بازنشستگی روانی و تعلیم و تربیت

We assess the validity of differences in eligibility ages for early and old age pension benefits as instruments for estimating the effect of retirement on cognitive functioning. Because differences in eligibility ages across country and gender are correlated with differences in years of schooling, which affect cognitive functioning at old ages, they are invalid as instruments without controlling for schooling. We show by means of simulation and a replication study that unless the model incorporates schooling, the estimated effect of retirement is negatively biased. This explains a large part of the “mental retirement” effects which have recently been found.

In many countries people are living longer and retiring earlier than in previous decades. They are spending a larger proportion of their lifetime in retirement, depending on earlier savings or pension benefits to finance consumption. Populations are aging and the burden of risk and provision of old age income security is shifting from public collective to private individual. Cognitive ability is important for both decision making in general and financial planning (Christelis et al., 2010) and senior saving behavior (Banks et al., 2010) in particular. Differences in cognitive performance amongst the elderly will increasingly determine their future consumption possibilities and drive inequality. Although there is a great deal of heterogeneity in individual trajectories, cognitive ability usually increases at a declining rate until middle age and declines afterwards. It is becoming increasingly recognized (Hertzog et al., 2008) that different cognitive ability profiles in later life are associated with different behaviors from pre-school (Case et al., 2009), through compulsory school (Glymour et al., 2008), choice of leisure activities (Scarmeas and Stern, 2003) and age of retirement (Adam et al., 2007). If cognitive ability is malleable and some of these relationships are causal then cognitive ability can be usefully thought of as a form of human capital (Ben-Porath, 1967) or health capital (Grossman, 1972). Cunha and Heckman (2007) develop a model of cognitive and non-cognitive skill acquisition through schooling investments, where the effect of schooling on cognitive ability is unambiguously positive. The relationship between retirement and cognitive functioning in a Grossman-like model is ambiguous. Bonsang et al. (2010) set up a standard Grossman model where individuals with higher cognitive functioning may perform work tasks and leisure activities more effectively. Work and leisure can involve different types of investment in cognitive functioning. Pension eligibility increases the cost of work relative to leisure investments. The net effect of retirement on cognitive functioning will depend on relative marginal productivities. Rohwedder and Willis (2010) follow instead the insights of the “use it or lose it” hypothesis, and the suggestions from psychological literature that cognitive ageing can be delayed by engaging in cognitively demanding activities. They argue that during retirement the brain is not stimulated to the same extent as while working in the labour market, and thereby loses cognitive ability. This argument is consistent with evidence from psychological literature if retirement from the labour market is associated with reduced mental exercise. Empirical evidence for retirement effects on cognitive functioning is mixed and depends on how cognitive abilities are measured and on the source of identification used. Coe et al. (2012) use employer-provided early pension benefit windows in the US to provide variation in retirement ages and find no effect on word recall and numeracy, or if anything a positive effect for blue collar workers. Several papers use cross-country variation in pension eligibility ages to find significant detrimental effects of retirement on short term memory or word recall (Bonsang et al., 2010, Rohwedder and Willis, 2010 and Mazzonna and Peracchi, 2012). In this paper we study the correlation between cross-country variation in pension eligibility ages and educational attainment, and the consequences of this for estimating the effect of retirement on cognitive functioning. In Section 2 we describe our data, which resembles as closely as possible that used in the cross-country studies discussed above. Section 3 states the arguments for instrument validity, illustrates the correlations of interest and presents a bias simulation. Section 4 presents a replication study and extension of Rohwedder and Willis. Section 5 concludes.

The relationship between retirement and cognitive functioning at older ages is ambiguous from a theoretical perspective, and empirical work has produced mixed results, but the weight of evidence from cross-country studies has found that retirement has a negative effect on cognitive functioning. We show that cross-country variation in pension eligibility ages are invalid instruments for retirement, unless the empirical model takes differentials in schooling levels into account. Ignoring the distribution of schooling levels across instrument-determined groups partly explains the large negative estimates in the influential study of Rohwedder and Willis (2010). Our study is about the empirical analysis of the effect of retirement on cognitive functioning, but our findings are of more general relevance for other studies that use similar identification arguments and where schooling may be a confounder. Studies that exploit cross-country or cross-state policy variation in other contexts may have otherwise good arguments for instrument validity which are undone because schooling is an omitted explanatory of the outcome of interest. Conversely, other studies of the effect of retirement on cognitive functioning that use different identification strategies or different instruments, (as Bonsang et al. (2012), who exploit panel variation and social security eligibility in the U.S. for identification) do not suffer from this particular bias. Throughout the paper we have considered the consequences of omitting education as an exogenous variable. Mazzonna and Peracchi (2012) suggests that the endogeneity of education may also be a concern. If cognitive functioning at old ages correlates with determinants of schooling level, as Banks and Mazzonna (2012) find, then controlling for education may not eliminate the endogeneity bias, and the estimation of a causal effect would require accounting simultaneously for the endogeneity of both education and retirement. This is left for future work.