1 Introduction

A common feature of international studies on housing and housing markets in risk areas is that the emphasis is placed on the devaluation of the house prices (Mueller et al., 2009; Koster & Van Ommeren, 2015; Zhang et al., 2010; Boelhouwer & Van der Heijden 2018; Durán, 2022), place attachment of residents in relation to natural or induced hazards (Bonaiuto et al., 2016; De Dominicis et al., 2015; Jansen, 2020), and coping strategies of the residents (Jansen et al., 2017; De Dominicis et al., 2015). Rather than focusing on the housing market or on the residents in a risk area, the present study explores ‘who’ actually moves to ‘which dwelling’ in a residential hazardous environment, with a special focus on mid-to-later life homebuyers of 50 years and older. After all, if ageing, sometimes risky and depopulating rural regions, are unable to retain their elderly or attract new elderly people, apart from the outflow of young and well-educated people, they will have even more issues. In fact, civic engagement in rural, sometimes depopulating areas, will depend to a large extent on the vital elderly, who have more time available than other age groups to be active in voluntary work and associations (Ubels et al., 2020).

In general, research on the ‘drivers’ of mobility that result in migration flows (Schewel, 2019) and on the relationship between relocation and life course stages or life events, have mainly been studied at the household level (Rossi, 1955), later completed with national socio-economic and demographic circumstances (Dieleman, 2001; Clark, 2013). However, few studies have investigated the individual characteristics and circumstances of mid-to-later life homebuyers who move to a dwelling in a risk area, the main topic of this paper. The ageing of the population in many Western countries, brought about a focus on residential mobility and housing needs of senior citizens in academic literature (Atkins, 2018). As a result, migration in the pre-retirement and retirement period, defined as ‘midlife migration’ and ‘later life migration’, gradually became an accepted research field in internal migration theory (Wulff et al., 2010; Stockdale & Catney, 2014; Atkins, 2018; Haacke et al., 2019; Stockdale & MacLeod, 2013). Wagner & Mulder  (2015) also stress the relevance of the life course approach to residential relocation, especially in relation to older adults, whose housing conditions depend on their health situation. Scholars agree that older people are inclined to stay in their familiar place and do not relocate until this is unavoidable, as they are attached to their home and environment (Haacke et al., 2019). Nevertheless, some triggers for moving can be distinguished, like individual characteristics, household composition and regional features that encourage the actual mobility behaviour (De Groot et al., 2011; Helderman et al., 2006). Stockdale & MacLeod (2013) refer to the pre-retirement phase of individuals aged between 50–65 years as a trigger for change, including a change in residential choice. This residential change at the pre-retirement and retirement stage could be, according to international studies in the US, UK, and Sweden, in favour of rural areas (Stockdale, 2006). Stockdale et al. (2013) also pinpoint the usefulness of research into the mobility of older migrants to rural areas. The prime reason for this is, that the ageing process in Western society is particularly visible in rural areas, which suffer from population decline and outmigration of youngsters in search for higher education and work (Champion & Sheperd, 2006; Bulder, 2018). Regarding moving to rural areas, Stockdale (2016) argues that birthplace return migration should also be considered. Swedish research confirms that retirement migration not only favours rural areas, but also the retiree’s place of birth (Lundholm, 2012). The question is whether persons at the pre-retirement and retirement stage still prefer the rural idyll or birthplace when safety risks are at stake. Several studies have shown that the age of individuals does influence the mobility decision to a risk area. For example, research by Zander & Garnett (2020) in an Australian sample, confirms that older people in their mobility decision assigning higher importance to almost every natural hazard, like floods, wildfires, earthquakes, and cyclones, than younger people.

This paper aims to contribute to the existing body of knowledge on the issue concerning the characteristics of the relocation of persons aged 50 years and older to a rural area, labelled as a risk area. In a study on midlife to later life home purchasers in the case of the Groningen earthquake region in the Netherlands, the following main question is formulated: “Which characteristics of mid-to-later life home purchasers, housing attributes and earthquake circumstances predict the housing transactions in a rural earthquake area in comparison to a rural non-risk area?” This main question results in research questions covering three steps that lead up answering the main research question.

  1. 1.

    Which socio-demographic characteristics of mid-to-later-life homebuyers are significant to understand the actual housing choice in the rural Groningen earthquake region instead of the rural non-earthquake Groningen region?

  2. 2.

    Which housing attributes are decisive when it comes to an actual home purchase in a risk area, struck by earthquakes, compared to a non-risk area?

  3. 3.

    Which earthquake circumstances of the original and new place of residence of the mid-to-later life homebuyer explain the housing transactions in the Groningen earthquake region?

The objective of this study is to identify significant characteristics of mid-to-later-life homebuyers and their home purchase in the rural, Groningen earthquake region. The paper is structured as follows. Section 2 contains an overview of the theoretical background and concepts of actual mobility in relation to rural areas and risks. Section 3 introduces the case study and data description, used variables and methodology. Section 4 presents the results and how they are related to existing academic research on the actual residential mobility to rural risk areas. Finally, the paper ends with conclusions (Sect. 5).

2 Theoretical background

2.1 Housing markets in risk areas

Most international research on housing and environmental risks focuses on the impact on housing markets and even more specific on house prices and value decrease of properties. For example, Porcelli & Trezzi (2019) stress in their study on the impact of earthquakes in Italian regions, that seismic events are considerable personal shocks at the local level, while the impact on the economic output and employment at the aggregated national level, is minimal. They even call for more research into the housing market’s response to earthquakes. Mueller et al. (2009) conclude in the case of wildfires, that especially after repeated wildfires in Southern California (US), housing demand fell immediately after each wildfire and house prices decreased by almost 23%, while, in the meantime, as new home purchasers move in, house prices recover. Zhang et al. (2010) also investigated the impact of three hazards, namely flooding, hurricanes and toxic chemicals, on housing values in Harris County, Texas. According to the developed model, hazard proximity and the perceived personal risks seem to compensate for the property value.

2.1.1 The housing market in the Groningen earthquake region

In contrast to the natural earthquakes in Italy, the earthquakes and subsidence in the province of Groningen in the north of the Netherlands are induced by gas extraction and have a considerable impact on building structures and residents up to the present day. In 2012 it became widely known that there is a relation between natural gas extraction and earthquakes due to soil subsidence. As a result of these earthquakes houses in the region have been physically damaged and various studies found empirical evidence that noticeable earthquakes led to house price decreases (Francke & Lee, 2015; Bosker et al., 2015; Koster & Van Ommeren, 2015). Induced earthquakes in the Netherlands, not only led to soil subsidence and therefore damage to the buildings, but also to a reduction in living comfort and risks of (fatal) injuries (Koster & Van Ommeren, 2015; Van der Voort & Vanclay, 2015). Boelhouwer & Van der Heijden (2018) conclude that the earthquakes in Groningen have a substantial impact on the housing market and liveability for the residents. Furthermore, some of the municipalities in the earthquake region experience an unbalanced population decline, which (might) manifest itself in fewer services, fewer amenities, youngsters leaving and, as a result, an ageing population (Bulder, 2018; Boelhouwer & Van der Heijden, 2018; Elshof et al., 2014; Haartsen & Venhorst, 2010). According to Boelhouwer & Van der Heijden (2018) subjective perceptions of the liveability and quality of life in the area, affect the reputation of the region and therefor play an important role in developments on the housing market. These perceptions of the Groningen earthquake region might result in a less attractive region to potential buyers of property, damaged or not. Moreover, Boelhouwer & Van der Heijden (2018) argue that variables measuring the damage of the dwellings, like the intensity and frequency of the earthquakes, and characteristics of the dwelling, such as year and type of construction, will influence the housing market and thus the migration in and out of the earthquake region. Jansen & Boelhouwer (2016) also conclude that the induced earthquakes, still have a significant impact on the residents, houses, and the regional housing market. Some dwellings are damaged beyond repair, causing the region to be less attractive for residents and migrants as a residential area. A combination of safety risks, fear, damage to the image of the region (Jansen & Boelhouwer, 2016) and the perceived quality of life (Boelhouwer & Van der Heijden, 2018; Jansen et al., 2017) might have negative consequences for the economic perspective of the area, and therefore, also have an impact on the regional housing market. That said, Porcelli & Trezzi (2019) found evidence in their study following 22 earthquakes in 95 Italian provinces that in most cases the influence on economic indicators is small. The papers of both Boelhouwer & Van der Heijden (2018) and Porcelli & Trezzi (2019) affirm there is a need for research on the sectoral responses of economic activity and behaviour after an earthquake, especially as reflected in the housing market.

2.2 Mobility in and to risk areas

After the earthquakes in Christchurch in New Zealand in 2010 and 2011, Dickinson (2013) researched patterns of residential mobility amongst 31 households that were obliged by official announcement to sell their property to the Crown. Here concepts of affordability and safety were highly valued by the mandatory residential movers purchasing post-earthquake property. Hunter (1998) concludes that, in a study on voluntary, internal migration at the national level in the United States, American counties with risky, environmental characteristics do not lose considerably more residents than areas without environmental hazards. However, Hunter (1998) also observes that these regions attract fewer new residents. Boelhouwer & Van der Heijden (2018) report an 8% reduction in satisfaction with the neighbourhood in the Groningen earthquake zone between 2012 and 2015. Consequently, they presume this may lead to outmigration, more vacancies, fewer businesses and services, employment decline and, finally, fewer settlers in this ‘unattractive’ area. To get insight in the residential mobility into the Groningen earthquake region, before and after the heaviest earthquake in 2012 (in Huizinge), Boes (2016) reported for the municipality of Loppersum, from 2009 until 2016, an 80% decline in the number of homebuyers originating from outside the province of Groningen. The reasons why homebuyers in general, after the earthquake in Huizinge, chose a house in the municipality of Loppersum, are diverse (Boes, 2016). The most frequently mentioned reasons were, in order of importance, the size of the house, work, relationship or divorce, having been raised in the area, family and the environment. Burningham et al., (2008) found in the case of flooding in the UK in the years 1998 and 2000, that homeowners are more risk aware than renters. Furthermore, Burningham et al., (2008) argue that homeowners tend to deny the flood risks from an economic perspective: they fear the possible negative impact on the insurance and value of their property. Both Mueller et al., (2009) and Burningham et al., (2008) suggest that new homebuyers might not be fully aware of the actual wildfire risks in the US or flood risks in the UK respectively in a high-risk area, due to a lack of information or denial of the risks. According to Jansen & Boelhouwer (2016), the likelihood that potential new homebuyers might be aware of and avoid the risks of the Groningen earthquake region, depends on the epicentre of the earthquake, the soil composition, and the year of construction of the dwelling. As a result of the earthquakes in the Groningen earthquake region one would expect an outflow of residents (Boelhouwer & Van der Heijden, 2018) and no new incomers. However, after the earthquake of Huizinge in 2012, according to an analysis of housing market transactions in the Groningen earthquake region over the years 2013–2015 houses are still being sold and the purchasers are mainly ‘local buyers’ (94%), originating from the earthquake region itself (Van der Kloet, 2018). De Kam & Meij (2017) also conclude the housing market in the Groningen earthquake region can be considered as a ‘closed regional market’. Outsiders of the earthquake region or province of Groningen neglected the region in this period. In contrast to ‘outsiders’, Burningham et al., (2008) argue that residents that had local experience with (minor) flood risks, indicate a lack of awareness, expressed by homeowners that consider their own dwelling as ‘not at risk’. Data on the migration flows in North-Eastern Groningen collected by Boumeester & Lamain (2016) show that 15% of the people who moved to the earthquake area in the period 2003–2008, relocated over a distance of 100 kilometres and more. In the period 2009–2014 this percentage decreased from 14.4 to 11%. Furthermore, they also conclude that, in line with De Kam & Meij (2017), the north-eastern part of Groningen has a ‘fairly regionally oriented’ housing market: approximately three quarters of the interregional removals in the earthquake area take place over a maximum of 25 kilometres.

2.3 Actual mobility of mid-to-later-life households

In research on the housing choice of mid-to-later life homebuyers and the resulting mobility or relocation, a meaningful difference to address is the intention to move and the actual moving behaviour (Haacke et al., 2019). In the present paper we focus on the actual housing choices of mid-to-later life home purchasers to buy a home and live in a rural risk area. The results of a study by De Jong (2020) on actual residential mobility of persons of 55 years and older, for instance, show that mid-to-later life homeowners are 0.66 times less likely to have moved in the last two years than tenants. Mellander et al., (2011) also conclude that ‘housing tenure’ has the most significant influence on the propensity to stay: homeowners are more likely to stay instead of moving. According to Helderman et al. (2006) homeownership is widely recognized as a barrier to migration, although De Groot et al., (2011) in their research find that homeowners are more likely to realise an intention to move than tenants who wish to relocate in the rental housing market. In addition, older people seem to be less mobile than younger people (Mulder, 2007; Lewicka, 2011), although Atkins (2018) found that the actual residential mobility differs in the midlife and later life phases of the life course: in the city of Perth (Australia) the actual mobility is concentrated in the pre-retirement stage (55–64 years) and amongst the older aged seniors (over 85 years). Haacke et al., (2019) also conclude in their study on older aged residents (age 60+) that age influences mobility: they found a mobility peak in the 65–70 age group.

In general research on residential mobility of older adults consider a life course approach (Clark, 2013) assuming that life events can cause a disbalance that might be resolved with a residential move. According to the Developmental framework of migration at later life of Litwak & Longino (1987), comfort and amenities influence the migration at the early retirement phase. At advanced age, residential mobility is often linked to a move closer to children and family to help with or receive care. Finally, the last move concerns institutional care. In the Theoretical model of elderly migration process, Wiseman (1980) introduced the distinction between local movers and long-distance moves and the associated triggers, such as a desire for amenities, need for assistance or a long-distance return migration. This latter form includes return to place of birth or area where childhood was spent. Apart from life events as triggers for move, De Jong (2020) argued that especially factors concerning the dwelling influence and predict the residential mobility of adults aged 55-years and older.

Based on the above, this study focuses on the actual home purchasing behaviour of mid-to-later life homebuyers in a rural, risk area, of which some parts even have to deal with population decline. It is important to realize that buying a home can be residential mobility, defined by Helderman et al., (2006) as movements on a shorter distance, namely a residential move over less than 35 km, but also internal migration, defined as changing home over longer distances. Helderman et al., (2006) argue that, under normal circumstances, a residential move at this shorter distance can be explained by housing motives, while a longer distance is mainly motivated by job change. A study of moving distance and housing attributes therefore might clarify who the homebuyers are behind the housing transactions.

2.4 Characteristics influencing residential mobility of mid-to-later life homebuyers in and to a rural area

To model the actual residential mobility process, Dieleman (2001) mentions housing choice at the household level and mobility behaviour as core elements of residential relocation. Next to the matching of households (‘who’) and dwellings (‘what’) Dieleman (2001) and Stockdale & Catney (2014) suggest that the context of the local housing market and regional differences in circumstances need to be included. In their study on residential mobility, Clark et al., (2006) emphasize that not only housing attributes, but also neighbourhood characteristics should be taken into account. Bijker et al., (2012) find that amongst motives for moving to a specific rural area, housing characteristics are the main motivation, directly followed by the physical qualities of the environment and personal reasons. Based on research on rural migration to less popular and popular areas, Bijker et al., (2013) also conclude that different people migrate to different areas, as the movers have different individual characteristics, motives, and values. Scott et al., (2017) stress in their study that, besides background characteristics of movers, the actual housing characteristics of the home should be incorporated as a new dimension in rural mobility theory. Scott et al., (2017) argue that the decision to move to the countryside is usually dominated by the rural idyll, while the importance of housing characteristics is often under researched; a case study into rural residential mobility in the Irish context shows, that almost 60% of the movers from an identical rural region, mention housing characteristics, in favour of social, economic, or physical factors, as the main driver for their move. Hjort & Malmberg (2006), for instance, indicate that studies in Sweden also reveal that housing and environmental requirements are of substantial importance for rural residential mobility in the country. In accordance with other residential mobility research, De Jong (2020) introduces three categories of variables to explain the relocation of later-life adults: individual characteristics, dwelling characteristics, and neighbourhood characteristics.

The present paper on relocation to a rural, risk area, alongside to the individual characteristics (Bijker et al., 2013) and housing characteristics (Scott et al., 2017), also refers to the need to incorporate regional circumstances (Dieleman, 2001), akin to work by De Jong (2020), who utilised the so-called neighbourhood characteristics. According to research on rural residential preferences in Ireland, Bullock et al., (2011, p. 703) observed that one of the main factors that determine the preference for a housing choice in a rural residence, rather than the design of the house, is the house location, defined as ‘social and physical characteristics of the rural destination’.

To answer the question of ‘who’ moves to ‘what’ and ‘where’ in these specific circumstances, requires exploring the influence of the risks of earthquakes on the actual housing choice and the housing market. The assumption that the actual housing choice might be influenced by individual characteristics, housing attributes and/or earthquake circumstances is based on the abovementioned theoretical background.

3 Methods

3.1 Data description and variables

To get a more profound understanding of residential mobility of mid-to-later life homebuyers in a risk area, this paper compares homebuyers in the rural Groningen earthquake area, with those who buy a dwelling in the rural non-earthquake municipalities of the province Groningen. To this end, we used a dataset of all housing transactions in the province of Groningen, registered by the Land Registry of the Netherlands (the so-called ‘Cadastre’) from 17 August 2012 until 2019. This date of 17 August 2012 is the day after the relatively severe earthquake measuring 3.6 on the Richter scale occurred in Huizinge (KNMI, 2020). It is nationally accepted as the starting point of the awareness of the potential damage to property from earthquakes in the Netherlands (Van der Voort & Vanclay, 2015). Our dataset includes characteristics of homebuyers of 50 years and older related to the purchased property in 22 rural municipalities of the province of Groningen. It is enriched with variables concerning the earthquake conditions in the region, like the Peak Ground Velocity (PGV). This PGV-value reflects the impact of earthquakes, based on the location of the homes and the epicentres, depth, and strength of the earthquakes. The PGV is a cumulative value calculated from earthquake data of the Royal Netherlands Meteorological Institute KNMI (KNMI, 2020; Koster & Van Ommeren 2015). During the research period 6,082 housing transactions have been recorded in the rural earthquake and non-earthquake area in Groningen, involving homebuyers of 50 years and older, indicating 27.4% of all the housing transactions in the province. The variables in our dataset consist of a combination of key variables concerning socio-demographic (personal) characteristics, housing attributes and circumstances concerning the location of the (previous) dwelling, related to possible population decline and/or earthquakes. As the data covers all actual housing transactions, there is no need to deal with possible biases of representativeness of a sample, which is a major benefit for the results of our analysis. Logistic regression analysis helps to identify and combine the different types of statistically significant characteristics of the midlife and later-life homebuyers, housing attributes and characteristics of the place of residence in the rural region of the province of Groningen. Each of the characteristics might be decisive.

3.1.1 Socio-demographic characteristics

The socio-demographic characteristics collected in the dataset include the variables age category, homeownership (single or multiple homebuyer(s)), birthplace region, starting homebuyer (also from a previous rental home step, yes/no), previous residence (located in the rural earthquake region, city of Groningen, rural non-earthquake region (= rest of province of Groningen), elsewhere in the northern Netherlands, (the surrounding provinces of Friesland or Drenthe), or Rest of the Netherlands or elsewhere, and finally, moving distance (in km) . Next to the individual characteristics, moving distance might be an important indicator or characteristic of the mid-to-later life home purchasers moving to the earthquake region or non-earthquake region.

3.1.2 Housing attributes

A meaningful insight in rural mobility research is the influence of physical housing characteristics on the choice for moving to a rural residence (Scott et al., 2017).Bijker et al., (2013) emphasize in their work on the migration to popular and less-popular rural areas the role of the average house price as a derivative of the surroundings and housing characteristics, like housing size and housing quality. Although Scott et al., (2017) consider this assumption a first step to include housing characteristics in the decision-making on rural migration, the description of the housing attributes should be more detailed. According to them it is not just the green and natural environment that motivates people to move to rural areas -as is often assumed in the rural migration literature- but more importantly, it is the wish for the extension of the internal and external private environment of the house, that pulls the movers to the countryside.

The housing attributes used in the present study are in every way more specific. Alongside variables regarding the individual house purchase price, other housing attributes have been used, such as housing type, lot size (mentioned by Scott et al., (2017) as the external private space), living space (mentioned by Scott et al., (2017) as the internal private space) and construction year of the purchased dwelling, recalculated as home age at the year of purchase. In line with the outcome of research of Bullock et al., (2011), which demonstrated that the preference for housing in a rural area mainly is determined by the social and physical characteristics of the destination, we use areas experiencing population decline, recognized by the Dutch government, as a characteristic of the housing location in the housing attribute ‘recognized area with population decline’, yes or no.). The fact that some rural parts of the province of Groningen also have been identified as a region with population decline is likely to play a role in the decision of a mid-to-later life homebuyer when purchasing a home in the rural area (Elshof et al., 2014; Boelhouwer & Van der Heijden, 2018). Of the recognized areas with population decline established from 2015 (revised on 1 January 2018), five municipalities belong to the earthquake area and five municipalities belong to the non-earthquake area (see Appendix 3).

3.1.3 Earthquake circumstances of the location of the dwelling

The last group of variables concerns the earthquake conditions homebuyers need to consider when purchasing a house since 16 August 2012. We define earthquake circumstances in this context as physical and economic circumstances that as result of the earthquakes, could influence the decision whether or not to buy a home in the earthquake area. For instance, we expect that the decision to buy a home in the earthquake region, will be influenced by the year of purchase of the home over the period 17 August 2012 until 2019. Over time people may have gradually become aware of the impact of the earthquakes due to the first formal acknowledgement of the risk of the earthquakes and the possible damage to the houses in the region by the government in January 2013 (Jansen et al., 2017). The government regulations, namely the Value Decrease Scheme (see Sect. 3.2) and the Damage Compensation Regulation, which came into effect in 2014, constitute the first, official confirmation of the possible damage or value decrease of the houses because of the earthquakes in the region. Next to the year of purchase, we expect that the Peak Ground Velocity (PGV) (Koster & Van Ommeren, 2015) the homebuyer was confronted with at the previous residence, influenced the decision of mid-to-later life homebuyers to purchase a home in the earthquake region or not. After all, the mobility decision is influenced by the experience people have with specific hazard risks (Zander & Garnett, 2020; Burningham et al., 2008).

In this way we use a combination of physical and/or technical circumstances in terms of the Peak Ground Velocity (PGV) and the influence of the earthquakes on the economic circumstances in terms of the number of homes sold on the housing market during the research period. One might expect that during the research period, variables like the year of the home purchase and the cumulative Peak Ground Velocity (PGV) of the location of the homebuyer’s previous dwelling, will have an impact on the prediction of buying a home in the earthquake region, or in the non-earthquake region .

3.2 Recognized earthquake area based on the Value Decrease Scheme

The housing market in the Groningen earthquake region can be characterized by a relatively high ratio of owner-occupied homes; 64% of the houses in the eleven municipalities is owner-occupied (CBS, 2020), while in the province of Groningen in total, this percentage is on average 54,8% (CBS, 2020). More specifically, the earthquake region has a relatively high proportion of single-family dwellings: the share of single-family houses varies from at least 70–95%, depending on the municipality (SPG, 2016). As result of the earthquakes, the Dutch Petroleum Company (NAM) launched in 2014 the so-called “Value Decrease Scheme” to compensate inhabitants who sold their houses located in the Groningen earthquake region for any loss of value. Initially the Value Decrease Scheme was applied to homes in eight recognized earthquake municipalities, afterwards eleven municipalities have been recognized as the Groningen earthquake region. Our control region, the rural non-earthquake area also consists of eleven municipalities in the province of Groningen (for specification, see Fig. 1 and Appendix 1).

Fig. 1
figure 1

The rural Groningen earthquake region versus non-earthquake region in the province of Groningen, situation in 2013

(Source: www.provinciegroningen.nl, design by Geodienst/University of Groningen, 2021)

Full size image

3.3 Research methods

Because our dependent variable is dichotomous, namely ‘Purchase of a home by a mid-to-later-life homebuyer in the rural Groningen earthquake region’ (value = 1) versus ‘Purchase of a home by a mid-to-later life homebuyer in the rural non-earthquake region’ (value = 0), we use logistic regression analysis. Independent explanatory variables are categorized into socio-demographic characteristics, housing attributes, and earthquake circumstances.

We ran logistic regressions on each of the three categories of the characteristics in line with their theoretical importance: (1) socio-demographic characteristics of the homebuyers, (2) housing attributes of the purchased homes and (3) the earthquake circumstances in relation to the place of residence and year of purchase. If an independent variable has a positive B coefficient, this indicates that an increasing value of the variable also increases the likelihood of the dwelling having been purchased in the earthquake region. The opposite is true for a negative B coefficient: in this case an increasing value of the independent variable means the probability of having bought a dwelling in the non-earthquake region, increases. The best fitted logistic regression models, based on the Hosmer and Lemeshow Goodness-of-Fit-Test, are presented in respect of the three categories of characteristics in three cumulative steps. The Hosmer and Lemeshow Goodness-of-Fit-Test in the SPSS Binary Logistic Regression analysis measures whether the model and used independent variables fit well with the data (Agresti & Finlay, 2009; Lyman Ott & Longnecker, 2010; Sieben & Linssen,2009). At first the model based on the socio-demographic variables is introduced (Model I). In the next step the housing attributes were added (Model II) and finally, the earthquake circumstances complement the last model, Model III. After each step the ‘goodness of fit’ and explanatory value of the model were estimated. We checked for multicollinearity in the final model of socio-demographic variables, housing attributes and earthquake circumstances of the location of the dwelling.

4 Results

4.1 Descriptive statistics

This section presents the outcomes of the performed logistic regressions in respect of the probability that persons of fifty years and older will purchase a home in the Groningen earthquake region based on the three types of characteristics. The dependent variable is the probability that mid-to-later life movers purchase a home in the rural Groningen earthquake region against the rural non-earthquake region in the province of Groningen.

In order to define the relevance of the significant socio-demographic characteristics, housing attributes and earthquake circumstances with regards to the choice to purchase a dwelling in the earthquake region or in the non-earthquake region, the description of these independent variables is reported in Tables 1 and 2, and 3, respectively. In the logistic regression models, all cases with missing values for the independent variables have been treated listwise, in other words, they were excluded from the analysis. The used dataset consists of information on relevant socio-demographic characteristics of the mid-to-later life homebuyer including age category, starting homebuyer yes or no, single or multiple homebuyer(s), previous residence in the Netherlands, previous residence in the Groningen earthquake region, birthplace region and moving distance (see Table 1). Table 1 clarifies that most of mid-to-later life house purchasers belong to the category of younger midlife homebuyers, especially in the earthquake region: 54.9% of the home purchasers are in the age group of 50–60 years, and afterwards in a decreasing trend, the older age groups. It is noteworthy, that both the share of single homebuyers and starting buyers is relatively high in the earthquake area. More than 75% of the 50 + aged homebuyers originate from the Groningen province, while 85.5% of the middle-aged and older home purchasers with a previous residence in the earthquake region, purchased their next home in the earthquake region. This internal relocation is also the case for homebuyers in the Groningen non-earthquake region (90.5%).

Table 1 Descriptive statistics of the housing attributes concerning housing transactions of mid-to-later life homebuyers in the rural Groningen earthquake and non-earthquake region from 17 August 2012 until 2019 (N=6,082).
Full size table

Of all housing transactions by mid-to-later life homebuyers in the province of Groningen over the period of 17 August 2012 until 2019, almost 65% had a previous residence in a non-earthquake region. To summarize the divergent known birthplace regions of the homebuyers, the number of birthplace regions has been reduced to five, of which the largest share concerns the ‘Other Groningen region’ which includes the city of Groningen. Next to the purchase price of the house, information about the lot size of the house, living space of the house, the age of the residence at the year of purchase, construction year, housing type and year of home purchase are also indicated in the used dataset of housing transactions in the province of Groningen (see Table 2). Although Bijker & Haartsen (2012) argue that the (average) house price alone shows the value buyers attribute to their home and environment, as well as to the size and quality, the abovementioned dataset gives us more specific information about the housing characteristics. Scott et al., (2017) also emphasize the importance of the home and living environment, as well as their submissive role in the rural migration research topics. The average moving distance corresponds with the results of Boumeester & Lamain (2016) on movements in the earthquake region over the period of August 2012 until mid August 2015: three-quarters of the movements take place within a radius of 25 km.

Table 2 Descriptive statistics of the housing attributes concerning housing transactions of mid-to-later life homebuyers in the rural Groningen earthquake and non-earthquake region from 17 August 2012 until 2019 (N=6,082).
Full size table

Table 2 reports the average house price and other housing attributes of the purchased dwellings in the rural areas of the province of Groningen in the specific research period. The average housing construction age at the time of purchase is predominantly almost 50 years. A detached house is the most common type of house purchased in the Groningen countryside (42.6%), while nearly a fifth consists of apartments. The homes purchased are evenly distributed among areas with or without population decline.

Table 3 presents the descriptive statistics concerning the earthquake circumstances of the house purchases by mid-to-later life homebuyers in the specific period from 17 August 2012 until 2019. Over 36% of all the housing transactions of 50 + aged home purchasers in the rural part of the province of Groningen took place -at that time- in the earthquake zone. The cumulative PGV-values in the data vary between zero and a maximum of forty, on average the PGV-value fluctuates around 9 to 10, which corresponds to a mean impact class of ‘moderate’. However, it is good to keep in mind, that the average is based on all housing transactions, in both earthquake and non-earthquake zones. The last indicator that might predict the probability of a home purchase is the year of home purchase: an upward trend in home sales by mid-to-later life homebuyers in the research period might indicate a recovering housing market in the rural Groningen region, with or without earthquakes. In fact, the entire Dutch housing market from 2013 to 2018 was characterized by a state of recovery after the 2008–2013 housing crisis (DNB, 2017; CBS, 2018).

Table 3 Descriptive statistics of earthquake circumstances concerning housing transactions by mid-to-later life home purchasers in the rural Groningen earthquake region and non-earthquake region from 17 August 2012 until 2019 (N = 6,082)
Full size table

4.2 Which characteristics determine a house purchase in the rural Groningen earthquake region?

The next step in our analysis is exploring which characteristics most strongly predict the dependent variable, i.e., a housing purchase by 50 + aged people in the recognized Groningen earthquake region or in the non-earthquake rural region in the province of Groningen. In this section we present three logistic regression models to this end. The first model, Model I in Sect. 4.2.1 exposes the prediction of a housing transaction in the Groningen earthquake region only based on the socio-demographic characteristics of mid-to-later life homebuyers. The second model, called Model II in Sect. 4.2.2 is based on the same socio-demographic characteristics included with the housing attributes of the purchased dwellings. The third and final model in Sect. 4.2.3, Model III shows the impact of adding the earthquake features around the housing transaction to the predictive ability of this model, in the sense that a housing transaction in the earthquake region will or won’t occur. This last stepwise model shows the added explanatory value of the different characteristics in connection with the prediction whether a housing transaction of mid-to-later life homebuyers will be made in the earthquake region or in the non-earthquake region.

4.2.1 Who purchases a dwelling in the rural Groningen earthquake region?

Model I shows the prediction of a housing transaction in the Groningen earthquake region versus non-earthquake rural region in the province of Groningen, based on the socio-demographic characteristics of a mid-to-later life homebuyer (see Table 4).

Table 4 Logistic regression model on the prediction of a home purchase of a mid-to-later life homebuyer in the rural Groningen earthquake region based on socio-demographic characteristics (Model I)
Full size table

The analysis shows the probability that mid-to-later life multi-person buyers purchase a home in the non-earthquake region increases, whilst single buyers of 50 years and older are more likely to buy a dwelling in the earthquake region. If the earthquake region is defined as a less-popular area, there are similarities in the residential mobility research outcomes of other studies. There seems to be a considerable overlap with, for example, results of Bijker et al., (2012) concerning the rural, less-popular areas selected at that time: three out of the four less-popular municipalities selected were subsequently classified as earthquake municipalities. In this way, our findings are in line with the conclusion of Bijker et al.,(2012) that movers to less-popular, peripheral areas more often are singles, including the oldest age groups.

Another significant socio-demographic characteristic of the home purchaser in our analysis, is ‘the previous residence’. If the household originates from one of the municipalities of the earthquake area, the probability of a move within that same rural earthquake region increases, while homebuyers from the non-earthquake region in the province of Groningen, and other provinces, even from surrounding provinces, choose in favour of the non-earthquake region in the province of Groningen. These outcomes are in line with previous research results concerning the existence of a ‘closed regional housing market in the earthquake region’, in other words, people already living in rural, risk areas are likely to be the purchasers of a house in the earthquake region (De Kam & Meij, 2017; Van der Kloet, 2018). De Groot et al., (2012) define them as ‘local movers’ and conclude that their connection to a specific rural residential area is an important driver in actually realizing housing in one’s favoured rural place. Stockdale (2006) also observed that less-popular rural areas particularly pull local migrants.

However, in contrast to the study by Bijker et al., (2013) on the relationship of age and migration to less-popular rural areas versus popular areas, we found no statistically significant indication that the age of mid-to-later life homebuyers influence the actual move to the earthquake region. According to the results of the study by Bijker et al., (2012) the physical qualtities of the environment are decisive into the choice of middle-aged people to move to a specific less-popular rural area.

Finally, it appears the birthplace region of the mid-to-later life home purchaser is a strong indicator for the prediction of a home choice in the earthquake region. It seems that mid-to-later life homebuyers with a birthplace in East Groningen, mainly consisting of non-earthquake municipalities, overall buy a dwelling in the non-earthquake region (see for specific details Appendix 2). However, households with a birthplace in Delfzijl and surrounding municipalities, overall earthquake municipalities, and the Other municipalities of the province of Groningen, partly earthquake and partly non-earthquake region, purchase a home in the earthquake region more often than people with a birthplace region outside the province of Groningen. This outcome regarding the place of birth is in accordance with the conclusions of Lundholm (2010) on migration in Sweden of people belonging to the age group of 55–70 years old: especially people born in the rural areas are oriented towards the rural region and susceptible to return to their parish of birth. De Groot et al., (2012) also conclude that the place of origin is a determining factor for the actual move of migrants to the preferred rural area.

All in all, the socio-demographic characteristics of the mid-to-later life home purchaser demonstrate a good fit of the model based on these independent variables (Nagelkerke R2 = 0.632). This result is similar to the conclusion of De Groot et al., (2012) and Elshof et al., (2014) that individual characteristics are important to actually realize one’s rural residential preferences.

4.2.2 Who purchases which dwelling in the rural Groningen earthquake region?

In the second step of our analysis, the housing attributes of the purchased dwelling are added to the logistic regression model (Model II, Table 5). This resulted in a minor increase of the Nagelkerke R2 (now 0.645) indicating that housing attributes do not add a lot of extra explanatory value to the model. Still there are some significant features of the dwelling, that influence the choice of the mid-to-later life home purchaser to buy a house in the earthquake region or not. Following Model II, an increase in home purchase prices means a decrease in the probability of mid-to-later life homebuyers selecting a dwelling in the earthquake region. In this respect, again it is interesting to draw attention to the research of Bijker et al., (2012), on migration to less-popular rural areas in the Netherlands. They also found that people, indicating the ‘low housing purchase price’ as an important motivation to migrate, moved more often to less-popular areas. The lot size of the house, defined by Scott et al., (2017) as the ‘external private space’, seems one of the predictors of rural mobility to the non-earthquake regions although it is a small significant effect (-0.010): the larger the lot, the sooner people of 50 years and older will opt for a home in a non-earthquake region of Groningen. The opposite is true for the age of the dwelling at the time of purchase. To obtain a continuous variable, this last variable is the transformation of the construction year of the dwelling. Model II shows that the older the dwelling at the time of purchase, the larger -although it is only a slight increase- the probability of a mid-to-later life homebuyer purchases a dwelling in the earthquake region. Lastly, the housing types significantly influence the probability of a home purchase by a mid-to-later life homebuyer in the earthquake region. We find that mid-to-later life homebuyers of an apartment or semi-detached house (including terraced/corner house) in comparison to purchasers of detached houses, are more likely to move in favour of the non-earthquake region.

Table 5 Logistic regression model of the prediction of a home purchase of a mid-to-later-life homebuyer in the Groningen earthquake region based on socio-demographic characteristics and housing attributes (Model II)
Full size table

4.2.3 Who purchases which dwelling under specific earthquake circumstances in the rural Groningen earthquake region?

Model III (see Table 6) includes variables concerning the earthquake circumstances in the area, based on the assumption that these additional characteristics account for the difference between moving of middle-aged and older people to a rural area with or without risks. Overall, the Nagelkerke R2 increased again, despite the missing values and, consequently, the reduction in the number of available housing transactions (N = 5,233). It reveals that the earthquake characteristics contribute to the explanatory power of Model III. Two earthquake circumstances have a significant effect: the year of purchase of 2016 compared to reference year 2018 and the experience of the homebuyer with earthquakes in the previous residence, measured by the cumulative Peak Ground Velocity (PGV) as the earthquake intensity of the specific location. So compared to reference year 2018, mid-to-later life homebuyers are less likely to purchase a home in 2016 in the earthquake region than in the non-earthquake region. If the mid-to-later life home purchaser experienced earthquakes in the previous residence, the higher the probability the next step of a home purchase will be in the earthquake region. Burningham et al., (2008) found in the case of flooding in the United Kingdom for example, that experience with flooding causes the assessment of local risks to be underestimated by residents whose dwelling is located in a potentially risk area. It could be that, in accordance with the results of Burningham et al., (2008), homebuyers who experienced earthquakes in the previous residence, are not unaware, but less concerned about their risk status. De Dominicis et al., (2015) and Armas (2006) respectively in two Italian flood risk areas and the earthquake area of Bucharest, Romania, attribute these feelings about the residential area to another observation: people with an intense attachment to a risk area use this as a defence mechanism, as they are feeling themselves secure, and consequently, tend to neglect or even deny the hazards.

Table 6 Logistic regression model of the prediction of a home purchase of a mid-to-later life homebuyer in the rural Groningen earthquake region based on socio-demographic characteristics, housing attributes and earthquake circumstances (Model III).
Full size table

After adding the earthquake characteristics to Model II, almost all socio-demographic and housing characteristics remained significant. One of the remarkable results in Model III is, that the independent variables ‘previous residence in Municipality of Groningen’ and ‘the location of the purchased house is in an area with population decline’, reach significance of p < 0.05. If people originate from the Municipality of Groningen, i.e., mainly the city of Groningen, the higher the probability mid-to-later life movers will purchase a house in the non-earthquake region in comparison to homebuyers from the Rest of the Netherlands or elsewhere. This is in line with earlier findings that earthquakes may result in an area’s negative reputation. The recognized areas with population decline are spread across the province of Groningen, in the earthquake and non-earthquake region (see Appendix 3). Focusing on the fact that the purchased dwelling is located in a region with population decline, the homebuyer chooses the combination with the non-earthquake region.

This final model, Model III, though, again confirms the fact that people aged 50 and older move from the earthquake region to the earthquake region. According to the Odds Ratio in the final model, the probability that mid-to-later life homebuyers, whose previous residence is in the earthquake region, is 5,2 times more likely to have bought a dwelling in the earthquake region, in comparison to homebuyers originating from the Rest of the Netherlands or elsewhere, controlling for the other variables in the model. Consequently, a local homebuyer’s market has been established. This is in line with the existence of regional movements, reported by De Kam & Meij (2017), Boumeester & Lamain (2016) and Van der Kloet (2018). Boelhouwer & Van der Heijden (2018) mention the negative impact of earthquakes on the housing market in the Groningen risk area, especially in respect of people from outside the region, who are therefore avoiding the region. In risk areas in-migration can be influenced by a negative image, as ‘outsiders’ view potential risk destinations as less attractive due to a possible lack of facilities (Hunter, 2005).

In a study on place attachment in the Groningen earthquake region, Jansen (2020) assumed that residents move within or between neighbourhoods, but stay in the region as they are predominantly, strongly attached to and/or born and raised in the region. Bonaiuto et al., (2016) observe in a research review paper on place attachment and risk areas that, once a person is affiliated to a social entity, in the sense of a person or a place, it’s rather difficult to leave the risky place, as people tend to move towards familiar persons and places they are attached to. Bijker et al., (2013) also explain the differences between choosing a house in a less popular rural area versus a popular rural area mainly based on the social characteristics of the place. Studies on highly attached individuals in relation to volcanic risks in Iceland and Indonesia, disclose they want to return to the risk area due to a social, spiritual, and economic dependence on the location (Bonaiuto et al., 2016).

4.3 Limitations

Although the dataset used is valuable, one limitation of the data on housing transactions in risk areas might be the fact that supply and demand on the housing market in a risk area is undoubtedly disturbed (Boelhouwer & Van der Heijden, 2018). In a crisis-situation it is plausible that the most beautiful homes are first to be sold in favour of less attractive homes (Jansen & Boelhouwer, 2016). Logically, in a difficult sales period in a seismic area, only the actual home sales will be recorded and the possibly less attractive, potentially damaged, ‘for sale’ homes remain invisible. Of course, the final home choice of mid-to-later life homebuyers is also determined by the composition of the housing stock according to housing type, the available housing supply in the (non)earthquake region at that time and the income of the homebuyer(s). Our housing transaction dataset of the Land Registry also does not register if a homebuyer really moves to the dwelling or wants to buy-to-let.

Another limitation of our study concerns the fact that the difference in risk and damage to the houses in the risk area is not evenly distributed. Bosker et al., (2015) demonstrated that houses with recognized damage compensation, before the actual home purchase price has been valued, will probably be sold at higher prices in advance. Finally, the missing values in our analyses are mainly due to a combination of a lack of data on the region of birth, the previous residence, and consequently also on the Peak Ground Velocity value (PGV value) of the previous residence. Given this divergent combination of missing values of different variables, we do not expect significant effects on the results.

Regarding the above-mentioned comments, our dataset lacks a few desired socio-demographic characteristics concerning the mid-to-later life home purchasers. Although Heijs et al., (2011), Coolen & Hoekstra (2001) and Jansen (2012) argue that traditional individual characteristics no longer explicate residential choices, more details on the traditional variables of the home purchaser like education, income data and the specific household composition (with or without in-living children) would have been a useful addition to the analysis.

5 Conclusions

This paper attempts to present a broader view on rural residential moves by comparing socio-demographic characteristics, housing attributes and earthquake circumstances concerning the housing transactions by mid-to-later life homebuyers in a rural risk area in contrast to a rural non-risk area. To retain the elderly and/or attract new vital, mid-to-later life people to rural, and certainly to rural risk areas, can be essential as they are able to contribute to the civic engagement in voluntary work and social associations.

The results of logistic regression analyses of all the housing transactions in the rural part of the province of Groningen, involving people aged 50 and older, show that predominantly the socio-demographic characteristics of the homebuyer and the earthquake circumstances of the (previous) home location, are decisive when it comes to purchasing a home in the risk area. Being a mid-to-later life single buyer with a former residence in the earthquake region, and born in the earthquake region, enlarges the probability of buying a home in the earthquake region. These outcomes may be explained in two ways.

First, the results could be explained by the increase in the number of older singles in general and their motives for staying close to their social networks. Mulder (2007), Coulter et al., (2016) and Wagner & Mulder (2015) draw attention to redefining short distance moves as residential mobility actions that are linked to families, social ties, and socio-economic position. Due to demographic trends like population ageing, the increasing numbers of single-person households at the mid-to-later life stage, the growing number of divorces and lone parents, in combination with physical care, childcare, joint custody arrangements and assistance from family and friends (Coulter et al., 2016; De Jong et al., 2016), might explain the residential move of single-person households, despite the hazards, to or within a risk area. Bijker et al., (2012) also demonstrated in their research on regional moves to rural, less-popular areas within the northern Netherlands, that a group of movers, including older singles and return migrants, is motivated by living close to relatives and social networks.

A second explanation of our results may be the possible existence of a strong place attachment, including social or interpersonal attachment to a rural risk area, the latter possibly even because of the risks. Jansen (2020) concludes that an unwillingness to leave the Groningen earthquake region is based on a strong place attachment, mainly concerning the elderly residents (55 and older). In their research into the dimensions and scales of place attachment, Hidalgo and Hernandez (2001) found that social attachment, as a part of place attachment, is of greater importance than physical attachment. This last observation, especially applied to a risk area, could mean that the importance of interpersonal attachment (Bonaiuto et al., 2016) compensates for the risk of physical damage to buildings and/or infrastructure and consequently may result in people staying in or returning to a risk area, despite the risks. Focusing on the physical features of the dwelling, in contrast to the conclusion of Scott et al., (2017) that housing characteristics form an important motive to move to rural areas in non-risk circumstances, in our study housing attributes hardly contribute to the prediction of a house purchase in the earthquake area. However, in our analysis, housing type and housing purchase price are statistically significant housing attributes. If the housing purchase price increases, the probability of buying a home by 50 + aged people in the earthquake region diminishes. These findings correspond with the outcomes of the research by Bijker et al., (2012) that low house prices are a significant incentive for single-person households to move to specific, rural, less-popular areas in the northern Netherlands. Our analysis further reveals that an important earthquake condition is the fact that a mid-to-later life homebuyer who has experienced earthquakes in the previous residence, is more likely to buy a property in the earthquake area (again). Instead of the expected ‘flight’, residents apparently still ‘fight’ and actually relocate in the risk area (Jansen et al., 2017). Hunter (2005) attributes this relocation in the case of natural and technological hazards to environmental factors, household characteristics and insiders’ risk perception. With this last concept, Hunter (2005) points to the danger that outsiders may assess the hazards of a risk area in relation to residential mobility completely different from the insiders, while the latter have an overall view of the situation.

All in all, our findings indicate, in line with earlier research, the existence of a local housing purchase market in the Groningen earthquake region. Local homebuyers apparently want to continue to live in the area, despite the risks. Understanding which characteristics of mid-to-later life homebuyers, of the chosen home and place of residence in the Groningen earthquake region are relevant, will help to extend the academic literature and to make policy decisions on residential mobility and relocation of these economically and socially important, ageing inhabitants of a rural risk area.

Bonaiuto et al., (2016) stress the need to explore the two competing goals a resident in a risk environment has to deal with: self-protection versus connection. Empirical investigation on the interaction between place attachment, social attachment and risks on the housing, neighbourhood, or region scale, might clarify the residential mobility processes in a risk area even further. As social and economic costs of seismic hazards in risk areas throughout the world are considerable, and according to future estimations even will increase due to climate change (Armas, 2006; Dominicis et al., 2015), future research might shed more light into ‘who’ and the reasons ‘why’ residents want to move out or remain in a risk area.