝑜𝑛𝑜𝑚𝑖𝑐 = 0.52).

Figure 3-21: Scatter plot between Social economic deprivation and Physical deprivation

Figure 3-22: Word cloud for Cromhoffsbleek-Kotman Neighbourhood

Figure 3-23: Word cloud for Wesselerbrink Zuid-Oost Neighbourhood


35

Figure 3-24 and 3-25 present the word clouds for the least socioeconomically deprived, Ruwenbos

(𝐼𝑆𝑜𝑐𝑖𝑜𝑒𝑐𝑜𝑛𝑜𝑚𝑖𝑐 = 0.13), and Oikos (𝐼𝑆𝑜𝑐𝑖𝑜𝑒𝑐𝑜𝑛𝑜𝑚𝑖𝑐 = 0.15).

3.4.2. Word Clouds for physical Deprivation

Figure 3-26 and 3-27 present the word clouds for the two highest values for physical deprivation, Boekelverd

(𝐼𝑝ℎ𝑦𝑠𝑖𝑐𝑎𝑙 = 0.67) and Buurtschap Twekkelo (𝐼𝑝ℎ𝑦𝑠𝑖𝑐𝑎𝑙 = 0.66).

Figure 3-28 and 3-29 present the word clouds for the least physically deprived areas, Marssteden(𝐼𝑝ℎ𝑦𝑠𝑖𝑐𝑎𝑙 =

0.36), and Boddenkamp (𝐼𝑝ℎ𝑦𝑠𝑖𝑐𝑎𝑙 = 0.37).

Figure 3-24: Word cloud for Ruwenbos Neighbourhood

Figure 3-25: Word cloud for Oikos Neighbourhood

Figure 3-26: Word cloud for Boekelverd Neighbourhood

Figure 3-27: Word cloud for Buurtschap Twekkelo

Figure 3-28: Word cloud for Marssteden Neighbourhood

Figure 3-29: Word cloud for Boddenkamp Neighbourhood


36

3.5. Illustrations

This section will present some of the extracted photos from the Instagram’s Geo Data Base which has been prepared for this research. Digging into each most and least deprived neighbourhoods reveals following images in each dimension:

3.5.1. Social Economic Deprivation status

The following pictures has been extracted from the least deprived neighbourhoods consists of Oikos,

Eilermarke, Ruwenbos. At a glance, photos in this category present a sense of happiness and certain level

of convenience among people’s lives.

On the other hand, the photos that have been extracted from the neighbourhoods of Wesselerbrink

Noord-West, Wesselerbrink Zuid-Oost, and Cromhoffsbleek-Kotman are significantly different with the

first category.


37

3.5.2. Physical Deprivation status

As it mentioned earlier, the most deprived neighbourhoods in Enschede are the ones which located in

suburban areas like Buurtschap Twekkelo, Buurtschap Broekheurne, and Boekelerveld. The extraction of

Instagram images, also confirms the final results.

On the other hand, the photos that are extracted for the neighbourhodos of Boddenkamp and Marssteden

shows the better status. These neighbourhoods benefit from the lower values of deprivation.


38


39

4. CONCLUSION AND RECOMENDATION

4.1. Conclusion

Based on three main research objectives proposed in chapter one, following answers have been found out

to cover the research questions:

To explore geo-tagged social media as an indicator for identification deprived areas at the

Neighbourhood level in city of Enschede.

o How the geo-tagged social media indicator incorporates with final indices of deprivation?

Results of Factor Analysis in section 3.1.2 reveal that geo-tagged social media indicator is

associated with factor of provision of services. Therefore, this indicator can be considered

as a physical indicator in order to quantify levels of deprivation at neighbourhood level.

o Is there a relationship between geo-tagged social media indicator and final deprivation

indices?

Based on proposed conceptual framework in section 2.2 and the results of correlation

analysis in section 3.1.1, there is a relationship between geo-tagged social media indicator

and final index of Physical Deprvation. This relationship is positive and significant but the

strength is not high. Moreover, the relationship between geo-tagged social media indicator

and social economic deprivation is not significant.

o What is the content of geo-tagged social media indicator?

The content of the geo-tagged social media was analyzed in a qualitative inquiry by verifying

the word frequency of positive and negative words used as the caption of Instagram

photos. The result of the frequency analysis was visualized in word clouds. These word

clouds were calculated for the neighbourhoods with highest and lowest deprivation

according to the social economic and physical dimension. For the social economic

dimension there was a significant difference between the words used in a deprived and

well-off neighbourhood, for physical dimension, the contents of fashion and fitness are

among the most observed words in the clouds.

Mapping social economic deprivation as well as physical deprivation at Neighbourhood level

o What is the social economic deprivation status at the Neighbourhood level?

In section 3.3 the geo-visualization for the social economic dimension of Enschede

deprivation mapped which provided a pattern showing that rural areas of Enschede have

lower social economic deprivation. Analyzing the individual indices also shows that in

deprived neighbourhoods the percentage of inactive people and people with non-western

origin are usually high.

o What is the physical deprivation status at the Neighbourhood level?

In section 3.3 the geo-visualization for the physical deprivation was presented. The map

reveals a spatial pattern showing that neighbourhoods which are within built-up areas have

a lower deprivation. The more distance from the center of Enschede causes the more

physical Deprivation. Based on correlation analysis deprived areas have a far distance from

secondary schools. Also, deprived areas have a small distance to cafeterias.


40

Understand the relationship between social economic deprivation and physical deprivation

o To what extent the social economic deprivation is correlated with physical deprivation at

the Neighbourhood level?

The results showed that at 5% significance level, the final index of social economic

deprivation is associated reversely with the final index of physical deprivation. However,

the correlation is not strong, the fact has been proved by correlation matrix presented in

section 3.1.

4.2. Research limitations

Due to statistical data limitation 16 Neighbourhoods (approximate 25% of the Neighbourhoods) had

missing values for some attributes for security reasons. Therefore the final indices for Social economic and

Physical dimensions was only calculated for 54 Neighbourhoods of Enschede municipality.

During the Factor Analysis was verified a low communality (0.269) for the indicator Growth of new houses.

As this value is lower than 0.3 it should have been removed from the analysis, but due to time constrains

the indicator wasn’t removed. Also in the correlation analysis was verified a 99% statistically significant high

correlation (0.946) between the indices Inactive people and Low Income, to further develop the framework

one of those two indices should be removed.

More precise results for the word cloud can be obtained by using website that takes in consideration

connectives from multiple language (in this case English and Dutch), that cleans for special characters and

that’s case insensitive.

Another limitation for this research refers to data extraction from Instagram’s API. The data extraction

process within Instagram API can be done by specific parameters which does not allow you to differentiate

certain groups of users like residents or tourists. Therefore the spatial distribution of geo-tagged photos are

the results of everyone who geo-tagged photo in Instagram application.

Moreover, this should not be neglected that the number of unrelated photos to deprivation are relatively

high in each neighbourhood and representative photos should be extracted cautiously.

4.3. Recommendation

As a recommendation for future work one could analyse the temporal patterns of geo-tagged social media,

verifying if exists any kind of correlation between the hours and week days of social media activity and

deprivation. Also other kinds of social media could be evaluated, such as Foursquare, which can provide a

rich data of Points of Interest. Analysis in variability of deprivation are lacking research as well, one could

do a temporal comparison of how the deprivation of Enschede changed within a specified time frame.


41

LIST OF REFERENCES

Baud, I., Sridharan, N., & Pfeffer, K. (2008). Mapping Urban Poverty for Local Governance in an Indian

Mega-City: The Case of Delhi. Urban Studies, 45(7), 1385–1412.

http://doi.org/10.1177/0042098008090679

CBS. (2015). Centraal Bureau voor de Statistiek. Retrieved June 9, 2015, from http://www.cbs.nl/

EEC. (1985). On specific Community action to combat poverty (Council Decision of 19 December 1984).

Official Journal of the EEC.

European Union. (2012). Measuring material deprivation in the EU- Indicators for the whole population and child-

specific indicators. Luxembourg.

Field, A. (2005). Discovering Statistics Using SPSS (3rd ed.). London: SAGE Publications.

Instagram. (2015). Instagram API. Retrieved June 9, 2015, from https://instagram.com/developer/

Kapiche. (2014). Text is Beautiful. Retrieved June 12, 2015, from http://textisbeautiful.net/

Li, L., Goodchild, M. F., & Xu, B. (2013). Spatial, temporal, and socioeconomic patterns in the use of

Twitter and Flickr. Cartography and Geographic Information Science, 40(2), 61–77.

http://doi.org/10.1080/15230406.2013.777139

Martínez, J. (2015). Lecture notes on Urban poverty Analysis. Enschede: University of Twente, Faculty of

Geo-Information Science and Earth Observation.

Martínez, J., & Dopheide, E. (2014). Indicators: from Counting to Communicating. Journal for Education in

the Built Environment, 9(1), 1–19. http://doi.org/10.11120/jebe.2014.00009

Meijer, M., Engholm, G., Gritter, U., & Bloomfield, K. (2013). A socioeconomic deprivation index for

small areas in Denmark. Scandinavian Journal of Public Health, 41(6), 560–9.

http://doi.org/10.1177/1403494813483937

Mpata Wekisa, E. (2014). Neighbourhood Deprivation in Enschede- Unfolding deprivation with GIS and

social media. Enschede: University of Twente, Faculty of Geo-Information Science and Earth

Observation.

NGR. (2015). Nationaal Georegister. Retrieved June 9, 2015, from http://www.nationaalgeoregister.nl/

Noble, M., Mclennan, D., Wilkinson, K., Whitworth, A., Exley, S., Barnes, H., & Dibben, C. (2011). The

English Indices of Deprivation 2010. Retrieved from http://eprints.ioe.ac.uk/2461/

Noble, M., Wright, G., Smith, G., & Dibben, C. (2006). Measuring multiple deprivation at the small-area

level. Environment and Planning A, 38(1), 169–185. http://doi.org/10.1068/a37168

Pacione, M. (1995). The geography of multiple deprivation in Scotland. Applied Geography, 15(2), 115–133.

http://doi.org/10.1016/0143-6228(94)00005-B

provincie Overijssel. (2015). Data Bank Overijssel. Retrieved June 1, 2015, from

http://www.overijssel.databank.nl/jive

Quercia, D., & Saez, D. (2014). Mining urban deprivation from foursquare: Implicit crowdsourcing of city

land use. IEEE Pervasive Computing, 13(2), 30–36. http://doi.org/10.1109/MPRV.2014.31

Townsend, P. (1987). Deprivation. Journal of Social Policy, 16(02), 125–146.


42

http://doi.org/10.1017/S0047279400020341

Wikipedia. (2015a). Application Programing Interface. Retrieved June 9, 2015, from

http://en.wikipedia.org/wiki/Application_programming_interface

Wikipedia. (2015b). Geotagging. Retrieved June 9, 2015, from http://en.wikipedia.org/wiki/Geotagging

Wikipedia. (2015c). Instagram. Retrieved June 9, 2015, from http://en.wikipedia.org/wiki/Instagram

Wikipedia. (2015d). JSON. Retrieved June 9, 2015, from http://en.wikipedia.org/wiki/JSON

Wikipedia. (2015e). Tag cloud. Retrieved June 12, 2015, from http://en.wikipedia.org/wiki/Tag_cloud

Yuan, Y., & Wu, F. (2014). The development of the index of multiple deprivations from small-area

population census in the city of Guangzhou, PRC. Habitat International, 41, 142–149.

http://doi.org/10.1016/j.habitatint.2013.07.010


43

ANNEX 1

Ge

oD

ataBase

DataSe

tC

lassFe

ature

type

Attrib

ute

/Field

De

scriptio

nY

ear

Sou

rce o

f data

BU

_Naam

Nam

e of ea

ch n

eighb

ou

rho

od

20

11

_V3

CB

S

AA

NT_IN

WP

op

ulatio

n o

f each

neigh

bo

urh

oo

d2

01

1C

BS

P_6

5_E

O_JR

%p

erson

s 65

years an

d o

ver2

01

1C

BS

i_age6

5N

orm

alized valu

e of %

perso

ns 6

5 yea

rs and

over

20

11

P_STER

FTM

ortality relative

20

11

CB

S

i_mo

rtN

orm

alized valu

e of m

ortality relative

20

11

GEM

_HH

_GR

ho

useh

old

size2

01

1C

BS

i_ho

usesiz

No

rmalized

value o

f ho

useh

old

size2

01

1

P_N

_W_A

L%

peo

ple w

ith no

nw

estern o

rigin2

01

1C

BS

i_no

nw

estN

orm

alized valu

e for %

peo

ple w

ith no

nw

estern o

rigin2

01

1

P_N

IETAC

T%

of in

active 2

01

1C

BS

i_inactive

No

rmalized

value o

f % o

f inactive peo

ple

20

11

P_LA

AG

INK

H%

ho

useh

old

with lo

w in

com

e2

01

1C

BS

i_low

Inc

No

rmalized

value fo

r % h

ou

seho

ld w

ith low

inco

me

20

11

WO

ZA

verage value o

f residen

tial area2

01

1C

BS

INK

_INW

Average in

com

e per capita

20

11

CB

S

Ho

useaff

Ho

use affo

rdab

ility2

01

1O

wn

i_ho

useaff

No

rmalized

value fo

r ho

use affo

rdab

ility2

01

1

AvgD

ist_bu

sA

verage Distan

ce to b

us statio

n

20

12

NG

R

i_bu

sN

orm

alized valu

e for A

verage Distan

ce to b

us statio

n

20

12

AF_D

AG

LMD

Average distan

ce to d

aily necessities

20

11

CB

S

i_dailyN

ecN

orm

alized valu

e for A

verage distance to

daily n

ecessities2

01

1

BEV

_DIC

HTH

Po

pu

lation

den

sity2

01

1C

BS

i_den

sityN

orm

alized valu

e for p

op

ulatio

n d

ensity

20

11

WO

NIN

GEN

_20

11

Total N

um

ber o

f ho

usin

g 20

11

20

11

CB

S

WO

NIN

GEN

_20

07

Total N

um

ber o

f ho

usin

g 20

11

20

07

CB

S

new

ho

use

Gro

wth o

f new

ho

usin

g 2

00

7-2

01

1O

wn

i_new

ho

use

No

rmalized

value fo

r grow

th of n

ew h

ou

sing

20

07

-20

11

AF_A

RTSP

RA

verage distance to

GP

20

11

CB

S

i_gpN

orm

alized valu

e for A

verage distance to

GP

20

11

AF_C

AFTA

RA

verage distance to

cafeteria2

01

1C

BS

i_cafeN

orm

alized valu

e for A

verage distance to

cafeteria2

01

1

AF_O

ND

VR

TA

verage distance to

secon

dary ed

ucatio

n2

01

1C

BS

i_secEdu

No

rmalized

value fo

r Average distan

ce to seco

nd

ary edu

cation

Instagram

Geo

-tagged so

cial med

ia2

01

4-2

01

5In

stagram

i_instagram

No

rmalized

value fo

r geo-tagged

social m

edia

20

14

-20

15

SocialEco

no

mic

Final So

cial Econ

om

ic Dep

rivation

Ind

ex2

01

1O

wn

Ph

ysicalFin

al Ph

ysical Dep

rivation

Ind

ex2

01

1O

wn

idIn

stagram's p

ost U

RL

20

14

-15

Instagram

Place

Man

ual Lo

cation

nam

e2

01

4-1

5In

stagram

time

Date an

d Tim

e of In

stagram p

ost

20

14

-15

Instagram

image

Image U

RL

20

14

-15

Instagram

captio

nA

dd

ition

al image descrip

tion

wro

te by In

sta users

20

14

-15

Instagram

Origin

s of d

ata:

http

s://apigee

.com

/con

sole/in

stagram

http

://ww

w.n

ation

aalgeoregister.n

l

http

://ww

w.cb

sinu

wb

uu

rt.nl/

Hea

lth and

living co

nd

ition

Edu

cation

Place po

pu

larity

Ensch

ede_

Dep

rivation

_20

15

Neigh

bo

urh

oo

ds

Final D

eprivatin

Ind

ices

po

lygon

Ensch

ede_

neigh

bo

urh

oo

d

Po

pu

lation

com

po

sition

Emp

loym

ent

Inco

me

Accessib

ility

Ho

usin

g

Po

int

Instagram

_20

15

Table 0-1: Meta Data


44

ANNEX 2

Table 0-2: Complete Correlation Analysis including City Center Neighbourhood


45

ANNEX 3

Figure 0-1: Percentage of persons 65 years and older at neighbourhood level in Enschede


46

Figure 0-2: Relative Mortality Rate at neighbourhood level in Enschede


47

Figure 0-3: Household size at neighbourhood level in Enschede


48

Figure 0-4: House affordability at neighbourhood level in Enschede


49

Figure 0-5: Average Distance to bus stops at neighbourhood level in Enschede


50

Figure 0-6: Average Distance to daily necessities at neighbourhood level in Enschede


51

Figure 0-7: Population Density at neighbourhood level in Enschede


52

Figure 0-8: Average Distance to General Practitioners at neighbourhood level in Enschede


53

Figure 0-9: Average Distance to secondary education at neighbourhood level in Enschede

Revision_PGD_April2016

Documents