COVID-19, Shifting Urban Growth Dynamics and Preferences for Regional Living in Australia

ABSTRACT

Stated preference surveys and qualitative engagement were used to examine the potential for change in Australian urban growth dynamics. We found that residents with strong preferences for large and mid-sized cities appear unlikely to change their preferences. However, 21 percent of individuals currently living in large cities were open to moving to mid-sized cities. 54 percent of respondents placed a high importance on quality of life, local healthcare, and affordable housing. Employment is a pivotal barrier to relocation. To encourage regional growth, we recommend: developing local employment, supporting remote work, investing in post-retirement living infrastructure, and increasing local educational services.

摘要

偏好调查和定性参与被用于发现澳大利亚城市增长动态变化的潜力。我们发现，对大中城市有强烈偏好的居民似乎不太可能改变他们的选择。然而，目前居住在大城市的人中有21%愿意搬到中等城市。54%的受访者高度重视生活质量、当地医疗保健和经济适用房。就业是搬迁的关键障碍。为了鼓励区域增长，我们建议：发展当地就业，支持远程工作，投资退休后生活基础建设，并增加当地教育服务。

KEYWORDS:

• Regional growth
• regional employment
• urban growth dynamics
• Australian cities
• migration

1. Introduction

Over the past two decades much of the debate on growth within urban systems has focussed on the growing dominance of the largest metropolitan centres in advanced and developing economies (Moretti 2012). As researchers have noted, cumulative causation has increasingly concentrated growth across the world in what have been referred to as “superstar cities” (Kemeny and Storper 2020), often at the expense of smaller urban centres. These spatially divergent patterns of growth have exacerbated interregional inequality, highlighted by the rise of populist politics in Europe and North America (Rodríguez-Pose 2018). Glaeser (2011) has advocated for the on-going development of these places, noting the largest urban centres are the major source of innovation within an economy, and a key driver of national prosperity. Glaeser (2011) has argued that further concentration of population and economic activity in these places is both inevitable and desirable, maximising both the welfare of individuals and the economic standing of society (Beer and Clower 2019). These sentiments have resonated with researchers working on networks of world cities (Turok and McGranahan 2013), as well as agencies such as the World Bank (2009) who have prioritised national economic growth.

Historically, major state capitals have been the centres of greatest population growth in Australia. Between 2011 and 2021, metropolitan areas (excluding ACT) all experienced greater growth than “rest of state”; people living in capitals cities increased by 17 per cent, while regional Australia grew 11 per cent (ABS 2021). In 2019–2020, 72 per cent of the population was concentrated in the five largest capital cities (AIHW 2023). Prior to the pandemic the Australian population was projected to increase by up to 24.6 million people by 2066, and approximately 55 per cent of this growth was expected to occur in Australia’s two largest cities – Sydney and Melbourne (ABS 2008). The largest Australian cities are under great strain, as reflected by expensive local housing markets, rising levels of housing unaffordability, growing traffic congestion, and overcrowded schools and hospital (Vij et al. 2021; PIA 2018). It has been argued that more spatially balanced patterns of population growth that encourage regional development could help alleviate some of these pressures on the largest cities (e.g. Archer et al. 2019).

There has been ongoing debate about the role of government in encouraging the growth of regions. Beginning in the 1960s and 1970s, Australian governments and researchers began to engage with issues around the apparent “over development” of some metropolitan centres (Neutze 1977, 1978), and the potential to encourage the growth of “new cities’ (Logan 1978). The Department of Urban and Regional Development (DURD), established by the Whitlam Government in 1972, identified a number of “new cities" to be developed across Australia, such as Albury-Wodonga and Bathurst-Orange. The failure of many of these planned cities to emerge as major population hubs has been attributed to an over-estimation of the capacity of governments to establish major new urban centres (Sorensen 2000; PC 2017; Daley et al. 2018). Conversely, some studies have argued that regional policy in Australia since the mid-1970s has been limited by the failure of governments to prioritise regional growth and regional issues (e.g. Collits 2004). There has been a perception that governments have been scarred by the failures of the programs administered by DURD and therefore remain reluctant to redirect substantial public funding to regions in need (Collits and Rowe 2015; Daley and Lancy 2011).

The COVID-19 pandemic has challenged established urban growth patterns, as lockdowns, restrictions on domestic and international travel, as well as disruptions to global supply chains have reshaped economies and employment practices. There has been growth in remote working in some sectors, while restrictions on personal movement and growing awareness of the health risks associated with higher population densities have seen some households relocate away from metropolitan areas. These dynamics have been evident in Australia. “The population of regional Australia grew by 70,900 people during 2020–2021, in contrast to a decline of 26,000 for the capital cities … This is the first time since 1981 that Australia's regional population grew more than the capital cities, due to changing migration patterns during the pandemic” (ABS 2022). This growth was especially noticeable in regional New South Wales and Victoria, where stricter and more severe lockdowns in Sydney and Melbourne prompted high levels of out-migration to surrounding regional centres, such as Wollongong and Geelong, respectively, though more recent evidence indicates a return to pre-pandemic trends (ABS 2023). Florida et al. (2021) argue the events of 2020–2021 have the potential to exert a long-term impact on cities, through social scarring – an on-going reluctance to accept crowded places; a long-term shift to remote forms of economic activity, including shopping and work; a move away from infrastructure – including housing – that supports higher density populations; and changed urban forms as a preference for social distancing becomes entrenched. These potential shifts in preferences offer new opportunities for long-term population decentralisation and dispersion, building on previous government attempts to encourage and sustain regional development and growth.

This paper draws on recent policy-oriented work (Vij et al. 2021; Crommelin et al. 2022) to examine evidence for change in the drivers of urban growth post COVID-19. Further, it asks whether this new set of processes challenges the role of the largest cities as preferred places of residence. These theoretical insights are then examined for their policy implications, particularly with reference to the emerging aspiration of some governments for a more even distribution of population (Vij et al. 2021).

The remainder of the paper is structured as follows: Section 2 describes the quantitative and qualitative datasets used by the study. Sections 3 and 4 present findings from our empirical analysis of the quantitative and qualitative datasets, respectively. Section 5 concludes with a summary of our key findings, and a discussion on policy implications. Finally, Appendix 1 provides population rankings for key Australian cities, Appendix 2 describes the econometric framework used for our analysis of the quantitative data, and Appendix 3 reports estimation results for the final econometric models.

2. Data

Two methods were used to collect data for this research: an online survey, and interviews and focus groups with regional city residents.

2.1. Online Survey

We designed an online survey instrument to measure individual preferences comprised of six sections, all of which are described in detail below. Participants were given a glossary of terms to assist in their interpretation of important terminology.

1. Current hometown and city: Respondents were asked about their current city of residence, other Australian cities they have lived in the past for more than one year, their duration of residence in each city, and their purpose for moving to that city.

2. Knowledge about key attributes of their current city of residence: Respondents were asked about their knowledge of different social, economic and environmental indicators of the city in which they live. In particular, for their current city of residence, they were asked about the following nine indicators: (1) distance to the coast; (2) population size; (3) unemployment rate; (4) urban centre classification across six different categories (e.g. agricultural city, industrial city); (5) average home sales value; (6) average annual full-time wage rate; (7) average daily commute time; (8) average cost of living for a single person, excluding rent; and (9) climate classification. At the end of this section, the correct information for each of the indicators was revealed, so respondents could compare their answers.

3. Stated preference experiment: Respondents were presented multiple stated preference (SP) experiments where they were offered a choice between living in two different cities, such as the example shown in Figure 1. The SP experiments were purposefully delivered after participants had reflected on their knowledge and experience of their current city in sections 1 and 2 to ensure participants had had an opportunity to consider key factors related to their choice of place of residence. Both urban areas shown within the SP experiments were described in terms of the nine indicators mentioned above. Respondents were asked to indicate the urban area that they would prefer to live in. As a baseline for comparison, each scenario also presented values for the same nine indicators for their current place of residence. Each respondent was shown 8 different scenarios, and the values of each attribute were systematically varied across scenarios across the ranges shown in Table 1, based on a block efficient design generated using the software Ngene (Choice Metrics 2009). Appropriate constraints were used to ensure attribute combinations make sense. The final experiment design comprised 144 unique scenarios split across 18 blocks. Respondents were randomly assigned to one of the 18 blocks. Respondents were presented a glossary with definitions for each of the attributes, and specific values where relevant (e.g. urban centre classification, climate classification), with examples.

4. Attitudes and perceptions: Respondents were asked about the importance of different factors when selecting and/or avoiding a city to live in, such as economic and education opportunities, quality-of-life, urban and natural amenities, and incidence of natural disasters. They were asked about the importance of these same factors for why they may not want to leave their current city of residence. They were asked about their perceptions of a mid-sized city (defined for the purposes of this study as having a population less than 100,000) and a large city (defined as having a population over 100,000) in terms of these different factors. Appendix 1 lists the top 50 Australian cities by population for reference. Finally, they were presented a list of policies that could support regional migration, such as support for employment, home ownership, etc., and asked to indicate the policies which would encourage them to stay/move to mid-sized cities.

5. General satisfaction: Respondents were asked about their satisfaction with different aspects of their life.

6. Socio-demographics: Respondents were asked about their age, gender, indigeneity, country of birth, education, employment, place of residence, household size and structure, and income.

Figure 1. Example scenario from the stated preference experiment.

Table 1. Range of attribute values used in our SP experiments to describe each urban area across different scenarios.

#	Attribute	Range of values
1	Distance to coast	on the coast; 37.5 km, 75km, 200km, 600km
2	Population size: mid-sized cities	5k; 10k; 25k; 50k; 75k; 100k
	Population size: large cities	250k; 500k; 750k; 1m; 1.5m; 2m; 2.5m; 5m
3	Average annual income per person	$55k, $60k, $65k, $70k, $75k, $85k, $90k, $100k
4	Unemployment rate	3%, 5%, 7%, 9%, 11%, 13%, 15%
5	Urban centre classification	industry city; service city; connected city; coastal lifestyle city; mixed function centre; agricultural city
6	Average home sales value	$240k, $400k, $560k, $650k, $720k, $880k, $1.04m, $1.2m
7	Average daily commute time (minutes)	20, 30, 40, 50, 60, 70, 80, 90
8	Average monthly cost of living for single person (excluding rent)	$1k; $1.1k; $1.2k; $1.3k; $1.4k; $1.5k
9	Climate classification	hot humid summer, warm winter; warm humid summer, mild winter; hot dry summer, warm winter; hot dry summer, cool winter; warm temperate; mild temperate; cool temperate; alpine

The survey concluded with an open text question to elicit any feedback from respondents about the survey itself. Overall, the survey received positive feedback and respondents’ comments highlighted that they were engaged with the topic.

The survey was administered in February 2021 through the web-based interface Confirmit (now known as Forsta) to a sample of 3,012 respondents aged 18 years and over from across Australia. Respondents were recruited through the market research company PureProfile. The data collection process was set up to ensure that the final sample represents the Australian population demographically (age, gender and income) and geographically (i.e. proportion of the population by state and proportion residing in a large or mid-sized city). For the sake of brevity, we do not include descriptive statistics of our sample. However, our sample is demographically and geographically representative of the Australian population.

We compare our sample distribution with the distribution of the general Australian population in 2020 as estimated by the ABS across these key demographic and geographic variables. Figure 2 compares the marginal distributions across different age groups. Figure 3 compares the joint distributions across different gender and age groups. The marginal distributions across both age and gender for our sample are quite similar to the 2020 ABS projections. The average age of our sample is 47.0 years, whereas the average age of individuals ages 18 years and above per the ABS is 46.5 years. Similarly, 50.9% of the sample is female, compared to 50.5% of the general population. However, when we compare the joint distribution, the differences are more prominent. Younger men and older women are slightly overrepresented in our sample compared to the Australian population. However, we control for these differences through appropriate reweighting procedures in our analyses.

Figure 2. Sample and ABS distributions across different age groups.

Figure 3. Sample and ABS distributions across different gender and age groups.

Figure 4 compares the marginal distributions for household income across different income quintiles, where the quintiles are defined such that the top quintile earns as much as the other 80% of households combined. As a ratio, households earning in the highest quintile on average earn 9.5 times the average bottom quintile income. As is apparent from the plot, the two distributions are in close agreement. Relatedly, average household incomes for our sample are $1,667 per week, whereas the average earnings for full-time adults in Australia in May 2020 were $1,714 per week.

Figure 4. Sample and ABS distributions across different income quintiles.

Finally, Table 2 compares the marginal distributions across different Australian states and territories. Again, the distribution for our sample is in close concordance with the corresponding distribution for the Australian population, as estimated by the ABS. Furthermore, based on the Urban Centres and Localities (ABS 2017) data, 83.4% of the Australian population was living in a large city (with a population greater than or equal to 100,000), and 16.6% was living in a mid-sized, city (with a population less than 100,000). Our sample has the same exact distribution.

Table 2. Sample and ABS distributions across different Australian states and territories.

State or territory	Sample	ABS (2021)
Northern Territory	0.6%	1.0%
Australian Capital Territory	1.8%	1.7%
Tasmania	2.1%	2.1%
South Australia	7.6%	6.9%
Western Australia	10.1%	10.4%
Queensland	20.6%	20.1%
Victoria	24.6%	26.1%
New South Wales	32.6%	31.8%

In summary, we conclude that our sample is roughly demographically and geographically representative of the Australian population. Therefore, findings from our formal analyses of this data should offer credible and robust estimates of preferences of the Australian population for settlement in urban and regional centres.

Traditionally, individual preferences for settlement in urban and regional centres have been analysed using observed patterns of migration and settlement in the real world. Observational datasets commonly suffer from missing variables and/or measurement errors, due to the absence of key information, or imprecise measurement (e.g. Yu 2001). Additionally, observational data offers limited statistical control over the explanatory variables of interest. The issue can further be compounded by multicollinearity, where two or more explanatory variables are highly linearly correlated, and there is insufficient variation in the data to reliably estimate the marginal effect of each variable of interest (Greene 2003). Finally, observational data can only measure preferences under past or current market conditions, and do not offer a basis for predicting preferences under structural breaks, such as the ongoing COVID-19 pandemic.

As an alternative, we used SP experiments to collect data from participating individuals to measure their preferences for settlement in urban and regional centres. SP experiments are not subject to the same sources of bias as real-world market data. As they offer considerable control – each of the variables affecting the decision being defined explicitly by the analyst, and there are no omitted variables and no measurement error (Mitchell and Carson 2013, Louviere et al. 2000). In addition, SP experiments offer a basis for predicting the impact of structural breaks, such as the COVID-19 pandemic, on future regional settlement and migration patterns, by incorporating these breaks as additional control variables. SP experiments have been used widely in the areas of regional development and residential location preferences (see for example, Audirac 1999; Aung and Vichiensan 2019; Bottero et al. 2017; Ivanova and Rolfe 2011; Jiménez-Espanda et al. 2022; Meester 2012).

2.2. Interviews and Focus Groups

We collected additional qualitative data to identify key factors shaping the lived experience of residents living in smaller regional cities, and how these experiences contribute to resident attraction and retention, to inform future policies supporting sustainable long-term regional growth. The qualitative data collection was designed to examine the perspectives of both residents and stakeholders across a diverse mix of regional cities. We looked at both the key drivers of migration decisions – employment, housing and amenity – and the factors that shape decisions of whether or not to stay in smaller cities, including local services and social connectedness. In addition, we considered how regional residents feel about the prospect of regional growth.

Five case studies were selected across four Australian states: Albury-Wodonga (border town between New South Wales and Victoria), Cairns (Queensland), Mildura (Victoria), Whyalla (South Australia), and Wollongong (New South Wales). These locations were selected to reflect a broad range of features, including size, economic structure and current growth rate. Table 3 provides detailed geographic and demographic information about the case study cities, as compared to the whole of Australia.

Table 3. Case study city summary data.

	Albury-Wodonga	Cairns	Mildura	Whyalla	Wollongong	Australia
Distance by road to state capital (kms)	553/327*	1,681	542	385	85
Population – 2016	89,007	144,733	33,445	21,501	261,896	23.4m
Population growth 2006–2016	13%	27%	11%	2%	12%	18%
Median age	38	37	39	40	39	38
Indigenous population	2.8%	8.9%	4.6%	4.8%	3%	2.8%
Migrant population	11%	22%	13%	19%	21%	26%
Median weekly household income	$1,213	$1,339	$1,022	$982	$1,318	$1,438
Median monthly mortgage repayments	$1,408	$1,668	$1,232	$1,213	$1,928	$1,755
Median weekly rent	$245	$300	$225	$178	$325	$335
Median weekly rent/weekly household income	0.20	0.22	0.22	0.18	0.25	0.23
Tertiary educated residents	12%	13%	9%	7%	14%	18%

Source: Crommelin et al. (2022).

* Distances are Albury – Sydney and Albury – Melbourne.

In each case study city, a resident focus group was conducted, involving between 7 and 14 local residents per city, resulting in a total of 43 residents. Participants were invited from a range of cohorts including existing residents, overseas migrants and domestic migrants. For three of the focus groups (Wollongong, Cairns and Whyalla), the participants included a mix of existing residents, overseas migrants, and domestic migrants; for two (Mildura and Albury-Wodonga) the participants were a mix of existing residents and domestic migrants.

Residents were recruited to the focus groups via social and local media advertisements and through local contacts. From each focus group, 2 or 3 individuals were invited to participate in an individual follow-on interview, resulting in eleven extended personal interviews. These were aimed at gathering additional lived experience at a greater depth. Interviewees were selected to represent a broad range of residential experiences including long-term residents, domestic and international migrants.

In addition, in each city, 4–6 stakeholder interviews were conducted, amounting to a total of 26 interviews. Stakeholders were approached from the following sectors: local government, major employers, anchor institutions, chambers of commerce, government service providers, migrant support groups, community and cultural organisations.

The resident focus groups, resident interviews and stakeholder interviews were conducted between November 2020 and May 2021. As this was a period of time characterised by COVID-19 outbreaks and lockdowns, the local engagement process was delayed and ultimately the decision was made to conduct many of the interviews online via Zoom. While focus groups were held in person, a number of these needed to be rescheduled or delayed, and it is likely that uncertainty around restrictions and health risks of meeting in groups reduced participation levels.

3. Survey Findings

We begin by reporting findings from a descriptive analysis of the data collected by our nationwide survey. Roughly 84 percent of our respondents indicated that they are currently living in a large city (i.e. population greater than 100,000), and 16 percent are living in a mid-sized city (i.e. populations between 5,000 and 99,999). Respondents were subsequently asked to provide the names of other Australian cities they have lived in previously for more than 12 months, the main purpose for residing in that city, and the length of their stay. Based on respondents’ answers, we identified different patterns with respect to settlement and migration patterns, and their relationship with city size, as shown in Figure 5. Roughly 71 percent of our respondents have always lived in a large city, whereas only 8 percent have always lived in a mid-sized city. 9 percent have migrated from a mid-sized city to a large city, while a lower 6 percent have migrated from a large city to a mid-sized city. In general, the majority of mid-sized city residents have lived in a large city at some point in their past, but the majority of large city residents have lived in large cities their entire lives. Across our sample, roughly 29 percent have lived in a mid-sized city at some point in their lives.

Figure 5. Migration patterns as a function of current city of residence.

Table 4 shows the mean and median length of stay in a single city as a function of its size. On average, across all city sizes, respondents indicated that on average they move cities every 18.9 years. However, numbers vary quite significantly between large and mid-sized cities. Large city residents have an average length of stay of 20.9 years, compared to 12.9 years for mid-sized city residents. Based on these responses, it appears that larger cities offer more stable living conditions that afford the opportunity for households to settle for longer periods. In contrast, mid-sized cities appear to be less stable, resulting in greater attrition of local residents.

Table 4. Respondents mean and median of length of stay in each Australian city.

	Total sample	Large cities	Mid-sized cities
Mean	18.9 years	20.9 years	12.9 years
Median	13.0 years	16.0 years	8.0 years

Figure 6 plots respondents’ willingness to continue living in a mid-sized city if they live in one already, or their willingness to move to one if they presently reside in a large city. Of those currently living in a mid-sized city, 82 percent indicated that they are willing to remain living in a mid-sized city and only 7 percent indicated that they are unlikely to remain. On the other hand, of those who are currently living in a large city, 53 percent indicated a willingness to move to a mid-sized city, and only 22 percent indicated that such a move would be unlikely. In summary, these statistics suggest that a majority of Australian residents are open to the prospects of living in mid-sized cities.

Figure 6. Likelihood of staying in or moving to a mid-sized city.

Table 5 shows how reasons for migration and settlement patterns have varied across individuals with different histories. Across our entire sample, employment and proximity to family are the two most popular reasons individuals choose to live in a particular city. However, there are some interesting differences when we examine this information further, based on migration histories. As one would expect, individuals that have always lived in either a mid-sized city or a large city have most frequently done so to be close to family. Reasons for moving to a mid-sized city tend to include better quality of life, more affordable housing, and better prospects to raise children. In contrast, reasons for moving to a large city tend to focus on employment and education opportunities.

Table 5. Reasons for settlement in different cities, as a function of migration histories.

Relatedly, respondents were asked about the importance of different factors in general, when deciding which city to live in, and their responses are shown in Figure 7. Interestingly, quality of life, quality of healthcare, crime rate, cost of living, and housing costs are rated as the five most important factors, and proximity to family and employment are only the sixth and seventh most important factors, respectively. In combination with responses to previous questions, it appears that while factors relating to quality of life are important determinants of where individuals choose to live, they are not always the precipitating factor that causes individuals to move to another city.

Figure 7. Importance of different characteristics when deciding on a city to live in.

Respondents were asked to select the characteristics they think best describe a mid-sized city and a large city. In total, eighteen characteristics were presented to respondents. Figure 8 plots the proportion of respondents that selected a particular characteristic to describe mid-sized cities and large cities. Consistent with responses to previous questions, respondents perceive mid-sized cities to offer better quality of life, and large cities to offer better access to urban amenities.

Figure 8. Proportion of respondents that used particular characteristics to describe mid-sized cities and large cities in general.

Respondents were presented eight different types of policies that governments may use to encourage settlement in mid-sized cities, and asked to indicate which of these policies would encourage them to move to a mid-sized city. The responses are shown in Figure 9. Interestingly, the most popular policies pertain to provision of high-quality health care and support for post-retirement living, indicating that the majority of respondents in our sample view mid-sized cities as potential places of residence for when they are older and retired from the workforce.

Figure 9. Proportion of respondents that agreed that the particular policy would encourage them to relocate to a mid-sized city.

Data from the hypothetical SP scenarios in the survey was used in conjunction with other demographic and attitudinal information collected as part of the survey to estimate latent class choice models (LCCMs) of individual preferences for regional settlement. LCCMs are finite mixtures of discrete choice models. They were first developed in the field of marketing sciences as tools to identify relatively homogenous consumer segments that differ substantially from each other in terms of their behaviour in the marketplace (Kamakura and Russell 1989). They have since emerged as a very popular form of discrete choice model, finding application in a wide variety of disciplines (for example, pedagogy: Burke et al. 2015; transport: Ardeshiri and Vij 2019; tourism: Chen et al. 2019; and economics: Motz 2021). In our case, LCCMs allow us to identify segments in the population that differ in terms of their preferences for different kinds of large and mid-sized cities. We describe the general LCCM framework in Appendix 2. We present detailed estimation results in Appendix 3.

We estimated a number of LCCMs with different model specifications, where we varied the explanatory variables, the functional form of the utilities, and the number of classes. Our preferred model identified four distinct segments, or classes, in our sample population that differed in terms of their preferences for regional settlement, their demographic characteristics, and attitudes towards living in different cities. To make the description of classes easy to follow, we ordered the classes in terms of decreasing preferences for living in large metropolitan cities, and increasing preferences for living in regional mid-sized cities. Table 6 summarises these key differences across the four classes. Here, we summarise the key findings from our analysis.

Table 6. Narrative summary of different segments in the sample population.

	Class 1 Metropolitan enthusiasts	Class 2	Class 3	Class 4 Regional enthusiasts
Share of sample population	16 percent	21 percent	54 percent	9 percent
Preferences for city size	Prefer larger cities	Preference for smaller cities, ceteris paribus	Neutral	Very strong preference for smaller cities
Sensitivity to other city characteristics	Care about wages, distance to coast, and access to urban amenities, insensitive to other attributes	Care about unemployment rates, insensitive to other attributes	Sensitive to wages, living costs, housing costs, commute times, unemployment, and distance to coast	Insensitive to any other city characteristics
Settlement and migration history	Highly likely to have always lived in large cities, and reluctant to leave large cities	Highly likely to be currently living in large cities, but have lived in mid-sized cities in the past, and open to moving to one again	Proportionally split between large and mid-sized city residents	Most likely to be currently living in a mid-sized city, or have lived in one in the past
Demographic characteristics	Younger, higher-income professionals that are single or part of a couple, and with no children	More likely to be a mix of young individuals living in single or shared households, and middle-aged individuals living in households with children. University-educated and employed full-time in high-wage managerial or professional jobs in white-collar sectors.	Older, lower-income individuals without a college degree that are employed part-time or retired

Class 1: Comprised 16 percent of the sample population. Individuals belonging to this class displayed a distinct preference for living in large cities. These individuals were highly sensitive to average wages when deciding where to live, as reflected by a high demand elasticity of 1.6–5.1 (see Table A5 in Appendix 2). The impact of other characteristics was not found to be statistically significant. Individuals belonging to this class tended to be young higher-income urban professionals living alone or with their partners in households without children. These individuals had lived in large cities for most of their lives, and expressed reluctance to move to a mid-sized city.

Class 2: Comprising 21 percent of the sample population, individuals belonging to this class displayed a preference for smaller cities. These individuals were sensitive to unemployment rates, but the size of the marginal effect was small, as reflected by a low demand elasticity between −0.8 and −0.5. The impact of other characteristics was not found to be statistically significant. Individuals belonging to this class tended to be a mix of young individuals living by themselves or in shared households, and middle-aged individuals living in households with children. They were frequently university-educated, employed in full-time managerial or professional jobs in white-collar sectors such as information, media and telecommunications. These individuals were likely living in large cities at present, but had lived in a mid-sized city in the past, and appeared open to the prospect of moving to a mid-sized city again under the right circumstances. They did not show a strong sense of attachment to their current city of residence. They did not view large cities as great places to live, work and study. Similarly, they did not think large cities offer easy access to nature, are easy to get around, attractive, safe or clean. Across all classes, they placed the greatest importance on physical attractiveness, the presence of iconic places and landmarks, and a multicultural mix of local residents. They reported policies encouraging relocation to mid-sized cities were most appealing when they supported home ownership.

Class 3: Comprising 54 percent of the sample population. Persons belonging to this class did not display a distinct preference for living in either mid-sized or large cities. Rather, their preferences were based on trade-offs across other city characteristics. Average wages and cost of living were the two most important attributes, followed by housing costs, commute times, unemployment rates, and distance to coast. Individuals belonging to this class disliked living in industrial or agricultural centres, and showed a preference for mild temperate climates. Individuals belonging to this class tended to be older lower-income individuals without a university degree employed part-time or retired. These individuals were also open to the prospect of living in a mid-sized city. One in three individuals belonging to the class had lived previously in a mid-sized city. They placed a high importance on quality of life. They reported that policies encouraging relocation to mid-sized cities were most appealing when they provide support for post-retirement living.

Class 4: Comprising 9 percent of the sample population, individuals belonging to this class displayed a very strong preference for living in smaller cities, to the point where their preferences were seemingly insensitive to other city characteristics. These individuals were most likely to be living in a mid-sized city, and/or to have lived in one in the past. In terms of their demographic characteristics, they shared a number of similarities with Class 3 in that individuals belonging to this class also tend to be older, lower-income, without a college degree, and/or employed part-time or retired.

In summary, we found that three-quarters of those surveyed by our study appeared open to moving to a mid-sized city under the right circumstances. Some 21 percent of survey respondents would be encouraged to move to mid-sized cities if they could offer comparable employment opportunities to large-sized cities. Fully 54 percent of survey respondents viewed mid-sized cities as excellent places to retire, and would be encouraged to move there if they could obtain support for post-retirement living.

4. Interview and Focus Group Findings

Findings from our analysis of the qualitative data reinforced and added to the trends evident through our quantitative analysis of the survey data. Broadly, most interview and focus group participants agreed that large cities provided better employment and education opportunities, while mid-sized cities offered greater quality of life, and the decision on where to live could usually be explained as a trade-off between these two competing factors.

Participants highlighted the lack of comparable employment opportunities in smaller cities, and obstacles to establishing a long-term career path. The issue was particularly acute for participants in professional roles.

Because the big cities have the big dreams and big opportunity and the perspective, so I think [I might move in the future]. There’s still the big city drag … it depends on what you’re doing. If you’re doing servicing, teacher, nurses, maybe Wollongong you can stay here for long, but if you’re like me and doing a lot of professional services, so the market is pretty level. So, you want to feel like a career achievement … it’s small. I would admit that kind of, I’m thinking about that, because I cannot [retire] for ages. (Wollongong Resident Interviewee 1)

Essentially, I've been told that to advance in my job, someone has to die or retire (Albury-Wodonga Focus Group Participant).

I’m quite fortunate, I’ve got a couple of uni degrees and an MBA. I guess coming to a town like this, it was – I don’t know what the right word is but – unappealing I guess, the jobs market, compared to what would be available in a Melbourne or Sydney or even Adelaide for that matter. So you can see why young people leave. I certainly didn’t move here for work, it was for family reasons and everything, which were the right decisions for us as a family, but […] I’ve learnt to be probably more complacent career-wise (Mildura Resident Interviewee 1).

Both the survey data and the qualitative data support the conclusion that “quality of life” factors are important in attracting new residents to smaller cities.

[It’s] the coastal lifestyle, and a place where you can get everything you want and need within five minutes. (Wollongong Stakeholder Interviewee 3)

I decided … that there was no quality of life [in the big city]. [So] … it was lifestyle. I did not want to be working in the big city, commuting an hour, raising children (Cairns Stakeholder Interviewee 4)

As this second quote suggests, we would argue that quality of life is the “it factor” – i.e. the drawcard which prompts people to explore whether they can make other considerations work. Shorter commutes, more affordable housing, easy access to nature and a close-knit community were the primary points of differentiation between regional cities and major metropolitan centres identified by our participants. The ability to connect with others, form close relationships, and enjoy the friendliness and sense of community was a significant aspect of quality of life for many participants.

But I think what did surprise me was the people. We were talking about this the other day of how nice they are, they’re friendly. You don’t seem to get that in the city and that did surprise me, because I wasn’t expecting that, having got used to city living, so that was a nice surprise. (Albury-Wodonga Resident Interviewee 3)

Yeah, look, for a young 26-year-old coming into a community, I had no connections at all. I did find the first six to 12 months quite challenging and it was – there were many occasions when I went, “I’m out of here, I’m out of here” before even […] my three years are up. But once you make a connection, that’s a critical point. (Whyalla Stakeholder Interviewee 2)

In Wodonga we’d made all those connections and networks and you could walk down the road, go to the supermarket, you’d run into people you knew, you felt that sense of belonging. (Albury-Wodonga Stakeholder Interviewee 3)

Lack of services was signalled as a significant barrier to regional living, with health, education and transport services being mentioned most often. While services such as health and education are important in terms of the amenities they provide to the community, they are also a source of employment. Regional centres experience particular issues and challenges due to their smaller size and relative geographic isolation, and health outcomes in these areas have typically lagged those in larger metropolitan centres. Concerns around the availability and quality of health care in regional areas was a recurring theme in the interviews and focus groups. While health services were generally rated as reasonable in most centres, access to specialist services and GPs were issues of concern. The health needs of older people are especially important for any centres hoping to attract retirees, and therefore any strategy to encourage this latent demand for regional living among retirees would have to address these shortcomings head on.

So that’s a real disadvantage. Places like Whyalla too, the health services – they’ll have some specialists fly over, might be available one or two days a week. Come over from Adelaide into a very good hospital. However, the ability to retain specialists in the community now is very difficult. So they’ll sort of fly in fly out. So again, getting access to that sort of specialist healthcare can be a problem. (Whyalla Stakeholder Interviewee 6)

Cities with a major tertiary education provider (Cairns and Wollongong) saw that the institution brought in domestic and international students, created employment and generated reasons for young people to stay in their hometown. Where a tertiary institution had a satellite campus, positive impacts were less apparent, and a lack of variety of course offerings was noted as an impediment to the success of the institution in adding value to the community.

What we’re trying to do is really have a stronger reliance between the tertiary educators in our two cities and our industry and businesses, so we can hopefully try and open up opportunities so our children don’t have to go away. (Albury-Wodonga Stakeholder Interviewee 3)

It was a bit disappointing when the government gave the $5,000 scholarships to people who had to leave home to study, whereas they really should be rewarding students who are staying local and studying. (Mildura Stakeholder Interviewee 1)

I think from an education perspective it is quite supportive in that sense because it’s got the university here and what we see with a lot of first generation migrations, they are often trapped in that survival mode or difficulty of getting qualifications recognised or retraining. But we do see that elsewhere it’s often the children or the second generation that have that opportunity to start again or go through the education system. So the fact that there is really supportive education supports at all stages and the pathway to transitioning to higher education, whether that be at TAFE or through [Registered Training Organisations] or the university, I think is a positive that does keep a lot of people here when they can access those opportunities. (Wollongong Stakeholder Interviewee 2)

Transport within regional centres was seen to be easier and more convenient than in large cities, with shorter commute times seen as a positive. However, for Whyalla and Mildura, remoteness and access to state capitals was a concern. That the remoteness of Cairns was not viewed in this way demonstrates the complex trade-offs that regional residents make when considering the costs and benefits of remote living. People in the cities studied were more reliant on private transport than public transport, and commented that public transport was not well-used because it was inefficient.

The downside with Whyalla of course is that the airport, which is – has two carriers. One is Qantas and one is Rex Airlines. […] The airport has about 80,000 passenger movements per annum. Port Lincoln airport, which has the two same carriers, would average about 180,000. (Whyalla Stakeholder Interviewee 6)

It’s not demand. It’s about having the right routes, and having the right size bus, and having the right timetable. So, you have all these big buses. When you get on a bus at Trinity you have to go via the university … JCU. Then you’ve got to go via a shopping centre, and then you go into town. So, nobody’s going to get on those buses. Not unless you’ve got nothing to do for the day, then get on the bus. (Cairns, Focus Group Participant)

I caught the bus home last night, and I was the only person on the bus. This is – it was a 45 minute bus trip to get home, which is six kilometres away. So, you know. That’s an issue, I think. (Wollongong Stakeholder Interviewee 3)

While young and middle-aged individuals were unlikely to move to a regional centre in the absence of appropriate education and employment opportunities, our survey data found that roughly one-in-two Australians viewed mid-sized cities as excellent places to retire, and would be encouraged to relocate if they could get support for post-retirement living in terms of healthcare, home ownership and access to other amenities. This highlights the potential for efforts to actively target retirees as migrants. These efforts would be building on recent trends in migration patterns; regional cities have attracted high proportions of older adults in the last 15 years and are emerging as popular retirement destinations (Bourne et al. 2020; Colman 2021; Hugo et al. 2015; Vij et al. 2021). However, it is well documented that regional and remote cities suffer from a lack of aged and health care facilities for older Australians and that in order to encourage healthy ageing outside of metropolitan areas, aged care services in regional and remote settings must be improved (ACSA 2004; Colman 2021; OPAN 2019)

So, fifth generation. My father had a taxi business here, so my brothers live here. I left when I was 19, went around the world, lived in Sydney for 30-something years. Came back six years ago. So, we came back to retire here, and I still catch up with people I went to kindergarten with. So, yeah, lots of old Cairns friends. (Cairns, Focus Group Participant)

Affordable housing was seen as a clear advantage of living in a smaller city by many interviewees and focus group participants, regardless of their age, and most felt that housing remained more affordable than in the big cities, despite rising prices. At the same time, however, increasingly limited supply – particularly of rental properties – has threatened this benefit. The COVID-19 pandemic has increased demand for housing in regional cities, with housing prices rising disproportionately in these areas (Coulter 2021). While at the time of conducting the research, the case study cities had few cases of COVID-19 (Wollongong had recorded the largest number, at 36 cases) there were two impacts that were noted by participants: rising housing costs and the disruption caused by border closures. In relation to housing costs, participants noted that affordability and availability of housing stock had declined, and that low-income earners were being displaced as previously affordable areas became gentrified. Border closures were particularly concerning for border cities (Mildura and Albury-Wodonga), creating impediments to ease and freedom of movement – one of the key attributes of regional living that participants enjoyed.

The recent example of border closures has put that into sharp focus, hasn’t it, because we don’t recognise the border from a community perspective. We just get on with it. But when you slam it shut when serious stuff goes down, people can’t get to work, families can’t see each other, so a big impact on the community and the economy. (Albury-Wodonga Stakeholder Interviewee 2)

The qualitative data from our interviews and focus groups, taken during the heat of the pandemic, reflects and reaffirms pre-pandemic views on what people value about regional living and the drawbacks of regional living. In essence, our study shows that the pandemic drew focus to some existing issues – for example that people in regional centres value affordable housing and ease of movement – but did not change what people are looking for when deciding on a place of residence.

5. Conclusions

This paper set out to understand whether urban growth dynamics in developed economies are changing in the wake of COVID-19, and associated shocks to the economy. Florida et al. (2021) have argued that in the future, cities may be subject to a new set of growth processes, including population-wide aversion to crowds. If realised, such changes would challenge the centre-piece of Glaeser’s (2011) argument around the benefits of urban density, and the increased concentration of capital and human resources in the largest cities.

Our analysis suggests that the pandemic has not had a major impact on the importance of different determinants of residential location. In the second year of the pandemic, when all major Australian urban centres had been subject to pandemic-induced lockdowns, residents reported the factors most likely to influence their residential location were those evident over a considerable period of time – job opportunities, access to education, quality of services and the cost of housing. In broad terms, the negative diseconomies of agglomeration faced by the largest metropolitan centres in Australia have generated a preference for life in smaller urban centres, but these preferences have not been acted upon because of concerns with respect to economic opportunities and the adequacy of services. Greater investment in the infrastructure of these places – including their connectivity to global markets – would give expression to their potential for growth. Additionally, poor access to services, especially in health and education, is an area where there is a scope for public sector investment that can drive contemporary growth patterns. Another barrier to regional migration was seen to be access to large metropolitan centres; this could be addressed in those cities with proximity to a major city through improved transportation infrastructure such as high-speed rail services, such as that being planned in New South Wales (New South Wales Government n.d.).

While life-stage migration emphasises the impact of key life events in determining migration (such as leaving secondary school, family formation, and retirement), Wulff et al. have shown that mid-life residential mobility has been on the rise for the last 40 years due to increasing divorce and separation, and “empty nests" (2010). The elements of place that attract migrants differ depending on the migrants’ life stage, therefore strategies for attracting different groups need to be nuanced. Based on our analysis, we posit there are three distinct paths, for three distinct but interconnected groups, that could be pursued by government policies seeking to achieve population decentralisation and encourage settlement in mid-sized regional cities. We further posit that these paths will function more effectively if taken together.

Firstly, government policies could focus on the development of local employment opportunities in regional centres, targeted at the attraction and retention of young and middle-aged working-age individuals. However, these policies would largely be a continuation of previous government efforts at driving regional growth, most of which have seen limited success. The ability of governments to be able to create the kinds of employment opportunities that would attract migrants from large cities and retain residents already living in smaller cities still remains questionable. Nevertheless, the widespread adoption of remote working arrangements during the COVID-19 pandemic, and their potential continuation after the pandemic, offers new opportunities for encouraging settlement in mid-sized cities that offer better quality of life, particularly in areas within commuting distance of major cities (Autor and Reynolds 2020), or those supported by high-speed internet services. Li et al. (2022) note that fast rail links between major cities and proximate regional cities could assist in population shifts, a finding that concurs with our qualitative findings on transportation between cities and regional centres. However, Li et al. advise caution; multiple mechanisms are required to develop sustainable regional growth and relying on a narrow approach such as simply providing transportation services is unlikely to yield long-term results.

The second approach to regional migration that is supported by our findings is the development of government policies that accelerate establishing infrastructure for post-retirement living in regional Australia. In 2019 Hugo et al. argued for the significant benefit of older Australians (baby boomers) moving into regional centres. The existing pattern showed that baby boomers were already inclined to migrate to regional centres (especially coastal regional centres), and they represented a substantial opportunity for those communities. Hugo et al. argued that baby boomers are well educated, wealthy, active and experienced citizens who do not represent a drain on resources; they have the capacity to bring economic and social benefits to regional communities. However, they also noted that with scaled migration of older citizens to regional centres there will be an increased need for health and aged care services and workers in regional settings; as we have shown (ACSA 2004; Colman 2021; OPAN 2019), the current support services for older Australians in regional and remote settings do not adequately service existing populations and would certainly require development to meet any growth. This migration pattern can bring with it financial benefits through employment and economic stimulation. Our empirical findings indicate that policies aimed at attracting older Australians to the regions are likely to appeal to a larger proportion of Australians. Further, the approach builds on existing demographic processes and preferences in relation to ageing and migration and will require a refinement in existing policies rather than a radical transformation. However, the approach is not without its own challenges. While it will bring services and related employment to the regions, it must be noted that the costs of service provision are higher in smaller urban areas due to lower economies of scale, and this strategy will likely not generate growth in high value-adding industries in these regions. Further, a large proportion of jobs in this model are lower-skilled and lower-paid jobs and do not address Li et al.’s (2022) call for higher-skilled, higher-paid regional employment. Li et al.’s (2022) point that regional economies cannot be dependent on one industry, whether that is mining, construction or caring, applies here. For a (regional) economy to thrive, diversification of industry is important.

This leads to the third approach that seeks to address the gaps in education service provision that our qualitative findings found; increasing educational offerings in appropriately-sized and located regional cities will encourage higher-skilled and waged jobs and diversification of industry. The Australian Universities Accord Interim Report (Australian Government, 2023) identified the establishment of Regional University Centres (RUCs) and New Tertiary Study Hubs (NTSHs) as Priority Action 1. RUCs and NTSHs are place-based, community-led solutions that allow tertiary students to be supported to remain in their hometowns while completing their education. The report indicates that RUCs (in remote and regional areas) New Tertiary Study Hubs (in outer metropolitan and peri-urban areas) could “improve participation, retention and completion for students" outside of CBDs. These hubs will not provide significant employment opportunities such as a full campus would, but they could encourage students to remain in their hometowns while studying, and thereby increase the chances of them remaining there post-education, potentially reducing the “brain drain” effect. This strategy of developing aged care and educational industries side-by-side addresses Li et al.’s concerns regarding economic diversification and promotes the retention of a broader age demographic within regional and remote settings.

In summary, we find that most Australians agree that smaller cities and regional centres offer better overall quality of life. However, employment continues to be a key determinant of where people live, as well as a key barrier to relocating to smaller cities and regional centres that cannot offer comparable employment opportunities. This continues despite, at the time of the survey, the population having experienced the brunt of the COVID-19 pandemic and associated lockdowns and working from home arrangements. Our findings demonstrate that the pandemic did not cause a significant shift in the attractions and detractions of regional living; rather it highlighted what is important in these communities and why they are desirable. There are three broad mechanisms through which governments can stimulate population growth in regional centres, focused on employment, post-retirement living and education. Increased adoption of remote working arrangements could help attract employed individuals with jobs that have some ability to be done remotely, especially to regional centres located in close proximity to a large city. Increased support for post-retirement living can help attract older adults looking for quality-of-life benefits offered by regional centres, in particular coastal centres. Increased investment in higher education in regional centres can help increase the retention of young adults. In the long run, increased in-migration and decreased out-migration of these demographic groups could help create local employment opportunities that attract other individuals in the workforce with jobs that cannot be done remotely.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research was supported by the Australian Housing and Urban Research Institute [grant number 32262] and the Australian Research Council's Linkage Grant [grant number LP170100940].

Appendices

1. Australian cities ranked by state and population

Table

ID	State / territory	City	2021 Census Population
1	NSW	Sydney	4,830,000
2	VIC	Melbourne	4,818,000
3	QLD	Brisbane	2,447,000
4	WA	Perth	2,098,000
5	SA	Adelaide	1,368,000
6	QLD-NSW	Gold Coast – Tweed Heads	694,000
7	NSW	Newcastle – Maitland	508,000
8	ACT	Canberra – Queanbeyan	491,000
9	QLD	Sunshine Coast	383,000
10	NSW	Central Coast	343,000
11	NSW	Wollongong	306,000
12	VIC	Geelong	290,000
13	TAS	Hobart	227,000
14	QLD	Townsville	179,000
15	QLD	Cairns	153,000
16	QLD	Toowoomba	142,000
17	NT	Darwin	127,000
18	VIC	Ballarat	112,000
19	VIC	Bendigo	103,000
20	VIC-NSW	Albury – Wodonga	98,000
21	TAS	Launceston	91,000
22	QLD	Mackay	84,000
23	QLD	Rockhampton	79,000
24	WA	Bunbury	76,000
25	NSW	Coffs Harbour	74,000
26	QLD	Bundaberg	74,000
27	QLD	Hervey Bay	58,000
28	NSW	Wagga Wagga	57,000
29	VIC	Shepparton – Mooroopna	54,000
30	VIC-NSW	Mildura – Buronga	54,000
31	NSW	Port Macquarie	50,000
32	NSW	Ballina	46,000
33	QLD	Gladstone	45,000
34	NSW	Tamworth	44,000
35	VIC	Traralgon – Morwell	43,000
36	VIC	Warragul – Drouin	43,000
37	NSW	Orange	42,000
38	NSW	Bowral – Mittagong	42,000
39	NSW	Dubbo	41,000
40	WA	Busselton	41,000
41	NSW	Nowra – Bomaderry	39,000
42	WA	Geraldton	39,000
43	NSW	Bathurst	37,000
44	VIC	Warrnambool	36,000
45	WA	Albany	35,000
46	TAS	Devonport	32,000
47	NSW	Lismore	29,000
48	SA	Mount Gambier	29,000
49	WA	Kalgoorlie – Boulder	29,000
50	NSW	Nelson Bay	28,000

2. Econometric framework

Discrete choice models based on the theory of random utility maximisation (RUM) have emerged as the predominant econometric framework for modelling individual choices (e.g. Train 2009). Among the large family of RUM-based discrete choice models, latent class choice models (LCCMs) have been used to capture heterogeneity in preferences and discover behavioural variations among the population.

We begin with a description of the class membership model, formulated in our case as the multinomial logit function: (A1) $P\left(q_{\mathrm{ns}}=1\right)={\displaystyle \frac{\exp \left({\mathit{z}}_{\mathit{n}}^{\mathbf{\prime }}{\mathit{\gamma }}_{\mathit{s}}\right)}{{\sum }_{{\mathit{s}}^{\mathrm{\prime }}=1}^S\exp \left({\mathit{z}}_{\mathit{n}}^{\mathbf{\prime }}{\mathit{\gamma }}_{{\mathit{s}}^{\mathrm{\prime }}}\right)},}$(A1) where $q_{\mathrm{ns}}$ equals one if individual $n$ belongs to class $s$, and zero otherwise; ${\mathit{z}}_{\mathit{n}}$ is a vector of individual-specific demographic and attitudinal characteristics, such as age, household structure and perceptions of different urban areas; ${\mathit{\gamma }}_{\mathit{s}}$ is a vector of class-specific parameters denoting sensitivity to the individual-specific characteristics; and $S$ is the total number of classes. Note that the appropriate number of classes is determined exogenously, by comparing predictive performance and behavioural interpretability across models with differing numbers of classes.

Next, we describe the class-specific choice model. Note that each survey respondent is presented eight stated preference scenarios, where they are offered a choice between living in two different urban areas, and asked to indicate their preference. The dependent variable of interest is the preferred choice, denoted $y_{\mathrm{ntj}}$, which equals one if respondent $n$ over scenario $t$ chose alternative $j$, and zero otherwise. For a given respondent $n$ and scenario $t$, the class-specific choice model predicts the probability that urban area $j$ is preferred.

Let $u_{\mathrm{ntj}|s}$ be the utility of urban area $j$ for scenario $t$ and respondent $n$, conditional on the respondent belonging to class $s$, specified as follows: (A2) $u_{\mathrm{ntj}|s}={\mathit{x}}_{\mathit{n}\mathrm{tj}}^{\mathbf{\prime }}{\text{β}}_{\mathit{s}}+{ϵ}_{\mathrm{ntj}|s},$(A2) where ${\mathit{x}}_{\mathit{n}\mathrm{tj}}$ is a vector of attributes specific to urban area $j$, such as climate, population size and average incomes; ${\text{β}}_s$ is the vector of class-specific parameters denoting sensitivities to these attributes; and ${ϵ}_{\mathrm{ntj}|s}$ is the stochastic component of the utility specification, assumed for the sake of mathematical convenience to be i.i.d. Gumbel with location zero and scale one across urban areas, scenarios and respondents. Assuming the decision-makers are utility-maximisers, the class-specific probability that urban area $j$ is preferred is given by the logit expression: (A3) $P\left(y_{\mathrm{ntj}}=1|q_{\mathrm{ns}}=1\right)=P\left(u_{\mathrm{ntj}|s}\ge u_{\mathrm{nt}{j}^{\mathrm{\prime }}|s}\mathrm{\forall }{j}^{\mathrm{\prime }}=1,\dots ,J\right)={\displaystyle \frac{\exp \left({\mathit{x}}_{\mathit{n}\mathrm{tj}}^{\mathbf{\prime }}{\text{β}}_{\mathit{s}}\right)}{{\sum }_{{\mathit{j}}^{\mathrm{\prime }}=1}^J\exp \left({\mathit{x}}_{\mathit{n}t{j}^{\mathrm{\prime }}}^{\mathbf{\prime }}{\text{β}}_{\mathit{s}}\right)},}$(A3) where $y_{\mathrm{ntj}}$ is the dependent variable of interest as defined previously; and $J$ is the number of alternatives shown to the respondent for any scenario, equal to two in our case. The reader should note that heterogeneity in the decision-making process is captured by allowing the taste parameters ${\text{β}}_{\mathit{s}}$ to vary across classes.

Equation (A3) may be combined iteratively over alternatives and scenarios to yield the following class-specific probability of observing the vectors of choices $y_n$: (A4) $P\left({\mathit{y}}_{\mathit{n}}|q_{\mathrm{ns}}=1\right)=\prod _{t=1}^T\prod _{j=1}^J[P\left(y_{\mathrm{ntj}}=1|q_{\mathrm{ns}}=1\right){]}^{y_{\mathrm{ntj}}},$(A4) where ${\mathit{y}}_{\mathit{n}}=y_{n11},\dots ,y_{\mathrm{nTJ}}$; and $T$ is the number of scenarios shown to a single respondent, equal to eight in our case.

Equations (A1) and (A4) may be combined and marginalised over classes, to yield the unconditional probability of observing the vectors of choices ${\mathit{y}}_{\mathit{n}}$, which in turn may be combined iteratively over decision-makers to yield the following likelihood function for the data: (A5) $L\left(\text{β},\mathit{\gamma }|\mathit{y},\mathit{w},\mathit{x},\mathit{z}\right)=\prod _{n=1}^N\sum _{s=1}^S\left[P\left({\mathit{y}}_{\mathit{n}}|q_{\mathrm{ns}}=1\right)P\left(q_{\mathrm{ns}}=1\right)\right].$(A5) The unknown model parameters $\text{β}$ and $\mathit{\gamma }$ may be estimated by maximising the likelihood function. All models for this study were estimated using the software package PandasBiogeme (Bierlaire 2020).

3. Estimation results

We estimated a number of LCCMs with different model specifications, where we varied the explanatory variables, the functional form of the utilities, and the number of classes. Our dataset comprised 3,012 individuals, each of whom were shown eight different choice scenarios. To facilitate comparison, Table A1 enumerates for each model the number of parameters estimated, the log-likelihood at convergence, the Akaike Information Criterion (AIC), the Bayesian Information Criterion (BIC) and Consistent AIC (CAIC). Based both on statistical measures of fit and behavioural interpretation, we select the four-class LCCM as the preferred model specification. In terms of fit, the four-class LCCM has the lowest AIC, BIC and CAIC, and therefore performs the best. In terms of the signs and relative magnitudes of the different model parameters and the accompanying behavioural interpretation of each of the latent classes, results for the four-class LCCM also proved to be the most satisfying.

Table A1. Summary statistics for LCCMs with varying numbers of classes.

Classes	Parameters	Loglikelihood	AIC	BIC	CAIC
2	86	−14,719	29,609	29,736	29,695
3	171	−14,519	29,381	29,634	29,552
4	256	−14,319	29,150	29,528	29,406
5	341	−14,272	29,226	29,730	29,567

Table A2 presents estimation results for the class-specific choice model of city preferences, and Tables A3 and A4 present estimation results for the class membership model for the four-class LCCM. To enable comparison of the marginal effects of different city characteristics on settlement choices, Table A5 further enumerates average demand elasticities with regards to each of the continuous variables in the class-specific model.

Table A2. Class-specific choice models of individual preferences for settlement in urban and regional centres.

Variable	Class 1		Class 2		Class 3		Class 4
	est	p-val	est	p-val	est	p-val	est	p-val
Large city constant	0.00	–	−0.125	0.01	−0.301	0.00	−∞	–
Population
Less than 50,000	−0.573	0.00	0.141	0.00	0.000	–
50,000–100,000	−0.525	0.00	0.117	0.02	0.000	–
100,000–500,000	0.441	0.00	0.000	–	0.000	–
500,000–1,000,000	0.247	0.00	0.000	–	0.000	–
More than 1,000,000	0.858	0.00	0.000	–	0.000	–
Other attributes
Distance from the coast (100 km)	−0.190	0.00	0.000	–	−0.215	0.00
Unemployment rate (%)	0.000	–	−0.029	0.01	−0.042	0.00
Average home sales value ($100,000)	0.000	–	0.000	–	−0.082	0.00
Average annual full-time wages ($100,000)	1.888	0.00	0.000	–	1.396	0.00	Deterministic class
Average daily commute time (hours)	0.000	–	0.000	–	−0.520	0.00
Average monthly cost of living per person ($1,000)	0.000	–	0.000	–	−0.813	0.00
Urban centre classification
Industrial centre	0.000	–	0.000	–	−0.482	0.00
Agricultural centre	0.000	–	0.000	–	−0.398	0.00
Climate
Hot humid summer, warm winter	0.000	–	0.000	–	−0.320	0.00
Mild temperate	0.000	–	0.000	–	0.194	0.00
Alpine	0.000	–	0.000	–	−0.482	0.00

Table A3. Class membership model.

Variable	Class 1		Class 2		Class 3		Class 4
	est	p-val	est	p-val	est	p-val	est	p-val
Constant			1.645	0.04	3.091	0.00	−0.787	0.33
Age
25–34 years			0.000	–	−0.553	0.01	0.000	–
55–64 years			0.000	–	0.000	–	0.612	0.00
Indigenous Australian
Yes			0.000	–	−0.783	0.14	0.000	–
Education
Master’s degree or equivalent			0.000	–	−0.472	0.07	−0.524	0.11
Graduate Diploma			0.000	–	−0.419	0.10	0.000	−
Bachelor’s degree or equivalent			0.000	–	−0.412	0.01	0.000	–
Household type
Couple family with no children			−0.633	0.05	0.000	–	0.000	–
Occupation
Managers			0.964	0.00	0.000	–	0.000	–
State of residence
Western Australia			−0.767	0.14	0.000	–	−0.830	0.02
Tasmania			0.000	–	0.836	0.17	0.000	–
Australian Capital Territory			0.000	–	0.000	–	0.925	0.02
Live in a large city
Yes			0.000	–	0.000	–	−0.534	0.00
Willing to move from large cities
No			−0.678	0.14	−0.566	0.00	0.000	–
Dwelling type
House	Base class		0.000	–	0.232	0.18	0.000	–
Semi-detached or townhouse			0.000	–	0.425	0.09	0.000	–
Dwelling payment
Monthly rent/mortgage ($1,000)			0.073	0.08	0.000	–	0.000	–
Characteristics of current city of residence
Mixed function urban centre			0.000	–	0.000	–	−0.534	0.00
Unemployment rate (%)			0.000	–	−0.102	0.05	0.000	–
Climate – hot humid summer, warm winter			0.000	–	−1.226	0.01	0.000	–
Climate – warm humid summer, mild winter			0.000	–	−0.372	0.06	0.000	–
Climate – warm temperate			0.000	–	−0.310	0.06	0.000	–
Settlement and migration history
Always lived in large cities			0.000	–	−0.874	0.00	−0.480	0.11
Have moved from mid-sized cities to large cities			0.000	–	−0.639	0.04	0.000	–
Have moved between large and mid-sized cities and currently living in a large city			1.188	0.25	0.000	–	0.000	–

Table A4. Class membership model (continued).

Variable	Class 1		Class 2		Class 3		Class 4
	est	p-val	est	p-val	est	p-val	est	p-val
Satisfaction level
With employment			−0.225	0.08	0.082	0.23	0.000	–
With financial situation			0.000	–	−0.202	0.01	0.000	–
With state you live in			0.000	–	0.000	–	0.203	0.07
With relationship with loved ones			0.261	0.09	0.000	–	0.000	–
Barriers to migrate
Comfort and sense of belonging			−0.997	0.00	−0.227	0.08	−0.400	0.01
Dependencies			0.625	0.00	0.000	–	0.240	0.03
Motivators to migrate
Access to urban amenities			0.000	–	−0.532	0.00	−1.145	0.00
Culture and heritage			0.386	0.04	−0.510	0.00	0.000	–
Employment and education			0.000	–	0.158	0.04	0.000	–
Quality of life			0.000	–	0.833	0.00	0.000	–
Characteristics used to describe mid-sized and large cities
Great place to live, work & study
Mid-sized city			0.000	–	0.000	–	0.824	0.01
Large city			−1.000	0.03	0.000	–	−0.84	0.00
Variety of art and cultural offerings; good nightlife	Base class
Mid-sized city			0.000	–	−0.336	0.22	0.000	–
Large city			0.000	–	0.715	0.01	0.597	0.09
Easy access to nature; easy to get around; attractive; safe; clean
Large city			−0.947	0.06	−0.654	0.01	0.000	–
Good place for investment; long life expectancy; good place to retire; high quality of life;
Mid-sized city			0.000	–	0.998	0.00	0.000	–
Large city			0.000	–	0.000	–	−0.909	0.01
Policies to encourage move to mid-sized cities
Employment guarantee for spouse/partner			−0.654	0.02	0.000	–	0.282	0.17
Support for home ownership			0.510	0.09	0.000	–	0.647	0.00
Support for post-retirement life			0.000	–	0.423	0.00	0.000	–
Income tax incentives			−2.316	0.11	−2.873	0.00	−0.299	0.14

Table A5. Class-specific demand elasticities for residing in mid-sized or large cities with respect to different city characteristics.

	Class 1		Class 2		Class 3		Class 4
	Mid-sized city	Large city	Mid-sized city	Large city	Mid-sized city	Large city	Mid-sized city	Large city
Distance to coast	−0.50	−0.25	–	–	−1.14	−1.85	–	–
Unemployment rate	–	–	−0.58	−0.77	−1.47	−3.06	–	–
Average home sales value	–	–	–	–	−2.03	−7.30	–	–
Average annual full-time wages	5.06	1.57	–	–	5.11	10.66	–	–
Average daily commute time	–	–	–	–	−1.62	−6.11	–	–
Average monthly cost of living per person	–	–	–	–	−4.88	−10.46	–	–