How strongly does ethnicity influence consumption? We ask this question in the context of car purchase decisions. In particular, we examine the concept of “ethnocentrism” here, which postulates that consumers normally consider goods from their own country/region superior to others. Given data availability, we focus on Asians and their car purchasing decisions in Southern California. The analysis below will not only allow us to answer this question, but it also demonstrates how GIS analysis can be used for car companies or dealers to reconsider locations. In the light of most recent reorganizations in the car industry and expected changes in the future, this type of analysis may prove especially helpful as a spatial reorganization of dealerships may occur.
To investigate our question of interest, we categorize brands by region of origin (e.g. Toyota and Honda as Asian, BMW as European, Ford and GM as American) and class (roughly corresponding to luxury, middle-class and economy). We then perform a simple spatial cluster analysis to identify dealership clusters, where a dealership cluster can be thought of as a set of dealers that are located close together and far from others. We construct trade areas around those dealership clusters, and calculate, for example, the share of Asian population in the trade area. We compare this share of Asian population with the share of Asian brand car sales in the cluster. We find that in most cases higher shares of Asian population imply higher shares of Asian car sales. However, only low income Asians really have a relatively strong affinity to Asian brands, while middle income Asians are undistinguishable from American buyers, and high income Asians seem to rather buy American brands.
How many clusters of car dealerships exist in Southern California? Unfortunately, this question turned out to be hard to answer precisely. The reason is that it depends under which conditions a certain dealership should be included in a cluster. For example, should a dealership be included if it is quite far from all the others in a cluster, but offers a different brand? Or should a dealership be excluded from a cluster if it is close by but another dealer already carries the same brand?
We decided to allow a statistical program to answer this question for us. All that we required was that not too many clusters would carry a brand more than once, that the maximum distance between dealers in the largest cluster would not be too large and that there would not be too many clusters that only had one or two dealers. We learned that all three criteria were very hard to fulfill simultaneously. We consequently chose a solution that provided us with reasonable values for most of the clusters. We ended up with 200 clusters, with 13% of clusters having more than one dealership carrying a particular brand, a 36 mile distance for the largest cluster (which is bearable since it is located in a thinly populated area) and about 5% of individual dealers being listed as their own cluster. Note that visual inspection of resulting clusters (see any of the maps provided) would have suggested to combine certain clusters. However, we deliberately did not change clusters based on visual analysis to investigate the quality of this automated procedure.
We then created trading areas for each cluster based on the simple assumption that any customer would only buy a car from the nearest cluster. This procedure defines the borders of trading areas for locations that are equally distant to two (or more) clusters of car dealers. We then calculate the share of Asian population as well as income per capita in the corresponding trade area of each cluster based on census data by block group. We also calculated the share of car sales in a cluster of a certain origin, namely Asian car sales as a share of total car sales. As a control, we also calculated the share of high end car sales in each cluster.
We visualized all this information in maps to perform quality control. We also used the data created and correlated Asian population shares with the share of car purchases of Asian brands in each cluster. We first, however, put our methodology to the test and checked the correlation between the share of high end car sales and per capita income. We did find that higher per capita income in a sales area is strongly correlated with sales of more luxurious brands. Visual inspection of the map, share of high end car sales and per capita income (see map below), confirms the calculations: larger dots are more prevalent in darker shaded areas and vice versa. As one might have guessed, especially European car brands benefit from this relationship, as can be seen from the map Car dealers by car brand origin & sales volume, per capita income by dealership cluster trade areas.
We then turned to our main question. Our preliminary results suggest that Asians are conservative car buyers: The correlation between total car sales volume in a cluster and total population in the cluster is of course strongly positive, while the correlation between total car sales volume and Asian population in the cluster is very weak. This correlation is outright negative for middle class Asians and somewhat positive for more affluent Asians. The data also indicates that Asians might rather buy lower class cars, even if they could afford higher class ones. Finally, our hypothesis was confirmed: in most areas, an increase in the Asian population share implied an increase in the dollar sales volume of Asian dealerships. Since most Asian cars are in the more affordable category, this is quite a striking result and is also confirmed by our visual analysis (see map below). It is especially true for lower income Asians, and the lower the income the stronger the effect. Careful inspection of the map Car dealers by car brand origin & class, & dealership cluster trade areas illustrates this finding. Interestingly, affluent Asians tend to buy American car brands, not European ones, which is contrary to both casual observation and other research and therefore deserves further investigation.
Our analysis demonstrates the usefulness of GIS for dealer location analysis. We expect that this type of analysis might become especially useful in the difficult situation the car industry faces as dealerships need to close. GIS here fulfilled two vital functions: it served to generate the data by location based on the available information. It was then used to ensure quality control of the results, for example, how the statistical tools assigned dealerships to clusters. While we decided against using the latter step to improve our analysis at this preliminary point, we will certainly do so in the future. Overall, GIS analysis proved useful in our context both for preparing business decisions as well as conducting scientific research.