Trends in the Intergenerational Transmission of Social Assistance in the Nordic Countries in the 2000s

A relationship between a person's social position and that of their parents has been shown in every known society. However, there is no consensus on a theory to explain this. There are many different, sometimes conflicting, theoretical models (Hout and DiPrete 2006; Piketty 2000: 432). At a very general level, the intergenerational transmission of social status can be explained as better off families passing on wealth, human and social capital, as well as values and aspirations to their children, and the children therefore directly inheriting better social positions or at least having better odds when competing for them (Hout and DiPrete 2006; Piketty 2000: 432). Moreover, the neighbourhood and reference group in childhood are known to play a role, as well as the (nursery) schooling system and the costs of higher education (Esping-Andersen 2004). One aspect of the overall social inheritance process is the intergenerational transmission of social disadvantages, which has always been of great interest to policy-makers and academics (Stenberg 2000).

Breaking the chain of intergenerational social disadvantages, or preventing it from forming, is a crucial policy target. The most (cost) effective policy against poverty and exclusion is commonly perceived to be preventing young adults from becoming excluded by breaking the vicious cycle of intergenerational disadvantage (see d’Addio 2007). Recipiency of social assistance, or welfare, is an often-used indicator for social disadvantages. However, empirical studies on the intergenerational transmission of social assistance are not that numerous. This is most likely due to a lack of suitable data rather than a lack of interest. Panel data with a long follow-up are expensive to compile and rare. Retrospective questions about parents’ economic situation during childhood are rather unreliable. Administrative registers are more reliable, but they have their own limitations. And countries where longitudinal administrative data-sets are available for research purposes are few in number.

Another difficulty is the comparability of different studies on intergenerational welfare. Estimates of the intergenerational transmission of social assistance are very sensitive to the length of the observation period, the time difference between the parent–child observation periods and the age of the child during the observation period (Moisio and Kauppinen 2011; Page 2004). However, some cautious comparisons between research results can be made, especially in the USA and the Nordic countries. In the USA, studies are based on panel surveys; Page (2004: 232) has drawn up a summary of these studies. Depending on what observation period is used (both for parents and children) for measuring welfare recipiency, the intergenerational correlation is estimated to be roughly between 0.20 and 0.30. In the Nordic countries, studies are based on administrative register data-sets. According to these studies, the parent–child association in social assistance recipiency seems to be weaker in the Nordic countries compared to the USA. The intergenerational correlation of social assistance is estimated to be 0.10–0.15 in Finland, Norway and Sweden (Lorentzen and Nielsen 2008: 34; Moisio and Kauppinen 2011; Stenberg 2000). A few quantitative studies outside the USA and the Nordic countries do exist, e.g. from New Zealand and Canada. The intergenerational correlation of social assistance is 0.37 in New Zealand (Maloney et al.2003) and 0.15–0.16 in Canada (Beaulieu et al.2005).

However, the research settings and data used vary from one study to the next, so the country differences observed in the above-mentioned studies could simply be due to the difference in research settings and data-sets. Also the difference between the USA and the Nordic countries is partly explained by institutional differences in welfare systems. Social assistance in the Nordic countries is means-tested and discretionary, but it covers the whole population. Experiencing short-term spells of social assistance are rather common especially among young adults (see Eardley et al.1996). But the intergenerational income mobility in general is much higher in the Nordic countries compared to the USA. In the Nordic countries, parent's incomes explain 10–20% of their offspring's income variations, while in the USA they explain 40–60% (Jäntti et al.2006; Jenkins and Siedler 2007).

Register-based studies on the intergenerational transmission of social assistance have increased in the Nordic and a few other countries. A growing number of studies based on longitudinal registers are also shedding light on the old question of the so-called welfare dependency hypothesis. That is, how much is a welfare system that is built to alleviate and protect against poverty actually causing poverty (culture), by creating incentives to remain unemployed, lowering the stigma of welfare use, and educating in how to use the system (Gottschalk 1996; Lewis 1996 (1966); Moffitt 1992). Beaulieu et al. (2005) found that there is a causal link between parental and child reliance on welfare in Canada. However, Edmark and Hanspers (2012) found that their sibling analysis provided no support for a causal effect of parents’ welfare use on children's future welfare use in Sweden.

While studies on the intergenerational transmission of welfare are not numerous, the temporal changes in the intergenerational transmission of social disadvantages are practically uncharted territory (see Wiborg and Hansen 2009). This is perhaps surprising given the central role that preventing intergenerational poverty has been given in the policy debate. Policy-makers usually demand information about trends, as changes can reveal the success or not of a chosen policy. Also the temporal change in intergenerational social mobility is seen as a key aspect of societal change and it has been the core topic in sociological social mobility studies for decades (e.g. Breen 2004).

This study considered trends in the intergenerational transmission of social assistance among young adults in three Nordic countries during the 2000s. To our knowledge, no previous comparative studies have been made on the temporal changes in the intergenerational transmission of social assistance. These three countries offer an interesting setting for a study. The Nordic countries are very similar, but there have been differences in their economic development during the last two decades. The three countries in question each have large welfare states and have similar cultures and institutional settings. Finland and Sweden faced a deep economic crisis in the early 1990s that led to a rapid restructuring of labour markets and tightened welfare state arrangements. Norway, on the other hand, was largely immune to the economic turmoil of the 1990s, thanks to its large oil reserves. Rates of unemployment and social assistance recipients did not rise in Norway in the way they did in Finland and Sweden (Jonung et al.2009).

Unemployment is often the reason for social assistance recipiency among young adults and the number of social assistance recipients is known to follow the changes in the unemployment rate (see Brännström and Stenberg 2007). That can be taken as the starting point for forming a hypothesis on how the trend in intergenerational social assistance relates to economic cycles. The queue theory (Thurow 1975) explains why long-term unemployment does not usually decrease at the same pace as overall unemployment during economic growth. It describes two types of mechanism why employers might discriminate against certain job applicants. The first relates to statistical discrimination in the screening of applicants, where the employer's assessment of job seekers’ productivity is affected by the perception of the average productivity of the group that job seekers belong to, such as long-term unemployed or an ethnic minority. The second mechanism stresses the (unofficial) recruitment channel that delivers information on open positions to the job seekers as well as delivers information about the job applicant to the employer and lowers the employer's uncertainty about the job seeker's productivity (Larsen 2003). When this theory is applied to intergenerational social assistance, we do not have to make an assumption that there is statistical discrimination towards job seekers whose parents were social assistance recipients. But we can assume that parents who are receiving social assistance are more weakly integrated into the labour market than average and that they have less (informal) recruitment channels and labour market arenas to help their offspring find a (first) job. So we can hypothesise that when the economy grows and the number of social assistance recipients declines, the decline in recipiency is stronger among those young adults whose parents were not social assistance (SA) recipients, and the intergenerational odds increase.

The institutional similarity and the differences in economic cycles in the three countries give an opportunity to study how different trends in the number of social assistance recipients relate to the trends in the intergenerational transmission of social assistance. The Nordic countries are also one of rather few countries where administrative registers are available for research on a large scale. Comparable administrative register data-sets were available from all three countries. The Finnish data is a 25% sample of all persons born between 1979 and 1988. The Norwegian and Swedish data each included an entire birth cohort born between 1976 and 1988. Ten mobility tables were constructed by linking the social assistance status of 20-year-olds with their parents’ social assistance status four years earlier. For testing country differences in the intergenerational transmission of social assistance, we combined all mobility tables into a layered mobility table that was modelled with log-linear and log-multiplicative layer effect models.

Nordic countries are small, export-oriented economies with large and extensive welfare states. Overall the tax-burden is 40–50% of gross domestic product and it provides universal and high-social security (including work pensions) with universal public services, including health care and social services, nursery schools and (tertiary) education. Social assistance is seen as a residual system that gives temporary aid in situations where universal social security is insufficient. Social assistance programmes are a last-resort form of means-tested economic support that is available to nearly all citizens as a guaranteed minimum income, corresponding to a reasonable standard of living. Social assistance often includes a basic cash benefit for daily living expenses and housing costs, and possible supplements to cover the special needs of the household and case-specific payments for occasional needs (Eardley et al.1996). However, in the last two decades, social assistance has become a more prominent part of income protection in the Nordic countries. This is partly because of structural changes in the labour markets, namely increased unemployment, and partly because of the restructuring of the welfare state. These two are highly interrelated developments (Timonen 2003). Long-term unemployment and labour market exclusion lead to more and more people being ineligible for earnings-related unemployment protection and social insurance, relying therefore solely on social assistance.

An economic crisis struck the Nordic countries in the early 1990s, especially Finland and Sweden, with Norway affected to a much lesser extent. Unemployment rose rapidly in Finland and Sweden, from 2–3 to 10% in Sweden and to 16% in Finland. In Norway, unemployment only rose from 6 to 7% (see Figure 1). Unemployment declined after the mid-1990s in Finland and Sweden, but remained on a higher level than before the 1990s. In contrast, in Norway, unemployment declined to a much lower level than before 1990. High unemployment eroded the fiscal base of the Finnish and Swedish welfare states and several welfare retrenchments and labour market deregulations took place in Finland and Sweden as a consequence (Lorentzen et al.2014). By the 2000s both the Finnish and Swedish economies and welfare states had undergone large structural changes. Institutional changes also took place in Norway, though on a much smaller scale (Kvist et al.2012).

One consequence of the welfare state restructuring was (applying also to Norway) that last-resort social assistance became a more prominent part of income protection. High unemployment together with restrictions in unemployment insurance protection resulted in an increase in the numbers of social assistance recipients in Finland and Sweden (Figure 1). In Finland, the number receiving social assistance increased from 6 to 12–14% during the first half of the 1990s – and declined to 10% during the second half of the 1990s. In Sweden, the increase–decrease of SA recipients was not as stark as it was in Finland during the 1990s: from 6 to 8% and then back to 6%. In Norway, there was only a slight increase in SA recipients during the early 1990s, while in the second half of the 1990s, the number of SA recipients decreased to a lower level than in the 1980s. However, in all three countries, social assistance became de-facto the only income support for large numbers of unemployed, especially for the young unemployed, who did not fulfil the eligibility conditions for earnings-related or basic unemployment benefits. What is interesting is that despite the strong appearance of similarity between the social assistance systems, the proportion of population receiving social assistance in 2008 is three times higher in Finland (6%) and two times higher in Sweden (4%) than in Norway (2%). One obvious explanation for this is country differences in unemployment. The unemployment rate in Norway is on a level where it could be described as having reached full-employment. Unemployment is the biggest explanation for the increase of social assistance recipients among those of working age (Marx and Nelson 2012). As can be seen in Figure 1, the proportion of population receiving social assistance follows closely the changes in the unemployment rate.

There are some differences in the social assistance systems in the Nordic countries, though in an international context they are very similar to each other. Social assistance systems in Finland and Sweden are classified according to Gough et al. (1997) into the citizenship-based but residual assistance regimes. These regimes are characterised by a high degree of codified rights to the recipients and relatively generous benefits. However, the Norwegian social assistance system is classified as a decentralised, discretionary relief regime that is characterised as less generous benefits and being highly discretionary (and having fewer recipients). However, the recasting of the welfare states in the 1990s has moved Finland and Sweden in the direction of a more residual system (Kuivalainen and Nelson 2012). And all three countries have made stronger attempts to tie work or activation obligations to social assistance recipiency after the early 1990s (Immervoll 2012).

It is difficult to estimate how much the reinforced activation policies in SA have influenced the number of recipients. The activation measures in these three countries do not directly have an effect on the number of recipients, since refusal to take part in an activation programme means a reduction in the SA benefit, not rejection of SA. However, the take-up rate in SA is known to be low and for Sweden it has been substantially hollowed out (see Bargain et al.2012; Kuivalainen and Nelson 2012), so the reinforced activation may have had a bigger impact on the number of recipients via a deterrent effect. On the other hand, high long-term unemployment and tightening sanctions have led to more young adults being ineligible for unemployment protection, leaving social assistance as their sole income protection. This might increase social assistance recipiency, though not necessarily, since social assistance is a so-called topped-up benefit that includes compensation for housing costs. Hence, a long-term unemployed person who loses eligibility for unemployment protection due to sanctioning is often already in social assistance.

Comparable administrative register data-sets were available from Finland, Norway and Sweden. The data-sets were ordered from and constructed in the National Statistical Offices and after anonymising, the data-sets were given to the researchers. The Finnish data is a 25% sample of all persons born between 1979 and 1988, while the Norwegian and Swedish data each included an entire birth cohort from 1976 to 1988. The data-sets contained information on the number of months that a household had received social assistance during a calendar year.

To be able to study the changes in the intergenerational transmission of social assistance in the 2000s, we need to construct a research setting that keeps age- and observation-period-related factors constant. Estimates of the intergenerational transmission of social assistance are very sensitive to the length of the observation period, the time difference between the parent–child observation periods, and the age during the observation period (Moisio and Kauppinen 2011). The data-set enables us to follow a single birth cohort across single calendar years and in this way control for the factors mentioned above. This data-set enabled us to compare year by year the social assistance recipiency of 20-year-old young adults in the period 1999–2008, together with information on their parents’ social assistance recipiency when those young adults were aged 16 years, that is, from the period 1995 to 2004.

Social assistance is a household benefit in the Nordic countries, so both parents act as recipients of SA. The parents of household in which the target person lived in the position of a child when 16 years old, are defined as the parents of the target person. The parents, or a single parent, are not necessarily biological parents. Those 20-year-olds who are still living with a parent(s) are treated as a separate household if they receive social assistance for themselves. This applies also as the (lawful) practice in social assistance regarding those aged 18 or older who are still living in their parent(s) household in the three study countries. The length of recipiency was grouped into four categories: 0 month, 1–3 months, 4–9 months and 10–12 months of SA during a calendar year. Ten mobility tables were constructed by linking 20-year-olds’ social assistance status with their parents’ social assistance status four years earlier. A 4 × 4 mobility table was constructed for each year from 1999 to 2008, forming a layered 10 × 4 × 4 mobility table for each country. In Norway and Sweden the data allowed for additional mobility tables (1996–1998), though these were used only for constructing the descriptive analysis in Table 1.

To test country differences in the intergenerational transmission of social assistance, we combined all mobility tables into a 3(country) × 10(year) × 4(parents SA) × 4(children SA) mobility table. We analysed the association in this layered, multidimensional frequency table with log-linear and log-multiplicative layer effect models (Xie 1992), using the LEM programme (Vermunt 1997). The aim of log-linear modelling in general is to produce as simple models as possible (based on a theory) that explains the associations in the frequency table. The idea is very similar to the standard chi-square test. The model fit is estimated with chi-squared values (G²) in relation to the degrees of freedom (df), a dissimilarity index (Δ) and a Bayes information criterion (BIC; Vermunt 1997: 74). A well-fitting model will have a Δ close to zero and a negative BIC that is as low as possible. A model with a negative BIC is preferred to a saturated model. The statistical significance test has a somewhat different interpretation here, since we are analysing population data and not a sample. Also, the usual limit of a well-fitting model (p > 0.05) is not sensible given the number of cases used (see Kuha and Firth 2011).

Table 1 presents proportions of social assistance recipients during the origin and destination years and the conditional probabilities (as well as odds-ratios) for the intergenerational social assistance uptake. The data is weighted by four in the Finnish table to represent the population. The Swedish and Norwegian data-sets are population level data. The size of the 20-year-old age group (All_N) is largest in Sweden, having increased from circa 98,000 to almost 118,000 between 1999 and 2008. The size of the 20-year-old age group has also increased in Norway, from circa 52,000 to almost 59,000. In contrast, the size of 20-year-old age group has decreased in Finland, from circa 70,000 to 67,000.

The proportion of social assistance recipients among 20-year-olds (SA20y) has decreased during the 2000s in all three countries. In Finland, the proportion of 20-year-olds receiving social assistance decreased in the period 1999–2008 from 19 to 12%, in Norway from 8 to 5% and in Sweden from 16 to 10%. The proportion of 16-year-olds whose parent(s) received social assistance in the period 1995–2004 decreased also (SA16y), although much less than rates among 20-year-olds. In Finland, the proportion of 16-year-olds living with a parent(s) who received social assistance decreased from 8 to 6% during the period 1995–2003, in Norway from 6 to 5% and in Sweden from 11 to 8%. These declining trends are parallel to an overall decline in social assistance recipients in these countries since the mid-1990s (see Figure 1).

In 2008, the intergenerational odds-ratio for social assistance (odds) was at the same level in Finland and Norway, but higher in Sweden. The intergenerational odds-ratio for social assistance became stronger in Finland and Sweden in the period 1999–2008, but not substantially in Norway. Those young Finns whose parent(s) received social assistance had an odds-ratio of 4.7 in 1999 for themselves having social assistance, compared to those whose parent(s) did not receive social assistance. In 2008, the odds-ratio had increased to 8.0. In Sweden, the intergenerational odds-ratio of social assistance increased from 10.5 to 13.9 during the same period. In Norway, the intergenerational odds-ratio of social assistance increased fractionally from 7.7 to 7.8.

When looking at the conditional (transition) probabilities, we can see that the proportion of young adults having social assistance has decreased both among those with an SA family and non-SA family background, though more so among the latter group. Among those whose parent(s) receive social assistance, the transition probabilities have decreased from 48 to 46% in Finland, from 57 to 49% in Sweden and from 39 to 26% in Norway in the period 1999–2008 (SA20y, 16y). The transition probabilities for social assistance among those young adults whose parent(s) had did not receive social assistance has declined from 17 to 10% in Finland, 11 to 6% in Sweden and 6 to 4% in Norway in the period 1999–2008 (SA20y, if not 16y).

The number of so-called two-generation recipients has also decreased in all three countries in the period 1999–2008. The two-generation recipients are defined as those 20-year-old young adults who are SA recipients and whose parents were also SA recipients four years earlier. The proportion of two-generation recipients of all 20-year-olds (SA16y& 20y_All) is 4% in Sweden, 3% in Finland and 1% in Norway in 2008. In all three countries, the proportion of two-generation recipients among young adults has declined in the period 1999–2008. In Finland and Norway, almost a quarter (24%) of young adults in SA are two-generation recipients (SA16y& 20y_SA20y), and in Sweden the figure is 40%. The proportion of two-generation recipients among young SA recipients is increased from 20 to 24% in Finland and from 24 to 40% in Sweden.

We might assume that the longer the parents are in social assistance, the stronger it predicts their children's social assistance recipiency. Table 2 presents the intergenerational mobility tables of social assistance in 2008 in Finland, Norway and Sweden. The intergenerational association between parent's and children's social assistance recipiency is strongly influenced by the length of the parent's recipiency in Finland and Sweden, but not in Norway. Among those young adults whose parent(s) received social assistance only for a short period (1–3 months), 36% had received social assistance in Finland and Sweden, and 25% in Norway. Among those whose parent(s) received long-term (10–12 months) social assistance, 62% received social assistance in Finland, 64% in Sweden and 25% in Norway. In Finland and Sweden, parent's long-term recipiency predicts long-term recipiency also for their children, but not in Norway. In Finland, 22% and in Sweden 31% of those young adults whose parent(s) had received long-term social assistance were long-term social assistance recipient themselves – compared to 7% among those whose parent(s) received social assistance only for a short period. In Norway, 5% of those coming from a long-term social assistance family were long-term social assistance recipients themselves, compared to 3% among those whose parent(s) received social assistance for a short period.

It seems clear that the length of social assistance recipiency plays an important role in the intergenerational transmission of social assistance. If we were to observe the trends in the intergenerational transmission of social assistance using simple odds-ratios, as given in Table 1, we might derive a distorted image of the trend. For this, we turn to model the full 4 × 4 transition tables with log-linear models and test whether the so-called fluidity has changed in these transition tables in the period 1999–2008. We use standard models and techniques that are used widely in social mobility research (Breen 2004; Erikson and Goldthorpe 1992). Table 3 presents the model fit estimates of the log-linear models fitted to the social assistance mobility tables. First, the layered 9 × 4 × 4 mobility tables are analysed separately for each country and after this, tables are combined into a single layered 3 × 10 × 4 × 4 mobility table that has three country-layers (C) and 10-year-layers (Y) for 4 × 4 parent (O) –child (D) mobility tables.

The first model (I) is the independence model and it gives us a baseline for studying the associations in the YOD tables in each country. The independence model assumes that there is no association between Y, O and D; hence, there is no association between the year and parent's and children's social assistance recipiency. The independence assumption did not hold: The model has 144 degrees of freedom with a likelihood-ratio value of 9686 (G²) and an index of dissimilarity (Δ) of 5.1% in the Finnish data. In the Norwegian data, the G² is 16,383 and the Δ is 2.5% and in the Swedish data the G² is 129,973 and the Δ is 7.7%.

The next model (II) is the so-called perfect mobility model that assumes that there is no association between a parent's and their child's social assistance recipiency, though the parent's and child's recipiency can vary between years. When comparing the model fit estimates to the independence model, we can see that the marginal distributions of social assistance recipiency among parents and children vary between years. This was already obvious from the figures presented in Table 1. Moreover, we can conclude by studying the model fit statistics that the intergenerational association is strongest in the Swedish mobility table, and weakest in the Norwegian mobility table.

The third model (III) is the so-called constant fluidity model that assumes that the level of recipiency varies between years and that there is an association between the parent's and children's recipiency, but that this association is the same from one year to the next. The model fit estimates tell that this model has already a quite good fit. In the Finnish data, the likelihood ratio is 138 with 81 degrees of freedom and the model misclassifies only 0.5% of the cases. The constant fluidity model also has a good fit to the Swedish data; the likelihood ratio is 303 and the model misclassifies 0.3% of the cases. In the Norwegian data, the constant fluidity model reproduces cell frequencies with an accuracy such that the difference between estimated and observed cell frequencies is not statistically significant (p-value is 0.164).

The fourth model (IV) is the so-called unidifference model that assumes that the OD sub-table (i.e. parent–child) has the same intergenerational association pattern across the years, but the strength of associations can vary in different years. This is done by applying log-multiplicative layer effects and assuming a uniform difference in the OD tables between years. Each OD mobility table is weighted by the β coefficient according to the strength of intergenerational associations in each year. In the Finnish and Norwegian data, the unidifference model reproduces the cell frequencies with such accuracy that the difference between the estimated and observed cell frequencies is not statistically significant. In the Finnish data, the p-value is 0.764 and in the Norwegian data it is 0.434. In the Swedish data, the difference is statistically significant (p-value is less than 0.001), however, the model fit is good: The model misclassifies only 0.2% of the cases in the Swedish data. By studying the coefficients of unidifference parameter β we can see that the OD association increases in the period 1999–2008 in the Finnish data from 1.00 to 1.28, and in the Swedish data from 1.00 to 1.07. In the Norwegian data, the β coefficient does not change i.e. indicating that the intergenerational association has not changed in Norway in the period 1999–2008. This result is in line with the conclusion made from the odds-ratios presented in Table 1. The intergenerational transmission of social assistance has become stronger in Finland and Sweden during the 2000s, but not in Norway.

In models V–VIII the same independence, perfect mobility, constant fluidity and unidifference models are tested against the combined data, where Finnish, Norwegian and Swedish mobility tables are pooled into a single layered mobility table. The final model VII is the unidifference model where the strength of association in the OD sub-table is allowed to vary a constant amount by country and year. The model has 331 degrees of freedom, the G² is 1751 and the Δ is 0.5%. Hence, the model has a very good fit; although the difference between estimated and observed cell frequencies is statistically significant, with a p-value less than 0.001.

Figure 2 presents the coefficients of the unidifference parameter β from model VIII for each country and year. The parameter β summaries trends and country differences in the intergenerational association (fluidity) of social assistance in the three countries for the period 1999–2008. The Finnish 1995–1999 mobility table is the baseline (1.00) to which all other OD tables are weighted according to the strength of association in that OD table. The coefficients β parameter supports the conclusion made from the odds-ratios and other descriptive figures in Table 1. The intergenerational transmission of social assistance is strongest in Sweden, but the intergenerational association has increased most in Finland. The intergenerational association was stronger in Norway than in Finland in the early 2000s, but during the 2000s the intergenerational association strengthened in Finland and by 2008, it had caught up to the level of Norway.

The two main results of our study are that the intergenerational association of social assistance, measured either as odds-ratios or fluidity, is stronger in Sweden than in Finland or Norway – and the intergenerational association has increased in all three countries during the 2000s. The trend in intergenerational odds-ratios is associated with the declining trend in the number of 20-year-old recipients in these three countries. When observing also the trends in the transition probabilities, it revealed that the probabilities of transitioning into SA have declined in all three countries. However, they have decreased more among those coming from a non-SA family compared to those coming from an SA family. This has strengthened the intergenerational odds-ratios.

Like in all studies on intergenerational mobility, it is difficult, if not impossible, to reduce the results into a single coefficient. We have studied the intergenerational transmission of social assistance with three estimates: odds-ratios, fluidity and transition (conditional) probabilities. Odds-ratios and fluidity have the advantage that they adjust for the differences and changes in marginal distributions, which is vital when studying temporal changes in intergenerational mobility. Transition probabilities have the advantage that they give a more nuanced picture of mobility. The problem is that these three estimates can give quite different pictures of the development of intergenerational mobility. The intergenerational odds-ratio and fluidity can show that the intergenerational association has strengthened at the same time as transition probabilities are seen to decrease, as long as the transition probabilities have decreased more in one group than in another. This is what we have observed in our study. At least for policy-makers it might be difficult to explain that the intergenerational transmission of social assistance has increased when those coming from a social assistance family have a smaller probability of ending up claiming social assistance. The methodological debate on how to measure the association between parents’ social class and their children's educational attainment shows how big an impact the choice of measurement can have on policy implications (see Hellevik 2002).

We have to also remember that the coefficients of intergenerational transmission of social assistance are very sensitive to the age when recipiency is observed (Page 2004). This is very crucial when interpreting the results of this study, since our data include only young adults. The intergenerational odds of social assistance recipiency are highest in the early twenties, after which the odds decline rapidly until the late twenties. Young adulthood is economically a precarious time of life and parents’ ability to help their offspring financially has a bigger importance during this phase than later in life, when the position in the labour markets has usually stabilised (Moisio and Kauppinen 2011). To explore whether age during the observation period influenced our results, we formed another data-set where social assistance recipiency was observed at the ages of 16 and 24 years and made the same analysis as with the data of 16- and 20-year-olds. The level of intergenerational association of social assistance was lower, as could be expected. However, results relating to the trends and country differences in the intergenerational transmission of social assistance were the same.

The trend in the intergenerational transmission of social assistance seems to relate to economic cycles in ways that we hypothesised in the introduction based on the queue theory of unemployment (Larsen 2003; Thurow 1975). The theory gives an explanation for why long-term unemployment seems to always decrease less than overall unemployment during an economic boom. According to our results, a similar phenomenon takes place when the number of social assistance recipients decrease. The probability of claimimg social assistance decreases less among those young adults who have a family background of social assistance compared to others and this increases the intergenerational odds-ratio. Parents can offer formal and informal recruitment channels to their children that are very important especially for those just entering the labour market. Parents who have claimed social assistance are more likely to be in unemployment than other parents, and in this way they can offer less recruitment channels to their children while they are seeking their (first) job. This gives an explanation as to why, when the overall amount of social assistance recipiency declines during an economic growth, social assistance recipiency declines less among those young adults coming from a social assistance family.

Unfortunately the observation periods in our data are limited, in the case of parents to the years 1995–2004, and in the case of children to the years 1999–2008; we therefore cannot study how the intergenerational transmission of social assistance develops when the economic cycle and the number of social assistance recipients moves in the opposite direction. The most recent economic downturn that began in 2008 with the financial crisis will soon give us register data with which to study how the intergenerational transmission of social assistance has changed when the number of social assistance recipients has increased. This will hopefully reveal more about the link between the number of social assistance recipients and the intergenerational transmission of social assistance.

Pasi Moisio is a Research Professor at the Minimum Income Unit, National Institute for Health and Welfare (THL) in Finland. His research interests and main areas of publication are poverty and income inequality, intergenerational socioeconomic mobility and minimum income protection. He is involved in several expert groups developing the legislation of social security and benefits.

Thomas Lorentzen is an Associate Professor at Department of Sociology at the University of Bergen, Norway. He works with longitudinal analyses of register data focusing on labour market outcomes and welfare receipt. He also teaches quantitative methods for the social sciences. Lorentzen is currently involved in several comparative register-data based analyses of the Nordic welfare states.

Olof Bäckman is an Associate Professor of sociology at the Swedish Institute for Social Research (SOFI), Stockholm University, Sweden. His research has mainly concerned poverty, unemployment and social exclusion in a life course perspective with particular focus on the youth-to-adulthood transition. How structural factors such as social policy, educational policy and the economic cycle intervene in processes of cumulative disadvantage is an important theme in his research.

Anna Angelin Ph.D. is a Social Policy researcher at School of Social Work, Lund University, Sweden. She is also Head of Research for EconSec Lund University Economic Security Institute. Her main research interests and areas of publication are marginalisation, social assistance and youth-child poverty.

Tapio Salonen is Professor in Social Work and Dean at Faculty of Health and Sciences, Malmö University in Sweden. His main research interests include poverty, marginality, participatory strategies and social policy.

Timo Kauppinen is Senior Researcher at the Minimum Income Unit of the National Institute for Health and Welfare (THL) in Finland. His research interests and main areas of publication include, for example, social assistance receipt and educational outcomes among young people, and urban and housing research.