1 Introduction

The 2008 world economic crisis has caused severe and long-lasting challenges for the public health sectors of many European countries (McKeeargue 2010; Mladovsky et al. 2012; Thomson et al. 2014). Public health budget cuts coupled with increasingly higher healthcare costs and the parallel weakening of labour markets are likely to have had an impact on health levels of citizens particularly in countries where the crisis hit harder (Baumbach and Gulis 2014; Karanikolos et al. 2013; Stuckler et al. 2009).

Secondary data analyses of national representative data from Spain, Greece and Italy, three countries that experienced economic turbulences more severely than others in the European region since 2008 (Tridico 2013), showed increasing rates of suicides (Branas et al. 2015; De Vogli et al. 2012; Lopez Bernal et al. 2013), mortality (Benmarhnia et al. 2014; Vlachadis et al. 2014) and poor mental health (Álvarez-Gálvez et al. 2018; De Vogli et al. 2013; Drydakis 2015; Moya et al. 2015) following the crisis. In particular, several studies have focused on the negative impact of the crisis on self-rated health (SRH) (Ferrarini et al. 2014; Hessel et al. 2014; Huijts et al. 2015; Zavras et al. 2012)—most commonly measured with a single item (e.g., “How is your health in general?”)—since it is a robust predictor of morbidity (Perruccio et al. 2012; Rosso et al. 2008), healthcare utilization (Dominick et al. 2002; Tamayo-Fonseca et al. 2015), hospitalizations (Kennedy et al. 2001; Nielsen 2016), and mortality (Cesari et al. 2008; DeSalvo et al. 2006; Jylhä 2009; Singh-Manoux et al. 2007).

The mechanisms behind the connection between negative economic conditions and poor SRH have been largely examined in epidemiological (e.g., Marmot 2002) and economical studies (e.g., Rivera 2001; Simou and Koutsogeorgou 2014). Economic crises can negatively impact the health status of men and women due to the occurrence of mental disorders associated with unemployment and reduced income (Aguilar-Palacio et al. 2015; Drydakis 2015), or via the increasing incidence of addiction problems and unhealthy behaviours including the consumption of cheaper and less nutritious food (Brinkman et al. 2009; Pieroni et al. 2013; Bonaccio et al. 2014), smoking (Gallus et al. 2015) and alcohol use (Men et al. 2003). Moreover, during prolonged periods of economic uncertainty, families may reduce their healthcare spending (Terraneo et al. 2014) while overburdened healthcare services may fail to provide adequate assistance to everybody in need (Karanikolos et al. 2013). In addition, socio-economic inequalities exacerbated by the crisis can also negatively affect SRH due to increased stress levels associated with social comparisons (Kawachi and Berkman 2000; Mansyur et al. 2008; Wilkinson 2002). Thus, during an economic backlash all these factors may contribute to the worsening of SRH via biological, socioeconomic, life style and medical resources pathways.

Nevertheless, in a review of empirical studies assessing the impact of the 2008 crisis on health outcomes in Europe, Parmar et al. (2016) underlined how evidences regarding SRH were mixed: some reported an improvement in SRH during the crisis (Aguilar-Palacio et al. 2015; Bartoll et al. 2015; Regidor et al. 2014) while others a decline (Hessel et al. 2014; Reile et al. 2014; Vandoros et al. 2013; Zavras et al. 2012). The main limitations of such previous epidemiological studies are the relatively short time intervals for health data and the lack of adjustment for pre-existing trends (De Belvis et al. 2012; Parmar et al. 2016; Stuckler et al. 2010). Furthermore, the heterogeneity of results that was further observed within countries may reflect differences between age groups. Elderly for example are more likely to be exposed than other age groups to the negative effects of an economic crisis since they are in higher need of healthcare assistance and are more exposed to poverty, social exclusion and poorer health (De Belvis et al. 2012; Feinglass et al. 2007; Hurd and Rohwedder 2012; Piumatti et al. 2018a; Sargent-Cox et al. 2011). Moreover, also young people are considered a category at risk during economic downturns because they are vulnerable to marginalization in the labour market and youth unemployment rates are particularly sensitive to the economic climate of a country (Sarti and Zella 2016; Scarpetta et al. 2010; Verick 2009).

In order to overcome such limitations from previous research, the current study focused on national secondary data from a sample of 97,252 Italian adult residents divided along nine cohorts in an accelerated longitudinal design (Istat 2018a) to examine longitudinal trends in SRH between 2004 and 2015. Latent growth modeling (LGM) (Bollen and Curran 2006; Hancock and Lawrence 2006) within the framework of structural equation modeling (SEM) was adopted to determine initial levels and rates of change in SRH at the individual level in different age groups: 18–30, 31–40, 41–50, 51–60, 61–70 and 71–81. The main advantages of using LGM in SEM to assess longitudinal variation in SRH are to estimate the range of individual differences in change over time and to determine what type of trajectory best fit the data (i.e., linear or nonlinear) (Kline 2015). A further novelty of this study is to have extended the adoption of LGM into an accelerated longitudinal design. This consists of temporally overlapping repeated measurements of independent cohorts forming adjacent segments. Such technique allows to determine whether similar trends are observed for consecutive cohorts and to approximate longer longitudinal trends on the basis of shorter time periods (Duncan and Duncan 2009).

In sum, a clearer understanding of the link between recession and health of the exposed populations may help countries to develop strategies in support of those most affected that may also serve for future scenarios of financial crisis, as pointed out by the World Health Organization (WHO 2009a, b). Accordingly, the aim of this study was to assess whether longitudinal trends in SRH are similar and consistent across different age groups in Italy preceding and following the 2008 world economic crisis.

1.1 The Italian Context

According to a recent report of the central Bank of Italy (Banca d’Italia 2018) the percentage of Italians ‘at-risk of poverty’ increased from 19.6% to 22.9% in the 2006-2016 period and it is now substantially higher than in other European countries with a similar size such as France (13.6%) or Germany (16.5%) (Eurostat 2018a). In the same period, although total national health spending remained stable (Fig. 1; OECD 2018a, b), household expenditure for healthcare decreased immediately following the hit of the 2008 crisis (Istat 2011). Surveys showed that this decrease was indeed due to financial reasons connected to the economic crisis (Cercle Santé Sociale and Europ Assistance 2011; Freni Ricerche Sociali e di Marketing 2011).

Fig. 1
figure 1

Source: OECD (2018a, b)

Annual growth rate in nominal gross domestic product (GDP), unemployment rate, youth unemployment rate and health spending (% of GDP) in Italy for the 2003–2015 period. Notes Nominal gross domestic product (GDP) is GDP given in current prices, without adjustment for inflation. Unemployment rate is the number of unemployed people as a percentage of the labour force, where the latter consists of the unemployed plus those in paid or self-employment. The youth unemployment rate is the number of unemployed 15–24 year-olds expressed as a percentage of the youth labour force. Unemployed people are those who report that they are without work, that they are available for work and that they have taken active steps to find work in the last 4 weeks. Health spending measures the final consumption of health care goods and services (i.e., current health expenditure) including personal health care (curative care, rehabilitative care, long-term care, ancillary services and medical goods) and collective services (prevention and public health services as well as health administration), but excluding spending on investments.

Full size image

Looking at Fig. 1, the anni horribiles for the Italian economy after the crisis could be placed in the years 2008–2009 and 2011–2012 when the annual growth rate in nominal GDP decreased the most and the unemployment rate—especially among youths—started to increase substantially. However, studies that looked at SRH changes in the Italian adult resident population after 2008 focused on different periods of time and reported mixed findings. Using multi-national data from the European Union Statistics on Income and Living Conditions (EU-SILC), Abebe et al. (2016) observed a decline in fair or poor SRH status in Italy in the 2005–2007 period while it remained stable between 2008 and 2011. Minelli et al. (2014) analysed longitudinal data collected by the Bank of Italy in the 2006–2010 period and found how differentials in SRH status spread in times of economic strain for those looking for job opportunities. Similar results were obtained by Sarti and Zella (2016) relying on 2007–2010 Italian data from the EU-SILC project that showed how unemployed and precarious workers had the higher risk of worsening their health status during those years. Other studies focused on the effects of recession on mental health outcomes in Italy and noticed how both suicides and attempted suicides due to economic reasons raised after 2008 (De Vogli et al. 2012) while deaths due to mental and behavioural disorders increased especially among the elderly (De Vogli et al. 2013). A review by Mattei et al. (2014) concluded that there is a reliable link between the economic recession and health and mental health of Italians but that more research is needed to better understand this phenomenon and guide social and political interventions at the national level.

During the 2004–2015 period, total national health spending in Italy followed similar trends than other European countries such as France, Germany or Netherlands, although it remained substantially lower (see Fig. 2; OECD 2018a). The National Health Service (Servizio Sanitario Nazionale, SSN) in Italy is a tax-funded decentralized system (with three levels: national, regional and local) providing universal healthcare coverage to all citizens. Funding to the SSN largely derives from public expenditures for nearly 78%, which is higher than the average 72% across OECD countries (OECD 2015). Out-of-pocket payments, regarding especially pharmaceuticals, outpatient care and dental services, account for much of the remaining financing (i.e., 18%, comparing to the average 20% across OECD countries), while less than 3% of the total healthcare expenditure is covered by private health insurances (OECD 2015). Overall, the Italian decentralized SSN system yields good health indicators with high quality care and significantly lower spending levels than many other European countries (OECD 2015). Italy is at the top of OECD countries with the highest life expectancy (83.3 years) alongside Spain (83.4), Switzerland (83.7) and Japan (84.1) (OECD 2018b). Nevertheless, between 2000 and 2011 the share of out-of-pocket payments decreased on average by 1.2% in the European area and by 5.1% in Italy (OECD 2015). As a result of a consistent reduction in central funding (Mangano 2010), Italian regions with large health care deficits raised local taxes to recover deficits and increased co-payments to reduce pharmaceutical expenditures (Mladovsky et al. 2012). Accordingly, as pointed out above, the number of households postponing or even giving up some forms of medical care for financial reasons has raised (Cercle Santé Sociale and Europ Assistance 2011; Freni Ricerche Sociali e di Marketing 2011). In this current scenario in Italy, it is thus very important to evidence whether periods of economic downturns at the national level correspond to worse SRH trends among the normal adult population.

Fig. 2
figure 2

Source: OECD (2018a, b)

Health spending (% of GDP) in Italy, France, Germany and Netherlands for the 2003–2015 period. Notes Health spending measures the final consumption of health care goods and services (i.e., current health expenditure) including personal health care (curative care, rehabilitative care, long-term care, ancillary services and medical goods) and collective services (prevention and public health services as well as health administration), but excluding spending on investments.

Full size image

1.2 Research Objectives and Hypotheses

The current study aimed to assess whether longitudinal trends in SRH vary before, during and after the 2008 world economic crisis in Italy. In particular, longitudinal changes in SRH at the individual level were examined according to different age groups to evidence consistencies or differences in SRH trajectories across the crisis and the life-span. Although the nature of the analyses carried out here was mainly explorative, the following hypotheses were formulated: (1) youngest (i.e., 18-30) and oldest (70+) cohorts are the ones where the worst SRH can be observed in correspondence with the years of greater economic uncertainty; and (2) worst trends in SRH can be observed immediately after the crisis (i.e., 2008–2009) and following the most severe periods of economic downturns (i.e., 2011–2012).

2 Method

2.1 Sample and Measures

The current study used secondary data available from the Italian National Institute of Statistics (Istat). Since 2004, each year Istat collected information from national representative samples of Italian adult residents using questionnaires from the EU-SILC project with the aim of obtaining comparable cross-sectional and longitudinal data on income, poverty, social exclusion and living conditions across European countries (Eurostat 2018b). In each baseline assessment year, a stratified multistage sampling design was adopted (Official Journal of the European Union 2003), that is a probability method based on dividing the target population into strata and then using a hierarchical structure of units within each stratum (Jain and Hausman 2006; Lohr 2008). More specifically, stratifying a population means dividing it into non-overlapping subpopulations, called strata. Independent samples are then selected within each stratum. Longitudinal EU-SILC data are collected over four-year periods. The first baseline data collection took place in 2004. Accordingly, Italian data from 9 overlapping longitudinal cohorts were obtained (N = 116,137), the first one comprehending the 2004–2007 period and the last one comprehending the 2012–2015 period. Figure 6 in the Appendix illustrates the adopted rotational design pattern.

SRH was measured with the following questions: “How is your health in general?”. Response options were coded 1 = very good, 2 = good, 3 = fair, 4 = bad and 5 = very bad (values were reverse coded for the current analyses to ease interpretation). SRH encompasses different dimensions of mental, physical and subjective health status (Altman et al. 2016; Singh-Manoux et al. 2006). The operationalization of this question as it was analysed here is aligned with previous research (Fayer and Sprangers 2002; DeSalvo et al. 2006; Jylhä 2009). Since LGM requires to have the same continuous dependent variable (in this case SRH) measured at least on three different occasions for each individual at the same intervals, in every cohort only participants reporting SRH at least two times out of four assessments were retained for the analyses while the remaining missing values were treated using the full information maximum likelihood estimation method (FIML) which adopts the expectation–maximization (EM) (see below the Analyses section for a full explanation of this approach). This decision was made considering also that it is preferred to have a sizeable portion of cases with at least three assessments of observed information so to over-identify a linear trajectory (Curran et al. 2010). Excluding participants that reported missing values on more than two observed measures of SRH across four assessments ensured this goal. This left a total of 97,252 individuals, 84% of the total original sample. Given this selection procedure, at baseline assessments the missing rates ranged between 0 and 1.7% across all cohorts. Moreover, the percentage of participants with at least three longitudinal observed measures of SRH ranged between 81 and 90%.

Table 4 in the Appendix reports descriptive differences between selected and non-selected participants at baseline. The latter were in general older, with lower educational level, more likely to not being married, and more likely to suffer from a chronic disease or to report limitations due to health problems. Baseline SRH levels were also significantly lower among non-selected participants. Table 1 reports cohorts’ socio-demographic information (i.e., gender, age, marital status) along with the incidence of chronic diseases and limitations due to health problems at baseline. Overall, the percentage of participants declaring to suffer from a chronic disease or to experience limitations in their daily activities due to health problems substantially increased from the first to the last cohort. Moreover, the percentage of individuals with a primary or lower educational level diminished in favour of the percentage of individuals with at least upper secondary education. These variations can be linked to demographic changes that took place within the Italian society between 2004 and 2015 (Istat 2018b), including the declining rates of married couples (Istat 2018c).

Table 1 Baseline sample’s socio-demographic characteristics by cohorts. Values are percentages
Full size table

2.2 Analyses

Within LGM two latent factors are estimated: the intercept, that is a constant for any individual across assessments representing the point where the individual trajectory of the observed measure intercepts the vertical axis; and the slope of an individual trajectory. The intercept and the slope are random estimates in LGM capturing how every individual growth curve differs from the overall trajectory (Curran 2003; McNeish and Matta 2018). The specific advantage of modeling individual longitudinal development using LGM within the SEM framework is a greater flexibility to estimate different shapes of growth by freely estimating specific slope factors so that change may better corresponds to the unique characteristics of the population under study (Curran et al. 2010). This straightforward flexibility of LGM, coupled with the possibility of adopting FIML and EM to deal with missing values, make it particularly suitable for the specific aim of the current study—namely to assess longitudinal trends in SRH between 2004 and 2015 across different age groups and consecutive cohorts—in comparison with other mathematically equivalent models such as the mixed effect approach (Curran 2003; McNeish and Matta 2018). In fact, LGM have been adopted in similarly designed studies (Duncan and Duncan 2004; Duncan et al. 2006).

Figure 3 depicts the LGM structural path diagram applied to estimate the intercept (i.e., average initial level) and slope (i.e., growth) factors of SRH for different age groups (i.e., 18–30, 31–40, 41–50, 51–60, 61–70 and 71–81) within each cohort. The current cut-offs to define age bands were chosen to provide a wide range of age categories, also considering the distribution of age and the large available sample size in each cohort. Three alternative model’s solutions were tested in each age group (Phan, 2011):

Fig. 3
figure 3

Latent growth curve model for self-rated health. Notes Observed variables are depicted as squares: t1SRH, t2SRH, t3SRH and t4SRH = self-rated health at assessment years 1, 2, 3 and 4. Latent variables are depicted as circles: Intercept (I; average initial level) and Slope (S; growth)

Full size image
  1. 1.

    A no-growth model where no slope component was assumed. The equation for this model can be written as

    $$y_{it} = \alpha_{i} + \varepsilon_{it}$$

    where yit is the repeated measure under analysis (in our case the SRH score) for individual i (i = 1, 2, …, N) at Time t (t = 1, 2, 3 4,) and αi denotes the intercept for individual i, namely an individual’s SRH level at the first time of measurement.

  2. 2.

    A linear growth model assuming a linear pattern of change across assessments by fixing slope parameters to 0 at Time 1, to 1 at Time 2, to 2 at Time 3 and to 3 at Time 4 (i.e., corresponding to the years between assessments). The equation for this model can be written as

    $$y_{it} = \alpha_{i} + \lambda_{t} \beta_{i} + \varepsilon_{it}$$

    where λt represents slope parameters at Time t (t = 1, 2, 3 4,) and it is defined as λt = t – 1 and βi denotes the slope of the latent trajectory for individual i (i = 1, 2, …, N).

  3. 3.

    A nonlinear growth model where the form of change across assessments was not specified a priori and slope parameters were fixed to 0 at Time 1 (i.e. λ1= 0, reading from the equation reported above), freely estimated at Time 2 and Time 3 and fixed to 3 at Time 4 (i.e. λ4= 3) to allow a separation of the intercept and slope components at baseline assessment and provide a scale of measurement for the slope.

Individual intercepts (αi) and slopes (βi) are assumed to follow multivariate normal distributions with means (μα, μβ), variances (σ2α, σ2β) and covariance (σα,β) (Grimm et al. 2011; McArdle and Nesselroade 2003). Comparisons between each solution were conducted by the means of Chi square (χ2) difference tests. Overall model fit was evaluated using the following fit indexes: The Bentler comparative fit index (CFI) (Bentler 1990), the Tucker Lewis index (TLI) (Bentler and Bonett 1980) and the Steiger–Lind root mean square error of approximation (RMSEA) (Steiger 1990). Models with CFI and TLI values over or equal to 0.90 and RMSEA values below or equal to 0.08 are considered acceptable in terms of fit (Curran et al. 2010; Preacher et al. 2008; Wu et al. 2009). Since the RMSEA may falsely indicated a poor fitting model when having small degree of freedoms, using different indexes at the same time provided a more reliable assessment of model fit (Kenny et al. 2015). Error variances for SRH at each assessment were freely estimated. Missing values were estimated using FIML, this procedure adjusts the likelihood function so that each case contributes information on the variables that are observed. FIML relays on the missing at random assumption and it assumes also the multivariate Gaussian distribution for the underlying variables (Dempster et al. 1977; Kline 2015). In addition to LGM analyses, average SRH was calculated for each participant across assessment years so to test for mean differences between cohorts based on univariate analyses of variance (ANOVAs) with Bonferroni post hoc adjustments. All analyses were carried out in Stata 15 (StataCorp. 2015. Stata Statistical Software: Release 15. College Station, TX: StataCorp LP).

3 Results

Absolute values of skewness and kurtosis for SRH across cohorts were respectively below 1 and 4 suggesting that this variable was reasonably normally distributed in every assessment year (Kline 2015). Little’s test for data missing completely at random (MCAR) applied to each set of four longitudinal SRH assessments per cohort was always significant indicating data were not MCAR. This is not surprising if we read the results of Table 4 in the Appendix: higher age, lower education, not being married or suffering from chronic diseases and limitations due to health problems are all factors that can explain why participants did not take part to every assessment. Nevertheless, as described above, the ML procedure in Stata to deal with missing values in LGM produces less biased estimates than other methods when data are not missing at random (Little and Rubin 1989; Schafer and Olsen 1998). This estimation procedure was further supported in the context of the current analyses considering the large sample size available, the portion of participants with at least three observations out of four in each cohort (i.e., between 81 and 90%) and the fact the normality assumption was met.

Table 5 in the Appendix reports the full set of comparisons between fitted latent growth models for SRH by cohorts and age groups while Table 2 only reports results for the best fitting models along with standardized estimates for the intercept and slope factors. In addition, Fig. 4 depicts the standardized estimated individual growth rates (i.e., slopes) in SRH by cohorts and age groups based on the results from the best fitting latent growth models.

Table 2 Results of latent growth models for self-rated health by cohorts and age groups. Standardized results are shown
Full size table
Fig. 4
figure 4

Estimated individual growth rates (i.e., slopes) in self-rated health by cohorts and age groups based on results from latent growth models. Y-axis reference lines indicate zero. Standardized results are shown

Full size image

Pre-crisis cohorts, namely the 2004–2007 and 2005–2008 ones, were the only ones where the two youngest age groups (i.e., aged 18 to 30 and 31 to 40) exhibited descending trajectories in SRH. Nevertheless, in the two subsequent ones collected around the beginning of the crisis, the 2006–2009 and the 2007–2010, only the oldest groups (i.e., aged 71 to 81) significantly declined in SRH across time. Indeed, with the exception of the 2008–2011 period, participants from the oldest age group (i.e., 71–81) always exhibited a significant longitudinal decrease in SRH across all cohorts. On the other hand, SRH trajectories were consistently stable among participants aged 41 to 50 across all cohorts except for the 2010–2013 period where SRH significantly declined.

The first post-crisis cohort, 2008–2011, appeared to be the ‘healthiest’ since it is the only one where no significant decline in SRH was observed at the individual-level for any age group, followed by the 2006–2009 and the 2007–2010. The rest of the post-crisis cohorts reported fluctuating trends in SRH. In the 2012–2015 cohort the 18–30 age group was the only one among all groups and cohorts to show a positive increase in SRH across time despite the fact that in this same cohort SRH was significantly declining from age 51 onward. Finally, the 2010–2013 cohort was the one with the worst SRH trends, with all age groups except the one aged 31 to 40 exhibiting a significant decline in SRH across 4 years.

Table 3 reports the results of ANOVAs with Bonferroni post hoc adjustments testing for mean differences in SRH (calculated as average SRH across assessment years) between cohorts. To further ease the interpretation of these last analyses, Fig. 5 depicts longitudinal observed mean scores in self-rated health by cohorts. Overall, it appeared to be a positive trend in SRH across cohorts, especially when looking at the significant differences between pre- and post-crisis periods. On average, every cohort after the 2005–2008 period reported significant higher levels of SRH. This tendency seemed to be interrupted by an abrupt decline in SRH taking place in 2011. Nevertheless, average outcomes in SRH did not return to pre-crisis levels. On average, SRH significantly decreased only once between the 2011–2014 and 2012–2015 cohorts.

Table 3 Comparisons of average longitudinal self-rated health between cohorts based on univariate analysis of variance with Bonferroni post hoc adjustments
Full size table
Fig. 5
figure 5

Longitudinal observed mean scores in self-rated health by cohorts (N = 97,252)

Full size image

4 Discussion

This study examined longitudinal trends in SRH between 2004 and 2015 across different age groups in Italy. Data were obtained from a national representative sample of 97,252 Italian adult residents from nine cohorts in an accelerated longitudinal design. The specific aim of the analyses presented here was to assess whether longitudinal trends in SRH were similar and consistent across different age groups in Italy preceding and following the 2008 world economic crisis. Two hypotheses were formulated: (1) youngest (i.e., 18–30) and oldest (70+) cohorts are the ones where the worst SRH can be observed in correspondence with the years of greater economic uncertainty; and (2) worst trends in SRH can be observed immediately after the crisis (i.e., 2008–2009) and following the most severe periods of economic downturns (i.e., 2011–2012). Results partially confirmed these hypotheses. First, worst longitudinal health trends were observed only few years after the strike of the crisis, namely between 2010 and 2013, while the years immediately after 2008 registered better health trends than precedent ones. Second, health levels among the oldest cohorts (70+) appeared indeed to be more prone to the negative influence of national economic downturns. On the other hand, youngest cohorts (i.e., 18–30) exhibited more fluctuating trends across years: they were more likely to show declining SRH in the years preceding rather than following the crisis, and during the worst four-year period for health trends they were not affected as much as participants from older cohorts. Moreover, youngest participants were the only ones across all cohorts and age groups to report a significant positive trend in SRH, namely in the 2012–2015 period.

In contrast with what reported by Abebe et al. (2016), SRH declined in Italy in the years preceding the 2008 crisis. This tendency was especially true among participants aged 40 and below or 71 and above and took place during a rather favourable period of economic stability characterized by a slow but steady decrease of the unemployment rate, also among youth (see Fig. 1). Subsequently, until 2012 SRH appeared very stable across age groups. It is noteworthy that between 2008 and 2011 SRH did not decline among any age group, and in general between 2006 and 2012 the only decline was observed among the elderly. SRH declining across the 4-year assessments among the oldest portions of each cohort (i.e., aged 71 to 81) and remaining more stable for participants aged 30 to 50 were indeed the most consistent results. The finding about a higher longitudinal stability of SRH from early adulthood throughout middle-life is aligned with previous research findings (Andersen et al. 2007; McCullough and Laurenceau 2004). Moreover, the fact that SRH tends to decline in older age is not only aligned with previous research findings (Cullati et al. 2014; Dening et al. 1998; Orfila et al. 2000; Sargent-Cox et al. 2010), but it specifically echoes the results of past research that adopted LGM to explore this issue (Cullati 2015; Piumatti 2017; Rohlfsen and Jacobs Kronenfeld 2014; Sacker et al. 2011). Nevertheless, we know that several factors can still impact SRH trajectories, yelding different outcomes across the life-span (Cullati et al. 2014; Pinquart 2001). Accordingly, heterogeneity in health changes in later life during times of economic crisis in contexts such as Italy should be further explored to evidence whether for specific portions of the elderly population positive SRH may be preserved across time if not even improved as some pointed out (Ferraro 1980; Idler 1993).

The clearer correspondence between economic downturns at the national level and declining trajectories of SRH is evident after 2011. In that year political and economic turbulences in Italy reached their peak with the government forced to resign after failing to gain a full majority in the Chamber of Deputies during a budget vote. These events were following a period during which financial markets and international institutions such as the International Monetary Fund and the European Central Bank were warning Italy about its public debt. The next government introduced a consistent package of austerity measures and spending cuts including raising taxes and retirement age. National plans were made to cut funding for investments in healthcare infrastructures from over 1 billion to 236 million euros, for research funding in the public health care sector from 91.9 to 18.4 million, and for disease prevention and health promotion from 29.6 to 5.9 million (Mladovsky et al. 2012). Even resources for semi-automatic defibrillators in public places were expected to be reduced from 4 to 2 million euros (Mladovsky et al. 2012). Italian regions had to introduce co-payments for visits to public and private accredited specialists and hospital emergency departments deemed inappropriate so to compensate reductions in central funding: 10+ euros for visits to doctors and analysis and 25+ euros for interventions in emergency wards not justified by urgent situations (De Belvis et al. 2012; Mladovsky et al. 2012).

A 2011 survey clearly summarises how Italians perceived the situation at that time (Cercle Santé Sociale and Europ Assistance 2011): 12% compared to 57% in 2009 were willing to pay more taxes to receive better health service. Indeed, in 2011 Italians diminished their household expenditure for healthcare (Istat 2011). Moreover, 2011 data show negative trends regarding health-related behaviours in Italy, including a reduced consumption of fruits and vegetables and a lower frequency of physical activity (Costa et al. 2012). Concurrently, in certain areas of the country such as the North, rates of first admission for hearth attack increased substantially in 2011 as a possible consequence of the stress related to precarious economic conditions (Costa et al. 2012). Results from the current study are thus consistent with those from such previous surveys and evidence once more a link between austerity measures of a country and the health of its citizens (Brand et al. 2013; Habibov and Afandi 2015; Karanikolos et al. 2013; Skroumpelos et al. 2014).

Despite the abrupt decline in SRH observed in 2011, there seems to be a slight overall positive trend in Italians’ SRH in the longer period. In fact, among participants from the cohort exhibiting worst SRH trends (i.e., 2010–2013), the estimated average initial level of SRH was the highest compared to the other cohorts, nevertheless they also declined on average more than any other cohort. This is due to the fact that post-crisis cohorts analysed here exhibited on average better health outcomes than pre-crisis ones. Reading from results of previous research (Burgard et al. 2013), we know that when examining the health consequences of recession at the aggregate level we are more likely to observe positive or stable health longitudinal trends. Evidences from previous economic crises have indeed pointed out that the number of individuals whose health is negatively affected by recessions is often exceeded by the portion of the population who reap benefits (Granados 2005; Riva et al. 2011). By contrast, this is happening when at the individual level we observe a slight decline in SRH. The explanation for this may be the higher mortality and non-responses rates among the participants with the poorest self-rated health (Andersen et al. 2007) and the compensatory mechanisms within individuals’ social capital networks to face adverse events such as unemployment (Piumatti 2016; Piumatti et al. 2018b; Saltkjel et al. 2017). Previous studies have in fact reported that while at the overall country level we may observe stable or positive trends in SRH during time of crisis even in the most affected regions, contrasted trends in health outcomes may still take place across specific portions of the population (Clause-Verdreau et al. 2018; Lersch et al. 2018; Saltkjel et al. 2017).

Individual social capital nets may have been especially determinant for youngest participants to compensate or alleviate the negative effects of the crisis in Italy. Indeed, in Mediterranean countries such as Italy young people are more likely to reserve a strong interdependence with their families but not with the society at large (Tsekeris et al. 2015). National surveys show that the percentage of Italian young adults (aged 18 to 34) living with their parents has increased from 78.5% in 2008 to 84.1% in 2017 and it is currently the fourth highest in the European area (where the mean percentage is 66.7%) after Slovakia (84.2%), Macedonia (84.7%) and Croatia (87.8%) (Eurostat 2019). In the absence of social safety valves, Italian young adults continue to live with their parents until they reach a stable enough position in the job market and a certain degree of economic independence (Alfieri et al. 2015). The support Italian young adults receive within their own families plays thus an essential role for their well-being in conditions of economic instability (Paleari et al. 2002; Piumatti et al. 2016). On the other hand, the situation concerning elderly people in Italy is radically different, with a share of people aged 65 or older living alone close to 31% (Eurostat 2015). In a recent survey conducted by Istat in Italy, 25.9% of interviewed elderly people declared to not have a strong net of social support, while 18% perceived their social support nets as weak, and about 50% have an intermediate situation (Istat 2017). Despite the fact that elderly people living alone are more likely to receive weak social support and are often in fragile health conditions, they are nevertheless often involved in providing informal care or assistance to relatives and non-relatives, in Italy as in the rest of Europe (Istat 2017). Concurrently, due to the low level of public service provision for home-based elder care in Italy, many families have turned to migrant care workers to provide care to their frail older members (Di Rosa et al. 2012). In Italy, as in other European countries, this widespread phenomenon is posing additional financial challenges especially for households with elderly members suffering from multiple ambulatory care-sensitive chronic conditions (Williams 2012). Therefore, findings from the current study confirm once more the impellent need to develop adequate economic and social policies in response of economic turbulences in Italy so to support elderly people in the current and future scenarios of crisis.

4.1 Limitations and Implications for Future Research

This study was not without limitations. First, selection biases cannot be excluded since for the type of longitudinal analyses conducted here participants which were absent more than twice across each four-year assessment were excluded. Looking at the differences between selected and non-selected participants (see Table 4 in the Appendix), it is likely that these latter could have contributed to observe worse longitudinal trends in SRH. Accordingly, the decline in SRH especially in the worst performing age groups (i.e., aged 61 and above) could have been underestimated. Second, several other confounding factors other than age that are known to be related to health outcomes in times of crisis have not been treated here. This study was mainly focused on examining long-term SRH trends to assess how these have changed before and after the 2008 crisis across different age groups. However, future studies should look at other specific determinants of SRH to study how the economic crisis in Italy has exacerbated social disparities in health between advantaged and disadvantaged groups. At this regard, the same longitudinal model applied here may be extended with a focus on the most critical period for SRH observed here (i.e., 2010–2013) by further adding time fixed and time-varying covariates and evidence whether different trajectories may be observed for different groups according to specific socio-demographic factors: not only age but also gender, socio-economic and working status. On a related note, other more recent methodological approaches to deal with ordinal responses could also be adopted, such as mixture latent auto-regressive models (Bartolucci et al. 2014) or latent Markov and growth mixture models for ordinal individual responses (Pennoni and Romeo 2017), to replicate and corroborate the findings of the current study. Finally, although SRH is a robust predictor of a wide range of health-related outcomes (Nielsen 2016; Perruccio et al. 2012; Singh-Manoux et al. 2007; Tamayo-Fonseca et al. 2015), future studies may look at the extent to which changes in SRH reported during the most uncertain economic periods in Italy correspond to changes in true health (Lindeboom and van Doorslaer 2004; Schneider et al. 2012).

Table 4 Comparisons between selected and non-selected participants across demographics and baseline self-rated health. Values are percentages unless stated otherwise
Full size table

5 Conclusions

To the best of the author’s knowledge, this is the first study to use data from an accelerated longitudinal design covering 11 years of economic instability in Italy to examine SRH trends in the normal adult population. The results discussed here contribute to understand how citizens’ health levels may be affected by prolonged periods of economic uncertainty and austerity. In particular, this study underlined the importance of studying this phenomenon at the individual level since longitudinal stability at the aggregate level may mask consistent within-country differences across time such as different longitudinal trends in health outcomes between different age groups.