Retail impact assessment (RIA) is a common methodology employed by local authorities in the UK to determine the impacts of new shopping centre developments on existing town and city centres. In the UK, initial approaches utilised retail modelling techniques such as retail gravity models, the most notable example being the Haydock study in 1964. There was widespread opposition to increasing quantification in planning and the use of mathematical models in the late 1960s and early 1970s, as model assumptions were increasingly questioned and different agencies brought different model results to the enquiries. In the 1980s the Department of the Environment advised against the use of mathematical models in impact studies. However, RIA has continued since the 1980s using alternative but still largely quantitative approaches. The aim of this article is to begin the research task of critically evaluating this standard RIA approach (widely used by retail consultants for planning enquiries) against traditional mathematical modelling procedures. Using the new Silverburn retail centre in Glasgow, UK, we shall explore the potential impact of the new centre based on these alternative methodologies and draw some conclusions on the relative strengths and weaknesses of each.
(ProQuest: ... denotes formulae omitted.)
Spatial interaction models (SIMs) have a long history in geography and related disciplines, especially following Wilson's derivation of the models using entropymaximising techniques (Wilson, 1967; 1974). Since then they have been used successfully in a range of applied settings, especially as retail location and impact assessment models. Birkin et al. (2002; 2010) review the progress made with applied SIMs in retail location planning. However, it is noticeable that this progress and success has been largely confined to applications within the private sector. A key question is what happened to these models for public sector usage, especially in the UK which is the focus of the study presented here.
Breheny (1988) expressed surprise (on behalf of the academic community as a whole) that spatial interaction models re-emerged in the UK for retail site location or impact assessment in the private sector during the early 1980s. Indeed, spatial interaction models are now integral to the site location methodologies used in many UK blue-chip organisations, such as Tesco, Sainsbury's, Boots, Thomas Cook, etc. (Birkin et al., 2002; 2010). Breheny's surprise was because these models had fallen out of fashion in the public sector in the late 1970s. According to Breheny, this was a combination of disenchantment and Government warnings that the use of such models was not appropriate at planning enquiries. In the view of Guy (1991), there was widespread opposition to increasing quantification in planning generally and the use of mathematical models during the 1970s. Norris (1990) even suggested that this was because SIMs involve 'optimistic guesswork' which is 'hidden by the workings of the model' (p. 108). Despite some SIMs continuing to be applied in the public sector, notably the models of the Unit of Retail Planning and Information (Alty, 1979), the step-by-step retail impact assessment (RIA) methodology became more popular. Even though there are many different versions of RIA, all the examples given in the bibliography here follow the same five main steps, which we review later.
The aim of this article is to provide a critical review of the RIA and SIM methodologies. Although critical comments have previously been made regarding both methodologies (England, 2000; Guy, 2007), very few empirical case studies have been presented in the academic literature. In this article, comparative analysis will be undertaken in relation to a major new out-of-town retail development at Silverburn near Glasgow. Silverburn was chosen as it is the most recently opened large out-oftown regional shopping centre in the UK. This allows the case study to be supported by richly specified consumer data from the National Shoppers Survey (NSS) provided by Acxiom Ltd (available since 2005), as well as by a variety of retail directory sources.
In Section 2, we review the need for impact assessment and the history of the two main methodologies. The data sources and the Silverburn study area are described in Section 3. The implementation of the two methodologies is described in Section 4, providing the basis for an assessment of the strengths and weaknesses of each approach in Section 5. Finally, some reflections and concluding comments are offered in section 6.
Retail impact assessment
A retail impact assessment (RIA) is a means of establishing the likely trading impact of a proposed new retail development on existing and committed retail developments (England, 2000). RIA is very important for planning policy and decision-making and it is an area of urban planning which evokes many different emotions from planners, politicians, developers and planning inspectors. Attitudes vary from enthusiastic to uninterested, confused, sceptical and even antagonistic (England, 2000). That said, RIA is a statutory requirement at most planning enquiries. PPS6 in 2005 made it clear that
[i]mpact assessments should be undertaken for any application for a main town centre use which would be in an edge-of-centre or out-of-centre location and which is not in accordance with an up-to-date development plan strategy. Where a significant development in a centre, not in accordance with the development plan strategy, would substantially increase the attraction of the centre and could have an impact on other centres, the impact on other centres will also need to be assessed. (ODPM, 2005, para. 3.20)
The importance of accurate impact assessments has been stressed by Norris (1990, 104). He argues,
if a wrong decision is taken with regard to superstore location then little harm will result, except maybe to the developer and to one or two nearby stores. However, if a decision to allow a major regional shopping centre turns out to be wrong, then the continued health of some of the existing town centres could be seriously curtailed and alternative or more suitable locations would be effectively sterilised.
Generally, a RIA is required when a proposed development is of a scale likely to have a significant impact on the trade of existing or committed retail outlets or centres and the surrounding area. Many planning authorities insist on such an impact study for all major new developments, and these applications in the UK are usually carried out by private-sector planning consultants on behalf of the prospective developer. An RIA will normally be considered essential for superstores, retail warehouses, retail parks, shopping complexes and shopping centres which are individually or cumulatively over 2500 m2 in gross retail floorspace, and it may occasionally be essential for smaller developments, for those which are likely to have a significant impact on smaller centres (ODPM, 2005, para. 3.23).
The conventional approach to RIA is to think about the impact on retail businesses and centres in terms of trade lost or diverted. However, impacts arising from a new shopping centre upon existing shopping centres can be identified by examining three types of impact.
* Economic impact: this type of impact concerns the changes in retail turnover or trading patterns in shopping centres, including employment impacts. This includes an assessment of the potential decline in the competitive position of a number of existing centres. BDP (1992) discuss the negative economic impacts such as loss of trade in existing centres and diversion of retail investment from existing centres. On the other hand, the improvement in scale, structure and diversity of centres, and potential employment gains and income multiplier effects, can be positive economic impacts (Guy, 2007). Other positive impacts might include the provision of a wider range of shopping facilities (improved choice for consumers), improved competition resulting in cheaper products, and, in those centres adversely affected by new developments, reinvestment and refurbishment resulting in revitalisation of centres as part of a competitive response (SG, 2007).
* Social impact: this type of impact is relevant to demographic and behavioural change and the implications for shopper profiles for existing and new centres and the role of social inclusion/exclusion (England, 2000, 10-11; SG, 2007). The increasing disparity in access to retail between social groups is a potentially negative impact, but the improvement of access to retail in areas of social exclusion is a positive social impact (Guy, 2007). The negative social impacts tend to adversely affect those on lowest incomes and/or the least mobile, but such impacts will depend upon the precise location of new developments and arrangements made for supporting access (Guy, 2007).
* Environmental impact: there are two types of environmental impact which are identified in the literature: transport impacts (traffic volumes, car parking, etc.) and impacts on the built environment (aesthetics of new buildings, loss of green land, etc.; England, 2000, 12-13; Oughton et al., 2003). Negative environmental impacts also include increased vacancy levels and/or poor quality shops and services which are left in other centres. On the other hand, 'planning gain' associated with remediation of contaminated land and visual improvements are positive environmental impacts (BDP planning, 1992 cited in Guy, 2007).
New, out-of-town, regional shopping centres first began to appear in the UK in the 1970s. Brent Cross in London was the first major regional shopping centre to start trading in 1976. These developments were accompanied by a number of wellknown impact studies (Bennison and Davies, 1980; Schiller, 1986; 1996; Reynolds and Schiller, 1992; Rowley, 1992; Howard and Davies, 1993; Donaldsons, 2005). These studies generally report significant negative impacts on neighbouring town and district centres due to these regional shopping centres. Furthermore, their out-of-town locations increase concerns of environmental damage and social equality (especially the issue of access for non-car owners: Crosby et al., 2005). For example, Roger Tym and Partners (1993) note particular adverse impacts on Dudley and Stourbridge of the Merry Hill shopping centre. For these centres, they report a loss of the most important multiples, an increase in vacancy rates, a marked reduction in shopping flows and falling rents. Similarly, in their study on the Metro Centre, Howard and Davies (1993) identify strong negative impacts in the southern peripheral streets of Newcastle City Centre already damaged by the trading effects of Eldon Square (a city centre regeneration scheme).
A focus on in-town shopping centres started in the UK after 1999. Oughton et al. (2003) studied the economic impact of the Oracle Centre, as a regional in-town shopping centre, located in the heart of the town centre of Reading. Their study seeks to improve the general understanding of what happens to surrounding areas when a new regional centre enters a traditional town centre retail environment. Crosby et al. (2005) also studied the impact of the Oracle centre on retail activity in the town centre using land-use data and the results of a retailer survey. The Oracle shifted the retail focus of the town centre to the south and east. Major retailers in the town centre surveyed experienced a real decline in sales volume in the year after the Oracle opened. Moreover, the Oracle acted for change, shifting the prime pitch, weakening peripheral areas and increasing turnover rates (Crosby et al., 2005).
As noted in the Introduction, two main methodologies have dominated the UK literature and practice regarding RIA: manual models and gravity or spatial interaction models. Manual models can be subdivided into two approaches: the 'step-bystep' approach and the 'market penetration' approach. The main difference between the two approaches is over how to estimate the turnover of the new centre. While this is estimated by the floorspace or the likely retail mix of the planned regional shopping centre in the 'step-by-step' approach, the turnover in the 'market penetration' approach is estimated from the available expenditure of the population in the catchment area.
Before discussing the UK history in more detail, it is useful to note that England (2000) argues that there is no clear 'preferred' approach to assessing retail impact in North America or Europe and methodologies are often less developed in comparison to the UK. However, a recent summary of impact studies in the US suggests a very similar methodology to the step-by-step approach discussed below (Civic Economics, 2008). Early approaches in the UK (typically 1960s and 1970s) tended to utilise retail modelling techniques, principally SIMs (or gravity models as they are often called in planning circles), which seek to allocate expenditure from origin zones (often census tracts) to destination zones (shops or shopping centres) based on the trade-off between centre attractiveness and distance travelled. Gravity or spatial interaction models were popular in the public sector in the 1960s and 1970s. The Haydock model is a good illustration of the use of the spatial interaction models where the first large-scale model was produced in 1966 (Hodges and Bennet, 1973; Foot, 1981). The model was built to study the shopping patterns in north-west England, with the area divided into 244 residential zones and included the 47 largest shopping centres. There was a problem that no retail floorspace data for these centres was available, so an index of the number of different types of stores in each centre was constructed (Foot, 1981). It was predicted that Haydock could expect to attract about £47.5 million (at 1961 prices) for the year 1971, not as much as Manchester or Liverpool, but nevertheless it seemed to show that the area could support a regional shopping centre. However, the model showed that many centres within 8 or 10 miles of Haydock would be adversely affected. Wigan could expect nearly half of its expected sales for 1971 to be diverted to the new centre, and about one-third of the sales expected for Warrington and St Helens was predicted to be diverted to the new centre at Haydock (Foot, 1981).
As noted in the Introduction, the use of spatial interaction models declined in planning enquires during the 1970s. Guy (1991) notes that planners lost faith in quantitative methods and retail models in particular and indeed began to outsource impact assessments to private consultancies. Also, it could be argued that many planning jobs were lost in the 1980s and local authorities had little choice but to outsource.
It became apparent that disagreements relating to factors such as zone size, calibration techniques or travel cost measures could result in radically different results from SIMs. It was also a problem when different agents (i.e. planners, retailers, consultants, developers, etc.) evidenced different results based on very different forms of the same theoretical models. The models became problematic at public inquiries where it was difficult to examine and test model assumptions and the effects that these had on impact results. Finally, the UK Department of the Environment advised against the use of mathematical models in impact studies (Norris, 1990; England, 2000). This antipathy towards SIMs has not gone away. Aberdeenshire Council (2004, 10) comment:
Classic gravity models have so many drawbacks in methodology, data, forecasts and assumptions and have been so frequently criticised at inquiries that there can be no longer any confidence in the exclusive use of such models.
That said, during the 1980s and 1990s in Scotland the URPI 'Markets' model was used for evaluating impacts of developments in, for example, Strathclyde and Central regions. Similarly, applications of URPI models appeared in certain English local authorities during the 1970s and 1980s (i.e. Alty et al., 1979).
From the mid 1980s, there was a shift towards a more 'practical' approach to assessing and evaluating retail impact with the result that, in 1992 the retail and planning consultancy firm Drivers Jonas was able to conclude that 'there appears to be a broad measure of consensus that retail impact studies should include percentage impact calculations but should deal in qualitative as well as quantitative terms' and that 'there is remarkably little dispute as to the stages to be undertaken in calculating percentage impact' (Scottish Office, 1992). The Drivers Jonas report was specially commissioned by the Scottish Office to investigate the effectiveness of common approaches to retail impact assessment and to discover key elements which might form the best possible type of assessment.
There has been little alteration to the steps undertaken to compute retail impact since the early 1990s: the so-called 'step-by-step' approach. This approach is reported to have two key advantages over other generic techniques (SG, 2007):
* They allow clear statement of assumptions and calculations. This allows the challenging and testing of assumptions by all parties (planning authorities, applicants and third parties) that support the decision-making process.
* They do not need the development of large-scale regional or strategic retail models before impacts can be assessed. This reduces the cost and time required to assess the developments and also allows RIAs to be flexible to be able to respond to changes - both local factors (such as additional floorspace) and also to the quickly changing structure of retailing.
However, it is clear that many of these studies use techniques that also raise serious questions concerning the reliability of revenue estimates given the heroic assumptions often also made. For example, catchment area size for new shopping centres is often simply estimated based on analogy with centres elsewhere in the country. More worryingly, revenue forecasts for a new centre are commonly based on national floorspace/ turnover ratios for key multiples entering these centres. This works reasonably well for average-sized centres, but can seriously overestimate revenues in small centres and underestimate revenues in large, new centres. A good example is provided by ISG Development Southern Ltd (2008) who examined the impact of a new centre in Stowmarket in the UK (a small market town). Using national floorspace/turnover estimates for new store revenues provided more predicted revenue than available expenditure in the entire region. Despite this revelation, the study carried on regardless with its estimates (clearly very positive ones on the potential of the new centre!).
The rest of the article will continue to look more critically at the assumptions made in both methodologies.
The study area and data sources
Silverburn is an out-of-town shopping centre opened near Pollok, Scotland, on 25 October 2007 as a planned regeneration strategy for the 1970s neighbourhood shopping centre in Pollok. This centre was chosen as the case study since it is one of the most recent regional shopping centres to be opened in the UK. To understand and model changing consumer behaviour, we can draw on the shopping patterns in Scotland evident from the NSS, a data source rarely available to the academic community and itself only available since 2005. Pollok is a predominantly workingclass housing estate located on the periphery of the city of Glasgow. This area has experienced severe social and economic problems for many years and was regarded in the 1990s as an area in need of 'priority treatment' (see Sim, 1984; McWilliams, 2004). The new shopping centre at Silverburn incorporates over 1 million square feet of net floorspace, including many of the large UK multiple retailers such as Debenhams, Next and Marks & Spencer (and a large Tesco store for both food and non-food goods) and it includes 4500 free parking spaces. Richard Low, Retail Property Holdings Managing Director, comments:
Silverburn will be the first of the next generation retail developments in the UK as it provides the very best of the High Street shopping environment within an enclosed architecturally-inspiring space benefiting from direct motorway access and plentiful free parking. I am reassured that retailers and restaurant operators, who have signed up in such large numbers, recognise the quality of the environment we have created. As things stand, six weeks before centre opening, we only have three units still available for lease. (Shopping Centre Magazine, 2007)
Recent planning guidance in the UK has not favoured large out-of-town shopping malls as UK planning laws since 1996 aim to protect existing town centres. In the case of Silverburn, there was a great deal of local anxiety over its impacts, not so much in Pollok where it was part of the regeneration activity, but in surrounding centres, especially Barrhead and market towns such as Paisley and Kilmarnock (Silverburn Centre, 2008). An independent study by retail consultants CACI predicted that Silverburn would become the fourth most important centre in Scotland behind Glasgow, Edinburgh and Aberdeen (making it number two behind Glasgow City Centre in the study area of this article). Using a spatial interaction model approach, CACI expected that wealthy families would visit from suburbs such as Bearsden and Newton Mearns (wealthy suburbs to the south of Silverburn: see figure 1), and from more rural areas to the south of the centre (Shopping Centre Magazine, 2007). Their conclusion was that Silverburn would be able to achieve a 5.42 per cent market share in its catchment area, progressing to 7 per cent over time. Director Ken Gunn is quoted as saying,
The scheme will attract good quality but currently underserved shoppers from an extensive retail catchment, and this strong market potential leads CACI to believe that Silverburn will sit within the top five retail destinations in Scotland. (Shopping Centre Magazine, 2006, 18)
The study area for the analysis in this article comprises of the five large postal areas: Glasgow, Falkirk, Motherwell, Kilmarnock and Paisley. The entire area covers 17,327.5 km2 (22 per cent of the area of Scotland). The sub-region is broken down into 428 postal sectors; Glasgow contains 185, Falkirk 43, Motherwell 40, Kilmarnock 70 and Paisley 90. The Census of Population tells us that 2,479,515 people live in 1,120,853 households in this sub-region, the most populated area in Scotland (2001, census data). Figures 1a and 1b show the study area and the expenditure across the different postal sectors.
To evaluate the impact of Silverburn, it is necessary to update 2001 Census data to 2008. Table 1 shows how this can be achieved based on national estimates from the Scottish office. (Note: the total number of households in the 40-km catchment is only 90,000 households less than the entire study region, mainly excluding the outlying PA 'island' postal sectors.)
These postal area updates are then applied to the postal sectors that make up each postal area. The second data requirement is the household expenditure within the catchment area of Silverburn for non-food goods. Based on the regional entry 'Scotland' in the UK Family Spending Survey for 2008, this was estimated at £55 per household per week. This results in an annual expenditure of £2860 per household for 2008.
The third type of data required relate to the existing shopping centres in the region. According to 'Management Horizons Europe, UK Shopping Index, 2008' (MHE Index; MHE Retail Ltd, 2008), there are 88 existing shopping centres and retail parks located in the study area (8 in Falkirk, 42 in Glasgow, 12 in Kilmarnock, 13 in Motherwell and 13 in Paisley). Glasgow City Centre is the largest shopping centre in the UK and therefore dominates the hierarchy in the region. Table 2 shows the principal centres and the MHE index of attractiveness. This index takes into account both store sizes and store fascias (the more multiple retailers present then the higher the score).
We also make extensive use of customer data for the region. Data from the NSS has been provided by Acxiom Ltd for research use in the School of Geography at Leeds. This major questionnaire survey captures demographics and lifestyle behaviours, but also tracks detailed shopping habits of each respondent, including major retail destinations. Each year, the NSS provides data on around one million UK households. This gives us a good snapshot of shopping habits prior to the opening of Silverburn and the catchment areas of existing centres (crucial information for the calibration of the SIMs in the 'Modelling shopping flows' section below). A thorough review of the strengths and weaknesses of this source of data is provided by Thompson et al. (2010). We shall explore some of these in the remainder of the article.
Exploring the impact of Silverburn
The step-by-step RIA approach
The common methodology used by planners and consultants has been summarised as follows (Drivers Jonas, 1992).
* First, identify the catchment area of the proposed retail development, the area from which the centre draws the majority of its revenue. The information about that ideally should come from a household survey to show the existing shopping patterns in the area, although often this information is not available. The catchment area may be subdivided into drive time isochrones, and these isochrones may be further subdivided into primary, secondary and tertiary zones for greater spatial accuracy in assessing impact.
* Second, estimate the expenditure within the catchment area, derived from the population and its per capita expenditure.
* Third, estimate the turnover of existing shopping centres. This information can be estimated by using national average turnover/floorspace ratios for individual firms (or the centre as a whole), or by using survey data and proportioning centre turnover on the basis of the customer patronage seen in the survey (i.e. as a percentage of the total spend).
* Fourth, estimate the turnover of the new shopping proposal. This is normally estimated on the basis of company average turnover/floorspace ratios for retailers known to be entering the centre (or trading performance of similar centres where such data are not available).
* Fifth, estimate the amount of spending in each existing centre which will be diverted to make up the new centre's turnover, and the locational source of that spending.
Finally, express the amount of diverted trade from each shopping centre as a percentage of the estimated pre-impact turnover of that centre. This traditional RIA methodology is summarised in Figure 2.
In the rest of this section, we operationalise these steps for Silverburn.
Identify the catchment area of Silverburn Shopping Centre. The catchment areas of the major shopping centres in the study area were calculated based on the results of the Acxiom NSS for the years 2005, 2006 and 2007. The catchment area of each centre was demarcated as a buffer, which includes all postal sectors where the proportion of trips that go to that centre is above 15 per cent. The buffer zones of these centres were approximated to the nearest kilometre. The catchment area of Silverburn was estimated based on the catchment areas of existing centres in relation to their MHE score. The MHE Index for Silverburn was calculated based on the attractiveness score of Braehead (because of the similarity of size and type of centre and the fact that a value for Silverburn is not yet available from Management Horizons). Using average turnovers for the different types of stores in each centre, Silverburn's index was estimated to be 198. This is 30 per cent higher than that of Braehead due to the presence in Silverburn of more international retailers and more magnet stores such as Debenhams and Tesco (and the largest Next store in Scotland). Based on this index score, the catchment area of Silverburn is estimated to be a 40 km radius from its centre. This makes Silverburn the second largest and most important centre in the region. Two more scenarios of the catchment area of Silverburn were also used (estimating the catchment area at 30 min and 40 min drive of the centre) to see how the results of the RIA differ with different estimations of the catchment area of the new centre. Catchment areas of out-of-town centres have been typically defined as 30-min drive times in previous studies (Bennison and Davies, 1980; Roger Tym and Partners, 1986; 1996; Howard and Davies, 1993; Donaldsons, 2005).
Estimate the expenditure within the catchment area of Silverburn. The expenditure within the catchment area was estimated based on the number of households and per capita expenditure. This expenditure calculation is described in the 'data sources' section above. The estimated expenditure within the catchment area of Silverburn (40 km) for the year 2008 equals approximately £3067 million. The expenditure for the whole study area was estimated at £3637 million.
Identify the competing centres of Silverburn. The competing centres include, logically, all of the centres which are located within the buffer zone of 40 km from Silverburn and those which are located just outside this zone to compensate for edge effects. To identify these centres, the straight-line distance between Silverburn and all shopping centres in the study area was calculated. The shopping flows, taken from the NSS for the years 2005-2007, were also utilised to identify the competing centres. From these surveys, it was possible to identify 20 centres as significant shopping destinations in the area (making up 85.57 per cent of the shopping flows within the estimated catchment area of Silverburn). Fifty-three other centres are present, but only account for 14.43 per cent of the shopping flows in Silverburn's catchment area. The study has included these 20 centres for comparison as they each have more than 1 per cent of the flows within the catchment area of Silverburn. It can be seen that the number of centres to include, and their proportions of the flows within the catchment area of Silverburn, are very sensitive to the size of the catchment area.
Estimate the turnover of the existing/competing centres. The percentage of flows which go to each centre in the whole study area was calculated based on the results of NSS for the years 2005, 2006 and 2007, data that were used to estimate revenues. For example, from the NSS data, it can be seen that 25.99 per cent of the non-food shopping flows go to Glasgow City Centre. This percentage was multiplied by the available expenditure within the study area (£3637 million) to estimate the turnover of Glasgow at £945.2 million. Table 3 shows the percentage of shopping flows which go to the top 20 existing centres in the study area and their estimated turnovers.
Estimate the turnover of Silverburn Shopping Centre. The estimated turnover for Silverburn was calculated as follows.
1. First, it is necessary to identify all the non-food retailers in Silverburn. Sixtysix retailers were defined as non-food shops in the centre (Silverburn - Centre map 2009). We make the simplifying assumption that food outlets (except Tesco) make no contribution to non-food retail floorspace, although they do influence the overall attractiveness of the centre (captured in the existing MHE index). We estimated the turnover of Tesco at £4 million from the non-food goods.
2. Next, we estimate the turnover of each retailer in three categories, using a combination of corporate intelligence (e.g. company reports) and internet research. For major stores (Debenhams and Boots) with known floorspace, turnover was estimated by using a national turnover per square foot for that retailer. Second, the average store turnover of the retailer across the country was calculated for other retailers when the information about the size of the shop and the turnover per square foot was not directly available. The annual turnover for each retailer, for the year 2008, was divided by the number of the stores that retailer has, and this gives the turnover of that retailer per store. Finally, when the information about total revenue and number of stores was not available, the turnover was estimated at £0.5 million for national chains and at £0.25 million for local chains or specialist stores.
3. The summed turnovers of all non-food retailers in Silverburn yield the estimated turnover for Silverburn Shopping Centre.
The estimated turnover for Silverburn for the year 2008 (design year) is approximately £139.8 million. However, we argue that the use of national averages will underestimate the turnovers in a large, regional centre as we expect major retail brands to perform at levels considerably above the norm. If the top retailers obtain 50 per cent more than national averages, then the turnover of Silverburn increases by £19.709 million to be estimated at £159.5 million. If those top retailers perform 100 per cent above national averages, then the turnover of Silverburn is estimated to be £179.2 million. Clearly Norris's (1990) 'optimistic guesswork' haunts this methodology too.
Estimate the diverted trade from each centre. This step requires estimating the amount of spending in each of the existing centres which will be deflected to make up Silverburn's turnover. To estimate this, we have calculated the proportion of trade captured by competing centres in the catchment area of Silverburn, and used these ratios to estimate the deflections. For example, the percentage of flows which go to Glasgow City Centre within the 40-km catchment area of Silverburn is 30.09 per cent. This percentage figure was thus multiplied by the estimated turnover of Silverburn (£139.8 million). This means Glasgow is predicted to lose £42.07 million from its revenue following the opening of Silverburn.
The impact of Silverburn on existing centres. This last exercise simply expresses the amount of diverted trade from each shopping centre as a percentage of the pre-impact turnover of that centre. For example, the diverted trade from Glasgow City Centre is estimated above at £42.1 million. This gives a percentage figure of 4.45 per cent. In other words, Glasgow City Centre is predicted to lose 4.45 per cent of its revenue to Silverburn. According to the results of this exercise, the Motherwell and Coatbridge Shopping Centres are estimated to be the most affected centres with a loss of 4.66 per cent of their revenue. Fourteen other centres are estimated to lose between 4 and 4.64 per cent of their revenues.
It is likely that, in reality, the impact of Silverburn on nearby smaller centres is larger than the impact on some of these large centres. Unfortunately, small centres are not always included in the NSS and some of these are actually close to Silverburn, especially Barrhead and Pollok Shopping Centres. To recap, no likely trade diversion to these small centres can be estimated with this technique because these centres were not referred to in significant levels by the respondents in the NSS. This is another major problem with RIA based on surveys: smaller centres are often underrepresented as the questions focus on primary destinations only (in comparison to SIMs, which can estimate deflections across all sizes of centres).
Modelling shopping flows in the study area using spatial interaction models
The SIM to be used in this article takes the following form:
Smij = expenditure by household type m in residence zone i at destination j
Omi = level of consumer expenditure of household type m in residence zone i
Ami = a balancing factor to ensure that
which is calculated as:
Wj = the attractiveness of destination j,
αm = a parameter reflecting the perception of a destination's attractiveness by household type m,
dij = the distance between origin i and destination j, and
βm = the distance decay parameter for household type m.
Birkin et al. (2002, 152) observe that these models allocate flows of expenditure between origin and destination zones on the basis of two main hypotheses:
* flows between an origin and destination will be proportional to the relative attractiveness of that destination vis-à-vis all other competing destinations, and
* flows between an origin and destination will be proportional to the relative accessibility of that destination vis-à-vis all other competing destinations.
The estimated expenditure in each postal sector, for the year 2008, is used to form the demand side in the model (Oi). The ranking scores for all shopping centres in the study area, taken from the 'Management Horizons Europe, UK Shopping Index, 2008' are used to form the attractiveness of each centre in the model (Wj). The distance decay term is based on travel times between postal sector centroids and each individual shopping centre (dij) (Mercator GeoSystems Ltd., 2010).
In order to generate the most appropriate spatial interaction model for this study, a progressive approach to model disaggregation was adopted (models not disaggregated by person type m). The first basic aggregate model contained a single beta value for the entire region. For a second model we disaggregated beta by postal area (five values). In our third model, we disaggregated beta by postal sector, producing 428 unique beta values with a distinct pattern of variation against population density, so that residents of the most remote rural areas show the greatest propensity to travel, through either inclination or necessity. The SIM was also then disaggregated by different sociodemographic variables for more robust results; age, ethnicity, income and car ownership (adding the disaggregation by person type m). The model first was disaggregated by age producing five groups; < 25 years old, 25-29 years old, 30-44 years old, 45-64 years old and finally those 65 and above. The next model disaggregated households by their ethnic background: white and non-white. The third model disaggregated households by their annual income. Households were divided into three groups; low income (less than £20,000 per year), medium income (£20,000-£39,999 per year), and high income (£40,000 or above per year). The final model disaggregated households by car ownership. Two groups were recognised based on car ownership: car owners and non-car owners. These separate models formed 60 sub-models which were finally combined in one final best-fitting model. In all cases, the SIMs were calibrated against the observed data for shopping flows in the study area taken from NSS data for the years 2005-2007. A variety of calibration techniques were used to fit the beta values (see Knudsen and Fotheringham, 1986, for a review of techniques). The best-fitting demographic model was disaggregated by car ownership (as well as the 428 zonespecific beta values).
For this car ownership model (for example), households were classified into two groups: car owners and non-car owners as noted above, and two separate models were run for these two groups. The results of the observed shopping flows taken from the NSS data proved (as one would expect) that car owners travel longer distances compared to non-car owners who seem to shop more locally; thus, a lower beta value was used for car-owners as opposed to non-car owners (see Figures 3a and 3b). The key point to note from these figures is the different spatial extent of the catchment areas in the two maps.
To assess the performance of a spatial interaction model, it is important to validate the ability of the model to replicate known data sets. R2 is one of the most commonly used goodness-of-fit statistics which was used in this study to show the model performance. The R2 value was 0.72 for the basic aggregate model and it went up to 0.82 for our final combined model after disaggregating by different socio-demographic variables. However, the R2 value for the predicted flows to just the 21 shopping centres included in the NSS was above 0.90, for the final disaggregated model.
Once the SIM has been calibrated and deployed, the 'catchment area' of a retail centre can be illustrated from the proportion of expenditure which is captured in each residential zone (the 'market share'). Figure 3 shows the SIM predictions of the catchment area of Silverburn. In relation to an established centre, the SIM is capable of providing a good representation of centre catchments against the observed survey data - see Figures 4a and 4b for an example.
Silverburn was predicted to have revenue of £126.5 million using the best performing aggregate model. The impact of Silverburn on the existing centres within the catchment area was very similar based on the results of both models. Three centres were predicted to lose more than 10 per cent of their trade following the opening of Silverburn; Barrhead, Glasgow-Pollok and Glasgow-Darnley Retail Park. Twenty other centres were predicted to lose 5-10 per cent of their trade.
A comparison between the results of the traditional RIA methods and SIM
This section aims to compare the results of the impact of Silverburn on the existing centres in the study area based on the two methods employed. This analysis will follow the same order of steps as given in the traditional RIA methodology in comparing the results:
The catchment area of Silverburn
This was estimated to be a 40-km radius based on the traditional RIA methodology. The SIM results show that Silverburn actually draws 99.1 per cent from its revenue from this area as predicted by the basic aggregate model and 96.28 per cent based on the final combined model. The results of the SIM also show that Silverburn draws about 68.37 per cent of its revenue from a zone of up to 10 km from the centre site and about 81.5 per cent from a zone of up to 15 km. This reflects one of the strengths of the SIM over the traditional RIA methodology: the SIM is not constrained by arbitrary catchment area boundaries as in the traditional RIA methodology. Furthermore, the SIM can be disaggregated by different socio-demographic variables as shown above. This gives a better representation of the catchment area of that centre based on known consumers' characteristics. The results also showed that RIA final results of the impact of the new centre are very sensitive to how wide the catchment area is, since the number of competitors and their proportion with the catchment area differs with different estimations of the catchment area. This was clear when comparing the results of the study with the other scenarios of the catchment area when it was estimated at 30 and 40 min drive of the centre. The number of competitors decreased and their proportion increased when using a buffer of 30 min drive of the centre as a catchment area of Silverburn compared to the results based on this study when the catchment area of Silverburn was estimated at 40 km of Silverburn.
The expenditure within the catchment area of Silverburn and within the whole study area was estimated using the same methodology. The estimated number of households (in each postal sector for the year 2008) was multiplied by the expenditure by households (£55 per household per week) and then multiplied by the growth ratio between the base and design year. The available expenditure within the 40 km catchment area of Silverburn was estimated at £3066 million and at £3637 million within the whole study area.
The designation of competing centres
The traditional RIA methods, based on the NSS, have estimated the likely impact of Silverburn mainly on the 20 largest centres within the catchment area of Silverburn, as flows to smaller centres were under-represented in the answers given by customers in those surveys to the question of 'major shopping non-food destination'. The SIM, in comparison, is able to give all 88 shopping centres an index score and therefore include them fully in the spatial analysis.
Turnovers of the existing centres
The estimated turnovers for the existing centres were very different based on the results of the two methods (see Figure 5). Turnovers of the large centres were very high based on the traditional RIA methods compared to the predicted turnovers based on the SIM results. For instance, the turnovers of Glasgow, Ayr, Falkirk, Braehead, Irvine and Hamilton shopping centres were estimated by the traditional RIA method to be nearly double the estimated turnover by the SIM, and substantially higher for other centres including Stirling, Irvine, Greenock and Motherwell shopping centres. On the other hand, the turnovers of the small centres - Wishaw and The Forge shopping centres - were predicted by the SIM to be higher than the estimated turnovers estimated by the traditional RIA methods. Many apparently minor centres are ignored completely by the RIA - another 68 shopping centres identified by Management Horizons are excluded by the RIA, but not by the SIM.
There are essentially two issues to address here. One is that the tendency for RIA to concentrate on a limited number of competing centres has a distorting effect, as many of these flows are reallocated to the major centres, making them appear too large. A similar process is at work in the NSS data collection process. It seems that local centres are often conflated, either by the respondents themselves or within the data management process, so that for example Renfrew Blythswood might become simply Blythswood. Again this tends to inflate the larger centres and erode the influence of the smaller ones. Hence, the results of the RIA are thus only as good as the survey results which underpin the assessment. One recommendation can be made here in order to improve the results of the RIA: to find a way to capture smaller centres in the results of the surveys more effectively. Another is to consider the full range of competing centres, and not just the major centres in direct proximity to a new development.
Figure 5 shows the estimated turnovers of the existing centres in the study area based on the results of the two methods.
Turnover of Silverburn
The estimation of the turnover for most of the retailers in Silverburn depends on the average turnover for each retailer per store in the traditional RIA method, while the turnover of Silverburn based on the SIM is predicted based on an estimated attractiveness given the size of each store. The main potential weakness of this stage of the RIA is that retailers may be trading better or worse in Silverburn than the average size of that retailer across the country would suggest. Furthermore, the information about the total floorspace and total revenues for small retailers is not available, so the study has estimated the turnover for national chains at £0.5 million and at £0.25 million for local chains or specialist stores which is problematic. The turnover of Silverburn in the two other scenarios of the RIA was estimated at £159.5 million and £179.2 million, respectively, as a result of increasing the turnovers of the big retailers by 50 per cent and by 100 per cent as discussed in step number 5 in the RIA methodology above. The impact of Silverburn on the existing centres based on these additional scenarios has increased by the same percentage of the increase in the turnover of Silverburn in these scenarios.
In a similar way, however, it is possible to be critical of the spatial interaction modelling approach in relation to variable selection. It is sometimes said that such modelling is as much an art as a science with regard to the choice of variable to use. For example, the choice of attractiveness term is crucial, and can be shown to produce very different results (Clarke and Clarke, 2001). The estimation of turnover within the SIM is heavily dependent on the Wj index score. The SIM results show that the turnover of Silverburn is estimated to be £126.5 million for the aggregate model and an attractiveness score based on the number of stores in Silverburn and their size. The estimate is £188.2 million (5.17 per cent of the available expenditure in the study area) based on the final disaggregated model and with the attractiveness scores for Silverburn adjusted based on the results of the shopping mix in comparison to that of Braehead. In the future it is hoped that new data will be available from Management Horizons to give a value for Silverburn which uses the exact same methodology and data as for all other centres in the region.
The impact of Silverburn
The impact of Silverburn shows the diverted trade from each centre to Silverburn as a percentage of the pre-impact turnover of that centre. The results show that RIA methodology has estimated a very similar impact on all existing centres. The impact on the largest 16 centres was between 4 and 4.66 per cent. Three centres were predicted to lose more based on the SIM (Paisley, Braehead and Glasgow City Centre). The impact of Silverburn on the existing centres was thus very different based on the two methods. In the SIM, Barrhead, Glasgow-Pollok and Glasgow-Darnley Retail Park were predicted to be most affected by the opening of Silverburn (11.8, 10.4 and 10.3 per cent, respectively). Fifteen other centres were estimated to lose more than 5 per cent of their revenues based on the results of the SIM (recall that these centres were not covered in the analysis based on the traditional RIA methodology as they were not in the survey data). The impact of Silverburn on distant centres, based on the SIM results, was very low compared to the impact estimated by RIA method.
Thus, the RIA methodology predicts a rather 'flat' impact on existing centres while the predicted impact based on the SIM is high for near, small centres and low for distant, large centres. Figure 6 compares the impact of Silverburn on the existing centres based on the traditional RIA method with the impact based on the SIM.
This article has compared two methodologies for looking at the impact of Silverburn on existing centres in the central belt of Scotland: the traditional RIA step-by-step methodology and a disaggregated spatial interaction model (SIM). Despite the growth in popularity of RIA, our results show that each stage of the traditional RIA approach is problematic. With respect to the turnover of competing centres, then dangerous bias can be introduced through the consideration of a restricted set of centres, and the means by which relative performance is evaluated may not be straightforward - in this case, we found that self-reported consumer data are not necessarily fully reliable. Second, the estimation of turnover for the new centre is not convincing, and the idea that this can be captured based on average floorspace/turnover ratios for individual companies is surely flawed. However, perhaps most damning of all, given that the objective is to assess impacts, is the tendency to provide undifferentiated deflections from competing centres to the new destination.
In comparison, one of the strengths of the SIM over the traditional RIA methodology is that the SIM can include all centres in the region more effectively. By including all centres directly in the model, the impact on local centres can be seen to be very high compared to RIA methods. Given evidence from around the UK of what actually happens when regional centres are opened, we would argue that this mirrors reality far more accurately. If the Wj factor is deemed to be realistic then the SIM is more robust for both revenue calculations and catchment area delimitation. However, the SIMs are very sensitive to that Wj value. Again dramatic differences in turnover can be seen when alternative Wj values are used. Estimating the exact value of Wj for a new centre needs careful data comparison to all existing centres in the region.
We recognise that these conclusions are to some degree argumentative in the absence of hard evidence about the actual impact of Silverburn, which first opened in 2007. The next stage of the research agenda is to revisit the two methodologies when new NSS data is provided by Acxiom (which will show the impacts of Silverburn within the shopper surveys), and when new data for shopping centre attractiveness is provided by MHE. These updated data sets will provide new evidence for model calibration and model development. In the meantime, work will continue with multiple disaggregations of the SIM based on the socio-demographic characteristics of the consumers within the study area, together with further estimations of the sensitivity and robustness of the key model outputs to uncertainties in the underlying data.
Despite outlining various criticisms of both approaches above, we believe the SIM approach provides a more coherent framework for retail network planning than does the RIA approach. While RIA is typically geared to local analysis of impacts, the spatial interaction models are much better at representing the mutual interactions between competing centres across a whole range of spatial scales. In particular, the interaction models are better able to handle the impacts on smaller centres which are often missed out on surveys of retail consumer behaviour which often ascertain only principle retail destinations for individual shoppers. Thus, the RIA analysis can only be as good as the detail provided in the surveys. Retailers therefore stand a much greater chance of evaluating the effect of new store developments on individual outlets using the SIM approach.
aberdeenshire council (December 2004), 'Assessing the Impact of Retail Developments in Aberdeenshire: A guide for Applicants, planning and services', http://www.aberdeenshire. gov.uk/planning/supplementary/documents/retail_impact.pdf (accessed 8 May 2008).
alty, r., mackie, s., moseley, j. and rainford, p. (1979), The South Yorkshire shopping model, Unit of Retail Planning and Information, URPI U12, Reading.
bdp planning and oxford institute of retail management (1992), The Effects of Major Out of Town Retail Development: Literature Review for the Department of Environment, London, HMSO.
bennison, d. j. and davies, r. l. (1980), 'The impact of town centre shopping schemes in Britain: Their impact on traditional retail environments', Progress in Planning, 14, 1-104.
birkin, m., clarke, g., clarke, m. and wilson, a. g. (1996), Intelligent GIS: Location Decisions and Strategic Planning, Cambridge, Geoinformation International.
birkin, m., clarke, g. and clarke, m. (2002), Retail Geography and Intelligent Network Planning, Chichester, Wiley.
birkin, m., clarke, g. and clarke, m. (2010), 'Refining and operationalising entropy maximizing models for business applications', Geographical Analysis, 42, 422-45.
breheny, m. (1988), 'Practical methods of retail location analysis: a review', in N. Wrigley (ed.) Store Choice, Store Location, and Market Analysis, London, Routledge.
civic economics (2008), 'A guide to retail impact studies', www.CivicEconomics.com.
clarke, g. p. and clarke, m. (2001), 'Applied spatial interaction modelling', in G. P.
clarke and m. madden (eds) Regional Science in Business, Berlin, Springer, 137-58.
crosby, n., hughes, c., lizieri, c. and oughton, m. (2005), 'A message from the Oracle: the land use impact of a major in-town shopping centre on local retailing', Journal of Property Research, 22, 245-65.
donaldsons (2005), 'Retail impact assessment in support of outline planning application to Wolverhampton City Council for development - retail core retail expansion, Wolverhampton', http://www.wolverhampton.gov.uk/NR/rdonlyres/006D5582-1152-4A8EB147- 8D4648E62675/0/RIAApril05.pdf (accessed 3 November 2008).
drivers jonas (1992), Retail Impact Assessment Methodologies: Research Study for the Scottish Office, Edinburgh, The Scottish Office.
england, j. (2000), Retail Impact Assessment: A Guide to Best Practice, London, Routledge.
estimates of households and dwellings in scotland (2008),' Table 1: Household estimates for Scotland by local authority area, June 1991-2008', http://www.gro-scotland.gov.uk/files2/ stats/household-estimates/he-08/he-08-all-tables.pdf (accessed 23 November 2009).
foot, d. (1981), Operational Urban Models: An Introduction, London and New York, Methuen.
general register office for scotland (2001), '2001 Census: Information about Scotland's people', http://www.gro-scotland.gov.uk/statistics/theme/population/estimates/index. html (accessed 20 July 2009).
guy, c. (1991), 'Spatial interaction modeling in retail planning practice: the need for robust statistical methods', Environment and Planning B: Planning and Design, 18, 191-203.
guy, c. (2007), Planning for Retail Development: A Critical View of the British Experience, London and New York, Routledge.
hodges, s. and bennet, d. (1973), 'Practical problems of gravity modelling', http://linkinghub. elsevier.com/retrieve/pii/S002463017380007X (accessed 20 December 2010).
howard, e. b. and davies, r. l. (1993), 'The impact of regional, out-of-town retail centres: the case of the Metro Centre', Progress in Planning, 40, 89-165.
knudsen, d. and fotheringham, s. (1986), 'Matrix comparison, goodness-of fit, and spatial interaction modelling', International Regional Science Reviews, 10, 127-47.
mcwilliams, c. (2004),' Including the community in local regeneration? The case of Greater Pollok Social Inclusion Partnership', Local Economy, 19, 264-75.
mercator geosystems ltd. (2010), 'Drive time and mapping: drive time matrix of Scotland'. mhe retail limited (2008), 'Management Horizons Europe Retail: UK Shopping Index'.
norris, s. (1990), 'The return of impact assessment: assessing the impact of regional shopping centre proposals in the United Kingdom', The Regional Science Association, 69, 101-19.
odpm (office of the deputy prime minister) (2005), 'Planning Policy Statement 6: Planning for Town Centres', http://www.communities.gov.uk/documents/planningandbuilding/ pdf/147399.pdf (accessed 2 November 2008).
ons (office for national statistics) (2008), 'Family spending and family expenditure surveys 1997-2007', http://www.statistics.gov.uk/downloads/theme_social/family_ spending_2007/familyspending2008_web.pdf (accessed 28 July 2009).
oughton, m., crosby, n., lizieri, c. and palmer, s. (2003), 'The local economic impact of regional in-town shopping centres: a case study of the Oracle Centre, Reading', Department of Real Estate & Planning, The University of Reading Business School.
reynolds, j. and schiller, r. (1992), 'A new classification of shopping centres in Great Britain using multiple branch numbers', Journal of Property Research, 9, 122-60.
roger tym and partners (1986), Greater Manchester Shopping Study, London, HMSO.
roger tym and partners (1993), Merry Hill Impact Study, London, HMSO.
roger tym and partners (1996), The Gyle Impact Study (Retail Impact), Edinburgh, HMSO.
rowley, g. (1992), Urban Spatial Growth and Metropolitan Areas: A Consideration of the Dynamics, Components and Developing Problems of British Cities. Department of Town and Regional Planning Occasional papers 110, University of Sheffield.
schiller, r. (1986), 'Retail decentralisation - the coming of the third wave', the planner, 72, July, 13-15.
schiller, r. (1996), 'Town-centre winners and losers', Estates Gazette, 13 July, 134-35.
scottish office (1992), 'Retail impact assessment methodologies: research study for the Scottish Office Environment Department (Drivers Jonas)'.
sg (scottish government) (2007), 'Town centre and retailing methodologies: final report', http://www.scotland.gov.uk/Publications/2007/12/24105030/7 (accessed 1 October 2008).
shopping centre magazine (2006), 'Glasgow's glut', Shopping Centre Magazine, 13 July, 8.
shopping centre magazine (2007), 'Silverburn approaches 100 per cent letting target', http:// www.shopping- http://www.shopping-centre.co.uk/news/fullstory.php/aid/2035/Silverburn_ approaches_100_per_cent_letting_target.html (accessed 15 February 2009).
silverburn centre (2008), 'Silverburn Centre', http://www.worldlingo.com/ma/enwiki/en/ Silverburn_Centre (accessed 1 October 2008).
silverburn - centre map (2009), 'Silverburn Online', http://www.silverburn.com/website/ CentreMap.aspx?ParentSectionId=386adbb6-7ac7-415a-b78e-318d9ba43589&Current SectionId=9bb6df92-b49d-48e9-9957-4d5803d05632&ContentInstanceID (accessed 10 August 2009).
sim, d. (1984), 'Urban deprivation: not just the inner city', Area, 16, 299-306.
thompson, c., stillwell, j. and clarke, m. (2010), 'Understanding and validating Acxiom's research opinion poll data' (working paper 10/6), Leeds, University of Leeds.
wilson, a. g. (1967), 'A statistical theory of spatial distribution models', Transportation Research, 1, 253-69.
wilson, a. g. (1974), Urban and Regional Models in Geography and Planning, Chichester, Wiley.
Hamzah Khawaldah is a research student, Mark Birkin is a Professor of Business Geography and Graham Clarke is a Professor of Spatial Analysis and Policy at the School of Geography, University of Leeds, Leeds LS2 9JT; email: email@example.com; firstname.lastname@example.org; email@example.com
Paper submitted September 2010; revised paper received and accepted June 2011.…