Table 6. OLS regression models for number of retailers by type within 500 metre radius of CBS cells regressed on built and socio-economic environment characteristics Variables Total Chained Discount Non-
chained Specialty Healthy Foreign House
value/m2 0.286*** 0.120*** 0.012*** 0.178*** 0.235*** 0.105*** 0.096***
Density
(pp/100m2) 0.314*** 0.035*** 0.022*** 0.198*** 0.237*** 0.000 0.092***
Log(Income) -0.079** -0.005 -0.001 -0.104*** -0.040* 0.001 -0.0172 Log(no.
working people) 0.101*** 0.023*** 0.029*** 0.025*** 0.064*** 0.021*** 0.026***
Crime rate 0.006*** 0.001*** 0.001*** 0.006*** 0.002*** 0.002*** 0.003***
% tourism
businesses 0.006*** 0.002*** 0.001*** -0,004 0.008*** 0.002*** 0,001 No. stations
and stops 0.140*** 0.007*** 0.003*** 0.067*** 0.103*** 0.009*** 0.002 No. university
campuses -0.010* 0.007*** -0.010*** 0.015*** -0.006 0.000 0.004 Constant -1.379*** -0.577*** -0.306*** -0.232* -0.991*** -0.644*** -0.489***
N 5764 5764 5764 5764 5764 5764 5764
Notes: * p<.05; ** p<.01; *** p<.001; Variables in bold are dependent variables with log-transformation Table 7. Logistic regression models for one or more food retailers of each category within 500 m radius of CBS cells regressed on built and socio-economic environment characteristics
Variables Chained Discount Non-
chained Specialty Healthy Foreign House value/m2 0.662*** 0.145** 0.391*** 0.460*** 1.706*** 0.351***
Density (pp/100m2) 0.545*** 0.454*** 0.556*** 0.650*** 0.288*** 0.322***
Log(Income) 0.582*** -1.093*** -1.685*** 0.039 0.523** 0.542***
Log(no. working
people) 0.170*** 0.270*** 0.150*** 0.250*** 0.653*** 0.275***
Crime rate 0.006** 0.007*** 0.005*** 0.020*** -0.007** 0.014***
% tourism businesses 0.016*** 0.018*** -0.04 0.017*** 0.039*** 0.011**
No. stations and stops 0.304*** 0.173*** 0.234*** 0.396*** 0.096*** 0.141**
No. university
campuses 0.173*** -0.260*** -0.30 -0.184*** 0.019 - 0.30*
Constant -8.158*** -1.595* 2.114* -6.877*** -13.182*** -8.034***
N 5764 5764 5764 5764 5764 5764
Pseudo R square 0.305 0.145 0.210 0.308 0.399 0.133
Notes: * p<.05; ** p<.01; *** p<.001
4. Results
Initial observations provide the settings for the next section, in which regression results are used to understand the association between the distribution of these stores and chosen independent variables. Two regressions are used to explore two concepts: presence and quantity of food retailers. Logistic regression is used for investigating the former, due to its use of dummy variables. OLS regression is used for the latter, due to the fact that there are multiple dependent variables, and independent variables are continuous.
It is interesting that while the two concepts of presence and quantity are used in different studies, no article has explicitly distinguished the two concepts and how they might be related to each other. Certain overlapping can be seen here; yet, it appears that these two are complementary. Presence of food retailers, or as termed by Smoyer-Tomic et al. (2008) and a few others as exposure to food retailers, assumed that the presence of of at least one retailers is representative for, e.g., local food environment (Morland & Evenson 2009).
On the other hand, the quantity of food retailers, or as referred to by some authors as availability (Apparicio et al. 2007; Zenk, Schulz, Hollis-Neely, et al. 2005), pays more attention to the number of food retailers in a defined geographical limit instead. Despite some differences between the specific research goals, these two concepts seem to be used for the common purpose of measuring the community nutrition environment. Either one of these two, depending on the specific research design, should be efficient for exploring the relationship between food retailers and urban environment. Still, both are used in this thesis to testify how the two are related to each other. This can also be seen as a contribution of the thesis to the existing literature.
Main regression results are presented in Table 6 and Table 7 (income models), with additional observations being drawn from Table 8 and 9 (social housing models), and Table 10 and 11 (non-Western models). The OLS regression provides both general and type-specific patterns. Logistic regression concerns presence of at least 1 retailer; thus, it is less explorative than OLS regression, concerning the exact number of retailers. Hence the logistic is considered here as supplementary to the OLS, helping illustrate the relationship between the two concepts. Since it is not necessary to consider the presence of at least one retailers of any type, the total variable is omitted for the logistic model.
It is necessary to point out how the results can be interpreted. OLS regression is pretty straightforward, using coefficient to illustrate the relationship. For example, the coefficient
increase in disposable household income would result in a decrease of 8% the number of all food retailers within the buffering distance.
Logistic regression, in its simplest form, helps to predict which of the two categories, in this case the presence and non-presence of food retailers, is likely to belong to a given certain information. Hence, the relationship is measured by log of odds ratio. For example, in Table 7, the log odds are 0.662, which means for every 1000-euro increase in the house value per square metre, the expected change in log odds is 0.662, which means in turn means higher chance that there is at least one food retailers within the limit of the buffered cell.
4.2.1. General patterns
Taking all types into account, controlling for tourism rate, a number of factors show extreme statistical significance. The same significance applies to population density and employment. Areas with higher house value per square metre, higher proportion of social housing, higher population density, and higher amount of working people have more food retailers available within 500 metre radius. Regarding the built environment, house value per square metre show high sensitivity to number of food retailers. Every 1000-euro increase in house price can be associated with that of 29% percent number of food retailers of all types.
This fits with the observation that most retailers’ distribution follows the centric pattern, in which the centroid lies at the historical canal districts, where understandably, property value is notoriously less affordable. Such a distribution reflects the retailers’ intention to strategically opt for locations with better profit prospect (Margheim 2007).
Social housing is slightly positively associated with the total number of shop in each buffered cells. While this appears to be divergent from the centric pattern, the reason might be the concentration of a large number of non-chained grocers and foreign food shops at certain areas, such as Saphartiparkbuurt, that, despite higher proportion of social housing, is far from being of lower socio-economic status like those in South-East or North.
When it comes to socio-economic variables, the positive associations with population density, and number of working people do not come as a surprise. 1-unit increase in number of people per 100 square metre can lead to 31% more availability of food retailers of all types.
It is common knowledge that retailing strategy would highly appreciate areas with higher density, which promise larger customer base, as well as exposure to potential ones. 1% change in number of working people equals almost 10% increased quantity. This shows that areas
4. Results
with higher amount of working people, who are not necessarily local residents, have higher food stores availability. This, again, seems to fall in line with the centric pattern.
The relationship with income is a negative one, though it remains hard to interpret the reason behind this. An increase of 1% income would result in a decline of 8% in number of food retailers. A further look into each category of food retailers might be helpful in this case, since it is likely that each category has its own dynamic when it comes to income. Higher crime rate is positively associated with exposure to food retailers. This may, once again, be explained by the centric pattern. Since areas close to centre are prone to higher level of insecurity, concentration of food retailers at this area might contribute to such a positive relation with crime rate. However, this explanation does not necessary apply to all areas, since certain areas in South-East or New West are known for having higher proportion of non- Western residents with higher crime rate.
Also of interest is the inverse relationship between the total number of retailers and proportion of non-Western residents. This seems to contradict the inverse relationship with income and direct relationship with social housing, yet in accordance with the centric pattern.
In addition to results on food retailers’ quantity, logistic regression provides insights into different exposure patterns of different types. Overall, most variables are proved to have statistically significant relationship with exposure to food retailers. Increase in house value, population density, and number of working people all results in increase of exposure of each type. There are a number of deviances when it comes to income and crime rate that remains difficult to explain, which suggest the need to look further into each category. Certain differences also arise between the two models, which would provide either complementary or contradicting interpretation. The next section turns to explore pattern of each category.
4.2.2. Type-specific patterns Chained supermarkets
Quantity of chained supermarket are raised by increase in every factor included, exception for income. However, the implications from these associations might not be that straightforward. House value and population density increases the amount of this type, which shows that these are spatially linked to areas with higher property value as well as higher number of residents. This fits with the centric pattern in general, and also proves that chained
Interestingly, income has no significant relationship, while both social housing and non- Western proportion have inverse ones. This seems to confirm the fact that chained supermarkets tend to favour developed areas. In addition, employment and crime rate’s positive association with chained supermarket also reflect the notion that these stores are centrically concentrated. The former shows that chained supermarkets remains locationally close to not only areas with high inhabitant rate but also business and commercial rate. The latter, while differentiating from results from others studies, also fits with the centric concentration.
In comparison, the presence of chained supermarket also increases with rises in house value, population density, employment and crime rate. However, while no significance is found between incomes and quantity, increase in income would increase the exposure of chained supermarkets. In contrast, non-Western and social housing have no significance in the models. These discrepancies, to a certain extent, prove that certain differences remain between these two concepts. One might argue that chained-supermarkets exposure is more dependent on income level, since it requires certain level of social-economic status to afford shopping here, or in another word, for the chained supermarkets to decide to locate in that area; but once the customers can afford such a threshold, then income’s influence is not significant anymore. Then, other factors would take their role in determining the location of these stores, as have reported above. This also seems to fit with the notion that chained- supermarkets in Amsterdam has a selective base of customers, mostly for middle- and higher class groups, but still remains not too high-end.
Discount supermarkets
Discount supermarkets have been expanding their territory in Amsterdam, due to their competitive pricing and claimed commitment to quality non-compromise. Thus, it is expected that these stores would favour less advantaged areas, e.g. with lower property value to lower the surface cost, and higher number of residents with lower socio-economic status since this is their target customer group. The results do not fully reflect such an assumption. Discount supermarkets are higher in number in areas with higher house value, higher population density and higher number of working people; whereas no significance is found for income, social housing and non-Western. This seems to contradicts with the assumption.
4. Results
A number of arguments could offer an explanation for such contradiction. It might be the case that while staying close to central pattern, these stores strategically choose for areas that still remains a high level of less advantaged residents. An additional rational for this is since discount supermarkets has attracted a larger group out of their initial target groups. A retail report by BBC stated that more than 31% shoppers at Aldi and Lidl are currently from the middle- or upper-middle class, while two years ago, this figure was as little as 12%
(Shadbolt 2015). Thus these retailers may want to locate in areas that can attract a broader customer demographics.
Crime rate slightly influence the amount of these stores, which can be interpreted either as a centric pattern or association with less advantaged areas. However, considering that the coefficient is small, and the insignificance of crime rate in the logit model, it appears that crime rate hardly represents a greatly influential factor for this category.
Logistic regression shows almost similar results, in which all variables showed positive relationships. However, when it comes to income, a significant negative association is found, indicating that those areas with lower income level tend are exposed to more discount supermarket, in addition to the direct relationship with social housing and non-Western. This fits well with the assumption. Still, high sensitivity to income suggests that currently, these discounters represent food supply choice for those with more restrained budget for food.
Non-chained grocers
The case of non-chained grocers is more difficult to interpret, since while the category is justified, certain amount of diversity within cannot be fully addressed. The quantity of these stores would increase in areas with higher house value, population density, number of working people and crime rate. Non-chained grocers are independently owned, with less extensive retail surface and product assortments than chained supermarket. It seems unlikely that the tendency to locates in the central areas means that they cater to customers of higher social-economic status; it is more likely that the aforementioned features give them more flexibility when it comes to location options, thus allowing them to remain close to central areas.
Somewhat in a similar way to discount supermarkets, the location of this category is sensitive to income variables. A 1% decrease in income would lead to a decline of 10% number of non-chained grocers. Thus, it is likely that these shops tend to operate in areas with
rationale behind this is since a number of these stores are owned by non-Westerns residents, they are located in areas with higher proportion of non-Western inhabitants. This might represent a wish to stay close to communities with similar background, either for commercial or personal reasons. In the case of immigrant entrepreneurs, spatial concentration offers both demand and supply advantages (Kloosterman & Van Der Leun 1999).
Logistic regression reveals similar trends for this category. Independent grocers correspond positively to all independent variables, yet negatively with income, which is then supported by the positive relationship with both social housing and non-Western. One is to infer that these shops somehow function similarly to the discount shops, in a sense that they provide an option for more economical food shopping. Still, due to their flexibility in comparison to discount stores, they are larger in quantity, as well as responding more to the income level of the target customer groups.
Specialty food shops
The amount of specialty food shops is associated positively with all the dependent variables, with the exception of income. As house value, population density, employment and crime rate increase, the exposure to these stores is higher by 18%, 20%, 6.4% and 0.2%
respectively.
Another deviance regarding income is shown here. As income gets higher by 1%, the number of specialty food shops decrease by 4%. The difference between income’s relationship with this type’s presence and quantity is of interest. Direct interpretation suggests that higher income areas are not necessarily more exposed to specialty food shops, but an increase in income would decrease the quantity of such shops. The reason behind this is unclear, but one might argue that income does not affect exposure to these type, but among areas that are exposed, the number of shops decrease as income rises. Supporting the relationship with income is the inverse association with non-Western proportion. In contrast, there are a slight positive association with social housing. A possible explanation is that due to the complex process of social housing development in the Netherlands, residents in social housing, in certain cases, does not necessarily have lower income.
If gentrification is considered as a process of neighbourhood characteristics transformation, driven by migration, in situ social mobility, and demographic change (Hochstenbach & van Gent 2015), the decrease here seems to fit with the claim that specialty
4. Results
food shops have either disappeared or driven to less advantaged locations in competition against chained stores, due to the fact that the inflow of higher-income residents (Hochstenbach et al. 2015) might have raised the property price, as well as threatening the customer base of these stores. This, however, seems to contradict the observation that some of these shops have tried to switch to a more high-end consumers group. The reason might be that these changes are not reflected in the locations, but rather the price and branding of the shops.
Healthy food shops
Healthy food shops, among all categories, can be considered the most high-end ones, since this category includes shops selling organic, biologische food with much higher price.
House value, employment and crime rate all have positive relationship with this category, which seems to correspond with the centric pattern. Interesting deviance from general patterns is the relationship with population density and income. There is no significance found in the relationship between income and number of healthy food shops; however, inverse relationship exist when it comes to social housing and non-Western. This seems to support the assumption that these stores are associated with areas with higher socio-economic status. Similar situation applies to population density. This, in contrast, seems to echo the assumption that these shops do not rely on appearance to the mass for profitability.
Turning to logistic model, similar trends exist. Among these trends, the extreme sensitivity to house value indicates a strong centric pattern. However, a number of differences arise. Positive relationships exist between house value, population density, income, and employment. It appears that these stores respond extremely well to areas with higher income level, and lower proportion of social housing and non-Western, where consumers are able to afford less competitive pricing of these stores. When it comes to crime rate, different patterns arise. Areas with higher crime rate would have higher number of healthy food shops, yet the odds of having at least one is lower. The rationale behind this difference is difficult to find.
One potential explanation is that less secure areas have less chance to be exposed to this type, but among those that are exposed, the number of shops increases as crime rate increases since these shops ted to locate close to the centre.
Foreign food shops
The case of foreign food shops also sees similarities to general patterns with the exception of income. House value, population density, employment and crime rate are all positively associated with quantity of foreign food shops. This strongly stays close to the interpretation of the general. However, once again, no significance is found between quantity and income. According to the logistic regression results, in a similar way to healthy food shops and chained supermarket, the exposure is extremely responsive to income. While similar to the case of healthy food shops, the reason behind this is hard to find. This might indicate that foreign food shops in the case of Amsterdam are indeed “exotic” food shops. Rather than catering to the residents of non-Western background, whose also tend to have lower socio- economic status, these stores are spatially associated with areas where residents are better off financially. One might argue that in this case foreign foods are desired not so much as foods, but as symbols, as markers of distance (van der Veen 2003). However, income is not related the amount of foreign shops, as no significant association is found. Still, other variables follow consistent pattern with the logit regression results.
A number of differences arise between OLS and logistic regression. Exposure to foreign shops are not related to social housing, while quantity decreases with the rise of social housing.
This appears supportive of the ‘exotic’ claim mentioned above. There are higher number of foreign shops in as non-Western decreases, whereas exposure and non-Western are directly related. It is likely that the among areas with at least one foreign shops, those with less non- Western residents will have higher amount of these shops.
In summary, most of the results stay close to the observed general pattern, with the exception of some deviances. In addition, the two concepts of presence and quantity, while providing some helpful insights, also bring up a number of contradictory results. The next chapter turns to further interpret these findings, and link them back to a broader context of Amsterdam as well as current academic debates.