Cognitive function, numeracy and retirement saving trajectories

Keywords: Numerical ability, Pensions, Retirement saving

2.1 Measurements of numeracy and literacy

As numeracy measures are a key element of what follows, we briefly outline the key measures here although the measure used is identical to that described in more detail in Banks and Oldfield (2007). In addition, subsequent to that paper, data on literacy has been collected for ELSA respondents so we will briefly discuss the correlation with numeracy and include literacy in much of our subsequent analysis.

The 2002 ELSA questionnaire asked respondents up to five basic questions involving successively more complex numerical calculations.⁹ The six possible questions are presented in Appendix 1. Answers to all questions are entirely unprompted (i.e. respondents are not given a menu of possible answers to choose from). Each respondent initially receives questions 2, 3 and 4. If all of these are answered incorrectly the respondent receives question 1 and that is the end of their numeracy module. Otherwise the respondent receives question 5. If the respondent reports a correct answer to any (or all) of questions 3, 4 and 5, they receive the final and most difficult question q6 that requires an understanding of compound interest. Since more able individuals receive more questions in this design the number of questions answered correctly is a straightforward measure of numerical ability that can be derived simply from this module. This is the measure summarised by Steel et al. (2003) in their initial descriptive analysis of the ELSA data.

The numerical ability measure we derive from the ELSA data is instead designed to place individuals into one of four groups according to their broad numerical ability. This has the advantage of allowing us to choose groups that have some prevalence in the population since a simple counting of correct answers does not take into account the relative difficulty of the questions and furthermore may lead to some clusters where there are many observations, with relatively few individuals at the extremes. Hence for our analysis we choose to define numerical ability in four broad groups according to which of the questions were correctly answered. This coding is indicated in Appendix 1.¹⁰

Our measure of numeracy is a very specific measure, focusing on an individual’s abilities to carry out simple numerical calculations accurately. We do not have access to broader measures of financial literacy such as those used by Lusardi and Mitchell (2007) that capture other dimensions such as knowledge of financial products, knowledge of different types of risks/returns and use of advice. However, given that our measure is strongly correlated with the composition of asset portfolios, in particular the positive relationship between numeracy and the propensity to hold stocks and private pensions (Banks and Oldfield, 2007), it is likely that our measure is correlated with financial literacy and knowledge.

Figs. 1a and 1b, taken from Banks and Oldfield (2007), show how this measure of numerical ability varies with age, sex and education. We use a simple classification of the population into three very broad education groups – Low education is defined as having no academic qualifications, medium education is defined as having O-levels or equivalent and high education is defined as having A-levels (or equivalent) or higher.

Across all education groups, numerical ability as defined by these measures is greater for men than for women, and greater for younger individuals than for their older counterparts. These results mirror those in the more aggregated analysis in Steel et al. (2003) who simply use the average number of correct answers by group. The association between numerical ability and education is clearly evident in the data – as would be expected, groups with higher education have higher numerical ability. Despite this strong correlation, however, there is still a reasonably good distribution of the population across the four numerical abilities within each education group. This is particularly true for the younger sample members who will form the sample of ‘retirees’ on whom some of our later analysis will focus. There are individuals with low numerical ability in the highest education groups and individuals with high numerical ability in the low and medium education groups. The presence of such variation is important if we are to look at the separate effects of education and numeracy on retirement saving outcomes.

Notably, the age pattern in numerical ability is much stronger for the more educated groups, both for men and women, particularly at the top end. Thus the differences across education groups, whilst still present, are less marked among the oldest members of the ELSA sample in comparison to the 50–59 year olds. It should be repeated, however, that our numeracy data are currently cross-sectional in nature, and the presence of both differential mortality (the rich and cognitively able living longer than their poor and less able counterparts) and the presence of cohort differences in numeracy will mean that these correlations may not be the true age profiles. Whilst the true age profile will only be revealed in longitudinal data, it is possible to speculate on the sign of such biases. Differential mortality would reduce the extent to which we observe a decline with age in our sample and, to the extent that compound and simple interest was more likely to be taught to the older members of our sample, cohort differences in the nature of education would work the same way. In these circumstances the age related decline observed in Fig. 1 may even be an underestimate of the true decline with age.¹¹

One further comment on cohort effects is warranted. While Fig. 1 shows substantial differences between the levels of numerical ability of men and women in the cohorts currently aged 50 and over taken together (i.e. the whole ELSA sample), this difference is unlikely to be indicative of differences between men and women for future cohorts. Younger generations of working-age women have more similar educational and labour market circumstances to their male counterparts and, if such circumstances lead to higher levels of numeracy, one might expect future generations of older women to be more comparable to men.

Of course, to argue that such differences in numeracy across individuals are the relevant ones for analysis of retirement savings and retirement saving trajectories, we need to believe that these differences, collected in 2002, are appropriate indicators of the previous lifetime differences across individuals, or at least the differences that have been present across the portion of life where individuals have been working in the labour market and making their consumption and saving decisions. Our data contain no data on numeracy levels prior to 2002, and therefore prior to age 50 for the youngest sample members, and nothing prior to even older ages for the others. While there is a wealth of research suggesting that numerical ability deteriorates with age, the evidence indicates that the magnitude of the deterioration is relatively small until at least retirement age (see Schaie, 1996; Office for National Statistics, 1997; Maitland et al., 2000; Schaie et al., 2004). Schaie (1996, p. 270) analyses longitudinal changes in a number of measures of cognitive ability, including numeracy, and finds that “these data [the Seattle Longitudinal Study] suggest that average decline in psychological competence may begin for some as early as the mid-50s, but that early decrement is of small magnitude until the mid-70s are reached.”

The implication of these studies for our results is that we can confident that, for most of our sample at least, the measure of numeracy that we have access to is indicative of the level respondents had throughout their working life. Moreover, in most of the empirical work that we present we focus on those immediately before and immediately after retirement age, that is before the largest falls in numerical ability manifest themselves. We must be careful, however, when analysing the behaviour of the oldest in our sample, which we touch on below in comparing wealth trajectories across cohorts. The deterioration in numeracy will bias these results, although the direction of the bias will be clear. We return to this issue when discussing the results that use multiple cohorts.

Turning to the measure of literacy, once again there is substantial covariation in cognitive abilities along this dimension with both the other measures of cognitive abilities and with education.¹² To summarise the key issue for our purposes, Table 1 shows the proportions within each numeracy group that fall into each of the three literacy groups. A number of features are worthy of comment. Firstly, it can be seen that the literacy test was a relatively low level test – around two thirds of the sample were in the highest group (receiving a perfect score in the test) and only around one in seven were in the bottom category. These fractions show systematic variation with age and education along the same lines as the numeracy data described above, but this analysis is not presented here. A second feature of Table 1 is the strong correlation between the literacy and the numeracy scores. Indeed it is only really within the bottom two numeracy groups that there are substantial proportions of individuals receiving less than perfect scores in the literacy tests.

The final issue that warrants discussion before we turn to the results of our empirical analysis is that of individual decision making versus collective decision making by a couple. All of our analysis is carried out at the individual level. Therefore, where both members of a couple are in the sample, they each enter the analysis. However, many of the outcomes on which our analysis will focus in the following section are defined most meaningfully at the level of the couple. Indeed, for some measures such as wealth, they are only available at this level due to the way that they were collected in the ELSA survey instrument. As a result of this, we often control for whether the individual is a member of a couple or not and whether or not their spouse or partner is in employment.

Related to this is whether we should consider the possibility that the decisions of an individual in a couple might be impacted by the level of numerical ability of their spouse. After all, one may not need to be particularly numerate if one is married to someone who is so and they are taking an active role in the management of household saving and consumption decisions. When we analyze outcomes that are defined at the level of the couple (such as income and wealth), our strategy will be to allocate each individual the highest numeracy, literacy and executive scores within that couple, effectively assuming they benefit from full sharing of the abilities and knowledge of their partner. In models for individual outcomes, i.e. when we evaluate the accuracy of individuals’ work expectations, their expectations over the adequacy of their retirement income and their subjective life-satisfaction (Tables 8–10) we revert to using individual numeracy scores.¹³

Of course, to the extent that there is a perfect correlation between the abilities of each member of a couple, this may be an inconsequential assumption. Table 2, however, shows that the correlation in numeracy scores between men and women in couples in our data is far from perfect. Looking at the third row of this table as an example, of the 37% of men that are in numeracy group 3, only 11.46 percentage points are married to someone who is also in that numeracy group. Roughly half of the spouses (19.52 percentage points) are in numeracy group 2, with substantial fractions in both the lowest and the highest numeracy groups as well. Similar patterns exist for each of the levels of numeracy, whether one chooses to look along the rows (i.e. indexed by the male’s numeracy level) or down the columns (i.e. indexed by female’s numeracy level). Of course the fact that numeracy scores are higher for men than for women mean that there is more sample above the diagonal than below.

The effect of this ‘household’ assumption on the correlation between numeracy and literacy is in fact to increase the correlation between the two dimensions even further. Table 3 replicates the analysis of Table 1 but this time uses the maximum levels for numeracy and literacy within a couple as opposed to the individual scores. Over 90% of the top numeracy group defined this way are also in the top literacy group, and the proportion of the lowest numeracy group falling into the lowest literacy group rises from 39% (at the individual level) to 43% when defined at the couple level.

Box 2. Construction of broad cognitive function categories.

Table 4.

4.1 Replacement rates

Financial wealth is only one element of retirement resources, and what is more, it is an element that is of differing relative importance across the socio-economic groups – other sources of retirement resources include private and state pensions as well as other welfare benefits. In order to test whether there is more ‘smoothing’ across pre- and post-retirement states it would be more appropriate to look at some measure of replacement rates for net income, or else changes in consumption expenditures. Net incomes (including asset and pension income) are completely measured in the ELSA survey so replacement rates are straightforward to compute. When it comes to expenditure changes, the only item of non-housing expenditures measured consistently across the three waves of data that we use is food spending. Expenditure on other non-food items was collected in 2004 and 2006, but we do not observe enough retirements between those years to be able to facilitate a detailed analysis.

Our analysis of replacement rates simply compares levels of income and food consumption before and after individuals are observed to retire, regardless of how old they were when they retired. As such, our sample is limited to the 700 individuals that are observed retiring in the four-year window and for whom we have full information on all the financial and cognitive variables required for our analysis. Table 5 presents unconditional summary statistics for the distribution of net income replacement rates for retirees between 2002 and 2006 across the four numeracy groups and for the distribution of changes in weekly food expenditures. For income, the distribution is somewhat more compressed for those with the lowest numeracy but other than that there are no marked differences across groups. On the spending side the differences are more apparent – with spending falling on retirement by more for the less numerate.¹⁸ Indeed food spending rose at the median for those in the most numerate group. This pattern – the positive gradient between food spending replacement rates with respect to numerical ability – is true for all three quantiles of the distribution presented in the table.

We also investigated the extent to which replacement rates vary with numeracy after controlling for characteristics such as age of retirement, level of wealth, work status of spouse etc. The regression results on most individual characteristics (which are not presented but are available from the authors on request) are largely in line with our expectations. Replacement rates tend to be lower for those who retire earlier, and are higher for those whose spouse remains in work (since our measure of income is at the level of the couple¹⁹). We find no significant correlation, however, between either measure of replacement rates and numerical ability. This provides some evidence (though it should be noted that this analysis is based on the relatively small sample of 700 individuals who we observe retiring between 2002 and 2006) that the variation in replacement rates across numeracy groups can largely be explained by characteristics that are correlated with numeracy rather than numeracy itself.

One way in which these relatively weak differences between groups in terms of changes in living standards around the time of retirement can be reconciled with the sharper differences in financial wealth trajectories, would be for the other determinants of economic well-being – namely private and state pensions and other state benefits – to be changing differentially across groups. Table 6, which looks at the percentage of income that comes from the state (either state benefits or pensions) by numeracy group for the same sample of retirees that we used for the evidence in Table 5. Almost 80% of income for the least numerate group comes from the state compared to just 30% for the most numerate group. So whilst the least numerate group have less wealth and are less likely to invest in risky assets, these numbers suggest that in fact, for this group, this may be an entirely sensible strategy because in retirement, they will simply rely on the state and enjoy a relatively high replacement rate. If this was the case we would not expect changes in expectations and changes in subjective measures of well-being to be correlated with numerical or cognitive ability. We turn to this topic in the next sections.

4.2 Expectations of retirement and retirement incomes

In common with a number of ageing studies, ELSA collects quantitative information on subjective expectations of a number of future events using the ‘per cent chance’ methodology. Individuals are asked to assess the chances of future events on a scale of 0 to 100, where 0 means there is no chance and 100 means the individual is certain the event will occur. Whilst ELSA collects expectations data on mortality probabilities, housing values, future health outcomes and inheritances/bequests, we focus here on two particular dimensions of expectations. First we look at the expectations of future work for those currently in work. For the large majority of those working at our baseline sample, these expectations can be assessed directly against the subsequent outcomes over the period that the expectations were referring to. Second, we look at individuals’ subjective assessment of the chances that their financial resources will, at some point in the future, prove inadequate to meet their needs.

Taking employment probabilities first, we divide up the sample that were employed at wave 1 into groups according to their self-reported chances of employment at future ages.²⁰ For those that crossed the age referred to in the question between 2002 and 2006 we calculate the fraction that actually were in work, split by the two broad numeracy groups representing the bottom two and top two numeracy categories respectively.²¹ In Fig. 4 these employment outcomes are plotted against the employment expectations for each of the two groups.²² Once again, whilst there are differences between the groups, and these differences go the way that one might suspect – the relationship for the higher numeracy group is slightly steeper on average, representing a stronger correlation between expectation and outcomes – the differences between groups are not particularly striking.

In order to investigate this relationship further and to control for other factors, Table 7 presents a simple model of whether the individual is working in wave 3 as a function of their wave 1 expectation. To facilitate a straightforward interpretation of the marginal effects with so many interaction terms we use a linear probability model, although the results are qualitatively similar if we use a probit model. On the right hand side, in addition to the usual demographic variables included in previous models, we also include the expectations of working reported in wave 1. We run the regression for everyone age under 65 (first two columns of numbers) and also for those aged under 65 who are working at wave 1 (second two columns of numbers). To assess the extent to which individuals in different numeracy groups are better at predicting their work probability, we interact work expectations with the numeracy dummies.

The coefficient on expectations in 2002 reveals that there is a positive and strongly significant correlation between expectations of working in 2002 and the probability of working in wave 3: everything else equal, for every percentage point increase in expectations in 2002 for those aged under 65 in this group, there is a 0.6 percentage point increase in the probability of working at wave 3. The coefficients on the interaction terms of expectations in 2002 with numeracy in Table 7 show how these correlations vary by numeracy group, with group II being the reference group. There are no significant differences between numeracy groups in their ability to predict the probability of working at wave 3.

Turning to the second set of coefficients in Table 7, we might expect those who were not working in wave one to be better able to predict their work probability in wave three since exiting the labour market is often a permanent change in state (although there could of course have been differences across numeracy groups as to the extent to which individuals understood this). To evaluate whether or not this is true, in the second two columns of numbers we restrict our sample to those aged under 65 who were working in wave 1. Whilst the correlation of expectations and outcomes that we previously noted is still observed, we find that on average this group are less able to predict their probability of working than for the whole sample: for every percentage point increase in the expectation of working at wave 1 there is an increase of 0.3 percentage points in the probability of working at wave 3 (this compares to a coefficient of 0.6 for the whole sample). Furthermore, we find here, as we found above when our analysis used the entire sample rather than simply those in work in wave 1, that there is very little difference between numeracy groups in their ability to predict the probability of working.

Turning to the second expectation that we analyse, namely the per cent chances that future incomes will be insufficient to meet future needs, it is clear that there is no concrete benchmark or outcome against which we can assess the expectations of different groups of the population. Instead, we therefore look at the stability and the correlation of these expectations over time within individuals. If a group of the population are consistently surprised (or underprepared) at retirement then we might expect to see systematic revisions of this expectation for that group.

It is certainly true that, on the surface, there is less stability in the financial expectations of the less numerate groups than in their more numerate counterparts. Looking at the lowest two groups together, the correlation across waves of their expectations of adequate financial resources is 0.3 for (both for the entire under-65 sample and 0.25 for those observed retiring between waves). The corresponding numbers for the top two numeracy groups are 0.44 for the entire under-65 sample and 0.43 for the retiring sample. But how much of these differences are simply a result of confounding factors?

To assess this in more detail, Table 8 provides an analysis of the correlations between expectations in 2006 and those collected in 2002, with the correlation allowed to depend on numeracy and a substantial set of controls for other factors. In addition, we split our analysis into three age groups to capture the fact that future financial insecurity may change as individuals move towards, into and through, retirement. Table 8 shows that (everything else equal), for the reference group there is a positive and significant, but somewhat small, correlation over time between financial insecurity expectations. Looking at differences in this correlation across numeracy group reveals that the biggest differences are found in the youngest age group where we find that relative to group II, groups III and IV have more stable expectations over future financial insecurity. For older age groups, there is also some weak evidence that these expectations are more stable for those with more numerical ability, though many of the differences between the coefficients of interest are either insignificant or only marginally significant at the 5% level.

While the relationship between wealth and the level of concern about future financial insecurity is negative for those aged 50–59, it is perhaps surprising there is only weaker evidence of this for the older groups. While those in the wealthiest 40% of the population are less likely to be concerned about the future than those in the bottom wealth quintile, there are no statistically significant differences (at the 5% level) between those in either the third or second quintiles and those in the bottom quintile in either the 60–69 or 70+ age groups.

To look specifically at the issue of ‘shocks’ to expectations around the time of retirement, Table 9 presents the same analysis for both the all-age sample and the sample of only those observed retiring between waves. The intertemporal correlation in expectations is no different for the reference group in the retirees samples than it is in the whole sample, although it is estimated less precisely for the former as a result of the considerably smaller sample size. There is a weak (statistically significant at the 10% level) effect of being in the lowest numeracy group on the change in expectations on retirement, but no significant differences between the other three groups. For the retiring sample, the correlations between wealth and the level of financial insecurity are, if anything, slightly more pronounced than in the all-age sample or either of the two younger aged samples in the previous table.

4.3 Subjective well-being around the time of retirement

Our results so far suggest that although numeracy is correlated with financial wealth accumulation behaviour we find only weak evidence to suggest that it matters for more fundamental outcomes such as replacement rates or accuracy of expectations. In this subsection, we turn to direct measures of subjective well-being to assess whether these differences in numeracy are correlated with changes in welfare on retirement. We are able use two measures of subjective well-being that were collected in both the 2002 and 2006 waves: The first is a single item taken from the CASP19 measure of quality of life²³ which asks how often individuals feel “satisfied with the way my life has turned out”. We translate the four possible answers (often, sometimes, not often, never) into a 4 point scale from 1–4 with the highest value relating to the ‘often’ category. The second measure is one dimension of the General Health Questionnaire 12-item measure of mental health (GHQ12) which asks how often the respondent has recently “been feeling reasonably happy, all things considered”. Again, we translate the four possible responses (not at all, no more than usual, rather more than usual, much more than usual) into a scale from one to four with the highest category representing the highest level of ‘happiness’. Since we are interested particularly in changes or shocks around the time of retirement we once again run a simple regression of the level of subjective well-being on its previous value for both the whole sample and only those retiring between waves one and three.

Table 10 reports results from the CASP19 measure of life satisfaction and Table 11 contains results from the corresponding analysis for the GHQ12 happiness measure. Broadly speaking, the former analysis shows higher life satisfaction for women, for those in couples and for higher wealth groups, for both the whole sample and the sample of retirees. These patterns are mirrored in Table 11 for the current happiness measure although the differences between groups are smaller and are not typically significant for the sample of retirees. Nevertheless, these cross-sectional correlations are all in accordance with the now common finding in economists’ empirical work on happiness and subjective well-being.

There are no consistent patterns emerging across numeracy groups, or any other dimension of cognitive function for that matter, either in the cross-sectional level of well-being or in the persistence of well-being between waves. Of particular interest, there is no evidence of differential shocks to life-satisfaction on retirement across numeracy groups, although it should be noted that given the size of the retiring sample, the power of these tests is limited. Moreover, there is no evidence of any less persistence in this measure in the retiring sample than there is in the all-ages sample, perhaps not surprising given the largely retrospective and evaluative nature of the question. Looking at the measure of ‘current’ well-being, as measured by the GHQ12 measure in Table 11, the correlation between changes on retirement and numeracy are once again weak and suggest that the more numerate group display greater stability over time in their answers. This, if anything, indicates a pattern opposite to that suggested by the analysis of the CASP19 measure. This reinforces the contention that there is little evidence of systematic and consistent gradients in the stability of subjective measures of well-being with respect to numeracy among those who retire, at least given the data currently available.

Derivation of numeracy classification variables

1. Aguiar M, Hurst E. Life-cycle prices and production. American Economic Review. 2007;vol. 97:1533–1559. [Google Scholar]
2. Ainslie G. The Breakdown of Will. Cambridge: Cambridge University Press; 2001. [Google Scholar]
3. Ameriks J, Andrew C, Leahy J. Wealth accumulation and the propensity to plan. Quarterly Journal of Economics. 2003;vol. 118:1007–1048. [Google Scholar]
4. Banks J, Blundell R, Tanner S. Is there a retirement-savings puzzle? American Economic Review. 1998;vol. 88:769–788. [Google Scholar]
5. Banks J, Emmerson C. Public and private pensions: principles, practice and the need for reform. Fiscal Studies. 2000;vol. 21:1–63. [Google Scholar]
6. Banks J, Emmerson C, Oldfield Z, Tetlow G. Prepared for retirement? The adequacy and distribution of retirement resources in England. London: Institute for Fiscal Studies; 2005. Report no. 67. [Google Scholar]
7. Banks J, Marmot M, Oldfield Z, Smith JP. Disease and disadvantage in the United States and in England. Journal of the American Medical Association. 2006;vol. 295:2037–2045. doi: 10.1001/jama.295.17.2037. [DOI] [PubMed] [Google Scholar]
8. Banks J, Muriel A, Smith JP. Attrition and health in ELSA and the HRS. Longitudinal and Life Course Studies. 2010 forthcoming; doi: 10.14301/llcs.v2i2.115. [DOI] [PMC free article] [PubMed] [Google Scholar]
9. Banks J, Oldfield Z. Understanding pensions: cognitive function, numeracy and retirement saving. Fiscal Studies. 2007;vol. 28:143–170. [Google Scholar]
10. Benjamin D, Brown SA, Shapiro JM. Who is Behavioral? Cognitive Ability and Anomalous Preferences. mimeo: Harvard University; 2006. [DOI] [PMC free article] [PubMed] [Google Scholar]
11. Bettinger E, Slonim R. Patience among Children: Evidence from a Field Experiment. Journal of Public Economics. 2007;vol. 91:343–363. [Google Scholar]
12. Borghans L, Duckworth A, Heckman J, Weel B. The economics and psychology of personality traits. Journal of Human Resources. 2008;vol. 43:972–1059. [Google Scholar]
13. Browning M, Lusardi A. Household saving: micro theories and micro facts. Journal of Economic Literature. 1996;vol. 34:1797–1855. [Google Scholar]
14. Burks SV, Carpenter J, Goette L, Rustichini A. Cognitive skills affect economic preferences, strategic behavior and job attachment. Proceedings of the National Academy of Sciences. 2009;vol. 106(no. 19):7745–7750. doi: 10.1073/pnas.0812360106. [DOI] [PMC free article] [PubMed] [Google Scholar]
15. Bozio A, Crawford R, Tetlow G. Institute for Fiscal Studies Briefing Note No. 105. London: Institute for Fiscal Studies; 2010. The history of state pensions in the UK: 1948–2010. [Google Scholar]
16. Deaton A. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Baltimore and London: The Johns Hopkins University Press; 1997. [Google Scholar]
17. Frederick S. Cognitive reflection and decision making. Journal of Economic Perspectives. 2005;vol. 19:25–42. [Google Scholar]
18. Gottfredson L. Intelligence: Is it the Epidemiologists elusive “fundamental cause” of social class inequalities in health?”. Journal of Personality and Social Psychology. 2004;86(1):174–199. doi: 10.1037/0022-3514.86.1.174. [DOI] [PubMed] [Google Scholar]
19. Gul F, Pesendorfer W. Self-control and the theory of consumption. Econometrica. 2004;vol. 72:119–158. [Google Scholar]
20. Koenker R, Hallock K. Quantile Regression. Journal of Economic Perspectives. 2001;vol. 19:143–156. [Google Scholar]
21. Kirby K, Winston GC, Santiesteban M. Impatience and grades: delay discount rates correlate negatively with college GPA. Learning and Individual Differences. 2005;vol. 15:213–222. [Google Scholar]
22. Laibson D. Golden eggs and hyperbolic discounting. Quarterly Journal of Economics. 1997;vol. 112:443–477. [Google Scholar]
23. Lusardi A, Mitchell O. Baby boomer retirement security: the roles of planning, financial literacy and housing wealth. Journal of Monetary Economics. 2007;vol. 54:205–224. [Google Scholar]
24. Lusardi A. Information, expectations and savings for retirement. In: Aaron HJ, editor. Behavioural dimensions of retirement economics. Washington: Brookings Institution Press; 1999. [Google Scholar]
25. Maitland S, Intrieri R, Schaie W, Willis S. Gender differences and changes in cognitive abilities across the adult life span. Aging, Neuropsychology and Cognition. 2000;vol. 7(1):32–53. [Google Scholar]
26. Marmot M, Banks J, Blundell R, Lessof C, Nazroo J, editors. Health, Wealth and Lifestyles of The Older Population in England: The 2002 English Longitudinal Study of Ageing. London: Institute for Fiscal Studies; 2003. [Google Scholar]
27. Melzer D, Gardener E, Guralnik J. Mobility disability in the middle-aged: cross-sectional associations in the English Longitudinal Study of Ageing. Age and Ageing. 2005;vol. 34:594–602. doi: 10.1093/ageing/afi188. [DOI] [PubMed] [Google Scholar]
28. Netuveli G, Wiggins R, Hildon Z, Montgomery S, Blane D. Functional limitation in long standing illness and quality of life: evidence from the English Longitudinal Study of Ageing. British Medical Journal. 2006;vol. 331:1382–1383. doi: 10.1136/bmj.331.7529.1382. [DOI] [PMC free article] [PubMed] [Google Scholar]
29. Office for National Statistics. Adult Literacy in Britain. London: Office for National Statistics; 1997. [Google Scholar]
30. Parker AM, Fischhoff B. Decision-making competence: external validation through an individual-differences approach. Journal of Behavioral Decision Making. 2005;vol. 18:1–27. [Google Scholar]
31. Peters E, Vastfjall D, Slovie P, Mertz CK, Mazzocco K, Dickert S. Numeracy and decision making. Psychological Science. 2005;vol. 17:407–413. doi: 10.1111/j.1467-9280.2006.01720.x. [DOI] [PubMed] [Google Scholar]
32. Pickering A. A Simpler Way to Better Pensions. London: Department for Work and Pensions; 2002. [Google Scholar]
33. Sandler R. Medium and Long Term Retail Savings in the UK. London: The Stationery Office; 2002. [Google Scholar]
34. Schaie K. Intellectual development in adulthood. In: Birren J, Schaie K, editors. Handbook of the Psychology of Aging. 4th edition. San Diego: Academic Press; 1996. pp. 266–286. [Google Scholar]
35. Schaie K, Willis S, Caskie G. The Seattle longitudinal study: relationship between personality and cognition. Aging, Neuropsychology, and Cognition. 2004;vol. 11:304–324. doi: 10.1080/13825580490511134. [DOI] [PMC free article] [PubMed] [Google Scholar]
36. Shiv B, Fedorikhin A. Heart and mind in conflict: the interplay of affect and cognition in consumer decision making. Journal of Consumer Research. 1999;vol. 26:72–89. [Google Scholar]
37. Smith JP, Willis R, McArdle JJ. Financial decision making and cognition in a family context. ECONOMIC JOURNAL. 2010 doi: 10.1111/j.1468-0297.2010.02394.x. this issue. [DOI] [PMC free article] [PubMed] [Google Scholar]
38. Steel N, Huppert F, McWilliams B, Melzer D. Physical and cognitive function. In: Marmot M, Banks J, Blundell R, Lessof C, Nazroo J, editors. Health, wealth and lifestyles of the older population in England: The 2002 English Longitudinal Study of Ageing. London: Institute for Fiscal Studies; 2003. [Google Scholar]