Policy context: social pension programme in Korea

By 2050, 38.2 per cent of Korea’s population will be aged sixty-five or older (Ministry of Health and Welfare, 2012). At 49.6 per cent, Korea’s later-year poverty rate is the highest of any OECD country, and four times higher than the OECD country average (OECD, 2015). The yet-to-mature public pension system has exacerbated this spike in Koreans’ later-year poverty. Its twenty-year-old contributory public pension scheme, the National Pension Scheme (NPS), covers less than 50 per cent of its working-age population (Jones and Urasawa, Reference Jones and Urasawa2014) and the income-replacement level remains only about 20 per cent (National Pension Research Institute, 2016). In 2008, the basic old-age pension (BOAP) was introduced to complement the NPS to address the problem of poverty among older people. The BOAP is a means-tested, non-contributory pension for older people whose income is below a specified threshold.

However, BOAP, a non-contributory pension called a social pension, was not successful in addressing poverty among older people. The electoral pledge by President Park also fuelled calls for an increase in benefit amount. In July 2014, the government replaced the social pension programme with the Basic Pension Scheme (BPS), which doubled the monthly benefit levels while keeping the coverage of 70 per cent of the older population intact. It provided a maximum monthly benefit of 200,000 KRW (about USD 166.18 as of 2020) to single persons and 320,000 KRW (about USD 265.89) to couples in 2014Footnote ¹ . This expansion, known as one of the largest social welfare reforms in Korean history, provides an interesting opportunity to examine how the increase of benefit amounts affected elderly’s material hardships.

Social pension and material hardships

The outcomes of welfare policy programmes can be evaluated with indirect indicators, such as income or resources, or direct measures, such as expenditure or quality-of-life indicators (Ringen, Reference Ringen1988). As a non-monetary, direct measure of material well-being, material hardship measures a household’s actual living standard by focusing on the affordability of necessities (Guio et al., Reference Guio, Gordon and Marlier2012) such as food, housing, utilities, medical care, clothing, and consumer durables (Beverly, Reference Beverly2001). Cumulative research has shown income, poverty, and material hardship partially overlap; in many countries, those are completely distinct between the groups of people who are considered poor based on income and those deemed poor based on hardship (cf. Weon and Rothwell, Reference Weon and Rothwell2020). The discrepancy is not surprising given that a material hardship can occur when a household is above the national poverty line, just as households below the national poverty line are not necessarily deprived (Whelan et al., Reference Whelan, Layte and Maitre2004). Material hardships are not always of monetary nature (Israel, Reference Israel2016), but pertain to a household’ needs or resources. A person is considered deprived when she cannot achieve a particular doing or being because she does not have the financial resources (Saunders and Wong, Reference Saunders and Wong2011; Guio et al., Reference Guio, Gordon and Marlier2012).

Social pension can increase someone’s resources in old age by supplementing the market income, thereby reducing material deprivation for people who would otherwise not have enough financial resources to finance their lives. Given that older adult households are more likely than the working age population to have lower fixed (regular based) incomes, the BPS benefit, paid on a regular basis (monthly), can redress the structural constraints in consumption for purchasing the basic needs for older adults and reduce material deprivation.

A growing body of research supports the role of public transfers on material hardship among older adults in international contexts. Public transfers have improved food insecurity (Case and Menendez, Reference Case and Menendez2007), poverty, or consumption in India (Unnikrishnan and Imai, Reference Unnikrishnan and Imai2020), in Mexico (Galiani et al., Reference Galiani, Gertler and Bando2016), in China (Wu and Ramesh, Reference Wu and Ramesh2014; Zheng and Zhong, Reference Zheng and Zhong2016), and in the USA (Lu et al., Reference Lu, Shelley and Liu2021). Notten and Guio (Reference Notten and Guio2020) showed that the universal social transfer could substantially reduce material hardship among older households in EU countries. The effects of social transfer are found to be greater among low-income countries or where people reported severe material hardship.

In the Korean context, many up-to-date studies have focused on the effect of BOAP, the previous form of BPS, and found that it reduced difficulties in paying tax and for adequate heating (Shin and Do, Reference Shin and Do2015), increased consumption (Shin and Do, Reference Shin and Do2015; Jung et al., Reference Jung, Pirog and Lee2016; Koh and Yang, Reference Koh and Yang2021), economic satisfaction (Kang and Moon, Reference Kang and Moon2013), or non-financial well-being (Pak, Reference Pak2020). A few studies have examined the effect of BPS and found that it brought a 9.3 per cent decrease in the poverty rate and a 47,000 KRW decrease in the poverty gap (Lee et al., Reference Lee, Ku and Shon2019), as well as significant increases in total expenditure of about 144,300 KRW (116 US dollars) (Kang et al., Reference Kang, Park and Ahn2022)

Nonetheless, the impact of the social pension on material hardship may not be substantial if it causes behavioural changes, such as a decline in private transfers from children or a decrease in labour supply among older adults. These two potential factors are relevant to the Korean context. Korea’s older population tends to have high reliance on private transfers relative to those in counterpart countries (Whang et al., Reference Whang, Lee, Jin, Lee and Ahn2021) and the immature national pension system and confucianism culture highlights the responsibility of children to take care of their parents in old age. On average, private transfers comprise approximately 27 per cent of household incomes for elderly individuals (Koh and Yang, Reference Koh and Yang2021). While a line of research found no significant crowding effect from public transfer (Lee et al., Reference Lee, Ku and Shon2019; Pak, Reference Pak2020), others accumulated evidence of the crowding out of private transfers in Korea, thereby making insignificant changes to disposable income (Jung et al., Reference Jung, Pirog and Lee2016) or making the overall impact of BOAP on economic well-being less significant (Koh and Yang, Reference Koh and Yang2021). If the crowd-out effect occurs, the public transfers may not effectively improve material hardship. Furthermore, a social pension could affect the labour supply among the older population. Employment rate among the older population in Korea, 34.1 per cent of Koreans aged sixty-five and above, is the highest among OECD countries in 2020 (OECD, 2022b). Koh and Yang (Reference Koh and Yang2021) found that the effects of BOAP on the probability of being employed and on working hours among the elderly were low and statistically insignificant. However, another study indicated that a social pension increased the likelihood of retirement and reduced working hours (Unnikrishnan and Imai, Reference Unnikrishnan and Imai2020).

Present study

Even though many studies found a positive effect from the social pension in poverty reduction, little is known about the effect of BPS, which raises an interesting question about setting up empirical evidence of newly implemented policy in Korea. We first asked to what extent the BPS, affects the material hardship and life satisfaction among older adults as a growing research emphasises the importance of the simultaneous examination of material well-being and subjective well-being (Lloyd-Sherlock et al., Reference Lloyd-Sherlock, Saboia and Ramírez-Rodríguez2012). Next, we examined if and to what extent the effects of BPS on the material and non-material well-being varies by income level. This study contributes to the literature that investigates whether social pensions in Korean and international settings have focused predominantly on income increases and poverty reduction (Jensen, Reference Jensen2004; Kakwani and Subbarao, Reference Kakwani and Subbarao2007; Gasparini et al., Reference Gasparini, Cruces, Bergolo and Ham2010; Amuedo-Dorantes and Juarez, Reference Amuedo-Dorantes and Juarez2015; Lee et al., Reference Lee, Ku and Shon2019) by examining the effect of the BPS on material hardship among older adults. Furthermore, a detailed examination of the social pension on material hardship and well-being by income level can provide a better picture of whether and to what extent the cash income may help people at the lower end of the income distribution range (Israel, Reference Israel2016).

Parametric approach

To estimate the BPS impacts, we use the following set of regression models:

$${Y_i} = {\beta _0} + \rho {D_i} + \gamma {X_i} + {\varepsilon _i}$$

where Y _i is an outcome measure for observation i, X _i is a set of covariates, and ϵ _i is a random error term for observation i. Most importantly, D _i is a dichotomous eligibility indicator; i.e. D _i = 1 for eligible population, those whose age is above the cutoff (i.e. sixty-six); otherwise, D _i = 0. Then ρ estimates the treatment effect of the BPS near the cut-off point (x _c = 66). Depending upon the types of dependent variables, we employ Ordinary Least Square regression (material hardship indicator and life satisfaction measure) and Logistics regression approaches (four domain-specific hardships measures). For an intuitive interpretation of the logistic regression results, we report marginal effects of each independent variable, which estimate percentage points change in the risk of a dependent variable in accordance with a unit increase in the independent variable.

Regression discontinuity design

One of the drawbacks of conventional parametric regression approaches is internal validity. That is, due to the potential existence of unobservables as well as endogeneity, results from such parametric models are easily biased (Manski, Reference Manski1993; Keele and Minozzi, Reference Keele and Minozzi2013). The main advantage of RDD over other competing approaches is that it is closer to a natural experimental design as individuals close to a given cut-off are likely to be very similar, and the only thing that makes them different is the treatment assignment below and above the cut-off (Imbens and Lemieux, Reference Imbens and Lemieux2008; van der Klaauw, Reference van der Klaauw2008; Lee and Lemieux, Reference Lee and Lemieux2010). RDD enables us to take into account both observed and unobserved heterogeneity in the estimation of the treatment effect (the impact of the programme) because one below the cut-off could be a counterfactual for those who are above the cut-off, and vice versa, if they are close enough to the cut-off. In this study, we adopt the fact that observations just below and above the cut-off (age sixty-five) are likely to be very similar to each other concerning observed and unobserved characteristics; hence, the mean difference in the values of the outcomes identifies a Local Average Treatment Effect (LATE) of the BPS for the subpopulation of people at or near age sixty-five.

Because the RDD estimates LATE of the treatment of interest, we limit the sample to those who are close enough to the cut-off. Then an important issue is how to define the term ‘close enough’ through a bandwidth selection process. In general, the selection of bandwidth in nonparametric estimations aims to balance precision and bias. Using a broader bandwidth is likely to yield more precise effects because of using more data for the regression, but this also leads to bias and vice versa. Although it is advisable to check for the appropriate functional form, in practice, it may not be as critical when one is modelling using data very close to the cut-point – especially if there are very large samples in this interval. This is because the smaller the analytic bandwidth, the more likely it is that the slope of the regression line is approximately linear in this smaller interval (McCall and Bielby, Reference McCall, Bielby, Smart and Paulsen2012).

Hence local regression models focus on their subsample to estimate the LATE. A distinct difference between the nonparametric RDD approach and other conventional local regression models is that the former determines its bandwidth for the local estimation based upon its given data. Of many bandwidth selection procedures, we adopt mean squared error (MSE) optimal bandwidth selector, which is the default bandwidth selector for the RD treatment effect estimator. We also note that the results with the MSE are robust and similar to other bandwidth selection procedures, such as coverage error-rate (CER)-optimal bandwidth selector.

We utilise the ‘sharp’ RDDFootnote ² to estimate discontinuity in outcomes at the age cut-off. Our main estimating equation is given by

$${Y_i} = f\left( {{x_i}} \right) + \rho {D_i} + {\varepsilon _i}$$

$${D_i} = \left\{ {\matrix{ {1{\rm{\;}}\left( {if{\rm{\;}}{x_i} \ge 66} \right)} \cr {{\rm{\;}}0{\rm{\;}}(if{\rm{\;}}{x_i} \lt 66)}} } \right.$$

where Y _i is an outcome measure for observation i, f(x _i ) is an unknown smooth function of the running variable, i.e. age, and ϵ _i is a random error term for observation i. Most importantly, D _i is a dichotomous eligibility indicator; i.e. D _i = 1 for eligible population, those whose age is above the cutoff (i.e. sixty-six); otherwise, D _i = 0. Then ρ estimates the treatment effect of the BPS near the cutoff point (x _c = 66).

The analyses were conducted in the following order: to check design validity for RDD modelling, we first graphically check the distribution pattern among key variables. Second, parametric and RDD modellings are conducted with the entire sample, then followed up by the same set of analyses with three split samples by income level equalised by household members.

Sample characteristics

Table 1 presents summary statistics for the sample used for the parametric model. The largest proportion in the sample had food insecurity (7.9 per cent), followed by utility (2.7 per cent) and housing (2.3 per cent). A clear, descriptive pattern indicates that individuals in the lower-income group tend to have more food insecurity (18.2 per cent), utility insecurity (3.8 per cent), and housing insecurity (5.4 per cent). Members of this group tend to be older, female, with limited education, ad single. They also had a higher incidence of disability (22.0 per cent) and chronic disease (85.8 per cent).

Table 1 Descriptive characteristics of the variables in use (45 or older)

Note. Mean or Percentage reported

Standard deviations in parentheses

Analytic results

Parametric approach

Table 2 indicates that recipients of the BPS were likely to have less sense of material hardship in aggregate form (−0.069, p< 0.05), and less hardship in housing (−0.021, p<0.05) and food insecurity (−0.055, p< 0.01). The recipients also tended to be satisfied with their lives (0.13, p< 0.01). In terms of income (Table 3), we did not find any discernible influence of the benefit on material hardship in aggregate form, except in the lower income group at a marginal degree (−0.150, p<0.1). The positive effect of the benefit persists in housing and food insecurity for the lower-income group only (b = −0.060, p<0.05 and b = −0.133, p<0.01 respectively). The benefit was associated with more life satisfaction (b = 0.198, p<0.01). Private transfer is associated with lower level of material hardships (b = −.011, p<0.01), housing (b = −.002, p<0.01), food insecurity (b = −.006, p<0.001) and more life satisfaction (b = .02, p<0.01). Working status is not statistically significant.

Table 2 Empirical model estimates, parametric approach (OLS and Logit)

Note. ^†p<0.10 * p<0.05 ** p<0.01 *** p<0.001

Standard errors in parentheses

^M Marginal effect estimated for Logistic models

Table 3 Heterogeneous treatment effect estimates by income, parametric approach (OLS and Logit)

Note. ^†p<0.10 * p<0.05 ** p<0.01 *** p<0.001

Standard errors in parentheses

^M Marginal effect estimated for Logistic models

Nonparametric approach

To test the validity of the RDD, we check the main assumption that individuals do not control the variable that is used to determine treatment eligibility (i.e. age sixty-six). We confirmed no sudden change in the number of respondents in the study at age sixty-six since such a discontinuity might imply that individuals had control over their recorded age (Figure 1). We then checked that there were no discontinuities at age sixty-six in the following variables: gender, education, marriage, disability, chronic disease, private transfer, working status, number of employed family members, living arrangement, urbanicity (region).

Figure 1. Validity check.

Largely consistent with the results from the parametric analysis, the BPS benefit is likely to reduce the material hardship in the aggregate by 0.095 (Table 4) in the RDD model. With respect to domain-specific hardship, the BPS benefit reduces the risks of housing hardship, utility cut-off/bill payment delay, and food insecurity by 1.6, 1.7, and 5.0 percentage points. Considering the lower risk of experiencing those hardships (2.3 per cent, 2.7 per cent, and 7.9 per cent, respectively), the effect sizes are substantial. The benefit also was associated with greater life satisfaction (b = 0.150).

Table 4 Treatment effect estimates, nonparametric approach (RDD)

Note. ^†p<0.10 * p<0.05 ** p<0.01 *** p<0.001

Standard errors in parentheses

^M Marginal effect estimated for Logistic models

Figures 2 and 3 showed the plot for local squared polynomial regression discontinuity estimates. Confirming the results from Table 4, the results indicate a noticeable difference in the slopes for the regression line between the two groups in terms of material hardship in aggregate and life satisfaction.

Figure 2. Local squared polynomial regression discontinuity estimates – Material hardship, aggregated.

Figure 3. Local squared polynomial regression discontinuity estimates – Life satisfaction, aggregated.

Lastly, Table 5 shows the results for each income group. Across all groups, the cash benefit was likely to lead to less material hardship in aggregate. The positive effect of the benefit in housing and food insecurity persisted only in the lower-income group (b = −0.062, p<0.01 and b = −0.139, p<0.001 respectively).

Table 5 Heterogeneous treatment effect estimates by income, Nonparametric approach (RDD)

Note. ^†p<0.10 * p<0.05 ** p<0.01 *** p<0.001

Standard errors in parentheses

^M Marginal effect estimated for Logistic models