Eighty years of urban development in New Zealand: impacts of economic and natural factors

Abstract

We analyse impacts of economic and other factors on long-run urban growth in New Zealand. Growing cities must have preferred attributes (such as natural characteristics, social amenities and transport infrastructures) relative to other cities. We outline a theoretical model that includes distance-related effects on individual utility and thence population location. We test the model over 1926 to 2006 across 56 New Zealand towns using instruments dating from 1880 to deal with potential endogeneity. Three factors – land-use capability, sunshine hours and proximity to Auckland – are found to influence settlements’ long-run population growth. In addition, the proportion of population that is Māori is negatively correlated with population growth over the second half of the sample period. Supplementary evidence suggests that this variable relates to the importance of human capital for the growth of settlements over recent decades.

KEYWORDS:

• Population growth
• migration
• infrastructure
• amenities
• spatial equilibrium

JEL codes:

• H54
• R12

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Due to limitations in sample size, we do not investigate other potential sample splits.

2. Lutchman (2013) provides an in-depth discussion of the literature relating to transport infrastructure investment and regional growth. Apatov (2013) reviews literature examining the effect of higher education instititons (e.g. universities) on local growth.

3. E.g. Mera (1973), Blum (1982), Aschauer (1989), Munnell and Cook (1990), Glaeser, Scheinkman, and Shleifer (1995), Rappaport and Sachs (2003), Crescenzi (2005), Duflo and Pande (2007), Karlsson and Andersson (2007), Rappaport (2007, 2008, 2009), Rodríguez-Pose and Crescenzi (2008), Sterlacchini (2008), Glaeser and Gottlieb (2009), Graham, Gibbons and Martin (2009), Wang (2010), Grimes and Liang (2010), Duch-Brown, García-Estévez, and Parellada-Sabata (2011), Coleman (2012), Duranton and Turner (2012), Combes, Duranton, Gobillon and Puga (2012), Donovan and Munro (2013), Yu, De Jong, Storm, and Mi (2013), Ding (2013), Duranton and Puga (2013), Desmet and Rossi-Hansberg (2013), and Grimes and Young (2013).

4. Another possibly relevant issue, but one that we do not investigate further here, is differing attitudes to market outcomes between Māori and non-Māori individuals and firms (ANZ, 2015; Grimes, MacCulloch, & McKay, 2015).

5. For simplicity, we treat these characteristics as a scalar in this derivation, but the analysis is easily extended to a vector of local characteristics.

6. Given that the amenities are located at the centre, an individual's distance from the centre must be strictly positive.

7. Duranton and Turner's (2012) utility function accords distance travelled a positive elasticity, based on an argument that people travel in order to experience amenity services. By contrast, we consider that additional distance from the core city's amenities (i.e. lower proximity to the core city) reduces utility.

8. For instance, Desmet and Rossi-Hansberg (2013).

9. And hence an increase in net income at all locations given the nature of the constraints.

10. The difference GMM estimates for the lagged dependent variable fell below even the (downward-biased) fixed effects estimate of this parameter. This is consistent with a weak instrument problem biasing the results towards the OLS estimate of the parameter in first differences, which comes out at 0.42.

11. We thank one of the referees for this observation.

12. Note also that there are only six New Zealand cities with a university home campus (four of which had a university before 1926), so we have little variation to work with when trying to single out the effect of universities on growth even in the absence of collinearity issues.

13. The data are available at: http://motu.nz/our-work/population-and-labour/individual-and-group-outcomes/new-zealand-urban-population-data/ .

14. However Queenstown is omitted as it did not meet any of the four listed criteria.

15. To derive this measure, we averaged the LUC index values across all 2006 Census meshblocks within each Territorial Local Authority (TLA), weighted by meshblock land area (and we transformed the variable so that higher values corresponded to better agricultural land). Each town was then assigned the average LUC of the TLA that it falls within. A detailed description of the LUC index can be found in Lynn et al. (2009).

16. LUC may also correlate with land costs, though the direction of the relationship is unclear: on the one hand, less productive land at the city fringes will result in lower land prices for consumers (all else being equal), since the land has a lower value in its best alternative use; but if LUC is so low that the land is unsuitable for agriculture (e.g. wetlands and steep terrain) it will also generally be unsuitable for urban construction, corresponding to a lower land supply and therefore higher prices.

17. The Pearson correlation coefficients between these climate variables and LUC are all very small, so we do not interpret sunshine hours as affecting agricultural productivity nor LUC as reflecting climate amenities.

18. We use the following seven regional classifications: Auckland (within 200km of Auckland); Greater Auckland (all other North Island towns north of Lake Taupo); Wellington (within 200km of Wellington); Greater Wellington (all other North Island towns south of Lake Taupo); Christchurch (within 200km of Christchurch); Greater Christchurch (all other towns in Canterbury, Marlborough, Tasman or West Coast regions); and Dunedin (Otago and Southland). The first four regions are in the North Island and the last three are in the South Island. See Table A3 for a list of towns by region.

19. We attempted to gather data on hospitals from the 1926 SNZ Yearbook as another amenity measure, but the definition of ‘hospital’ at the time was too broad.

20. Data were not available for all towns and we had to approximate using the proportion Māori of the nearest neighbouring town. The resulting variable is quite coarse, with only 13 unique values. As a robustness check, we also estimated our regressions with the percentage Māori population in 1881. There are 28 unique values for the 1881 measure, and it has the advantage of being more clearly exogenous. Neverthless, there is strong persistence of Māori population proportions over time with the correlation coefficient between the 1946 and 1881 Māori proportions being 0.59. The 1946 and 1881 measures produced the same qualitative results; we retained the 1946 measure in order to be consistent with our other covariates (which are all observed between 1926 and 1966).

21. We control for: (log of) total population, share of Equivalent Full Time students (at universities and polytechnics separately) as a proportion of the working age population, unemployment rate, year effect, and area fixed effects. Results of the full regressions are available from the authors on request.

22. Separate regressions show a decline in vocational (degree) qualifications of 0.14% (0.15%) for each 1 p.p. increase in the Māori population share. The vocational (bachelors) [combined] results are significant at the 5% (10%) [1%] levels, respectively.

23. Our other climate measures – rainfall and average summer and winter temperatures – were never significant in any combination when included with and without sunshine hours.

24. We set the weights to zero for towns not in the same island, i.e. we assumed that North Island towns exert no influence on South Island towns, and vice versa. We also calculated the Moran's I statistic using the inverse of distance (rather than the inverse of squared distance) with very similar results.

25. This is the case also for the GMM estimates.

26. I.e. log population_{(t – 2)}, being the first lag of log population_{(t – 1)}.

27. Each of these variables was significant at the 10% level in the 1966–2006 POLS estimates that excluded the lagged dependent variable.

28. Donovan calculated the life index as (), where is the average rent paid by households in the TLA adjusted for housing quality (number of rooms, etc.), and is the average household income in the TLA, adjusted for observable characteristics such as education level and household size. This index reflects a spatial equilibrium approach in which people pay high rents relative to wages so as to access positive local amenities. The business index is defined as (), with household rent proxying for commercial rent. This index also reflects a spatial equilibrium approach in which firms that choose a highly productive locality can pay higher wages and must pay higher rents to reflect the more productive location.

Additional information

Funding

Ministry of Business, Innovation and Employment, Resilient Urban Futures.