The effect of long-term care public benefits and insurance on informal care from outside the household: empirical evidence from Italy and Spain

In Italy, the cost and design of public LTC-related services is highly fragmented and is shared simultaneously between the State, the regions, and the municipalities [23]. Cash benefits are the most important pillar of the Italian LTC public intervention in terms of expenditure and number of older people affected [23]. The main cash benefit, established by the law of 11 February 1980 [24], is the indennità di accompagnamento made available by the Social Security in the whole country to severely disabled people needing the permanent help of a relative to carry out the activities of daily living. There exist also some cash benefits provided by the regions, provinces and municipalities [25]. Public home help for personal care and domestic tasks as well as institutional care are managed by municipalities in coordination with the National Health Service.

Whereas health care services for elderly are free of charge in Italy, public home help for personal care and domestic tasks is means-tested and users can pay up to its full cost. There is a wide variation in the co-payment modalities as they are defined by municipalities [25]. The indennità di accompagnamento is universal and not means-tested and was set at €508 per month in 2015 independently of the age and place of residence of the recipient. It was granted to 363,868 individuals at the end of 2015 [26]. This cash benefit, which was initially thought as a measure to support informal caregivers, now serves as well to remunerate private home help, in particular help given by migrant workers [23].

To be declared eligible for the indennità di accompagnamento, an individual needs to be assessed by a health commission in a specialized centre or clinic after an appropriate period of observation or hospitalization [27] as 100% disabled and dependent, i.e. in need of continuous assistance or unable to walk without the permanent help of a relative. For other regional and municipal LTC services and cash benefits, eligibility criteria are not homogeneous, and each region has a specific dependency classification system taking into account mainly activities of daily living (ADL) limitations and to a lower extent instrumental activities of daily living (IADL) limitations [25].

Finally, the private LTC insurance market is rather thin in Italy [28]. Some insurance companies offer private LTC coverage, with products consisting of a life annuity in case of permanent or full dependency.

In Spain, the regions and the municipalities offer universal LTC public coverage following the 39/2006 Law [29]. The prestaciones y servicios para la autonomía y la dependencia, i.e. the dependency benefits and services, are granted to all individuals recognized as dependent regardless of their age, geographical location and financial situation. This subsidy can be either in kind in the form of formal care or financial as a percentage of formal care cost. According to the law, in kind subsidies have priority over the financial ones. In kind formal care can be provided at home, in nursing homes and in so called “day” or “night” centres. Financial subsidies can only be used to purchase formal LTC (if publicly provided LTC is not available) or to purchase specific personal assistance services.² In 2015, 745′720 individuals received the prestaciones y servicios para la autonomía y la dependencia, either as in kind or financial subsidies [31].

The system is financed via general taxation and means-tested co-payments of the users. The average co-payment is estimated to be €304, €412 and €662 per month for moderately, severely and major dependent respectively, which represents about 50% of the total cost [32]. The assessment of users’ participation to the total cost of LTC is complex due to regional heterogeneity [32, 33].

Severity of dependency is evaluated by a socio-medical team following a visit and an interview at the place of residence of the person applying for public benefits. The evaluation tool is a unified scale that has been approved in 2011 by the Spanish government under the Royal Decree 174/2011 [34]. To be eligible to public services and subsidies, an individual has to be declared at least as moderately dependent, i.e. needing help to perform several ADL at least once a day or needing limited or not continuous help to be autonomous. Initially, only individuals recognised with major and severe dependency were covered by the public LTC system and it was not until mid-2015 that moderately dependent individuals became eligible to public coverage [31].

Finally, in Spain, the private LTC insurance market is rather small with 37,225 insured in 2015 [35]. However, the market is quite dynamic and shows high growth rates. Between 2012 and 2015, the number of insured experienced a growth rate of 29%, probably due to a low starting point [36]. Private insurance benefits can be either in the form of a pre-determined lump-sum or in the form of an annuity. Their eligibility criteria are tighter than the public ones, as private companies only recognize severe dependency corresponding to individuals needing a very high or permanent amount of support to stay autonomous.

We summarize the main characteristics of the Italian and Spanish LTC financing systems in Table 1.

We use the SHARE database to empirically study the effect of both public support and private LTC insurance ownership on the receipt of informal care in both Italy and Spain. SHARE is a multidisciplinary, longitudinal and cross-national micro-database containing information on health-related variables, labour market variables, economic variables and other variables (including education, housing, social support and family structure) of a representative sample of European individuals aged 50 years or older and their spouses. The first wave of SHARE was released in 2004. SHARE follows the design of the U.S. Health and Retirement study and the English Longitudinal Study of Ageing. For more details on the survey, readers should refer to Börsch-Supan and Jürges [37].

For the purpose of our study, we use data from the sixth wave of the SHARE database. The fieldwork of the sixth wave was completed in 2015, released in 2017 and contains information about 68,231 individuals from 18 different European countries. We discard the use of data from other waves as the Spanish public LTC system was not fully in place until mid-2015.

The subset of SHARE regarding Italy and Spain contains 10,949 observations, 5313 corresponding to Italy and 5636 to Spain. A restriction to individuals having at least one mobility, ADL or IADL limitation leaves us with 5097 observations, 2417 from Italy and 2680 from Spain. In addition, due to missing values for some variables, 236 and 336 observations are lost in the Italian and Spanish samples, respectively (19, respectively, 10 are lost for missing information on limitations). Thus, our final sample includes 4525 observations, 2181 corresponding to Italy and 2344 to Spain. Finally, for models including the control variables Net wealth and Regional dummies, additional missing values leave us with a total of 3760 and 3932 observations, respectively.

In this section, we present the variables used in our analysis, in particular, informal care receipt and LTC coverage, along with their descriptive statistics.

Table 4 provides a summary and description of the set of variables considered in the econometric models. Sample mean values are reported separately for Italy and Spain.

There is a significant overlap on the dependent variables Informal care and Informal care for IADL, as more than 90% of those who receive help by family members living outside the household receive it as care for IADL. While SHARE distinguishes private mandatory and voluntary LTC insurance, we merge them into one type of private insurance (variable LTCI private) to work with a variable maximizing the number of individuals privately covered. The informal care and insurance related variables’ sample means reflect the trends commented earlier in Sections “Informal care receipt” and “LTC coverage”. Additionally, Italians are less likely to receive formal home care when compared to Spaniards which is consistent with the observed differences in public LTC coverage. In both countries the average household is composed of 2.2 members and surveyed individuals have on average about 2 children. Roughly 63% of the individuals are married, around 18% are widow and between 20 and 25% of respondents live with their children.

In our econometric analysis, we run three probit models on the three binary variables defining informal care provided by family members from outside the household (see Table 4). This type of regression is suited when the dependent variable takes only two values. More formally, for each country, we model an individual’s probability of receiving informal care by the following equation: where \({\text{IC}}_{i}^{j}\) with j = 1, 2, 3 corresponds to the three dummy variables defining informal care in an aggregate way (j = 1), for ADL (j = 2) and for IADL (j = 3). While the superscript j is linked to the three regressions, the subscript i is linked to the observations, i.e. the responses from the surveyed individuals. \({\text{LTC}}_{i}\) refers to the two variables defining public and private LTC financing and \(X_{i}\) to the independent variables in Table 4 selected as control variables. Assuming the error term \(\varepsilon_{i}\) is normally distributed with mean zero, Eq. (1) can be estimated using a probit model.

We consider as control variables for the final model only those variables in Table 4 which fulfil two criteria. The first is to decrease the Akaike information criterion (AIC) [41] and the second is to be statistically significant at the 10% level. These criteria ensure that the selected variables improve the model’s goodness of fit without raising substantially the risk of overfitting. For the degree of dependency, only the variable with the highest explanatory power among the three mobility, ADL and IADL limitations is included, because of the large collinearity existing between them (i.e. ρ > 0.6). The selection of covariates is performed in the joint regression model with IC¹ as dependent variable (first column of Table 6). We retain this selection across the other models to have a unified specification.

Controls were tested under different forms (linear, binary and categorical) and we retained the one improving the most the AIC. Alternative specifications including more controls were also tested for all regressions. Results did not change substantially, validating our method for the selection of covariates. Finally, we also investigate the relationship between LTC financing and the probability of formal care use. In this case, our dependent variable is the formal home care utilization variable and LTC financing, informal care receipt by family members and the selected controls are used as explanatory variables.

The numerical results from the model calibration are presented in Table 5.

In Italy, public LTC support is found to impact significantly and positively informal care (\({\text{IC}}^{1}\)) in the first regression. The opposite result is found for Spain, where public LTC coverage has a significant negative impact on informal care provided by family members. This result would tend to confirm our initial hypothesis that different public LTC financing typologies may have a different impact on informal care. On one hand, in Italy, public LTC is characterized by a mixed system consisting of a universal national cash benefit granted to severely dependent (the indennità d’accompagnamento), complemented by a very heterogeneous set of additional cash and in-kind benefits provided and regulated at the regional and municipal levels. On the other hand, in Spain, while either in-kind or financial benefits are also granted to moderately dependent, they depend on formal care consumption. Hence, in Spain, benefits are received conditionally on the receipt of formal care which is likely to provide disincentives for informal care. This does not necessarily happen in Italy where cash benefits, the most important pillar of public LTC support, are not conditioned on the receipt of formal care and have more strict eligibility criteria. In Italy, cash benefits can be used to compensate informal caregivers which may explain the positive relationship between public coverage and informal care. Hence, such cash benefits would fulfil their initial role thought to support informal caregivers [23]. These first findings would tend to support the hypothesis of LTC public support decreasing the receipt of informal care for Spain but would reject it for Italy.

When informal care is defined only in terms of help with ADL or with IADL (second and third columns in Table 5), our results do not importantly change. When care is defined as informal care for IADL, the main difference is that the effect of private LTC insurance becomes statistically significant at the 10% level in both countries. Additionally, the estimate corresponding to public LTC suffers from an important reduction in Spain when informal care is defined as help with ADL. As help with ADL is a more intense form of care, this result is consistent with the findings of Bonsang [12] and Bolin [11] showing that formal and informal care are weaker substitutes if the intensity of care is high.

In the fourth column, we run a regression with the general definition of informal care as the dependent variable by additionally controlling for the individual’s net assets including housing wealth. More specifically, we include three dummies related to the country-specific quartiles of the sample wealth distribution. When including this variable, the sample size is substantially reduced due to the presence of missing values. Our results show that the coefficients corresponding to public LTC benefits do not suffer relevant changes, whereas those corresponding to private LTC insurance are reduced and become non-significant. Furthermore, we find that wealth has a non-linear effect on informal care.

The fifth set of results contains the regression models using formal care as dependent variable. As laid out above, we control for informal care receipt and for the rest of the variables except wealth (results do not change when we include it). As suspected, in Spain, where public LTC support is conditioned to formal care receipt, we find a positive association between LTC coverage, both public and private, and formal home care. In Italy, where the system is mixed, we find as well public and private LTC coverage to be positively associated with formal home care. Hence, on one hand, in Spain, LTC public benefits increase formal care and decrease informal care. On the other hand, in Italy, LTC public benefits increase both formal and informal care receipt.

Concerning the rest of the control variables, formal care is not significantly correlated with informal care in Italy, while it is significantly positively correlated with the three categories of informal care in Spain. A causal interpretation for this control variable’s marginal effect is beyond the scope of this paper. Indeed, we include formal care utilisation in the regression as an independent variable to control for the effects of public support and private insurance in both countries by formal home care availability. Having been in the hospital is significantly and positively associated with the three categories of informal care as well as with formal care. The number of IADL limitations is positively and significantly related to the receipt of both informal and formal care and a better health has the opposite effect on both variables. The number of members in the respondent’s household has a very significant increasing negative effect on both informal care from outside the household and formal care. The remaining variables proxying co-residential informal care (i.e. the marital status or having a co-resident child or caregiver) are not included as they become non-significant once we control by Household Members. Having a daughter, which can be considered as a proxy of informal care supply [12], is significantly positively related to the probability of receiving informal care from outside the household. Age is positively and strongly related with having received both informal and formal care, with the exception of informal care for ADL in Spain (p value of 0.102). In most cases, being a woman cannot be significantly related with care receipt. Finally, individuals speaking Catalan are less likely to receive informal care from a family member living outside the household and more likely to receive formal care than the rest of Spaniards.

We also computed the correlation matrix between independent variables and performed variance inflation factor (VIF) checks on all regressions. No major correlations or high values on these tests were found, indicating the absence of major multicollinearity.

In this subsection, we check if the differences found between Italy and Spain in the effects of LTC financing on informal care receipt by family members from outside the household are statistically significant in addition of having different signs. To do so, we run the regressions of Table 5 where informal care is the dependent variable without splitting the data in two country samples. We include a country dummy for Italy and the interactions LTCI public*Italy and LTCI private*Italy. We remove the dummy variable Catalan as we focus on the international differences. The results of this model are displayed in Table 6.

The effect of public LTC coverage in Spain is still significant at the 1% level when informal care is defined in general (columns 1 and 4) and as help with IADL (column 3). When informal care is defined as help with ADL, the effect of public LTC coverage is reduced substantially and the negative effect in Spain is only significant at the 10% level, which is consistent with the previous section’s findings, i.e. the effect of public LTC coverage is lower when informal care is defined as help with ADL. Concerning private LTC insurance, the results found previously are maintained as the effect of this variable is negative and not significant in general.

Looking at the country dummy, we find, as in the descriptive statistics, that despite the important differences between the Italian and Spanish public LTC systems, there are virtually no differences in the probability of receiving informal care. The country dummy is not significant in any case.

Finally, regarding the other control variables, no substantial changes are observed.

As stressed earlier, important heterogeneity at the regional level is present in the Italian and Spanish public LTC systems. In Italy, the main public LTC benefit is granted at the national level but regions and municipalities also fund additional forms of cash and in kind public support. In Spain, public LTC support is regulated at the national level by the law 39/2006 but Autonomous Communities are left with a great deal of discretion for determining co-payment rates and some characteristics of the benefits granted.

To address such heterogeneity, we again run a selection of the regressions of Table 5 and include a set of binary control variables corresponding to the regions in Italy and the Autonomous Communities in Spain. In Table 7, we report the coefficients corresponding to public and private LTC coverage. We do not estimate the model where wealth is included as a control as this alternative specification does not substantially affect the coefficients corresponding to public and private LTC coverage (see Table 5).

The inclusion of regional fixed effects does not substantially change the results reported in Table 5. The coefficients corresponding to public LTC coverage keep the same sign and increase their significance, despite that the sample is slightly reduced due to missing observations (missing regional information). The coefficients corresponding to the effect of private LTC insurance ownership keep the same sign but are non-significant at the usual confidence levels in all models. Finally, we also find that regional fixed effects are important determinants of informal care supply in both samples as pseudo \(R^{2}\) are larger in all regressions.

As a way to make sure that those who declare in the survey to have LTC coverage, either public or private, receive indeed an indemnity, we decide in the following to only consider those individuals who are strongly dependent and declare to have two or more ADL limitations. By considering the subsamples for Italy and Spain, we focus on those individuals who are most likely to be eligible for LTC benefits since by definition of the eligibility criteria only those with a high degree of dependency are eligible for LTC benefits. Our choice of the criterion having two or more ADL limitations is based on the Spanish public LTC benefits eligibility rule as laid out in Section “LTC financing in Italy and Spain”. This criterion is also in line with the practice of other European countries having a public LTC insurance scheme such as France [42] and Germany [43].

Table 8 shows the results of the different regressions with the selected subsamples. From the model specification, we removed the formal care, hospital, health and demographic control variables as well as wealth following the Bayesian information criterion (BIC)⁴ [41], with the objective of maximizing the degrees of freedom of this second econometric estimation given the reduced number of observations in the subsamples.

The results of this second set of regressions confirm the findings of the previous section. In Italy, we observe a significantly positive effect of public LTC coverage on informal care across all three categories of informal care while in Spain, this relationship is consistently negative. Regarding private insurance, we obtain a non-significant negative effect on the probability of receiving informal care in Spain. Note that in Italy, we could not include this variable in the regression analysis as no individuals declaring to own private LTC insurance remained in the subsample. In the regression with the pooled sample, the effects of public LTCI and the interaction are maintained and the country dummy is negative and highly significant, showing that the probability of receiving care for this subgroup of population is higher in Spain.

In addition, we find that the marginal effects of public LTC coverage on the probability of receiving informal care and their significance levels are larger than in the previous section in both countries despite the sharp reduction in the sample size. Regarding the control variables, with the exception of the number of IADL limitations, coefficients are larger in absolute value but no changes in their sign are observed.

We further control the robustness of our results using the bootstrap method. The bootstrapping technique, pioneered by Efron [44], consists of a Monte-Carlo simulation randomly drawing a large number of samples from the original set of observations, running the regression model and computing the distribution statistics of the obtained regression coefficients. This makes it possible to estimate the empirical distribution of a given estimator (or set of estimators) with the objective of checking the robustness of the analytical approximation of its values and confidence intervals. Since we work with a non-linear model (i.e. a probit model) and relatively small sample sizes, our context is propitious for the use of such resampling techniques [45].

For each country, we start by selecting 5000 random samples from the initial data. The number of observations in each of these samples corresponds to the size of the original dataset, i.e. 2181 observations for Italy and 2344 for Spain. Note that in random sampling, a same individual can be selected twice (i.e. sampling with replacement). Using the generated samples, the different probit models defined in Section “Empirical results” are estimated. As in Section “Controlling for regional fixed effects”, we do not estimate the model that includes wealth. This way, we obtain 5000 sets of the bootstrapped estimates. From these estimates, it is straightforward to extract their expected values and confidence intervals.

In Table 9, we first present the results of the bootstrapping for the parameters defining LTC coverage ownership in Italy. We provide the simulated coefficients’ two-tailed 95% confidence intervals and mean values.

In Italy, we observe that the 95%-confidence intervals of the parameters corresponding to public LTC coverage do not include zero (no sign change) which allows us to conclude on the robustness of the positive effect of public LTC financing on informal care. Moreover, the expected values of public LTC parameters are very close to the parameter estimates presented in Table 5. Therefore, the hypothesis that public support crowds out informal care is still rejected for Italy in the case of public LTC financing.

The bounds of the confidence intervals surrounding the estimates corresponding to private LTC insurance are of opposite signs making it impossible to judge on the trend of the marginal effect. Thus, no valid conclusions can be extracted for this parameter following the bootstrap exercise. Table 10 presents the results of the bootstrapping for the parameters defining LTC insurance ownership in Spain. In the case of the coefficients corresponding to public LTC coverage in Spain, the 95%-confidence intervals are below zero for all types of informal care, underlining the consistently negative sign of these estimates. Additionally, as in the Italian case, no relevant differences between the simulated expected values and the parameters reported in Section “Econometric analysis” are noticed and in any of these three cases. Therefore, the bootstrap results support the hypothesis that public support decreases informal care in the Spanish public LTC system.

The 95% confidence intervals of the coefficients corresponding to private LTC insurance contain both, negative and positive values, and thus no valid conclusions for these marginal effects can be extracted.