This article was downloaded by: [University of Stellenbosch]On: 31 August 2013, At: 03:12Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Housing StudiesPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/chos20

Homeownership and Long-Term CareJan Rouwendal a & Fleur Thomese ba Department of Spatial Economics, Faculty of Economic Science ,VU University , Amsterdam , The Netherlandsb Department of Sociology, Faculty of Social Sciences , VUUniversity , Amsterdam , The NetherlandsPublished online: 28 Jan 2013.

To cite this article: Jan Rouwendal & Fleur Thomese (2013) Homeownership and Long-Term Care,Housing Studies, 28:5, 746-763, DOI: 10.1080/02673037.2013.759179

To link to this article: http://dx.doi.org/10.1080/02673037.2013.759179

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.

This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

Homeownership and Long-Term Care

JAN ROUWENDAL* & FLEUR THOMESE**

*Department of Spatial Economics, Faculty of Economic Science, VU University, Amsterdam, The Netherlands,

**Department of Sociology, Faculty of Social Sciences, VU University, Amsterdam, The Netherlands

(Received December 2011; revised September 2012)

ABSTRACT We investigate the relationship between homeownership and institutionalization usinglongitudinal data from a Dutch community sample (N ¼ 2372) collected between 1992 and 2005,and find a negative effect of housing tenure on the probability of moving to a nursing home betweentwo subsequent waves. Our discrete time duration model is able to deal with time-varying covariateslike health and is flexible with respect to time effects. We have detailed information about healthstatus, presence of a partner and children, neighborhood, and housing. The effect of tenure remainssignificant after controlling for their impact. A variety of additional potential explanations related tohousing wealth and the price of long-term care are found to lack explanatory power. We thereforeinterpret our findings as the result of a strong desire among the homeowners to stay where they are—in their own property—and the better possibilities that they have—as owners—to realize this desire.

KEY WORDS: Long-term healthcare, aging, housing tenure, institutionalization

1. Introduction

Demand for long-term care is expected to increase substantially over the coming decades

due to population aging in many countries. Care-at-home, provided by relatives or

professional caregivers, is the most important substitute for institutionalization. Many

elderly like to stay in their homes as long as possible even when their health deteriorates,

but there seem to be important differences in the possibilities they have for realizing this

desire. In particular, it has been observed frequently that homeowners have a lower

probability of becoming institutionalized than renters, also when a large number of control

variables are included in the analysis (Muramatsu et al. (2007) and Gaugler et al. (2007)

for the US, Breeze et al. (1999) for the UK, and Nihtila & Martikainen (2007) for Finland).

In this paper, we explore the relationship between homeownership and the demand for

residential care using a rich dataset that allows us to consider a wide variety of variables

over time in a Dutch community sample of elderly. Our data show a substantially lower

transition rate of homeowners to institutionalized care, and we investigate the robustness

of this finding for the incorporation of detailed controls to parcel out compositional effects

q 2013 Taylor & Francis

Correspondence Address: Jan Rouwendal, Department of Spatial Economics, Faculty of Economic Science, VU

University, De Boelelaan De Boelelaan 1105, 1081 HV Amsterdam, The Netherlands. Fax: 31 205986004;

Tel.: 31 205986093; Email: jrouwendal@feweb.vu.nl

Housing Studies, 2013

Vol. 28, No. 5, 746–763, http://dx.doi.org/10.1080/02673037.2013.759179

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

that could mistakenly be interpreted as a causal link between housing tenure and the

demand for nursing home care.

In Section 2, we discuss the explanatory variables. In Section 3, we make some remarks

about the specificities of the Dutch situation to which our empirical work refers. Then, we

introduce the data and the model. In Section 5, estimation results are reported. Section 6

reports the results of some further explorations. Section 7 explains the conclusions drawn

from the study.

2. Institutionalization and the Demand for Long-Term Care

The primary focus of this paper is on institutionalization in nursing homes or related

institutions1 for the provision of long-term care. Differences between the transition rates of

homeowners and renters to nursing homes could result from compositional effects as well

as from structural reasons. Compositional effects occur when people who own their home

differ in the propensity to become institutionalized for reasons that are only indirectly

related to homeownership. There are many reasons why such effects may occur. We

discuss four of them, which are in our view the most important ones.

Health status is presumably the main driver of the demand for nursing home care.

A wide range of physical, mental, and functional health conditions is associated with

admission to long-term-care facilities (Fried et al., 2001; Gaugler et al., 2007; Geerlings

et al., 2005; Miller & Weissert, 2000; Thomese & Broese, 2006). Since there exists a

strong relationship between health and wealth at old age (Broese et al., 2003; Huisman

et al., 2005) and homeowners in general are wealthier than renters, they have on average

better health. Insufficient control for health may therefore easily lead to correlation

between homeownership and admission to nursing home that is not based on a causal link

between the two variables. It is therefore important to control for health status.

Somewhat related to this observation is the fact that homeowners live with a partner

more often than others and may also more often live closer to children,2 who could provide

the care that is necessary to postpone or even completely avoid the institutionalization that

would otherwise be necessary. Since much long-term care is provided by unpaid

caregivers, it is important to control for this possibility as well.

A third possibility for compositional effects is that long-term care is less available in the

locations where homeowners are concentrated. Even if public healthcare would guarantee

that nursing home capacity per 1000 inhabitants was approximately equal throughout

space, differences in population density could still cause a much larger physical distance to

the nearest nursing home, and this might have an impact on transition rates. Since

homeowners are more often located in areas with relatively low densities, this could also

lead to correlation between homeownership and transition rates. Moreover, if capacity

problems occur in nursing homes, it is unlikely that these are evenly spread over space.

Fourth, there is probably a relationship between the strength of neighborhood-based

social networks and the elapsed duration of stay in the current house or neighborhood.

Since residential mobility among homeowners is lower than that of renters, and the

difference is especially large for elderly people,3 this may also give rise to a correlation

between homeownership and institutionalization that is in fact caused by social network

effects.

The four effects that we have mentioned may of course all be present simultaneously

and many datasets lack the possibility to control for all of them at once. The data used in

Homeownership and Long-Term Care 747

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

this paper are exceptionally rich in this respect and enable us to take into account all the

compositional effects mentioned above. Moreover, they also allow us to control for

regional differences in the supply of care that may be correlated with differences in the

homeownership rate.

To anticipate our findings, a substantial effect of homeownership is left after all these

controls and we therefore direct our attention to the second class of explanations, those that

are closely related to homeownership itself. In Section 6, we therefore report the results of

some further explorations.

3. The Dutch Context

Long-term care in the Netherlands is targeted to chronic illnesses, and is mostly provided

by unpaid caregivers. The industry is dominated by not-for-profit facilities. Long-term

care in the Netherlands is provided by the welfare state, which means that it is available to

all. It is partly funded through (mandatory) collective insurance and partly from the

revenues of taxes. Those who make use of the system sometimes have to pay a (limited)

own contribution to its cost. The height of this contribution may depend on the receiver’s

income or, until 1997, wealth.

3.1. The Institutional Setting

The Netherlands is an example of a western European welfare state, with substantial

involvement of the government in healthcare. Since the 1950s, a dual system of long-term

care evolved that lasted until 1997. Intensive types of long-term care were originally

provided exclusively by what are called in Dutch verpleeghuizen or nursing homes. Such

care is very costly and difficult to provide by a private insurance system. For this reason,

the Dutch government decided to provide it through a collective insurance scheme.

Nursing homes therefore became—in principle—available to all, under the condition that

a person’s health condition was such that he or she qualified for this type of care.

The so-called verzorgingshuizen, a type of sheltered housing (often provided in large

apartment buildings) was designated for elderly who could, in principle, continue living

independently, but found it convenient to be able to make use of less intensive types of

care offered by the residential home, like housekeeping and the provision of hot meals. In

the early years (until the 1970s), people had to be in good health to be allowed to enter into

a sheltered housing. Those who did, in principle, had to pay the full cost of the care

provided. However, if they at any point in time lacked the wealth to do so, the public

healthcare system provided assistance from the general means. As a consequence,

entrance into sheltered housing was possible to all, but wealthy people had to pay a much

higher price until they ran out of resources. This became known as ‘eating one’s house’.

Starting in the 1980s, the national government increasingly restricted admission to both

types of residential care to those who were unable to receive sufficient care from other

sources, mostly care at home. Although there remained a distinction between sheltered

housing and nursing homes, in practice sheltered housing had become part of the public

healthcare system. In 1997, verzorgingshuizen were formally incorporated in the publicly

financed healthcare system (Staatscourant, 1996), and from that year onwards, there was

an income-dependent contribution to the cost of care in both nursing homes and sheltered

housing, which had effectively become nursing homes for those who needed less intensive

748 J. Rouwendal & F. Thomese

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

care. Wealthy patients no longer needed to pay the full cost until most of their wealth was

consumed.

The cost of long-term care provided in sheltered housing thus decreased substantially

for many homeowners. The large costs of this type of care before 1997 may well have

contributed to the lower propensity of homeowners to become institutionalized. If this

hypothesis is true, this reluctance should have decreased after the policy change in 1997.

3.2. Admission to Nursing Homes

The primary determinant for long-term care is a person’s health (Miller & Weissert, 2000).

In principle, this care can be provided in alternative ways, the most important distinction

being between care-at-home and institutionalization (see, for instance, Woittiez et al.,

2009). Care-at-home can be provided not only by professional workers but also by unpaid

caregivers, often relatives of the recipient. The availability of this substitute thus depends

on a person’s family and social networks as well as on the intensity of the demand for care.

Admission to nursing homes in the Netherlands depends on a professional, independent

assessment of the person’s health and the associated need for care.4 This assessment has

become increasingly dependent on the availability of care at home. Capacity problems in

long-term care emerged in the late 1980s and gave rise to long waiting lists for

professional care in both the nursing homes and the homes-with-care. These waiting lists

were an important cause for policy concern in the 1990s and early 2000s, while the

problem seems to have been mitigated since then. As long as people were on a waiting list,

they necessarily had to rely on a substitute for residential care: professional care given at

home (which is also part of the public healthcare system) or care provided by relatives,

most typically the partner, if present and able, or adult children. Problems caused by the

long lists were mitigated by giving priority to people with an acute need for receiving

professional care. For some others on the waiting lists, the substitute was quite

satisfactory, as they refused to accept admittance in a nursing or residential home after

being on the waiting list for some time (van Gameren, 2005). This suggests that, at least for

a part of those who were in need of long-term care according to expert judgment, good

alternatives to nursing homes were available.

3.3. The Dutch Housing Market

The Dutch housing market differs from that in many other European countries as well as

from the US through its large rental sector. According to Eurostat statistics, the percentage

of homeowners among pensioners in the whole EU was 60 percent in 1995, when the

Longitudinal Aging Study Amsterdam (LASA) panel started, as opposed to 43 percent

among the Dutch pensioners. The share of homeowners is less than 30 percent for low

incomes and more that 80 percent for high incomes (Rouwendal, 2009). This means that in

the Netherlands, the link between housing tenure (the probability of being an owner) and

income is stronger than in many other countries.

Much of the rental housing is social housing owned by housing corporations (35 percent

of the total housing stock). Almost all rental housing is rent-controlled since the Second

World War. Maximum rent increases are annually determined by the parliament and

follow inflation. This situation is generally expected to continue for the foreseeable

future.5 Maximum allowable rents are related to the number of quality points for which

Homeownership and Long-Term Care 749

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

a house qualifies. Location characteristics are hardly reflected in these quality points,

which implies that rental housing close to the center of urban areas (like Amsterdam) is as

cheap as rental housing in rural areas with the same amount floor space, number of rooms

and other characteristics that are associated with quality points. There is excess demand

for rental housing in many locations, especially in the urban areas. The allocation system

for rental housing differs over the country, but everywhere priority is given to those most

in need of affordable housing. Existing homeowners do not easily qualify as such, and

since 95 percent of the rental stock is subject to this system, mobility from the owner

occupied to the rental sector is therefore very limited.

4. Data and Method

4.1. The Database

Data are obtained from the LASA (http://www.lasa-vu.nl). LASA is an ongoing study on

physical, emotional, cognitive and social functioning of older adults, with a nationally

representative sample (Deeg et al., 2002). In 1992 (t ¼ 0), interviewers questioned 3805

respondents as part of the Living Arrangements and Social Networks of Older Adults

research program (Knipscheer et al., 1995), which used a stratified random sample of men

and women born between 1908 and 1937. The oldest individuals, particularly the oldest

men, were over-represented in the sample, which resulted in approximately equal numbers

of men (n ¼ 1859) and women (n ¼ 1946). The sample came from population registers of

11 municipalities: the city of Amsterdam and two rural communities in the west of the

Netherlands, one city and two rural communities in the south, and one city and four rural

communities in the east. These regions represented the differences in religion and

urbanization in the Netherlands at the time. Of the 6107 eligible individuals in the sample,

2302 (38 percent) refused cooperation due to a lack of interest or time, and another 734

were ineligible because they were deceased, or too ill, or cognitively impaired to be

interviewed.

In 1992–1993 (t ¼ 1, N ¼ 3107), 1995–1996 (t ¼ 2, N ¼ 2545), 1998–1999 (t ¼ 3,

N ¼ 2076), 2001–2002 (t ¼ 4, N ¼ 1691), and 2005–2006 (t ¼ 5, N ¼ 1257) LASA

performed the follow-ups. Between 1992/1993 and 2005/2006, 46 percent of the

respondents died, 5 percent were unable to participate in the study because of severe

physical or mental health problems, 14 percent refused to be re-interviewed, and 2 percent

could not be contacted because they moved to another country or an unknown address. In

each wave, the interviewers received a 4-day training course and the LASA fieldwork

manager supervised them intensively. The interviewer tape-recorded the interviews to

monitor and enhance the quality of the data obtained. The interviews took between 1.5 and

2 h. Between observations, mortality was regularly determined on the basis of death

registers.

We selected 2046 respondents with at least two observations between 1992/1993 and

2005/2006, who were living independently in 1992/1993, and included those who had

died between waves. Measurement differences between 1992 and 1992/1993 precluded

using the first wave (t ¼ 0). We will refer to the 1992/1993 wave as t ¼ 1 or wave 1,

and so on. We have at most four observations for each respondent. For 1062

respondents this is the case. A total of 486 respondents were observed three times and

496 twice.

750 J. Rouwendal & F. Thomese

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

4.2. Measurements

Institutionalization. At each observation, the respondent’s address was compared with the

previous address to see whether he or she had moved in the preceding period. If

respondents had moved, we asked in what type of housing they lived. Alternatives

included a home-with-care or a nursing home. Interviewers also observed the housing

type. Based on both measurements, a variable was constructed indicating whether or not

the respondent had moved to an institution between two waves. If a respondent had died

between waves, it was established whether he or she had been institutionalized before

death. Short stays in a caring or nursing home were not registered as institutionalization. In

our data, institutionalization is an absorbing state. Of the 265 transitions to either a home-

with-care or a nursing home reported in our raw data, 29 were to a nursing home. Mainly

because of this small share of immediate transitions to a nursing home, we collapsed both

transitions into one variable indicating institutionalization. We did not record whether

respondents moved from a home-with-care to a nursing home.6 This may imply that results

concerning the effect of the policy change underestimate a true effect.

Care-at-home. Respondents were asked if they received help with personal care (yes or

no). If they responded positively, they could indicate up to 12 sources of help, ranging

from the partner, children and other informal carers to a variety of formal carers. Personal

care is defined as having at least help with one of the following activities: to wash, to bath

or shower, dressing and undressing, to go to the toilet, to get up and sit down. The same

procedure was followed with respect to domestic tasks (no, yes), indicating activities like

preparing meals, doing groceries, cleaning the house, taking the garbage bags outside, and

also filling out forms. If any source of care-at-home was mentioned, we considered the

respondent to receive care-at-home.

Housing tenure. In 1992, respondents were asked if they owned their house. Alternatives

were own property, rented, sublet, and free of charge. The first alternative (own property)

was scored as homeownership. Outright owners have very low out-of-pocket user costs and

could therefore be expected to stay as long as possible in their home. Owners were asked

whether their house was free of mortgage. They could answer yes or no.

Health. To provide a sufficiently complete and concise overview of the respondent’s

(unobserved) health, we use five indicators that were assessed at each wave:

(a) Interviewers asked about the presence of seven chronic diseases: lung disease,

cardiac disease, arteriosclerosis, stroke, diabetes, arthritis, and malignant

neoplasm. We counted the number of chronic conditions mentioned.

(b) Self-reported functional ability is measured as the ability to perform six

activities in daily life (ADL), e.g., ‘Can you walk up and down stairs?’ The five

possible answers were: not at all, only with help, with a great deal of difficulty,

with some difficulty, and without difficulty. The six items constituted

hierarchically homogeneous scales at the observations (Loevinger’s H $ 0.59),

which were reliably measured (r $ 0.83). The scale ranges from 0 (no

disability) to 24 (severe disability).

(c) Cognitive ability is measured with the 30-item mini mental state examination

(MMSE; Folstein et al., 1975). The MMSE is a widely accepted scale for

Homeownership and Long-Term Care 751

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

measuring cognitive ability. It includes items such as recalling three words

previously learned, naming the current date, and doing a series of subtractions.

The scale ranges from 0 (minimum ability) to 30 (maximum ability) points.

(d) Depressive symptoms are assessed with the Center for Epidemiologic Studies

Depression scale, a 20-item scale (Radloff, 1977) which has been widely used in

older populations. Scores range from 0 to 60, and asks for such feelings and moods

as anxiety, happiness, or experiencing hopeless thoughts during the past weeks.

(e) Frailty is indicated by the respondent’s gait or walking speed, measured as the

number of seconds needed to walk 10 feet and back.

Availability of care provided by children. We included the total number of children and

the number of living children within 30min travel distance reported by the respondents. To

compensate for skewness, we used a log transformation in some of the analyses.

Neighborhood involvement. Respondents were asked for the number of years a

respondent has been living in the same neighborhood. There also was an extensive

identification procedure for the personal network of respondents. This included asking for

people in the neighborhood with whom the respondent had frequent and important contact.

For each of the people mentioned, respondents could indicate the frequency of contact,

ranging from never (1) to daily (8). We counted the number of neighbors identified in the

network and the number of neighbors with whom the respondent had at least a monthly

contact.

Special adjustments in the house. Respondents could say if they had any of the 18

adjustments, ranging from extra banisters to an alarm system or a special lift in the house.

Satisfaction with housing. This was measured with a direct question. Respondents could

indicate to be dissatisfied (1), not satisfied (2) or satisfied (3) with their current housing

situation.

Control variables. Education was measured in 1992, as the highest level of education

obtained. The nine response categories ranged between no school finished at all and a

university degree. To improve international comparability, the variable was recoded into

years of education.

Presence of partner. The partner is an important source of help, and the loss of a partner

can trigger institutionalization. Given the health disparities among renters and

homeowners, the former will be more often without a partner. At each observation, we

assessed the presence of a partner in the household. We use the presence of a partner at

each wave.

In 1992, sex and age were recorded.

Urbanization and region. Housing tenure may be associated with local and regional

differences, which also affect the demand and supply of long-term care. Urbanization may

affect the demand, which presumably is lower in less urbanized areas. We use a national

measure indicating the number of addresses per square meter. The scale ranges from 1

(,500 adresses) to 5 (.2500 adresses). Furthermore, van Gameren (2005) has

752 J. Rouwendal & F. Thomese

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

documented regional differences in the length of waiting lists for caring and nursing

homes in the Netherlands. He concludes that such differences exist and can be substantial.

In order to control for the impact of spatial factors, we have introduced dummies for the

three regions in which respondents were sampled.

4.3. Method

We concentrate on the transition from living independently to becoming institutionalized

in a nursing home. The statistical tools that are appropriate for investigating this

phenomenon are known as event history or duration analysis. These techniques focus on

the transition rate or hazard. In our discrete time data, this is the probability that a

household living independently in wave t is institutionalized in wave t þ 1, which we

denote as hðx; tÞ.The most popular tool for duration analysis is the proportional hazard model that

specifies the hazard function as the product of a function of time only (the so-called

baseline hazard) and a function of the covariates only: hðx; tÞ ¼ h0ðtÞh1ðxÞ. The latter

function is often specified as an exponential function (h1ðxÞ ¼ ebx, where b is a vector of

coefficients that have to be estimated). This specification implies that changes in the

covariates have the same proportional effect on the hazard at each time. The choice of the

baseline hazard is important, and since theory often does not imply specific guidelines for

its specification, flexibility in this respect was desirable. Cox (1972) showed that with

continuous time data, the specification of a baseline hazard function could be avoided

through the use of a partial likelihood technique.

The Cox model is less suitable for the discrete time data that we exploit here. We noted

already that with our discrete time data, the hazard is in fact the probability that a transition

takes place between subsequent waves of the panel. This implies that we can use

functional forms developed in the discrete choice literature to model this hazard (see, for

instance, Cameron & Trivedi, 2005, pp. 602–603). We follow this approach by using the

logit specification: hðt; xÞ ¼ eatþbx=ð1þ eatþbxÞ. In this specification, a separate

coefficient at is estimated for each wave. This offers similar flexibility in modelling the

time dependency of the hazard as does Cox proportional hazard model.7

A particular advantage of the discrete time hazard specification is that it can easily deal

with time varying covariates, whereas continuous time models, including the one

developed by Cox, become less easy to handle when the values of the x’s change over

time. In our logit-based duration model, the specification of the hazard becomes:

hðt; xÞ ¼ eatþbxt=ð1þ eeatþbxt

Þ, and this is as easy to handle as the model in which x is

constant over time.

To estimate the model, we take into account the panel nature of our data. The typical

pattern is that a respondent is first observed living independently and then moves to a

nursing home. The likelihood of such a sequence of observations for a respondent i who

becomes institutionalized between waves T and T þ 1 is li ¼QT21

t¼1 ð12 hðt; xitÞÞhðT; xiT Þ.When estimating the model, we also include right-censored observations referring to

respondents who are not observed to move to a nursing home. The likelihood of such

observations is li ¼QT21

t¼1 ð12 hðt; xitÞÞ, where T is the last wave in which the respondent

is observed. Right censoring may take place because the respondent dies without

becoming institutionalized,8 because the respondent drops out of the sample for other

reasons, or because the respondent was still living independently in the most recent wave

Homeownership and Long-Term Care 753

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

of the panel included in our data. It is also easy to deal with respondents who missed one

wave of the panel.

A final issue that has to be discussed is unobserved heterogeneity among the

respondents. Such heterogeneity may be thought of as an individual effect, ji, that should

be included in the hazard as: hðt; xÞ ¼ eatþbxtþji=ð1þ eeatþbxtþji

Þ. This heterogeneity

implies that some people have a greater probability to become institutionalized in the

course of time than others and, thus, are likely to drop out of the sample. We can model jias a random variable with a particular distribution (for instance: the normal distribution)

and estimate the model.9

5. The Robustness of the Impact of Homeownership on Institutionalization

5.1. Descriptives

Descriptive information on the variables included in the analysis is provied in Table 1. The

figures in this table refer to respondents in wave 1, which have been subdivided into

renters and owners. Table 1 shows that almost all variables are correlated to some extent

with the housing tenure. Homeowners use more care at home, are more often male, are

somewhat younger, higher educated, in better health, etc. It is interesting to note that the

share of homeowners differs substantially over the regions. In region 1, owners are over-

represented, whereas in region 2, the share of renters is higher than average. This reflects

Table 1. Renters and owners in wave 1 (percentages or means and standard deviations)

All Owners Renters

Use of care at home (%) 3.4 4.7 2.4Year of birth 1924 (8.4) 1926 (8.0) 1923 (8.3)Sex (% women) 53 49 55Education (# years) 9.0 (3.3) 9.7 (3.6) 8.5 (3.3)

Frailty 8.0 (3.6) 7.4 (2.9) 8.5 (3.9)Functional ability 28.1 (3.6) 28.8 (3.7) 27.7 (4.0)Depressive 7.3 (7.2) 6.3 (6.8) 8.0 (7.3)Cognitive functioning 27.4 (2.3) 27.7 (2.2) 27.2 (2.3)# Chronic illnesses 0.89 (1.0) 0.76 (0.9) 0.98 (1.0)

Living with partner (%) 71 82 63# Children 3.0 (2.1) 3.0 (2.1) 2.9 (2.2)# Children within 30 min 1.5 (1.3) 1.6 (1.4) 1.4 (1.3)

Urbanization (scale 1–5) 2.9 (1.5) 2.3 (1.3) 3.4 (1.5)Region 1 (%) 33 42 27Region 2 (%) 27 10 40Region 3 (%) 22 25 20Region 4 (%) 17 23 14

Years of residence in the neighborhood 23 (17) 26 (18) 21 (16)# Neighbors 9.9 (8.6) 11.6 (8.9) 8.6 (8.1)# Frequently contacted numbers 6.0 (5.6) 7.0 (5.9) 5.3 (5.2)# Adjustments 0.18 (0.4) 0.11 (0.3) 0.22 (0.4)Housing satisfaction 2.88 (0.4) 2.95 (0.2) 2.82(0.5)

754 J. Rouwendal & F. Thomese

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

differences between the urban areas, where most of the housing is rental, and the less

urbanized areas, where the share of owner-occupied housing is larger.

A similar table could, of course, be presented for the other waves. Such a series of tables

would show that the share of homeowners increases slightly over time. This is partly

caused by the selection process just noted, and also by the larger propensity of renters to

become institutionalized. A small number of respondents have moved from privately

owned to rental housing.10

Such a table would also demonstrate that all health indicators, apart from cognitive

ability, show a slightly decreasing performance over time in the population. The change in

the average value of these variables is, of course, suppressed by the selective attrition of

those with the worst health condition, either through death or being unable to fill out the

questionnaire.

Table 2 provides the raw data on institutionalization and housing tenure in our sample.

The figures in the table confirm that institutionalization occurs much less among the

homeowners than among the renters. Among those transiting to an institution, the share of

owner occupiers is much lower. Clearly, homeowners differ substantially from renters in

their propensity to become institutionalized. We will now consider first how much of this

difference can be attributed to compositional differences between owners and renters.

5.2. Estimation Results

Estimation results are presented in Table 3. Our most elementary specification is reported

in column (1): we only use a constant and a homeownership dummy. The significant

negative coefficient confirms the findings of Table 2. The second column reports our basic

specification in which we use dummies for waves t ¼ 2, 3, and 4,11 and the individual’s

year of birth, sex, and education as explanatory variables. Incorporation of these time and

cohort dummies implies that we cannot also include the respondent’s age as an

explanatory variable.12 The results of this second model suggest that the probability of

becoming institutionalized increased over the years. The year of birth has a strongly

significant negative effect, implying that younger cohorts have a much lower propensity to

become institutionalized. Women have a substantially larger propensity of becoming

institutionalized than men, while higher education has the opposite effect compared with

lower education. The coefficient for homeownership drops substantially, but remains

significant.

We noted above that the absence of control variables causes potentially serious bias

in the coefficient of the homeownership dummy. To see if and to what extent this is indeed

Table 2. Moves to nursing and residential homes

Between waves

1 and 2 2 and 3 3 and 4 4 and 5

From renting 51 (3.9%) 57 (4.9%) 36 (4.1%) 24 (3.7%)From owning 6 (0.6%) 14 (1.7%) 11 (1.6%) 6 (1.1%)Total 57 (2.5%) 71 (3.5%) 47 (3.0%) 30 (2.5%)

Note: The figures in parentheses give the moves as percentages of theassociated total numbers of observations.

Homeownership and Long-Term Care 755

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

the case, we introduce four sets of such control variables. We start with what are arguably

the most important variables in this respect: the health status indicators. Estimation results

in column (3) show that frailty (measured as low walking speed) is positively related to

becoming institutionalized, while high scores ( ¼ better performance) on functional

ability (ADL) and cognitive functioning (MMSE) indicators are negatively correlated with

institutionalization. Respondents reporting chronic illnesses and depressive symptoms do

not become institutionalized more frequently.13 The coefficients for the time dummies are

all much smaller and insignificant, which suggests that they mainly picked up the effects of

deteriorating health in the previous specification. The coefficient for the year of birth also

becomes smaller, but remains highly significant. Sex and education are no longer

significant, which may be interpreted as saying that our healthcare indicators control

effectively for health effects that are correlated with these two variables. The effect of

homeownership on institutionalization is reduced to half its value in the earlier model, but

it remains negative and significant.

In the next variant of the model, we introduce the presence of a partner in the household

and the logarithm of the number of children þ1.14 Column (4) reports estimation results.

Table 3. Estimation results of a discrete time transition model for institutionalization

(1) (2) (3) (4)

Constant 23.12 (0.08) 20.17 (0.33) 4.62 (0.83) 4.32 (0.88)Wave 2 0.44 (0.19) 0.16 (0.19) 0.14 (0.19)Wave 3 0.56 (0.21) 0.15 (0.22) 0.13 (0.23)Wave 4 0.72 (0.24) 0.24 (0.26) 0.09 (0.28)Year of birth 20.17 (0.01) 20.12 (0.01) 20.11 (0.01)Gender 0.35 (0.16) 0.18 (0.18) 0.05 (0.19)Education 20.054 (0.026) 0.008 (0.03) 0.02 (0.03)

Homeowner (y/n) 21.26 (0.18) 20.64 (0.19) 20.61 (0.20) 20.59 (0.23)

Health indicators# Chronic illnesses 20.03 (0.07) 20.01 (0.07)Functional ability (lo-hi) 20.081 (0.02) 20.078 (0.02)Cognitive functioning (lo-hi) 20.15 (0.02) 20.15 (0.02)Depressive symptoms (lo-hi) 20.00 (0.01) 20.00 (0.01)Frailty 0.027 (0.01) 0.030 (0.01)

Demographic indicatorsPartner present 20.43 (0.20) 20.41 (0.20)Log (# children þ 1) 0.15 (0.13) 0.11 (0.13)Supply controlsUrbanization Included, but nsRegion Included, but ns

Social network indicatorsYears in neighborhood 0.001 (0.004)N (respondents) 2372 2372 2372 2372N (observe) 7026 7026 7026 7026Loglikelihood 2756.7 2706.5 2703.6 2695.5

Notes: Robust standard errors are given in parentheses. The number of respondents refers to the number ofpersons that we observe in at least two subsequent waves and who provided enough information to includethem in our model. The number of observations is the number of times respondents provided sufficientinformation in subsequent waves to include this information into our estimation model.

756 J. Rouwendal & F. Thomese

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

Only the dummy, indicating the presence of a partner, has a significant coefficient. Again,

the coefficient for homeownership decreases somewhat, but it remains significant with a

(robust) t ratio of 2.8. We have experimented with several other indicators for the effect of

children [a dummy for having at least one child, the (untransformed) number of children,

the number of children living within 30 min of travel time and the number of children

living with the respondent].

However, the estimated coefficients for these alternative variables were never

significant. Since it has been documented that daughters are more important for giving

informal care than sons, we have also experimented with a dummy for at least one

daughter, the number of daughters, and the presence of a daughter within at most 15 min of

travel distance, and although the signs of the estimated coefficients were now consistently

negative, as expected, they were never significant.

We control for availability of long-term care by introducing dummies for the regions in

which our respondents were living and for the degree of urbanization of their municipality

of residence. Differences in supply due to differences in population density or regional

capacity problems should be captured by these variables. However, none of the

coefficients for the dummies that we introduce are significant. We introduce eight dummy

variables and the loglikelihood increases by approximately eight points; the implied

p-value is just over 5 percent. The coefficient for the homeownership dummy hardly

changes.

This remains the case when a variable indicating the social (neighborhood) capital of

the respondent is introduced. In column (4) of Table 3, we have taken the number of years

a respondent has been living in his or her present neighborhood, but similar results (an

insignificant coefficient for the social capital indicator) were reached when we included

the number of neighbors or the number of neighbors with whom the respondent has

frequent contacts.

In order to take better account of the panel nature of the data, all model specifications

have been rerun with a random effects logit model. The random effects play the same role

as the mixing in other specifications of the proportional hazard model and thus address the

dynamic sample selection problem created by sequentially removing respondents who

became institutionalized earlier from subsequent waves. This resulted in very modest

changes of the estimated coefficients. In particular, the coefficient for homeownership

hardly changed and was significant in all model variants.

A final concern that we addressed is that homeowners may first move to the rental sector

when their health deteriorates, and subsequently become institutionalized. This could bias

our results with respect to homeownership. We therefore constructed a new variable that

indicated if a respondent has been observed as a homeowner in the previous or present

wave and used this variable instead of the homeownership dummy that indicates

homeownership in the present wave. The effect of this change is that renters who were

previously observed as owners are now treated as if they still are the owners. The results of

the modified model were similar to those of the original one. In the most extensive model

(column (4)), the homeownership coefficient increases in absolute value to 20.68 (the

standard error remains 0.23). We conclude that our results are not biased by the selective

movement to the rental sector of the homeowners in poor health.

Summarizing, we have to conclude that after controlling for four potentially important

compositional effects, we still find a statistically significant negative effect of

homeownership on the institutionalization hazard. The effect is also significant from an

Homeownership and Long-Term Care 757

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

economic point of view. To see this, observe that the impact of homeownership is larger

than that of the presence of a partner. For the sample average, the institutionalization

probability is small for owners as well as renters, but when health problems occur, the

impact of homeownership becomes sizable. To illustrate, we provide some computations

based on the estimated model. When there are health problems and no partner is present,

the probability of institutionalization for renters is 3.6 percent, while the corresponding

probability for owners is 2.0 percent. With more severe health problems, the

institutionalization probability can be 6.3 percent for renters and 3.5 percent for owners.

It appears therefore that homeownership, or some mechanism closely related to it, is

responsible for a substantial part of the correlation between housing tenure and

institutionalization.

6. Further Explorations

Since our results are inconclusive with respect to the reasons behind the large difference

between the probability of owners and renters to become institutionalized, we have

considered a number of alternative explanations for this finding. We considered various

indicators of attachment to local networks: the number of neighbors, the number of

neighbors frequently contacted, and the number of years that the respondent lived in the

current home. None yielded a significant effect on the probability of becoming

institutionalized. There is, however, a substantial difference in the mobility of elderly

renters and owners after the age of 65 years. Although the mobility rates of both groups are

low, the average elapsed duration of stay for renters stabilizes, whereas that of owners

keeps increasing (see, for instance, Rouwendal, 2009).

Next, we investigated three possible causal mechanisms associated with wealth or, more

specifically, home equity that could explain the link between homeownership and lower

transition rates that remains after controlling for compositional effects. One limitation of

our dataset is that it contains little information about wealth. Homeownership itself is of

course strongly correlated with wealth and it is therefore possible that part of the

remaining correlation between homeownership and institutionalization is in fact due to

wealth. We explore some important ways in which such a relationship could manifest

itself. The first one is that being wealthy may have important consequences for the cost of

receiving long-term care. Moving to an institution is often associated with higher costs for

homeowners than for renters (Dietz & Haurin, 2003). In the west-European welfare states,

the patient’s own contribution to received (long-term) care is usually dependent on either

income or wealth. Moreover, in some countries, not all institutions providing long-term

care are covered by the public healthcare system in the same way. This was the case in the

Netherlands (to which our empirical work refers) until 1997. However, we did not find

significant effects of the higher price that the homeowners had to pay for long-term care

until a policy reform in 1997.15 The—on average—larger wealth of homeowners may not

only imply a higher cost of institutionalization, it apparently also provides them with better

means to postpone or avoid this event.

A second possibility that has attracted some attention in the literature is that elderly

people may postpone institutionalization by deliberately using the prospect of a substantial

bequest as an incentive for their relatives to provide the care they need (Bernheim et al.,

1985). The idea is that older people may expect something of their children or other

relatives in return for the wealth they bequeath to them. The house is usually the largest

758 J. Rouwendal & F. Thomese

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

asset of households, and a luxury house obviously signals a large bequest. The prospect of

receiving part of the inheritance may consciously or unconsciously increase the

willingness of relatives to provide services required by the older person in need of care.

This could imply that children of homeowners, who are the usual beneficiaries, provide

more informal care than children of renters do. However, we found no indication of a

strategic bequest motive through which homeowners get more care from children in

exchange for the bequest of the house.16

A third effect of wealth on the transition to institutionalized care may be caused by the

fact that owner-occupied housing is usually more luxurious than rental housing. There

exists a well-established association between housing tenure and investments in home and

neighborhood (Rohe & Stewart, 1996; Sampson et al., 1999). This may play out in two

ways. Homeowners may not only invest more in their homes at large, but also invest more

in special adjustments to the house that enable them to stay in their home longer. Second,

owners may feel more attached to their homes because of the stronger effort invested in the

living environment, compared with renters (Redfoot, 1987). On the other hand, it must be

noticed that homeowners do not live in houses that seem better suited for the needs of old

age than renters. They often live in detached houses with a garden, in houses with stairs,

and in large houses. Also, more renters live in a house that has been adjusted to specific

needs of the person living in it. The net balance seems positive, however, as homeowners

indicate to be on average somewhat more satisfied with their housing situation than

renters, see the last lines of Table 1.

7. Conclusion

By taking into account a wide array of possible explanations, this study has enabled us to

develop and test a large number of hypotheses on the mechanism behind a difference in

institutionalization rate between renters and owners. The LASA study is not only unique in

its breadth, but also offers a panel spanning over 15 years, allowing the incorporation of

historic change. We have found that after this information is used, a significant correlation

between homeownership and the probability of becoming institutionalized remains. The

remaining correlation does not necessarily represent a causal effect of homeownership on

institutionalization.

First, the possibility remains that even our detailed controls for health status did not

completely eliminate its effect. Health is notoriously difficult to measure and even though

our data offer better possibilities to control for it than most other sources, we cannot

exclude that some effect remains. Nevertheless, we think it is of some importance to see

that the introduction of so many controls does not remove the significant statistical

relationship. The remaining health effect must be due to aspects of health that are not

strongly correlated with those that we have included in our analysis. Note, also, that such

unmeasured health differences must overcome the opposite effects of some other

differences that we observe. For instance, living in a large house limits access to care-at-

home and makes adaptations more costly (as indicated by Easterlow & Smith, 2004), but

most elderly homeowners are able to continue living in such houses. This implies that

omitted effects must be quite strong. We were unable to come up with such effects that

were not yet included in the analysis.

Second, we already mentioned that our data contain limited information about wealth,

and also here some effect may remain that appears in the coefficient for homeownership.

Homeownership and Long-Term Care 759

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

However, at the same time, the strong relationship between wealth and homeownership

would lead one to expect a significant impact of the reduction in the costs of long-term care

for wealthy homeowners that was associated with the institutional change in 1997, and

would also make it more likely that we found a confirmation of a strategic bequest motive.

In both cases, we did not find a significant effect, which makes it less likely that the

remaining correlation between homeownership and institutionalization is due to a wealth

effect.

A third possibility that deserves mentioning is that our results are related to the

differences in mortality between homeowners and renters. This difference explains the

increasing rate of homeownership over time in our sample, which was documented in

Table 1. As institutionalization often signals an increased probability of death, the two are

closely related. However, our attempts of controlling for health status by a series of

indicators are, at the same time, attempts to control for differences in death rates between

renters and owners that are related to health. For this reason, we do not think that this is a

serious problem in our analysis. One would rather think that homeowners, who more often

reach advanced age than renters, would therefore have a higher probability of becoming

institutionalized.

Limitations of the data need mentioning. Foremost, sample attrition is a problem in

samples like ours. Although great care has been taken in the data collection to include

information from the more frail respondents through proxy interviews and post-mortem

information, we disproportionally lack data from respondents who are at high risk of

becoming institutionalized. This does not necessarily bias our results with respect to the

significance of explanatory variables, but it implies that our estimated probabilities of

institutionalization are biased downwards.

Our interpretation of the findings reported in this paper is that most elderly owner

occupiers actively try to stay in their house as long as possible. That is, they do not want to

move to any other dwelling, nursing, or residential home. A probable reason is that the cost

of moving house—and especially its non-monetary part, the effort involved in realizing

the move and becoming settled in a new environment—increases substantially with age

and with decreasing health.17 In other words, the costs of moving rise at the same time

when its potential benefits go up. The success of homeowners in delaying

institutionalization increasingly becomes an obstacle against institutionalization in itself.

There is a strong preference among the elderly for aging in place, even if objective

measures indicate that this place is becoming less suitable for them. It is then perhaps no

surprise that this effect is much stronger for homeowners who have to leave behind their

own property when becoming institutionalized, often for the remainder of their life, than

for renters. Moreover, homeowners appear to be more persistent in trying to realize their

desire to stay in the house in which they more often lived for a very long time than renters.

Being the owner, they probably have a larger say in the decision to stay where they are

than renters in otherwise comparable circumstances. As owners, they may have learned to

be more self-reliant in housing issues than renters. This may show up in differences that

we were not able to measure, such as the ability to arrange optimal care. For example, Bus

et al., (2012) found that cancer patients with high socioeconomic status were diagnosed

earlier and got better treatment than patients with low socioeconomic status, despite equal

access to diagnosis and treatment. According to the authors, the higher status patients were

better able to get the best out of the healthcare system. Such a capability may have only

760 J. Rouwendal & F. Thomese

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

been partially grasped by our measures of education and network characteristics of owners

and renters.

If this interpretation is correct, there are good reasons to facilitate the apparently strong

preference for aging in place among elderly homeowners. This may not only be in their

own best interest, but it will probably also have a mitigating effect on the demand for long-

term care, which is expected to increase substantially in the coming decades. Following

this suggestion could mean that more possibilities are offered for receiving long-term care

at home. This does not necessarily imply that more public money has to be spent: the often

considerable amount of home equity accumulated by the elderly can probably be used to

help finance such arrangements. The present-day (and future) elderly differ from earlier

generations in pension and other wealth, in expected remaining life time, and also in their

demand for care. It seems probable that the demand (willingness to pay) for personalized

care arrangements will increase and that creative arrangements to use the home equity to

make it effective are, in principle, attractive for all parties involved.

Notes

1 The Netherlands has two types of institutions where long-term care is provided: nursing homes and

residential care comparable with sheltered housing. The former provides the most intensive types of

care. The distinction will be explained in Section 3. In order to keep the terminology simple, we will

refer to both types of institutions as nursing homes, unless it is clear from the context that the

distinction matters.2 A surprisingly large number of adults live close to their parents. See Compton & Pollak (2009) for the

US and Hank (2007) for the European countries.3 See Rouwendal (2009, Figure 8).4 This was the case through the period 1992–2007 to which our data refer, although originally admission

to homes-with-care was open to all.5 This makes it unlikely that homeownership is used as a hedge against rent risk.6 Since the time between subsequent waves is approximately 3 years, it is possible that some of the 29

moves to nursing homes were in fact moves to a home-with-care followed by a move to a nursing

home.7 It may be noted that the logit-based discrete time hazard model does not belong to the family of

proportional hazard models.8 It may be noted that for all respondents who died during the time window of our study and who were

not institutionalized in the last wave in which they were observed, the partner or relatives were asked if

the institutionalization had taken place before the respondent’s death.9 This is often referred to as a random effects panel data model. Although for the logit model a fixed

effects approach is also possible, estimation is restricted to the observations for which different choices

are observed over time. In the context of the present paper, this means that estimation must be based

only on respondents who become institutionalized. Right-censored observations cannot be included in

the analysis, which is obviously undesirable. We have therefore not used this approach.10 Over the whole period of observation, 82 owners moved to the rental sector and 24 renters became

owners.11 The role of these dummies is comparable to that of a flexible baseline hazard in continuous duration

models.12 The reason is that there is a linear relationship between time, age, and year of birth.13 If we include the chronic diseases separately (lung disease, cardiac disease, arteriosclerosis, stroke,

diabetes, arthritis, and malignant neoplasm), we see differences in their impact on institutionalization,

but only one coefficient becomes significant (that of arteriosclerosis), while the coefficient for

homeownership changes hardly (0.60) and remains significant.14 This variable is equal to 0 when the number of children equals 0, and increases less than proportionally

with the number of children.

Homeownership and Long-Term Care 761

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

15 We investigated this issue by introducing a cross effect for being a homeowner before the policy

reform in 1997. The hypothesis to be tested was that homeowners had a lower propensity to become

institutionalized before 1997. However, the coefficient of this variable was not significant.16 We investigated this issue by including a cross effect of the number of children and the ownership

dummy into the model. The hypothesis to be tested was that homeowners with children would have a

lower hazard of becoming institutionalized because their children provide more care than those of

otherwise comparable renters because of the strategic bequest motive. However, the coefficient of the

cross effect was insignificant and had the wrong (positive) sign.17 Note that we controlled for the duration of stay in the house.

References

Bernheim, B. D., Shleifer, A. & Summers, L. H. (1985) The strategic bequest motive, Journal of Political

Economy, 93, pp. 1045–1076.

Breeze, E., Slogget, A. & Fletcher, A. (1999) Socioeconomic and demographic predictors of mortality and

institutional residence among middle aged and older people: Results from the Longitudinal Study, Journal of

Epidemiology and Community Health, 53, pp. 765–774.

Broese van Groenou, M. I., Penninx, B. W. J. H. & Deeg, D. J. H. (2003) Socio-economic inequalities in health

decline and mortality in an older population, Aging, 15, pp. 174–183.

Bus, P., Aarts, M. J., Lemmens, V. E. P. P., van Oijen, M. G., Creemers, G. J., Nieuwenhuijzen, G. A., van Baal,

J. W. & Siersema, P. D. (2012) The effect of socioeconomic status on staging and treatment decisions in

esophageal cancer, Journal of Clinical Gastroenterology, doi: 101097/MCG0b013e31824e8ff8.

Cameron, A. C. & Trivedi, P. K. (2005) Microeconometrics (Cambridge: Cambridge University Press).

Compton, J. & Pollak, R. A. (2009) Proximity and coresidence of adult children and their parents: Description and

correlates, Michigan Retirement Research Center working paper 2009-215.

Cox, D.R. (1972) Regression models and life tables (with discussion), Journal of the Royal Statistical Society B,

34, pp. 187–220.

Deeg, D. J. H., van Tilburg, T., Smit, J. H. & de Leeuw, E. D. (2002) Attrition in the Longitudinal Aging Study

Amsterdam: The effect of differential inclusion in side studies, Journal of Clinical Epidemiology, 55,

pp. 319–328.

Dietz, R. D. & Haurin, D. R. (2003) The social and private micro-level consequences of homeownership, Journal

of Urban Economics, 54, pp. 401–450.

Easterlow, D. & Smith, S. J. (2004) Housing for health: Can the market care? Environment and Planning, 36,

pp. 999–1017.

Folstein, M. F., Folstein, S. E. & McHugh, P. R. (1975) ‘ Mini-mental state ’: A practical method for grading the

cognitive state of patients for the clinician, Journal of Psychiatric Research, 12, pp. 189–198.

Fried, L. P., Tangen, C. M., Walston, J., Newman, A. B., Hirsch, C., Gottdiener, J., Seeman, T., Tracy, R., Kop,

W. J., Burke, G. & McBurnie, M. A. (2001) Frailty in older adults: Evidence for a phenotype, Journals of

Gerontology, 56A, pp. M146–M156.

Gaugler, J. E., Duval, S., Anderson, K. A. & Kane, R. L. (2007) Predicting nursing home admission in the U.S.:

A meta-analysis, BMC Geriatrics, 7, pp. 1–14.

Geerlings, S. W., Pot, A. M., Twisk, J. W. R. & Deeg, D. J. H. (2005) Predicting transitions in the use of informal

and professional care by older adults, Ageing and Society, 25, pp. 111–130.

Hank, K. (2007) Proximity and contacts between older parents and their children: A European comparison,

Journal of Marriage and Family, 69, pp. 157–173.

Huisman, M., Kunst, A. E., Bopp, M., Borgan, B. K., Borrell, C., Costa, G., Deboosere, P., Gadeyne, S.,

Glickman, M., Marinacci, C., Minder, C., Regidor, E., Valkonen, T. & Mackenbach, J. P. (2005) Educational

inequalities in cause-specific mortality in middle-aged and older men and women in eight western European

populations, Lancet, 365, pp. 493–500.

Knipscheer, C. P. M., de Jong Gierveld, J., van Tilburg, T. G. & Dykstra, P. A. (Eds) (1995) Living Arrangements

and Social Networks of Older Adults (Amsterdam: VU University Press).

Miller, E. A. & Weissert, W. G. (2000) Predicting elderly people’s risk for nursing home placement,

hospitalization, functional impairment, and mortality: A synthesis, Medical Care Research and Review, 57,

pp. 259–297.

762 J. Rouwendal & F. Thomese

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013

Muramatsu, N., Yin, H., Campbell, R. T., Hoyem, R. H., Jacob, M. A. & Ross, C. O. (2007) Risk of nursing home

admission among older Americans: Does state’s spending on home- and community-based services matter?

Journal of Gerontology, 62B, pp. S169–S178.

Nihtila, E. & Martikainen, P. (2007) Household income and other socio-economic determinants of long-term

institutional care among older adults in Finland, Population Studies, 61, pp. 299–314.

Radloff, L. S. (1977) The CES-D scale: A self-report depression scale for research in the general population,

Applied Psychological Measurements, 3, pp. 385–401.

Redfoot, D. L. (1987) ‘On the separatin’ place’. Social class and relocation among older women, Social Forces,

66, pp. 486–500.

Rohe, W. M. & Stewart, L. S. (1996) Homeownership and neighbourhood stability, Housing Policy Debate, 7,

pp. 37–81.

Rouwendal, J. (2009) Housing wealth and household portfolios in an ageing society, De Economist, 157,

pp. 1–48.

Sampson, R. J., Morenoff, J. D. & Earls, F. (1999) Beyond social capital: Spatial dynamics of collective efficacy

for children, American Sociological Review, 64, pp. 633–660.

Staatscourant (1996) Bijdragebesluit zorg, Staatsblad 486.

Thomese, F. & Broese van Groenou, M. I. (2006) Adaptive strategies after health decline in later life: Increasing

the person-environment fit by adjusting the social and physical environment, European Journal of Aging, 3,

pp. 169–177.

van Gameren, E. (2005) Regionale verschillen in de wachtlijsten verpleging en verzorging (Regional Differences

in Waiting Lists for Nursing and Residential Homes, in Dutch) (The Hague: Sociaal-Cultureel Planbureau).

Woittiez, I., Eggink, E., Jonker, J. & Sadiraj, K. (2009) Vergrijzing, verpleging en verzorging (Population Aging,

Nursing and Ccaring, in Dutch) (The Hage: Sociaal-Culureel Planbureau).

Homeownership and Long-Term Care 763

Dow

nloa

ded

by [

Uni

vers

ity o

f St

elle

nbos

ch]

at 0

3:12

31

Aug

ust 2

013