1Disability pension (henceforth shortened to DP) – i.e., compensation to persons who cannot work because of disability – is an important service provided by welfare states to raise the quality of life of disabled citizens (Dickson & Hyde, 2000). Denmark is no exception, with DP expenditures in 2014 amounting to 5.6 billion euros, corresponding to 52% of all public disability expenditures (Statistics Denmark, 2017). In Denmark entitlement to DP requires that one’s working ability is permanently reduced due to impairment. Thus, the DP criteria include both impairment and working ability.
2Previous research suggests that onset of disability may not only affect persons’ ability to work and lead to impoverishment (Braithwaite & Mont, 2009), but can also impact parenting (Clarke & McKay, 2014) and marital stability, at least for men (Singleton, 2012). The predictors of disability benefits (including pension) are well-documented (e.g., Canivet et al., 2013; Gustafsson et al., 2014). In contrast, the consequences of receiving disability benefits, in particular for non-work-related aspects of life, are less studied, and the findings of previous research may not permit causal interpretation (e.g., Chen & van der Klaauw, 2008; Øverland et al., 2008). Does receipt of a permanent social benefit in the form of a DP ease – as intended – the difficulties arising from a disability? The answer to this question is important not only for people with disabilities (Cross, 2013) but also for scholars interested in the impact of welfare state policies for vulnerable groups in society and for policy makers assessing whether such benefits should be extended or curtailed.
3The economics literature on DP has focused on the extent to which a DP crowds out work, and it finds that it is very much the case (e.g., Maestas, Mullen & Strand, 2013). It is worth remembering, however, that the purpose of a DP is to enable the disabled persons to enjoy life like other people as much as possible – including in the sphere of partnerships – without having to work. It is therefore no surprise that DP reduces the supply of work.
4Research has documented the negative impact of disability on employment, independent income and relationships. Compared to the non-disabled, disabled persons enter marriage later in life and are also more likely to exit marriage through divorce or separation (Clarke & McKay, 2014; Singleton, 2012; Teachman, 2010). Explanations for marital disruption among the disabled draw on social, psychological and economic theories of divorce that focus on the mutual emotional and practical support (e.g., affection, insuring against financial hardship) between partners. They argue that the risk of divorce is raised by an event that decreases the real or perceived benefits from marriage or increases the gains associated with exiting the relationship to be single or in favor of another partner.
5A permanent disability of a partner is such an event (e.g. Becker, Landes & Michael, 1977; Levinger, 1979; Peterson, 1979). From this perspective, the question is whether a DP can help to cushion the fall in the real or perceived gains from marriage caused by the disability. This question is particularly relevant in a country such as Denmark characterized by a high degree of gender equality and relatively low social and economic costs of divorce. Indeed, Denmark has one of the world’s highest divorce rates (Christoffersen, 2004; Statistics Denmark, 2017; Hansen, 2003). In sum, one would expect a permanent disability to increase the probability of divorce among the disabled, particularly in a Danish context marked by progressive laws and social norms regarding divorce. Previous studies (e.g., Singleton, 2012) that have examined the consequences of onset of disability for marital stability have not distinguished clearly between the effects of disability and the effects of DP. The effect of DP on the recipient, in particular on marital stability, thus remains an open question. We answer this question by tracing the effects of a DP independently of the disability of the person.
- 1 Divorce is far from taboo in Denmark and does not pose an obstacle to contracting marriage again. A (...)
6To do this we use an instrumental variable (IV) regression, exploiting inter-municipal variation in the granting of DPs as an instrument to obtain more accurate estimates of the causal impact of DP on recipients’ marital stability, which is defined as the absence of divorce. In Denmark, many couples live together, often for many years, without being married. Although there are no longer religious or moral reasons to get married, a marriage still means an intention to remain together and divorce therefore means that this project is abandoned.1 In the next section, we present an overview of the literature on the effects of social benefits on marriage. The overview indicates that the effects of permanent social benefits such as DP on divorce have not been examined previously. We are thus the first to investigate this problem. We then describe the statistical method and the model we use to assess the impact of DP on the stability of marriage. Next, we present our findings, which suggest that a DP per se stabilizes a marriage, especially in the short run, but also in the longer run. We conclude by discussing these findings from the perspective of the broader literature.
7Research on the impact of social benefits on inter-personal relationships is mainly from the US. It examines the impact of such benefits on both entry to and exit from relationships as well as on fertility, and presents a wide range of results. It is only partially relevant to our research question, as it mainly concerns younger people who receive various forms of social welfare assistance, typically single mothers. More important, the research has only examined the effect of temporary benefits, thus leaving a major gap in the literature on the impact of social benefits.
8In the US, interest in the impact of social benefits has often been driven by moralistic concerns that welfare benefits create poverty and social problems (Murray, 1984). Writing in this vein, Niskanen (1996) perceives social benefits as causal in all manner of social “pathology,” including aid dependency, poverty, children born out of wedlock, unemployment, abortion, and violent crime. The US federal support programme, Aid to Families with Dependent Children (AFDC 1935-1996), was effectively limited to single parents. The programme was suspected of reducing marriage rates and providing incentives to women to have children rather than work. However, research on the impact of AFDC shows varying results. For instance, some scholars found that AFDC counteracted as well as supported dissolution of relationships (Moffitt, 1997), while others found no effects or very small ones on relationships (Blackburn, 2003; Hoffman & Duncan, 1995; Hoynes, 1995). Others argued that AFDC’s effects on marriage varied according to labour market conditions and political context (Harknett & Gennetian, 2003). Concerns about AFDC’s effects on people’s life choices, in particular that the scheme encouraged women to become single mothers, became grounds for replacing AFDC with the more restrictive Temporary Assistance for Needy Families (TANF). A study of the TANF scheme in its earliest years found that it produced fewer divorces, but no more marriages (Bitler et al., 2004).
9In Britain, research on the impact of social benefits has looked at the “in-work benefit,” which is conditional on the presence of children in the household as well as a combination of income and household characteristics. For example, Francesconi and van der Klaauw (2007) found that the benefit has almost no effect on marriage stability, except in cases where the man has low income and there are small children in the household. In such cases, in-work benefits were found to increase the number of people who divorced, and the number who obtained jobs and kept them. Child support is an important element for marriage continuity, but in-work benefits were found to have almost no effect on continuity among those with many children (Francesconi & van der Klaauw, 2007). Another British study, Anderberg (2007), examined the impact of social benefits and tax rules on relationships and births. It found that the number of relationships was reduced by 7% for each £100 per week it cost to be a couple compared to being single, regardless of whether the cost resulted from reduced social benefits or tax legislation. Overall, Anderberg (2007) found that social benefits functioned as a form of income insurance, which facilitated easier movement into and out of relationships.
10In sum, much of the research suggests that social benefits have no or negligible effect on the frequency of marriage formation and dissolution. This research is valuable but leaves two important gaps. First, it is only about temporary social benefits, not permanent benefits such as the Danish DP. Second, the research is about younger persons, not about the group of 35-64 year olds, which is the group that most often receives a DP in Denmark. Both the duration of a social benefit and the age of the recipients can have a major influence on the outcomes being examined. A temporary benefit produces uncertainty, whereas a permanent benefit provides security. Furthermore, there are inter-generational differences in the nature of relationships. For example, young people seem to engage in more relationships of shorter duration; in contrast older people tend to have fewer but longer relationships (Waite & Lillard, 1991). Thus, though the above-mentioned results may be relevant for younger people weighing the pros and cons of entering into a relationship, they do not apply for older persons in well-established relationships. Our study context – Denmark – and data allow us to fill these gaps and thereby extend research in this area.
- 2 A flex job is a scheme where a person can work reduced hours and receive a full salary from the empl (...)
11Before describing our data and method, we outline relevant features of the Danish welfare state model for processing applications for sickness and disability benefits, in particular because the US system for applying for and deciding a disability benefit, which is the basis for most of the literature in this area (for details, see Maestas, Mullen & Strand, 2013), is different from the Danish system. In Denmark, a DP is the outcome of an extensive process in which municipal authorities play a decisive role. A DP applicant has at least 1 year of prior contact with the municipal authorities, and often long before the notion of such pension is discussed. If a person has worked for a number of years prior to entering a long period of sickness, the municipal authorities first arrange for rehabilitation measures from the date of receipt of the first sickness cash benefit. Depending on how successful the rehabilitation is, the person may return to a regular job or a less burdensome “flex job.”2 Only when all attempts at rehabilitation have failed is a DP taken up for consideration. Figure 1 illustrates the various stages in this process.
Figure 1. Stages in the processing of applications for sickness and disability benefits in Denmark
- 3 In the language of statistical analysis, risk only means probability, without any moral views on th (...)
12In Danish social policy, a DP is not a benefit that citizens can choose to apply for but one that municipalities can choose to grant citizens fulfilling certain conditions. We have identified a group of persons with specific characteristics to whom municipal authorities would be likely to grant a DP. This group constitutes the base population in our study and consists of married persons who, because of a health condition that has prevented them from receiving income from work, received in the period 1997-2006 long-term sickness benefits and are thus considered “at risk”3 for a DP. Right to sickness benefits presupposes that the person has been in work, and the group consists typically of persons who had been working for a number of years and then suffered a severe illness or accident. The long period of sickness may entail an onset of disability. Even if this is the case, a DP may not be awarded. With a DP the couple will have a lower income, as DP on average is 57.6% of the recipient’s former income. Without the DP the couple would have a much lower income, because social assistance is dependent on household income, and right to private invalidity insurance most often depends on DP, which means that most of the group with no pension will not have any right to other benefits either. In the face of such a sharp drop in income, couples without DP will have an incentive to divorce because the partner who cannot work will then be entitled to social assistance. It is, however, difficult to obtain a pension via pro forma divorces because municipalities often place the citizens under surveillance in order to detect whether they are cohabiting while at the same time receiving social benefits as single persons.
13Our data are from Statistics Denmark’s detailed population registers and consist of nine cohorts of married individuals in the period 1997-2006 ranging in number from 7515 to 11,929, giving a total of 85,986 observations. We define being on long-term sick pay (i.e., being on sick pay for at least 270 days in a year) as our “risk situation” because being in this situation increases the risk of sickness turning into disability. Therefore, the relevant outcome is whether those on long-term sick pay were awarded a DP shortly after. The first cohort of persons fulfilling this condition comprises those who in 1997-1998 received sick pay for at least 270 days. We identified the persons in this group who obtained a DP in 1999. For this first cohort, 1999 is called “year 1.” Persons who were in the risk group in 1997-1998 but who received a DP later than 1999 are excluded from the sample because the reasons for such an outcome are unclear. The second cohort consists of persons who in 1998-1999 received sick pay for at least 270 days, and we determine whether they obtained a DP in 2000, which for this group is called “year 1.” The same calculations are applied in 1999-2000 and so on up to 2005-2006, the ninth cohort.
- 4 Previous research (Bengtsson, 2002) shows that most of the people who apply for DP and are rejected (...)
14Table 1 provides descriptive data for the nine cohorts in our sample. It shows, for example, that persons in the at-risk population on average are 46.75 years old and have been married for 19.69 years. Their individual income before coming into the risk situation exceeds DP by 73.5%, which is quite normal for people of that age. When we compare these persons to their spouses on health related variables we find, as one might expect, that a greater proportion of at-risk persons (33%) used psychotropic medications compared to their spouses (16%). They also visited their general practitioner (GP) more frequently (10.8 visits on average compared to 7.3 visits among spouses). In most cases (72%), citizens on long-term sick pay were not considered for a DP, likely because they recovered and resumed work. Thus, 28% of the persons in the risk situation were assessed for their eligibility to receive a DP, and the vast majority of these – constituting 25% of the persons in the risk situation – were awarded pensions while the remaining 3% were not granted any pension.4 We compare the group that received a pension with the 72% of the clients who were not judged qualified to apply for a pension.
Table 1. Descriptive data for population of married clients at risk
|
|
|
Mean for groups are
|
|
Variable
|
Numbers of observations
|
Group mean (nine cohorts)
|
Awarded a DP in year 1
|
Without a DP in year 1
|
Still married in year 6
|
Divorced in year 6
|
|
DP year 1
|
85,986
|
0.25
|
1.00
|
0.00
|
0.26
|
0.25
|
|
Gender (female = 0, male = 1)
|
85,986
|
0.41
|
0.38
|
0.42
|
0.39
|
0.46
|
|
Age in years
|
85,986
|
46.75
|
47.73
|
46.43
|
46.99
|
45.95
|
|
Duration of marriage in years
|
84,915
|
19.69
|
20.50
|
19.42
|
20.11
|
18.21
|
|
Age difference
|
84,869
|
- 0.46
|
- 0.63
|
- 0.40
|
- 0.55
|
- 0.15
|
|
Gender x Age difference (interaction term)
|
84,869
|
1.12
|
1.08
|
1.13
|
1.04
|
1.35
|
|
Use of psychotropic medicine years-2, -1
|
85,986
|
0.33
|
0.39
|
0.31
|
0.31
|
0.40
|
|
Use of psychotropic medicine (partner) years-2, -1
|
84,869
|
0.16
|
0.17
|
0.15
|
0.14
|
0.20
|
|
GP use, avg. For years-4,-3, and-2
|
81,015
|
10.83
|
11.83
|
10.50
|
10.87
|
10.71
|
|
GP use (partner), avg. For years-4,-3, and-2
|
73,475
|
7.31
|
7.65
|
7.20
|
7.13
|
7.97
|
|
Partner has DP (= 1, else = 0
|
84,869
|
0.07
|
0.09
|
0.07
|
0.07
|
0.08
|
|
Average income/DP in years-4,-3, and-2
|
85,986
|
1.74
|
1.66
|
1.76
|
1.74
|
1.73
|
|
Average unemployment, year-4,-3, and-2
|
85,986
|
109.86
|
127.06
|
104.13
|
107.87
|
116.55
|
|
Min. 10% more unemployent than partner
|
85,986
|
0.62
|
0.59
|
0.63
|
0.62
|
0.64
|
|
Genderxmin 10% more unemployment than partner (interaction term)
|
85,986
|
0.28
|
0.25
|
0.30
|
0.27
|
0.32
|
|
Lives in municipality with medium density
|
85,986
|
0.15
|
0.14
|
0.15
|
0.15
|
0.14
|
|
Lives in the South Denmark region
|
85,986
|
0.23
|
0.24
|
0.23
|
0.23
|
0.22
|
|
Lives in the Zealand region
|
85,986
|
0.16
|
0.16
|
0.16
|
0.16
|
0.17
|
|
(Dummies for year 0=1997,…, 2005)a
|
|
|
|
|
|
|
DP year 1: 1 = the person is awarded a pension in year 1; 0 = else; Age difference = age of person in risk group–age of partner. Gender*Age difference (interaction term): If person at risk is male, the term = the variable “Age difference.” If person at risk is female, the term = 0. Psychotropic medicine year-2 and-1: 2 = person at risk used psychotropic medicine in years-2 and-1; 1 = person at risk used psychotropic medicine in year-2 or-1; 0= else. Person has minimum of 10% more unemployment: 1 = person’s average unemployment in years-4,-3, and-2 exceeds partner’s average unemployment in years-4,-3, and-2 by at least 10%; 0 = else. Gender*Person minimum 10% more unemployment (interaction term): If person at risk is male, the term = the variable “Person has minimum of 10% more unemployment”; if person at risk is female, the term = 0. Municipality with medium population density (municipalities are categorized into three groups by population density, low, medium, high): 1 = person at risk lives in municipality with medium density; 0 = person at risk lives in municipality with high or low density. Detailed variable definitions are provided in the appendix.
a The nine cohorts comprise 10,128, 7928, 7671, 7515, 8854, 9550, 10,510, 11,929 and 11,901 persons, respectively. Data derive from the following Danish registers: National Board of Appeals (DP awards), Statens Serum Institut (health data), and Statistics Denmark (all other data).
- 5 In any case, the same procedural delay can be assumed to apply for those with and without a DP.
15The dependent variable in the analysis, “still married after x years,” is a dummy variable coded 1 for those persons who are still married in each year of the nine-year period covered by our analysis. The main independent variable, for which we use the instrument, “awarded DP year 1,” is also a dummy variable. Divorce is measured by the date of the legal judgement and this date may safely be assumed to be relatively close to the actual decision to divorce. Divorce is relatively easy to obtain in Denmark. Today a divorce can be obtained immediately through the Internet if agreed to by both parties, and lacking such an agreement, after a six-month separation. In the period we examine, divorce could be obtained after six months separation if agreed to by both parties, and if not, after twelve months of separation.5
16In addition, we include the following independent variables as controls, drawing on the literature on other factors influencing marriage formation and dissolution. Gentleman and Park (1994) found in Canada that an age difference between a married couple where the man is 2 to 10 years older gives the most stability and least risk of divorce. Therefore, we control for difference in partners’ age. We control also for the number of children in the family (Waite & Lillard, 1991). Because men have long had the role of breadwinner in the family, Becker’s theory (1973) predicts that the man’s income influences the ease with which a relationship is established and maintained, while the woman’s income has the opposite effect. White and Rogers (2000) conclude on the basis of US data that both the man’s and the woman’s income are associated with more marriages, fewer divorces, better marriages, and enhance the children’s lives. We therefore control for income. We also control for the partners’ employment status, as research from a number of countries suggests, overall, that having a job is an important predictor of marriage longevity (Blekesaune, 2008; Cohen, 2014; Hansen, 2005; Jensen & Smith, 1990; Poortman, 2005; Schaller, 2013; Schoen, Rogers & Amato, 2006).
17Divorce is more common in the first years of a marriage and in younger age groups. It is a process that sorts out marriages with a bad match and preserves those which function well. So the duration of a marriage may be considered a measure of its quality and we control for this factor. Use of psychotropic medicine and use of general practitioner are used as proxy indicators for overall health prior to the risk situation and serve as controls for risk of disability. Region of the country and living in municipalities with high, medium and low population density are used to tap cultural differences, e.g. in social norms and attitudes towards divorce, and other geographic differences, including differences in risk of disability. As some variables, e.g. age difference between spouses, may have different effects for men and women, we also include them as interaction terms with gender. In addition, we include a series of year dummy variables to capture temporal variation in marital stability over the period.
18The main challenge in assessing the effect of DP on marital stability is to separate it from any concomitant effect the disability leading to the pension may have. To distinguish between the impact of disability and the impact of DP, we use a statistical method, IV regression, that isolates the impact of DP. Using Ordinary Least Squares (OLS) regression, the standard tool in the regression toolkit, has the disadvantage that the variation in the indicator “Disability Pension (DP)” would inevitably be accompanied by a variation in the indicator “disability,” as these two variables are strongly associated. We would therefore be unable to isolate the effect of the DP on risk of divorce.
19The solution lies in finding a variable (an “instrument”) that is associated with the variable “DP” but is unrelated to the variable “disability,” the use of which will therefore ensure that the estimated effect on marital stability is only an effect of “DP,” not of disability. The rate at which municipalities award DPs (“Municipal DP award rate”) is one such instrument. Use of this instrument in an IV regression exploits the fact that in Denmark the granting of a pension to disabled persons is not automatic but is partly the outcome of municipal decisions.
20The relationship between the onset of disability and granting of DP contains both deterministic and probabilistic elements. It is not only the person’s disability that determines whether the outcome will be a DP; there is also an element of discretion introduced by municipal decision-making processes, and differences in municipal practice. Inter-municipal differences in DP award rates in Denmark have been examined in a number of studies, revealing considerable variation across municipalities. The studies have concluded that the municipal administration of DP explains much of this variation (Bengtsson, 2002). In the present study, award rates in our population vary widely – between 5% and 50% – in the 271 municipalities. Figure 2 shows how the municipalities are distributed across four population quartiles, where DP award rates of 24%, 28% and 32% delimit the quartiles. This also means that the probability of being granted a DP is much higher in some municipalities. For example, in the two largest municipalities in our sample, Aarhus and Copenhagen, it is 28.6% in and 17.0%, respectively.
Figure 2. Distribution of Danish municipalities in the at-risk population across four quartiles of disability pension award rates. The at-risk population consists of those on long-term sick pay (i.e. on sick pay for at least 270 days in a year). Own calculations using data from National Board of Appeals
- 6 In the US, Maestas, Mullen and Strand (2013) used differences between case managers’ standards for (...)
21The award of DP is thus not a completely random occurrence. We argue, however, that it is partly determined by the person’s ability to work, as intended, and partly by other reasons that are unrelated to such ability and may be linked to traditions in the municipality and local opportunities for rehabilitative intervention. In other words, these reasons are related not to the disabled individuals whose pension case is being assessed, but rather to the municipality where they reside. If we assume that municipal award rates are unrelated to the incidence of disability and to the rate of marital stability in the municipalities, the effect of DP on marital stability can be isolated. To justify the assumption that municipal award rates are unrelated to the incidence of disability or divorce, we control for persons moving to the more generous municipalities, and for intermunicipal differences in divorce rates. This accounts for the possibility that individuals intentionally move to municipalities that are more likely to grant DP or have lower divorce rates.6 Figure 3 below illustrates the logic underlying our IV regression model with municipal award rates as the instrument.
Figure 3. Illustration of the logic underlying the instrumental variable regression
22To estimate the effect of DP on marital stability, we use the STATA biprobit command, which implements a binary probit analysis in two steps, with municipality used as a clustering variable. In the regression equation that estimates divorce (i.e., the absence of marital stability) as a function of a DP and certain background factors, “DP awarded year 1” is replaced by the predicted probability of obtaining a DP, which is determined by the total award rate in the person’s home municipality and the same background factors as in the principal equation.
23As mentioned above, we control for whether an applicant’s domicile is random by studying movements into and out of municipalities. We compare the number of persons who in the previous 5 years moved to a municipality that is more generous than the one they came from, with the number that moved to a municipality that is less generous. Because the two figures were almost identical, there is no evidence that people move to municipalities where it is easier to obtain a DP. We also checked that the instrument itself does not affect marital stability, by looking at the correlation between divorce rates among 40-59 year olds in general and municipal award rate. It turns out that this correlation is quite small (0.027), far from being significant – and moreover positive.
24We control for various predictors of divorce as mentioned in the Variables section. However, the model is limited in that it does not permit inclusion of background variables that occur later than the pension award and application process. Therefore, all background factors are measured in year–1 and earlier. This means that factors such as the presence of very young children in the household could not be included. Moreover, it means that when we look at marital stability five years after the possible granting of DP, we cannot take into account the effect of income and unemployment in the previous six years, but only in the years before then. All background factors found to be statistical significant are included.
25We tested for whether our decision to use the selected instrument was justified by conducting the standard test to check whether the IV regression was better than an OLS regression. Technically, this means using a Wald test to determine correlation between error terms of the model’s two equations. We found a correlation of –0.24 between the error terms, which is significant (P = 0.000) meaning that the IV regression is substantially better. The equations employed in calculating estimates are in short form (the full equations are given in the appendix). In Eq. (1), instead of DP year1 itself, we use the term that is estimated from the instrument, the municipal disability award rate, according to Eq. (2):
P(married)=DP year 1*coefficient+sum of back ground factors∗coefficients (1)
P(DPyear1)=disability award rate*d+sum of back ground factors∗coefficients (2)
26Table 2 compares disabled persons who were granted a DP to those who were not on two key variables: the duration of marriage and the divorce rate. It shows that a higher percentage of the group that received a DP in year 1 was still married immediately afterward (96.52% vs. 90.35%). This suggests, at first glance, that DP prolongs marital stability. But already from year 2, the percentage of new divorces is larger in the DP group than in the non-DP group. For example, the divorce rate in year 2 is 5.47% and 4.02% for the two groups, respectively. In year 6, the proportion of surviving marriages is nearly the same in the two groups, 77.27% and 76.83% for those with and without a DP in year 1, respectively – a difference of only 0.44 percentage points.
Table 2. Descriptive data on marriage or divorce in at-risk population, years 0-6
|
Still married if DP
|
|
New divorces in year 1-6 for
|
|
Year
|
Is not awarded in year 1 (%)
|
Is awarded in year 1 (%)
|
Difference: awarded-not awarded
|
|
Group without pension (%)
|
Group with pension (%)
|
Difference: without-with (%)
|
|
0
|
100.00
|
100.00
|
|
|
|
|
|
|
1
|
90.35
|
96.52
|
6.17
|
|
9.65
|
3.48
|
6.17
|
|
2
|
86.33
|
91.05
|
4.72
|
|
4.02
|
5.47
|
– 1.45
|
|
3
|
83.27
|
86.30
|
3.03
|
|
3.06
|
4.75
|
– 1.69
|
|
4
|
80.87
|
82.67
|
1.80
|
|
2.40
|
3.63
|
– 1.23
|
|
5
|
78.68
|
79.85
|
1.17
|
|
2.19
|
2.82
|
– 0.63
|
|
6
|
76.83
|
77.27
|
0.44
|
|
1.85
|
2.58
|
– 0.73
|
As the risk population is about 85,000 persons, even small differences are statistically significant.
27At first glance, the descriptive data thus suggest that a DP preserves marriage in the short run, but that the effect fades out over the following 5 years. The “fast” divorces may well include people who cannot afford to live together without the supplementary income from sick pay when it ends and prefer to divorce in order to qualify for full social assistance. The right to sick pay, based on employment in the past (a minimum of 3 months), is a benefit of limited duration (one year at the time of this study, with scope for extension only in exceptional cases). Most people who received long-term sick pay could not continue to receive this benefit in the event they did not obtain a DP. If they did not return to work and their spouse could not provide for them, they likely received social assistance.
28Apart from this early effect, the percentage of divorces in the non-DP group is actually lower in years 2 through 6. So the descriptive data give the – misleading, as we shall show – impression that having a DP leads to a higher divorce rate after the initial year.
29The results of the IV regression with the dependent variable “married after 5 years” are presented in Table 3. They suggest, overall, that a DP has a direct causal impact on marital stability even after 5 years after controlling for a host of other factors and, most important, net of the effect of the disability itself. Specifically, the results of the stage 2 probit regression shows that DP in year 1 increases the probability of still being married up to year 6 to a significant degree (b = 0.358, P < 0.001).
Table 3. Results of stage 2 of a two-stage probit regression of marital stability after 5 years on DP. Dependent variable: still married
|
Variable
|
Coefficient
|
Standard error
|
z
|
P
|
|
DP awarded year 1
|
0.358
|
0.078
|
4.60
|
0.000
|
|
Partner has DP year-1
|
–0.112
|
0.022
|
–5.12
|
0.000
|
|
Gender
|
–0.110
|
0.017
|
–6.34
|
0.000
|
|
Age of persons in risk group
|
0.563
|
0.134
|
4.19
|
0.000
|
|
Age2 of persons in risk group*
|
–0.019
|
0.005
|
–4.02
|
0.000
|
|
Age3 /100 of persons in risk group*
|
0.029
|
0.007
|
3.90
|
0.000
|
|
Age4 /10,000 of persons in risk group*
|
–0.016
|
0.004
|
–3.89
|
0.000
|
|
Age difference (persons in risk group–partner)
|
0.006
|
0.002
|
3.83
|
0.000
|
|
Gender age difference
|
–0.009
|
0.003
|
–3.30
|
0.001
|
|
Duration of marriage
|
0.033
|
0.002
|
15.56
|
0.000
|
|
Duration of marriage2 *
|
–0.001
|
0.0001
|
–9.29
|
0.000
|
|
Psychotropic medicine year-2,-1
|
–0.136
|
0.008
|
–16.46
|
0.000
|
|
Psychotropic medicine (partner) year-2,-1
|
–0.089
|
0.010
|
–8.60
|
0.000
|
|
GP use (partner) year-4,-3,-2
|
–0.003
|
0.001
|
–3.98
|
0.000
|
|
Average income/DP year-4,-3,-2, grouped
|
–0.021
|
0.009
|
–2.42
|
0.015
|
|
Average unemployment year-4,-3,-2/100
|
–0.030
|
0.004
|
–6.94
|
0.000
|
|
Min 10% more unemployment than partner
|
–0.085
|
0.016
|
–5.27
|
0.000
|
|
Gender ≥10% more unemployment than partner
|
0.063
|
0.021
|
2.98
|
0.003
|
|
Lives in municipality with medium density
|
0.043
|
0.015
|
2.88
|
0.004
|
|
Lives in the Capital region
|
–0.047
|
0.015
|
–3.03
|
0.002
|
|
Lives in the South Denmark region
|
–0.144
|
0.017
|
–8.44
|
0.000
|
|
Lives in the Zealand region
|
–0.110
|
0.017
|
–6.40
|
0.000
|
|
Year dummies
|
–0.113–0.000
|
|
–4.79–0.00
|
0.008 - 0.000
|
For variable definitions, see appendix. Age2: age*age; age3: age*age*age; age4: age*age*age*age; duration of marriage2 *: the same function is calculated for duration of marriage. We use here Taylor’s law, which says that any function in an interval can be approximated arbitrarily well with a polynomial if you just include a sufficient number of powers of x. The same function is calculated for duration of marriage, where duration2 is included. Test of the suitability of a biprobit model: Rho = –0.2333 with standard deviation = 0.0462; Wald test of rho = 0: X2 = 23.70; P = 0.0000; ATE = 0.0974, standard deviation = 0.0014. The results of stage 1 regression (dependent variable = DP awarded in year 1) are presented in Table A1 in the appendix.
- 7 This is as expected, since we are unable to control for the spouse’s disability here.
- 8 An interaction term between a variable and gender means that the variable has different effect for (...)
- 9 In this case, the interaction term coefficient is so large that we find opposite effects of unemplo (...)
- 10 In developing the model, we tested for other variables found to be important for marital longevity (...)
30Not surprisingly, whether the partner has a DP is also important (b = −0.112, P < 0.001) but the relationship is inverse: the probability of still being married in year 6 is lower.7 Being older relative to one’s partner makes divorce less likely, but only for women, as the interaction term8 exceeds the age difference term. (Note that male partners most often are older, e.g. a woman in the at risk population nearly the same age as her partner has a more stable marriage than a woman who is much younger.) The client’s economic condition is represented by income in years-4 to -2, average unemployment rate in years -4 to -2, and by a variable that indicates whether the person’s unemployment duration was 10% longer than the partner’s. If the latter is true, the marriage is less stable (b = −0.085, P < 0.001), but only if the person is a man with proportionately more unemployment. If the person is a woman, the relationship is reversed by the interaction term (0.060).9 Local differences in culture and changes over the nine years are taken into account by the last variables (geographical location and year dummies), but they are of no interest in this context.10
- 11 This is a technique to assess the effect on a group with a composition similar to ours. Bootstrappi (...)
31We were unable to control for a few factors mentioned in the literature. This includes for example, the number of children younger than 6 years (because most of them are born after year -1) and the number of children with disabilities (because Danish registers do not contain information on children’s disabilities). The size of effects was estimated by bootstrapping.11
32Table 4 presents the estimates of the average treatment effect derived from the regression results in Table 3. The results indicate that a DP reduces the risk of divorce by 5.8 percentage points after 2 years and by 9.7 percentage points after 5 years, counting from the year in which the DP is awarded. Given that the total risk of divorce for the at risk population is 16% in year 3 and 23% in year 6 (not shown in the table), it means that the risk of divorce is reduced from about 17% to 11% by year 3 and from 25% to 15% by year 6.
Table 4. Effects of disability pension after 2 and 5 years on marital stability (average treatment effect)
|
Difference between DP recipients and non-recipients
|
|
After 2 years (percentage points)
|
After 5 years (percentage points)
|
|
Married (in year 0) who are still married
|
+ 5.8
|
+ 9.7
|
|
Married men (in year 0) who are still married
|
+ 8.0
|
+ 12.2
|
|
Married women (in year 0) who are still married
|
+ 5.5
|
+ 8.1
|
Difference between DP recipients and non-recipients is statistically significant (P < 0.05).
33A comparison of the results in Table 4 to the descriptive data in Table 2 reveals the “naïve,” misleading nature of the latter. For the first year, both tables show a higher risk of divorce for the group that did not receive DP compared to the group that did receive DP (+6.17 and +5.8 percentage points, respectively, for Tables 2 and 4). After the first year, Table 2 shows more divorces each year in the group with pension (e.g., in year 2, 5.47% of those with DP versus 4.02% of those without DP). The results from the IV regression (Table 4) show the opposite: there is a higher risk of divorce among those without a DP and the difference continues to grow though not at a constant rate. Thus, after five years it is 9.7 percentage points, whereas the difference in Table 2 is just 1.17 percentage points (79.85%–78.68%).
34Table 2 gives a misleading impression because the group of DP recipients also differs systematically from the non-DP group by having more or greater disabilities. The IV regression removes this second difference from the comparison and shows that a DP per se (separated from onset of disability) means fewer divorces. Table 2 shows the combined effects of the disability and DP (for those who get it) on marriage and divorce. Many in the group who did not receive a pension are divorced over the following years, seemingly as an immediate consequence of a lack of DP. But in subsequent years more persons who received pensions were divorced, thereby (apparently) eliminating the difference.
35In contrast, Table 4 shows the effect of the pension itself when disability is kept constant. The immediate effect of a pension award here too is fewer divorces in the pension group than in the non-pension group right from the start. But, here the trend of fewer divorces in the pension group continues through the years. Table 4 shows that the difference between the two groups does not decline but increases. Thus, we see that the specific effect of a DP is a continued stabilizing effect on marriages. DP thus has an impact that counters the effect of the disability (Singleton, 2012).
36Table 4 also shows that the importance of DP for preserving marriage is larger for a man than for a woman, not only in a 2-year but also in a 5-year term. Among men, the difference in marital stability between DP recipients and non-recipients is 3-4 percentage points greater than it is among women, suggesting that a DP is more important for shoring up the marriage of male recipients. This suggests more generally that the role of breadwinner, an important element in traditional masculine identity, is still an important part of the husband’s role in marriage in a Danish context. Such an interpretation is supported by a study by Pierce, Dahl & Nielsen (2013), who, using a regression discontinuity model on Danish data, found that if the husband’s salary is less than the wife’s, the husband uses more medication for erectile dysfunction. In sum, the above results underscore the advantage of using a causal method like IV regression, which is able to distinguish clearly between disability and DP.
37Our analysis has contributed to the literature in a number of ways. First, most existing research on the effects of social cash benefits on marriage concerns temporary benefits and young people. We extend this research by examining the effects of a permanent social cash benefit (DP in Denmark) among older persons. Second, we focus on marital stability whereas most of the existing literature focuses on work and benefit dependency. Our main concern has not been that DP is a disincentive to work – in fact, the very aim of a DP is to enable the disabled person to live without working. Therefore, instead of integration into the labour force we have considered another equally important aspect of life, marital stability. Our analysis suggests that in a situation with a sizeable risk of disability followed by divorce, a DP acts to preserve the recipients’ marriage. Third, we contribute by distinguishing between consequences of disability and consequences of DP. This is by no means simple given the efforts to ensure that DP is awarded only to persons with disabilities. Therefore, there is a strong association between disability and DP. To estimate the effects of DP it is, nevertheless, necessary to factor out the effects of disability, so we used IV regression. In contrast to the misleading impression given by the descriptive analysis, the results of the IV regression indicate that a DP reduces the risk of divorce by almost 6 percentage points after 2 years and by around 10 percentage points after 5 years, counting from the year in which the DP is awarded.
38It makes sense to speak of the consequences of receiving DP. Disability is a complex phenomenon, but the granting of DP is a human decision. Even if it is obvious that disability has consequences, for various reasons those consequences may be hard to pin down. Disability may be many different phenomena – somatic, psychiatric and mental – and it is often unclear when exactly the onset of a disability occurs. Thus, Greve (2012) found that schizophrenia caused an income loss all the seven years before the first diagnosis. Hence, disability is a much more complicated concept and empirical phenomenon than DP and determining its consequences presents therefore an even greater challenge.
39Another question is whether our results, which are obtained from data from 1997-2006, are relevant for decisions about DP today. If they were an argument for relaxing the criteria and lowering the threshold for awarding DP in Denmark in the period 10-20 years ago, it can be argued that they provide even stronger grounds for doing so today, when access to DP has become more restricted with the reform of 2013.12 Hopefully, our analysis will contribute to thinking more broadly about DP than is often the case. Most of the debate concerns the question of disincentive to work. Even if this is an important question, we should not forget that DP may have wider consequences, including those for quality of life and social relations. The wide-ranging deleterious consequences of disabilities for quality of life are widely documented. Social welfare programmes, including the provision of disability benefits, in a number of countries seek to alleviate the negative effects of disability (Dixon & Hyde, 2000). Our study shows that such benefits help to improve the lives of disabled persons by stabilizing their marriage.
40A final question is whether these results are applicable to other countries. We would argue that they are generalizable to the other Nordic countries, Finland, Iceland, Norway and Sweden. These countries employ a similar approach to social security provision as that used in Denmark, have similarly comprehensive welfare state policies as well as congruent social norms and attitudes regarding divorce and the disabled (Dixon & Hyde, 2000; Sainsbury, Lawson & Priestley, 2017).
- 13 To investigate this we tried the same method for married persons who have received social assistanc (...)
41Our analysis also has a number of limitations. First, as already mentioned, IV regression cannot take into consideration events and conditions that occur after the risk situation. We define a risk situation as married persons experiencing a long-term sickness and trace the consequences of receiving a DP immediately after this risk situation, given the degree of disability. Even though we look at the consequences two and five years later, we are unable to take into consideration what happens in the meantime, e.g. if the persons experience an income loss, unemployment spell or changes in health after year 0. Nor can we take into account whether people have a child after year 0. The reason is that such changes could have been caused by the decision regarding DP. But evidently, in reality there will also be events that have entirely other reasons. They will produce a random variation in our model. Second, our findings cannot be considered the effect of a permanent social benefit on mature married persons in general, as they are linked to the concrete risk situation we use. It is likely that analyses for other risk situations will give different results.13 A third limitation is, as already mentioned, that the result concerns only persons in the grey zone, who would be awarded DP in some municipalities but not in others. In practice, this is not a serious weakness, as it is exactly these persons that will be affected by changes in the award threshold, and therefore the persons it is most interesting to know something about. A final limitation is that the result is linked to the period that we considered, 1998-2006. Since then the threshold for acquiring a DP has been raised. This fact must be taken in consideration when using the results in debates about DP today.