Valuing Cultural Diversity of Cities

Abstract

This paper revisits whether cultural diversity is a source of local production and/or consumption amenities. We adapt the analytical framework of Roback (1982, 1988) and Chen & Rosenthal (2008) to estimate the impact of cultural diversity on city wage and rent premiums from hedonic regressions. We focus on New Zealand which -with high residential mobility and ease of setting up business -is particularly suited to this framework. Additionally, our estimates are based on a very large data set: complete unit record census data on individuals and dwellings in 110 urban areas spanning 32 years. Controlling for observed and unobserved city characteristics, and for the potential endogeneity of diversity, we find that cultural diversity serves as a local positive production amenity and a weakly negative consumption amenity. The results are mostly robust to measuring cultural diversity by birthplace, ethnicity or religion; and to using a range of measures of diversity. We conclude that the presence of people from different cultural backgrounds enhances the profitability of urban firms. In contrast, a city's population has a weak preference for living near others who are culturally similar to them. The effects are stronger in larger cities. JEL codes J31; R21; R23; R31

Cities with Forking Paths? Agglomeration Economies in New Zealand 1976-2018

“…Our model follows Maré and Poot (2019), which in turn adapts Gabriel and Rosenthal (2004). To start, we assume that mobile workers choose their locations given preferences for housing, H, a composite consumption good, Y , and local amenities, f u (A).…”

“…13 Gross income appears to be an accurate measure of labour income for full-time workers aged 25-years or older. For these workers, Maré and Poot (2019) Figure 1 presents histograms for our dependent variables, where the top and bottom panels show incomes vis-á-vis rents and the left and right panels show aggregate data vis-á-vis spatial premia, respectively. We find broadly symmetric distributions with some evidence of fat tails, especially on the low-side.…”

“…Following Roback (1982), Gabriel and Rosenthal (2004), Y. Chen and Rosenthal (2008), and Maré and Poot (2019), we estimate a simple economic model that measures the relative advantages of locations to firms and workers in terms of the price of labour and floorspace. Although grounded in the existing literature, our study combines several useful attributes.…”

“…And, in contrast to Maré and Graham (2013), who estimate agglomeration economies for nine regions in New Zealand, we estimate agglomeration economies for 134 locations. Unlike Gabriel and Rosenthal (2004) and Martínez-Galarraga et al (2008), but like Maré and Graham (2013) and Maré and Poot (2019), we use micro-data to control for the observed characteristics of individual workers and dwellings. Together, these attributes enable us to derive estimates of agglomeration economies in production and consumption that are relatively robust and broadly comparable over time and space.…”

“…The first panel presents the common income and rent elasticities, w and r , whereas the second panel presents the associated estimates of agglomeration economies, E p and E c . The latter are computed from the elasticities, as per Equations 8 and 9, where we followMaré and Poot (2019) and assume the cost shares of floorspace in production and consumption are γ = 0.10 and α = 0.20, respectively.14 The third panel summarises the time-and location-specific elasticities that are included in each of the five variants of Model D. And, finally, the bottom panel summarises two model performance measures, R 2 and elpd, where we prefer the latter.15 Turning to the results, Model A-which uses aggregate data-returns income elasticities that are approximately twice as large as Model B-which uses micro-data. The elpd values indicate Model B performs worse than Model C, supporting the latter's use of Student's t-distributions to mitigate influential observations.16 Turning to Model D, we find Model D4-which allows rent elasticities to vary with time and location-has the highest elpd of the models we test, closely followed by Model D5.…”