 Hello everyone and welcome to this webinar on the impact of online shopping on the UK High Street. I'm Margarita Seraulo. I'm an outreach officer working for the UK Data Service based in Manchester at JISC and I'm presenting today with Lester Lager. He's a lecturer in Geographical Data Science for the Consumer Data Research Centre at the University of Liverpool. This is just a quick overview today. I will introduce the phase two of the big data network and the Consumer Data Research Centre in particular. And then Les will tell you more about his research on your resilience, including the data and methodology he used. And then we will finish off with some questions. Right, so the ESRC has created a big data network which is divided in three phases. As part of the phase two, it is investing in the establishment of data research centres with a focus on business and local government data. These three centres are the BLG Data Research as you can see. So the ESRC Business and Local Government Data Research Centre based at the University of Essex, the Consumer Data Research Centre which I'm going to show you more about later, and the Urban Big Data Centre based at the University of Glasgow. So these three research centres make data routinely collected by business and local government organisations accessible for research and they benefit to data owners and society and they always ensure that individuals' identities are safeguarded. So the UK Data Service has a big data team called Big Data Network Support. So the ESRC has invested in us to enable research to make the most out of these data for knowledge exchange and impact. And we support and coordinate the activities between the three data research centres. Our aims are to unify data discovery across the phase two of the Big Data Network collections. We encourage the sharing of information and expertise across these centres and coordinate user training capacity building in Big Data Analytics for researchers using the data. Just a few words about the CDRC in particular because Les does work for the CDRC. They combine expertise from four different universities or researchers within those universities at the University of Leeds, the University College London, University of Liverpool and the University of Oxford. They do provide data that cover a range of topics concerning the characteristics, constraints and outcomes of consumption. You can find the searchable data catalogue on their website and they do offer open, safe, guarded and secure data. So with different access conditions depending on the sensitivity of the information contained in the data set. They do offer some training as well in data analytics mainly on how to use R and GIS. And I strongly suggest you check their website out where there's more information about it and it's cdrc.ac.uk. And now I'm going to hand over to Les who's going to talk to you more about his research. Hello, my name is Les Dovegan and I'm going to talk today about irresilience of UK retail centers. This is a project I've been working on for about last 18 months or so. So I will start with a bit of a background on the retail sector in the UK and the evolution of retail town centers in the UK. Then I will introduce the irresilience concept and move on to the empirical analysis of and research outputs. Lastly, I will outline the value added by this study. So let's start with a bit of information about retail sector in the UK. Many commentators claim that Britain is a nation of shopkeepers. Shopping is a national pastime so there must be something to it. As the UK retail sector has been very successful over the past 25 to 30 years and it's an important part of UK's economy. There are almost three million people working in retailing at the moment which is about 10% of the UK workforce. Retail sales stood at 7.1 billion pounds per week last year in 2015 and this is an equivalent of 21% of UK's GDP as you can see is a massive chunk of that GDP. On average, we make about 200 shopping trips per year which comes to about just under four trips shopping trips per week. There are various shopping destinations but in this study I will look mainly at town centers and shopping centers. So let's start with the evolution of town centers. Town centers, they constantly evolve and this is viewed as a good thing. Sometimes that allows them to adapt to things like shifting consumer demand or external internal shocks. In the UK, the evolution has been driven by four major forces, the first of which is competition from out of center large retail development. This type of developments have largely resulted from the policies of touchers government and they lose approach to planning system. The so-called vitality and viability of UK town centers was only prioritized in the mid-1990s by adoption of the so-called sequential test which prioritized retail developments within town centers followed by edge of center, district centers and only as a last resort out of town sites. Then we've got the second force which is a shock of the economic crisis and the following era of austerity. In broad terms, the impact was complex but it was largely associated with sharp increases in vacancy rates. The average vacancy rate in England and well in the UK had increased from about 8% in the pre-crisis period sort of 2007-8 to 14.5% in 2012 and since then it was gradually declining and it stands at the moment at about 12%. Then we've got the changing demographics and consumer culture. These include some trends that have been going on for some time such as aging society or decreasing household size. However, really interesting here are the recent trends such as increased demands for value for money or progressive rise of convenience culture. That's why we can see in our high streets an increasing number of pound loans, savers, home bargains or convenience stores such as Tesca Express or Science British Local and then we've got the fourth force rapid growth of online sales. Online sales have essentially tripled over the past seven, eight years and now they exceed 15% of total UK retail sales and I have to say this is the highest figure in the entire world. This can be nicely symbolized by Amazon having become the country's eighth largest retailer. The major retailers they have adapted to that change relatively quickly. The traditional store-based retailers, we call them sometimes bricks-and-mortar transferred part of their business online which was relatively easy and this new business model is also is referred to as bricks-and-clicks. However, it is worth mentioning that this transformation wasn't as successful for many smaller and independent retailers. And what is really interesting here shown on this bottom graph is that the online penetration rates, they vary considerably by retail or service sector. The sales of products that can be easily digitized such as music, films, games or software take place mainly online nowadays. And at the other end of the spectrum, these types of retail services that provide some sort of experience such as health and beauty or leisure services. There are also products that we like to expect in person physically and also just groceries. So what is the research question? Well, I was interested in the response of UK town centers to the increasing online sales. So far most researchers have focused on supply factors and as a result the geography of online sales was little understated. So I've addressed the issue and answered some of other questions as well related to both supply and demand factors. And in particular, there were these two questions here. How the structure of traditional high streets is being impacted by consumers' propensity for online shopping? And a second question, how irresilience of retail centers can be measured in a meaningful, useful, helpful way? The novel thing about this research was that we employed big data from various sources in order to examine the propensity for online shopping at local level area. We estimated the exposure of retail centers to online shopping and we estimated retail catchments at a national extent. So let's have a quick look at the concept of irresilience. Retail irresilience is a novel concept, a concept that defines the vulnerability of British retail centers to the effects of rapidly growing internet sales. In other words, it's a concept that defines and measures the impact of online sales on more traditional retailers and traditional high streets in the UK. So as this is a novel concept, the study required a theoretical framework but also a robust method of irresilience can be measured. So the empirical study consisted of the following parts which are listed here. It was the estimation of catchment areas for UK retail centers. It was the obtaining of internet user classification at a small area level, that was the LSOA, lower super output area level. There was the assessment of the exposure of retail centers to online shopping estimation of retail supply vulnerability and I will talk about those in a minute in more detail. So let's start with the theoretical framework. So the irresilience is about both supply and demand factors. On the supply side we have connectivity which essentially is available infrastructure to get online and we proxy this by speed and rurality. On the demand side we have consumer behavior and essentially the decision of whether or not to shop online. That's very crucial whether or not to shop online. This decision depends on customer's engagement with information and communication technologies so-called ICT but also on the catchment demographics. In particular factors like age and socioeconomic status turn out to be very important. That decision of whether or not to shop online is also closely related to retail supply factors in this box here. And we looked at things like retail center attractiveness, safety, accessibility, shopping convenience. So the assumption was if the retail center offers a safe, convenient shopping environment, one which is really attractive, the center will be more resilient. So let's have a look at our data. First is applied and demand data. So overall there are about 1,300 town centers in the UK and twice as many shopping centers and retail parks. The map on the right shows spatial distribution of these town centers, 13,000 town centers depicted by the size and as you can see the size of London it's much bigger than any other center. Not even any other center is close to that size is completely. In terms of retail occupancy data we got it from local data company. Another supplier of such data is GoatXperien. This data provides detailed information on the location of all retail and service units and the different classifications. There is a broad classification which includes these four categories, comparison, convenience, services and vacant. Comparison retailing essentially is the non-food retailing, convenience retailing is the food related retailing, services and vacant which are our empty units. And there is also much more detailed classification with more than 100 categories and in here you will have categories like convenience stores, supermarkets, shoemakers, post offices, estate agents and so on and so on. We also used 4.7 million unit postcode level internet speed test results which are available from broadband speed tracker.co.uk and we also used distances from the nearest mobile mass the so-called base stations. In terms of demand data I have to say that the data were assembled at the 2011 lower layer super output areas so-called LSOA level which comprises over 34,000 zones in England and Wales of between 1,000 and 3,000 people. So each LSOA had about 1,000 to 3,000 people in it. So firstly a range of socio-demographic indicators from the 2011 census was collected including levels of education, employment sector, prevalence of full-time students, age structure and population density and secondly the Oxford Internet Institute Survey the so-called OXIS were employed. That was a sample for the whole Great Britain with a sampling method that enabled the projection of estimates to the whole country and comparison over time at LSOA level again. So one of the first steps was after the conceptual framework it was estimation of retail catchment areas. So the aim was to estimate catchment areas for a national network of retail centers. This is a complex task and one that requires a large degree of generalization and a substantial computational power to say the least. So traditionally there are two types of methods used to estimate retail catchments that are the simple methods such as buffers or drive distances or drive times. They are still popular and used by many retailers however they are unlikely to sufficiently capture the complexity of different attributes that may influence true catchment extent. So our preferred approach was to use spatial interaction models the so-called gravity or probabilistic models. This type of models they apply Newtonian laws of physics to the modeling of shoppers behavior based on the influence and the attractiveness of the store or retail center or network of retail centers and influence declines with distance between origin and shopping destination. These types of models they provide much more accurate catchment areas. However one of the major limitation is that there is a need for some sort of customer insight data in order to calibrate the model otherwise we end up using only arbitrary values which is not might end up with you know irrelevant output. So here is the half probability model we use bespoke version of it and in here you've got got the formula of that model and graphical representation just to help you to understand how the model works. So the simple version of proof I calculated so I calculated the probability P for each customer's location I in our case this where I saw us of using a particular center J. So we calculated probability P for customer's locations I of using retail center J and this turned out to be a function between retail center attractiveness A and a distance D between the actual retail centers and consumer domicile. Of course you can disaggregate that model parameters into things like large small medium town centers we used in about five or six parameters in that case this is just a simple version of it. So then we have some outputs and model calibration as well depicted here on these two figures. So we can see here Manchester on the top and Birmingham on the bottom so the output looks quite reasonable I have to say it looks quite good in fact in red we have the primary catchments in orange you have the secondary and yellow the tertiary catchment. So for the catchment for Liverpool sorry for Liverpool for Manchester city center it's really large and the catchments for the surrounding satellite towns are much much smaller. But in terms of model calibration I used axioms data on shopping flows. These are origin destination matrices derived from surveys on non-food primary shopping destinations okay non-food primary shopping destinations. By creating the probabilities for each of the SOA I mapped axioms data the real world catchments if you like and these are on the left hand side. And that was quite easy to compare against our model output which is on the right hand side. Overall the correlation was just around 90% which is quite good however the inner city location in the inner city location the response rate was relatively low and therefore we've got this fragmented picture as the error tends to be higher in that central area. And we published the results of this study in a journal of retailing and consumer services so if you want to have a look please do so. Then we've got the internet user classification. It is a purpose-built classification of internet use and engagement and we use k-means clustering method and we employed three major blocks of data. This were Oxford internet surveys, I've already mentioned those before, the so-called oxys, internet and ablic infrastructures and socio-demographic indicators from the 2011 census. And an example of such classification is shown in these maps for Liverpool and London. There are 11 major groups that engage with ICT in completely different ways so each of these groups which is here they engage in different ways with information communication technology or also what it means that some of these groups will be engaging in online shopping more than other groups. In London there is a quite distinctive spatial pattern but in Liverpool it's not that clear. And here I've got one of these rendered pictures of internet user classification. It looks like satellite picture or something in London by night but it's not. It's internet user classification. It's just different types of rendering. Then we've got consumers propensity for online shopping. So as you can see on this graph here on the right hand side nationally rates of online shopping quite to 53%. So this is our zero basically in fact and here we're showing the differences from the national mean. And so that clear difference is between different IUC groups. These are the same groups as I showed in the previous map. So for example the low density by high connectivity group which is here and the other two groups called constrained by infrastructure in French are most likely to engage in online shopping whereas uncommitted and casual users, emerging as all these three or four groups here less likely to engage in online shopping. So having the information for the whole country we could create catchment profiles based on that classification. And having catchment profiles for every single town center we could create index of high exposure. And this map here on the left hand side provides evidence that predominantly the secondary and tertiary little centers located around major metropolitan areas reveal the greatest exposure to the impacts of online sales. This trend has been reiterated in other parts of the country although to a smaller extent. So all the bigger metropolitan areas such as Manchester, Liverpool, Leeds will have similar trends but to slightly smaller extent. What's also very interesting is that none of the big centers, town centers, city centers was classified as highly exposed. They were all basically like commuter belt in southern East England. And then we've got vulnerability to online shopping. I used the evidence from literature and I calculated the composite measure of retail centers vulnerability based on supply factors. Essentially these were data on retail occupancy occupancy. So I look at the proportion of different retail types, broader, more detailed categories and then and calculated the indices. So what you need to know here that there were two types of impacts is the distinguished. So there was a positive impact for anchor stores and leisure units. As if a center had relatively many anchor stores and quite rich leisure units, this center was much more likely to be resilient as people were much more likely to come to this center and stay a bit longer. Also leisure units, leisure offer is very difficult to digitize. And there was a negative wave of the so-called digitalization retail. These were the types of retail services that have high online penetration levels. So just to recap, higher proportions of digitalization retail were associated with enhanced vulnerability, whereas higher proportions of anchor stores and leisure units indicated greater resilience to online sales. So by exploring and then intersecting the exposure and vulnerability indices, I could calculate the resilience scores for each retail centers. And here I've got two tables showing 10 most and 10 least resilience town centers in the UK. The final scores were scaled into the range between 1 and 100 for the clarity reasons. So it can be seen quite clearly what is the score and where that center is placed in the index between 101. And here we've got flow diagram of the way we calculated the resilience scores. So just to recap again, we had this, on the left-hand side, we had this OXIS surveys, sensors and internet infrastructure data. And based on this data, we came up with these internet user classifications, the so-called IUC, by using K-means clustering methods. And on the other hand, we had this local data company data, town center boundaries data and road network. So by using this data, I was able to calculate retail catchments for every town center in the country by using the bespoke half model. In stage two, using those two internet users classification and retail catchments, I could calculate index of high exposure to online shopping. And directly from this data, I could calculate index of supply vulnerability. And by intersecting those two indices, I came up with this irresilience measure that I showed you before. And the results of this study were published in GeoForm quite recently. So what is the value added to this study? Well, the study provides new insights into the debate on the impacts of online retailing on traditional brick and mortar retailers. It also tries to rebalance the current debates which has focused on supply factors. So I'm trying to show the importance of demand factors as well, in other words, geodemographics. The study investigates how the resistance to impact of online sales can be measured and what role local demographics may have in that context. Okay. And the study also offers valuable tools for various stakeholders. Basically, people who are engaged in revaluation of retail capacity models or people who are engaged in different schemes of town center, improving town center performance will find this tool pretty useful. And all these models, all tools are open source, so they're really available online. And the last slide, it's about next steps. So at the moment I'm working on updating retail center boundaries. As the official DCLG department for communities and local government boundaries, which we used were from 2004, which is quite old, but that's the official boundaries half at the moment. So the extent has changed in many cases quite significantly. So for this reason we are updating these boundaries, but also creating a new method which will be more robust and will be much easier to rerun the model and update town center boundaries, let's say, on an annual basis after each survey or something like that. Second project I'm working on at the moment is related to irresilience, is revaluation of retail catchments based on variable propensity for online shopping. So the current catchment model assumes uniform propensity for online shopping, which in other words doesn't account for it at all. Okay. So the irresilience project provides empirical evidence that variable propensity can have an impact on catchment extent. So it would be useful to rerun the model using our irresilience score. And you've got two maps in here showing that's how catchment for Gloucester, how this irresilience can impact potentially the catchment extent. And we also are working on evaluation of our model using consumer insight data. We have quite lots of data on the CDRC website. So one of these data set is about click and collect from one of the major retailers, and we will use that data to evaluate our models. And thank you very much. I'm going to thank you for presenting today. And I'm going to thank all the attendees. I hope you found it interesting. And have a good afternoon. Bye everyone. Thank you. Bye bye.