FACTORS DRIVING THE POPULATION GROWTH AND DECLINE OF PORTUGUESE CITIES

Despite the worldwide trend of urbanisation, data reveal that some cities are growing whereas others are losing inhabitants. To assess such dynamics in Portuguese cities, demographic, employment, housing, and climate variables were analysed as possible drivers of population change for the period 1991–2011. Panel data models show that higher shares of employment in the secondary and tertiary sectors, higher maximum temperatures, and a higher proportion of middle-aged vacant houses act as pull factors attracting inhabitants, whereas a higher unemployment rate is a push factor for cities. 1 CIEO – Research Centre for Spatial and Organizational Dynamics, University of Algarve, Portugal; 2 ISCTE-IUL Instituto Universitário de Lisboa, Portugal; 3 Landscape Dynamics and Social Processes Group, Instituto de Ciências Agrárias e Ambientais Mediterrânicas (ICAAM), University of Évora, Portugal. Corresponding author: Ana Paula Barreira, CIEO – Research Centre for Spatial and Organizational Dynamics, University of Algarve, Campus de Gambelas, Building 9, P–8005–139 Faro, Portugal. E-mail: aprodrig@ualg.pt


Introduction 1
Cities that have achieved economic success tend to attract new inhabitants, and thus 2 demographic growth is often related to economic growth (Begg 1999; Lutz 2001; 3 Storper and Manville 2006). Cities with economic growth usually show an increase in 4 the amount of accumulated wealth available for redistribution, better job opportunities, 5 and a higher quality of life (Bhatta 2010; Blanc, Cahuzac, and Tahar 2008) by creating 6 the conditions that are needed for businesses to become established (Hansen and 7 Niedomysl 2009) and for public amenities to be provided (Kourtit,Nijkamp,and 8 Scholten 2015). Jointly with economic opportunities, the provision of affordable cities. Although some pull and push factors of cities may have universal relevance, 12 some factors appear to differ territorially (e.g., Guimarães et al. 2016; Royuela, Moreno, 13 and Vaya 2010). Therefore, identifying the specific factors that attract inhabitants into 14 particular cities has become increasingly relevant for policy purposes. 15 The process of urbanisation has been prevalent in Portugal since the 1960s, as it 16 has in most other parts of the world. The percentage of the Portuguese population living 17 in cities was 23% in 1960 and 30% in 1981, subsequently rising to 43% in 1991 and to 18 44% in 2011 (i) (calculations based on INE -Portuguese National Statistics). However, 19 despite the city-based population increasing as a percentage of the national total, not all 20 cities have increased the number of their inhabitants. In fact, about one-fifth of 21 Portuguese cities registered the reverse process of population decline between 1991 and  ). The process of urbanisation is based largely on the fact that urban areas tend to 23 have more jobs for skilled workers and higher wages (Blanc, Cahuzac, and Tahar 2008) 3 as well as improved public services and higher levels of social well-being 1 (Panagopoulos, Duque, and Dan 2016). 2 Despite this tendency for population to be increasingly concentrated in or around 3 cities, growing cities go hand in hand with declining cities (Cadwallader 1991). As a 4 result of demographic transitions such as increased life expectancies and decreased birth 5 rates, data reveal the existence of cities with increasing proportions of older residents 6 and, more surprisingly, cities that are losing inhabitants. In fact, despite the rise of urban 7 agglomerations, Turok and Mykhnenko (2007) found that the growth of European cities 8 generally slowed down between 1960 and 2005. 9 Migration tends to be related to employment opportunities, and a causal 10 relationship between employment and population growth is commonly observed (IOM households tend to choose places with low unemployment rates (Bhatta 2010; Hunt 1 1993), with the need for employment being a driver of out-migration and city 2 population decrease (Hoekveld 2012). 3 Despite the lack of studies addressing the impact of sectoral employment on 4 population growth, an increase in the number of jobs in the industry and service sectors 5 is expected to be a factor favouring in-migration, as these sectors are concentrated in  Ham, and Burneika (2016). The former author found for local Canadian communities 14 that industry diversity has a positive effect on population growth whereas agricultural 15 and resource extraction sectors have a negative effect. The latter authors examined the 16 economic factors that explain population change in Lithuanian regions and found that 17 populations are likely to expand in areas with increasing employment in the services 18 sector and to decrease in areas with a high proportion of employment in the agricultural 19 sector. 20 Together with economic attraction factors, the affordability of housing has also 21 been found to be a major factor in migration flows (Sasser 2010). The number of vacant 22 houses also seems to play an important role in both attracting and repelling inhabitants. 23 Depending on the state of preservation (i.e., the condition) of the vacant houses, a city 24 can be either an appealing location or a place from which to move. Two types of 5 vacancy are identified by Fielder and Smith (1996), namely, "transaction vacants" and 1 "problematic vacants", with houses of the latter type being in poor states of repair and 2 expected to remain vacant. In more general terms, those cities that better reflect 3 households' housing needs more commonly show urban growth, with both affordability 4 and the availability of new houses playing important roles. According to the Second 5 State of European Cities Report (RWI 2010), the existence of affordable, high-quality 6 housing may constitute an advantage for city population growth, in particular for 7 smaller cities, thus partially negating disparities between small cities and large cities in 8 income and poverty. Bhatta (2010) points out that the lack of affordable housing impels 9 households to live in the areas surrounding cities (the outskirts), thus promoting urban 10 sprawl.

11
Other factors influencing the quality of life in cities are also vital for encouraging 12 in-migration or for preventing out-migration. One such factor is the process of de-  An out-migration of inhabitants may also occur when climatic factors challenge 23 the quality of life. Rappaport (2007) found that temperatures play a role in the way that 24 city populations evolve. By analysing the average maximum temperatures in January 25 6 and July for US counties from the 1880s to the 1990s, that author found that lower 1 maximum and higher minimum temperatures helped explain patterns of population 2 growth. Cadwallader (1991) discovered for US cities between 1975 and 1980 that the 3 level of climatic attractiveness was greater in growing cities. For Europe, Cheshire and 4 Magrini (2006), considering the former EU-12 countries, did not find any evidence that 5 weather influenced mobility between countries but did find it relevant for mobility 6 within countries, whereby "warm days" measured by an upper value of heat, dryness, or 7 sunshine favour higher city population growth. In contrast, Mykhnenko and Turok 8 (2008) studied Eastern European cities but failed to find any difference in population 9 growth between cities in the sunnier, southern parts of Europe and those located in the 10 cooler, northern parts.  population of these growing cities increased by 25%, meaning that a net total of 20 630,000 inhabitants moved to live in such cities between 1991 and 2011. 21 Despite the prevalence of cities that grew between 1991 and 2011, 31 cities 22 showed population loss in that period, with 14 cities displaying a persistent decline 23 (both decades). Moreover, during those two decades, Portugal's two biggest cities, 24 Lisbon and Oporto, experienced population declines of 17.4% and 21.5%, respectively.
Although the number of cities that are losing inhabitants is small, the fact that the 1 country's two main cities are included within the 31 shrinking cities means that these 2 shrinking cities overall lost 225,000 inhabitants between 1991 and 2011. 3 Considering the Portuguese division into regions according to the Nomenclature 4 of Territorial Units for Statistics (NUTS III), whereby 29 out of 32 regions have cities, 5 there is a diversity in the territorial distribution of cities growing and declining in 6 population (1991-2011). Variation is also found when the changes in the number of 7 inhabitants of cities and in the municipalities in which the cities are located are 8 compared. Figure 1 shows this territorially heterogeneous distribution of growing and 9 shrinking cities according to growing/shrinking regions and municipalities. 10 ( Figure 1 here) 11 Portuguese cities also exhibit a wide diversity in size. In 2011, 8.2% (13) cities 12 had fewer than 5,000 inhabitants, 20.3% (32) had 5,000-10,000 inhabitants, 60.8% (96) 13 had 10,000-50,000 inhabitants, 6.3% (10) had 50,000-100,000 inhabitants, and 4.4% 14 (7) had more than 100,000 inhabitants. The average size of Portuguese cities is 29,000 15 inhabitants, meaning that Portuguese cities are relatively small when compared with the 16 average size of European cities. Even the capital Lisbon, with 548,000 inhabitants in 17 2011, can be considered only a medium-sized city on a European scale. 18 The population growth between 1991 and 2011 varied widely between cities. 19 Caniço, on Madeira Island, increased by 240% (from 6,900 to 23,400 inhabitants), the 20 maximum positive change, whereas Oporto, the second-most-populous city in the 21 country, declined by 21.5% (from 302,500 to 237,600 inhabitants). On average, 22 between 1991 and 2011, the population in Portuguese cities grew by 9%, but when only 23 the growing (declining) cities are considered, they increased (decreased) on average by 24 25% (13%). Considering the average growth of the growing cities, 48 have values city, to 0.9% in 1991 for Chaves, a growing city. In Portuguese cities, the tertiary sector 11 dominates employment, and the secondary sector is the second-most-important sector, 12 although its importance decreased from 35.8% of all jobs in 1991 to 24.9% in 2011. 13 Urban employment in the primary sector decreased from 6.8% in 1991 to 2.4% in 2011.
14 In 2011, the average number of houses in a city was 15,900, 13.4% of which were 15 vacant (i.e., uninhabited). The oldest houses (>30 years of age) and the most recently 16 constructed houses (<10 years of age) were the age categories that included a higher 17 proportion of vacant houses (both 16%), whereas only 8.5% of the houses that are 10-  census, other data were also made available by city (e.g., population by age, 13 employment, and number of houses, among others); however, such data do not allow 14 comparisons to be made with the past as corresponding figures are unavailable for the 15 censuses of 1991 and 2001. Other geographical configurations are more common in 16 Portuguese statistics, including geographical areas according to the NUTS scheme, 17 municipalities, parishes, and other smaller territorial units. However, none of these 18 geographical configurations matches with the 'city', although the closest one is the 19 parish. Therefore, and because some of the variables used in the empirical models 20 described below are not available at the city level for all three censuses (1991,2001,21 and 2011), the 'city' in all cases is considered as the sum of the predominantly urban 22 parishes that compose it. The information on the parishes that are part of each 'city' is 23 available online as a result of the 2013 parish reorganization that was undertaken in 24 Portugal. Therefore, data presented regarding the 'city' are in fact an approximation 25 based on the boundaries of the constituent parishes. A comparison of the values of data 1 available for 'cities' with those of the urban parishes that compose them for 2011 shows 2 that the differences are small and not detrimental to the analysis.  from climatic data, which were retrieved from the website http://pt.climate-data.org. 17 The data refer to each of the current 158 Portuguese cities. 18 To explain the population growth of Portuguese cities, linear regression models 19 were estimated, with the dependent variable (Y) being defined as the percentage change 20 in city population between censuses (termed dPop). Since data are available from three    growth rates for should eliminate this endogeneity problem. 10 To assess the suitability of each model, the RESET test was applied.  Table 2 reports the coefficients of the explanatory variables for the random effects 6 models specified in the previous section (Model 1 and Model 2). The respective results 7 of the RESET test show that both models can be considered to be correctly specified. 8 Based on the usual significance levels (1%, 5%, and 10%), the results of Model 1 show 9 that the main drivers explaining city population growth are the variables unemployment

15
The results also corroborate the policies introduced by the governments of those  Table 1) to influence movements from rural and surrounding areas 21 into cities. The most relevant economic sector in explaining population growth is 22 employment in the tertiary sector (services), followed by employment in the secondary 23 sector (industry), which reflects the economic structure of the Portuguese cities whereby 24 more than half of the employment generated is in the tertiary sector (see Table 1). 25 To further examine the contribution of the employment profile to explaining tertiary sectors may be interchanged without consequences for city population growth.
14 Overall, the results of both models suggest that those cities that successfully made a 15 transition from an economy characterized by a higher number of jobs in the agriculture 16 sector to one relying mainly on industry and services became more attractive to citizens 17 as places to work and live compared with other cities. 18 The proportion of vacant houses in a city does not significantly affect its 19 population dynamics. However, cities with a higher share of middle-aged vacant houses 20 appear to be more likely to exhibit population growth. A possible reason for this is that 21 these houses to purchase or to rent are usually not as expensive as similar newer houses 22 and tend to be in better condition than older houses, so cities with a higher proportion of

22
As discussed above, Portuguese cities are very heterogeneous. On the one hand, 23 these cities display a wide diversity in size, with almost 30% containing fewer than 24 10,000 inhabitants. On the other hand, the phenomenon of suburbanization has led to 25 the growth of some cities but has promoted population decline in others, most notably in 1 the two biggest cities, Lisbon and Oporto. Thus, to examine whether the aforementioned 2 findings are generally applicable to the Portuguese context or, rather, they are valid only 3 for specific groups of Portuguese cities, Model 1 was re-estimated for three sub-samples 4 of the initial dataset.
5 Table 3 reports the results obtained for three distinct sub-samples of Portuguese 6 cities. In the first two models, the smallest cities (less than 10,000 inhabitants) and the 7 biggest cities (Lisbon and Oporto) were respectively dropped from the sample. Then, to 8 examine the robustness of the results with respect to the suburbanization process that 9 has affected the urban growth of some cities, in the third model all cities in the Lisbon 10 and Oporto metropolitan areas were excluded from the sample.

11
In general, the factors identified above as the main drivers of city population 12 changes remain important when each of the three aforementioned three groups of cities 13 is dropped from the sample. In all cases when a group of cities is dropped, a higher 14 proportion of employment in the tertiary sector and a higher share of middle-aged 15 vacant houses are, as in Model 1, significant determinants of population growth in 16 cities, and a higher unemployment rate is again a factor inducing population decline. In 17 fact, when compared with Model 1, there are only minor changes in the estimation 18 results, which concern only variables that were, or are now, marginally significant from  Note: ***, **, and * denote variables that are statistically significant at the 1%, 5%, and 10% 3 levels of significance, respectively.

5
When small cities are excluded from the sample (column 1 of Table 3), the 6 proportion of the population represented by working-age inhabitants becomes a 7 determinant of population growth. Thus, policies focused on the working-age 8 population, even when not related to the creation of employment opportunities, may 9 promote population growth in medium-sized and large cities. Moreover, the negative 10 effect on population growth of a high unemployment rate appears to be more relevant 11 for cities containing more than 10,000 inhabitants. In contrast, the creation of 12 20 employment in the secondary sector seems to be an important factor in the population 1 growth of small and medium-sized cities (column 2 of Table 3). Finally, a higher 2 maximum temperature appears to be a relevant driver of population growth only in 3 cities outside the Lisbon and Oporto metropolitan areas (column 3 of Table 3). Clearly, 4 these two larger cities have achieved a state where climate is no longer an influence on 5 their population dynamics. 6 7 Concluding Remarks 8 This study examined the determinants that make some cities more appealing 9 places to live compared with others. We tested whether explanatory variables related to 10 employment, housing, and climatic conditions are drivers of urban population growth 11 and decline, and found that greater employment opportunities in the secondary and 12 tertiary sectors, a higher share of middle-aged vacant houses, and higher average 13 temperatures favour population growth, whereas a higher unemployment rate helps to 14 explain population decline. 15 To increase their populations, cities must be active in generating job 16 opportunities. If those job opportunities are created mainly in the secondary and tertiary 17 sectors, then cities will have a higher probability of experiencing population growth. 18 Regarding the secondary sector, urban planners should contemplate promoting this 19 sector in designated areas where the impacts on the landscape and on the environment 20 are properly addressed. Technologies that are environmentally friendly and which lead 21 to higher productivity would help to change the usual negative image associated with 22 heavy industrial activity, and thus the implementation of high-technology industries 23 should be favoured. Such technology-driven industries, which are associated with promoting gains in the quality of life, should attract more workers and inhabitants to 1 cities with industrial potential.

2
The share of middle-aged vacant housing is a pull factor for Portuguese cities, 3 whereas old vacant housing, despite constraining population growth, does not have a 4 significant effect on population evolution. The irrelevance of old vacant housing as a 5 push factor of cities might be due to the underlying process of shrinkage in Portugal, 6 which is at an early stage. In other countries, where the shrinkage phenomenon is more in pushing away inhabitants, compared with the case of Portugal, which has encouraged 10 the adoption of house-demolition policies.

11
Cities with higher average temperatures emerge as being more appealing places to 12 live compared with other cities. However, cities in the same region reveal different 13 patterns regarding the capability of attracting inhabitants. This reinforces the importance 14 of the other factors affecting city population growth as revealed by the regression 15 models, namely, those related to the economic activity of the city and the affordability 16 of housing. 17 The process of urbanisation across the globe shows different patterns. While wider dichotomy between cities gaining inhabitants and cities losing them. This paper 24 highlights that distinguishing these two processes, one of growth and one of decline, is 25 22 not an easy task. Governments will need to conceive new policies and planning 1 approaches to deal with these different realities.

2
To further understand the differences found between the dynamics of population 3 growth and decline, future work would also need to consider some subjective indicators 4 of urban quality, such as inhabitants' emotional attachment to their cities and the level 5 of residential satisfaction. Such an analysis would help to more accurately define 6 suitable policies for dealing with the realities of city shrinkage and with the differences 7 in population decline/growth between cities.