American Journal of Law & Medicine

Retiree Out-of-Pocket Healthcare Spending: A Study of Consumer Expectations and Policy Implications

ii. Monthly Estimates by Age Cohorts

The distribution of monthly cost estimates across age cohorts, as illustrated in Figure Two F, offers another perspective on estimates by age cohorts. In contrast to responses discussed above with respect to future Medicare premiums, the median respondents in their forties are projecting almost the same average monthly costs as those just on the eve of retirement, suggesting that younger respondents are not projecting cost increases (or, alternately, might not understand the implications of real cost growth in healthcare costs (168)). While the median estimates of the younger cohorts are somewhat higher than those of the oldest cohorts, the differences are relatively small. As the extent of those increases is expected to be substantial, these responses may suggest an important source of consumer confusion, or at least misapprehension about likely healthcare costs in retirement. There does, however, seem to be somewhat greater uncertainty about retiree healthcare expenditures for at least some of younger cohorts. While the spread between the twenty-fifth and seventy-fifth percentiles does not widen markedly with the younger cohorts, the ninetieth percentile responses do, suggesting that at least a fraction of respondents in the younger cohorts are estimating substantially higher healthcare costs than their counterparts in the older cohorts.


iii. Other Features of Monthly Estimates

In contrast with the premium estimates discussed above, respondent estimates are relatively stable across treatment groups. As reported in Table Six, while Treatment C responses showed a modest narrowing in distribution similar to the narrowing noted above for their estimates of premiums, the treatment groups provided similar estimates, regardless of whether they were posed as simple questions about out-of-pocket costs in Treatment A or given a good deal of additional framing and anchoring in Treatments B and C. Median responses did not differ greatly across treatments.

Our survey questions on monthly cost estimates also included two extensions that explored the extent to which respondents expected their monthly costs to vary over the course of retirement. First, we asked all respondents to make separate monthly cost estimates for the last year of their lives. Respondents overwhelmingly estimated that they would have higher monthly costs in their final year. (169) The median estimate was $350 as compared with a median estimate for average monthly costs of $200 for all respondents. We also calculated the ratio of individual responses on this question to their average monthly cost estimates and determined that the median ratio was 1.46 or nearly 50% higher than the average monthly cost estimate. This is similar to, though not as extreme as, estimates based on expert data of a ratio of two, or nearly 100% higher. (170)

For respondents in Treatments B and C, we also compared individual respondent estimates of monthly costs at age eighty-five as opposed to monthly costs at age sixty-five to gauge their understanding of how costs are likely to change toward the end of their lives. While one quarter of respondents projected monthly costs at eighty-five at or below levels at age sixty-five, the median response indicated a projected increase of 33%, which is in line with the ratio of cost increases reported above in our literature review for costs experienced by the average member of an eighty-five-year old cohort, as compared to the average member of a sixty-five-year old cohort. (171) So, again on this dimension, the typical response was consistent with expert views. These results--in conjunction with the findings on overall monthly estimates above--suggest that at least some people do have some understanding of future healthcare expenditures, more so than we hypothesized we would find.

b. Lump Sum Estimates

As an alternative measure of retiree healthcare costs, we asked all respondents to estimate the amount of money that a person similar to the respondent would need to accumulate by the age of sixty-five in order to save enough money to pay for their total expected out-of-pocket costs for healthcare in retirement. (172) Our goal here was to solicit savings targets that the respondents would associate with the amount needed on the eve of retirement to cover expected healthcare costs in retirement. (173)

Many different factors could affect an estimate of lifetime spending: estimates of monthly spending, projections of life expectancy, and ability to toggle between monthly and lifetime estimates, considering real cost growth. Even if a respondent estimates monthly spending well, she might underestimate her life expectancy and the number of years of future spending. Further, recent financial literacy research illustrates that people have difficulty translating between periodic and lump sum payments, suggesting that our respondents might similarly struggle. (174) Accordingly, in designing our survey, we hypothesized that respondents' estimates of lump sum costs might be significantly further from expert views than their monthly cost estimates. As described below, while the lump sum responses were lower than monthly estimates compared to expert benchmarks, the overall structure of lump sum responses--and the bimodal distribution identified above--were similar to the responses for monthly costs

Survey responses on lump sum estimates are reported in Table Seven with results for all respondents on the top line, followed by responses broken out by age cohort and then treatment. The median lump sum estimate for all respondents was $50,000, with the twenty-fifth and seventy-fifth percentiles of responses ranging from $10,000 to $150,000.

i. Lump Sum Estimates Compared to Benchmarks

Our comparison of respondent estimates with expert estimates again center on 2020 benchmarks. As these benchmarks vary by gender, we distinguish between projected spending for male and female respondents. Figures Three A-B presents histograms of responses for male and female subsamples against their respective 2020 benchmarks. Our benchmark EBRI study does not include a twenty-fifth percentile estimate, so the left hand column of these figures reports the share of responses beneath the benchmark median. As with the comparable histograms for average monthly costs, Figures Three A-B shows that responses from both men and women again produced a bimodal distribution with a healthy share of respondents making lump sum estimates above the seventy-fifth percentile of the relevant benchmark (28.0% for men and 17.7% for women) and a disproportionate share of both subsamples reporting responses beneath the benchmark median (65.2% for men and 79.5% for women). While the general structure of these histograms is similar to analogous charts for average monthly cost estimates, (175) the women's responses fall further beneath benchmark metrics than the men's do. This is particularly true if one focuses on median responses. Whereas the men's median response of $60,000 is over 50% of the benchmark median for men of $109,000, the women's median response of $30,000 is less than a fifth of the benchmark median of $156,000 for women. (176) We discuss this gender differential in more depth in Section 3 below.

ii. Lump Sum Estimates by Age Cohorts

Results broken down by age cohort were also reminiscent of those we obtained for estimates of average monthly costs: The median responses of all age cohorts were at or close to $50,000, and the distance between the twenty-fifth and seventy-fifth percentile responses was also highly consistent across age cohorts, although the ninetieth percentile responses did tend to drift upward for younger cohorts, again suggesting greater uncertainty about future costs (see Figure Three C). On average, the younger cohorts seemed to be making lump sum estimates quite similar to those of older cohorts on the eve of retirement or in retirement. (177)


Interpreting responses of younger cohorts on these lump sum questions is difficult. Some respondents might have interpreted the question to solicit estimates of savings targets for someone reaching sixty-five today in which case adjustment for future real increases in healthcare costs would not have been appropriate. (178) It is also possible that respondents had difficulty in making adjustments to savings targets to reflect real increases in future healthcare costs. In this case, it is troubling that younger cohorts generally did not project high savings targets, especially if these projections influence retirement planning for individuals several decades away from retirement. (179) While there are undoubtedly complexities in interpreting responses of younger cohorts with respect to these savings targets, the fact that younger respondents did not estimate materially higher savings needs than older cohorts strikes us as a potentially important finding and worthy of further study. (180)

iii. Implied Lump Sum Estimates

One of the hypotheses that we wanted to explore with our lump sum estimates was whether respondents might do a relatively good job estimating monthly costs but then make some sort of systematic error in mental math that led them to make unreasonably low lump sum estimates. Such a cognitive error would be significant because it could lead individuals to set their targeted savings for retirement healthcare spending at too low levels, even if they did a relatively good job at estimating what their average monthly expenditures for retiree healthcare might be. To some degree, such errors may have occurred as the lump sum estimates reported above do fall further from expert benchmarks than did respondents' monthly cost estimates.

To explore whether this reduction in lump sum estimates might be the product of errors in respondents' expectations regarding their own life expectancies, we asked all respondents a series of questions about their own assessment of their life expectancies. (181) In reviewing responses to these questions, we found that respondents' estimates were reasonably close to SSA projections on life expectancy. Our median respondents reported an 80% likelihood of living past sixty-five, a 70% likelihood of living past seventy-five, a 50% likelihood of living past eighty-five, and a 10% likelihood of living past ninety-five. These responses underestimate the likelihood of surviving to sixty-five and seventy-five (which Social Security actuaries currently put at approximately 92% and 75%) by 10% to 15%, but somewhat overestimate the likelihood of living beyond eighty-five (which Social Security actuaries estimate in the range of 43%). (182) Thus, it does not appear that our respondents were systemically underestimating their life expectancies in a material way.

We further analyzed the relationship between respondents' monthly cost estimates and their lump sum estimates by generating for each respondent an "implied lump sum estimate," based on the monthly cost estimates that each respondent provided, his or her final year monthly cost estimates, and his or her reported self-assessed life expectancy. Using this information, we projected an expected cost cash flow for the person and then discounted that cash flow back to a valuation at age sixty-five, which represents the amount of money the person would need to exactly cover his or her self-reported expected monthly costs. We made these calculations using several different discount rates; the results reported here employ a 1.5% real discount rate. (183)

Rather than excessively discounting their lump sum estimates, respondents appear to have modestly adjusted upward their lump sum estimates as compared to implied lump sum estimates. The implied lump sum estimates, at least as we calculated them, were about 6% lower than respondents' actual lump sum estimates taken as a group ($47,000 median implied lump sum, as compared to $50,000 median actual estimated lump sum). Conceivably, respondents may have been adding a modest cushion of additional savings to make sure they would have enough for unanticipated costs. While these results are sensitive to our assumed discount rate, the results do not suggest significant downward errors in lump sum calculations as compared with monthly costs adjusted for self-reported life expectancies. (184)


In fact, we see some evidence of erroneous inflation of lump sum estimates when looking at the distribution of implied lump sum estimates by age cohort, as shown in Figure Four. The distribution of percentiles is much narrower in this figure than in the comparable figure (Figure Three C) for actual lump sum estimates. In particular, the ninetieth percentile estimates of the implied lump sum calculations are much lower (in Figure Four, we have superimposed circles indicating the ninetieth percentiles for actual lump sum estimates from Figure Three C and boxes to reflect the seventy-fifth percentile estimates). For example, the ninetieth percentile estimate of implied lump sums for the forty-five to forty-nine year old age cohort is about $180,000 whereas the comparable ninetieth percentile actual lump sum estimate is $750,000. In other words, far from excessively discounting their lump sum estimates, our respondents in many cases were offering lump sum estimates that were substantially higher than the savings levels actually needed to match their own estimated monthly costs and self-assessed life expectancies.185 A substantial number of respondents appear to have been engaging in mental math that suggested an unobtainably high savings target, rather than engaging in mental math that set unrealistically low savings needs. Such high targets could create a barrier to saving due to a sense of futility, as discussed further below.

iv. Other Notable Features of Lump Sum Estimates

As reported in Table Seven above, lump sum estimates showed extremely modest variation across treatments, with all three treatments having a median estimate of approximately $50,000 and only a modest narrowing of distributions from Treatment A to the other two treatments. So, as was true of monthly cost estimates, framing and anchoring had negligible effects on responses.

3. Estimates and Demographic Spending Variation

One of the complexities in interpreting respondents' answers is uncertainty about whether those reporting low or high estimated costs are, in fact, individuals who will actually incur below or above median healthcare costs in retirement. To tease out this question, we segmented our sample into a series of subgroups based on income, gender, self-reported health status, and financial sophistication (based on self-reported information on financial planning and familiarity with budgeting and healthcare insurance, as well as self-reported consultations with financial planners). We then analyzed whether this partitioning of the data produced differences in average monthly cost estimates or actual lump sum estimates that were consistent with expert evaluations of the relationships between these categories and out-of-pocket retiree healthcare costs. The results, which are summarized in Table Eight for average monthly costs, are mixed. (186)

Variations by income levels tracked expert evaluations. As discussed above, higher income individuals tend to pay more for retiree healthcare and also live longer, thereby increasing overall retiree healthcare costs. Our survey respondents seemed to be quite attuned to this effect. So, as reported in Table Eight, the median expected monthly cost of the lowest quintile respondents was just $100, whereas the median response for the highest quintile of respondents was $350. As shown in Figure Five, this differential was even more pronounced with respect to lump sum estimates where median estimates of the lowest income quintile were $10,000, as compared to $125,000 for the highest quintile. In terms of the effect of income on retiree healthcare costs, respondents' intuitions were directionally aligned with expert views, even if perhaps showing a stronger effect than experts might suggest is likely. (187)


We observed the opposite with regard to gender. As a result of having longer life expectancies and more expensive supplemental coverage, a typical woman retiring in 2010 has 40% higher expected out-of-pocket healthcare costs in retirement than a typical man (188) and higher expected annual costs. (189) Nevertheless, as reported in the second section of Table Eight, women largely estimated lower average monthly costs than men with a median estimate of $190 for women as compared to $217 for men. This difference was even more pronounced for lump sum estimates, where women's median estimate was $30,000 versus $60,000 for men. Thus, women underestimated average monthly healthcare costs as compared to men and compounded that underestimation in producing lump sum estimates, making their actual reported median estimates substantially below the benchmark median estimates for women, as drawn from our literature review.

Finally, we explored whether self-reported financial sophistication might be correlated with survey responses. We asked all respondents three questions to gauge financial sophistication. The most pronounced effect on monthly estimates came from our third question on financial sophistication: whether respondents had consulted with a financial planner about retirement. (190) The median estimate among respondents who reported a consultation with a financial planner was $300, as compared to $175 for respondents who said they had not. While one must treat self-reported responses of this sort with caution, these results do raise the possibility that personal interventions with respect to retiree healthcare costs may be effective in raising individual estimates of retiree healthcare costs.

In an effort to explore the interactions between these various correlates and respondents' estimates, we conducted a series of regression analyses. Table Nine A reports summary results for four rudimentary models examining the correlates of monthly cost responses. The first three columns of Table Nine A are quantile regressions (at the twenty-fifth percentile, median, and seventy-fifth percentile) and the fourth column is a trimmed Ordinary Least Squared (OLS) log formulation. (191) For each regression, we included gender, a dummy for age cohorts younger than fifty-five, income quintiles, health status, educational achievement, and a dummy representing consultations with financial planners as independent variables. While explaining only a small fraction of overall variation in monthly costs, all four models show consistent, statistically significant coefficients for the dummy for younger cohorts, income quintiles, educational attainment, and the financial planner dummy, suggesting that all of these variables are correlated with estimates of monthly costs. The coefficients for these variables also were intuitively coherent, with higher income quartiles having larger coefficients than lower income quartiles, and the magnitude of the coefficients for the financial planner dummy and educational attainment increasing for higher percentile regressions. The models also suggest that, once other factors are controlled for, respondents from younger cohorts do offer somewhat higher estimates of monthly costs than do the older cohorts, although the magnitude of those differences (on the range of seventeen dollars to eighty-two dollars in the quantile regressions) do not equal the projected increases in future health costs that experts predict. (192) While the female dummy has a negative coefficient in two of the models, in only one model is the coefficient statistically significant, casting some doubt on earlier results suggesting that women were making lower estimates across the board for monthly costs than men (once controls for educational attainment and household income are included).

Table Nine B presents similar regressions for lump sum estimates. (193) While most of the results were similar to those for average monthly cost estimates, the female dummy did, however, behave differently in the lump sum regressions, with consistently negative coefficients in all four regressions and statistically significant coefficients for three quantile regressions. So, negative effects of gender on cost estimates identified early seem to reemerge with these lump sum regressions, suggesting that women's estimates are more likely to fall short of men's when dealing with lifetime costs as opposed to monthly budgets. The financial planner dummy was again consistently significant in all models. The coefficients for income quintiles and younger cohorts were not as consistently statistically significant as with the monthly cost regressions, but they still retained the same basic structure as the analogous coefficients in the monthly cost regressions.

The regression models presented in the preceding tables should be viewed with some caution. To begin, survey responses on cost estimates are difficult to model as they skew towards higher numbers with a fairly large number of outliers. While quantile regressions, trimming, and log transformations of dependent variables are all designed to mitigate those complexities, these adjustments may not offer complete solutions. In addition, the explanatory power of the models is limited, with quite low adjusted R-Squares and Pseudo R-Squares in all cases. That said, the correlations with income, use of financial planners, and educational achievement seem reasonably robust, (194) Our survey responses, taken as a whole, did reflect the reality that wealthier respondents will likely spend more on healthcare costs in retirement as a result of a combination of progressive insurance premiums, higher consumption of healthcare services by the more affluent, and greater longevity. The strong performance of the financial planner dummy suggests that this kind of financial education may be associated with higher estimates of retiree healthcare costs, although causation here is unclear. While the negative correlation between the female dummy and cost estimates was not robust across all our regressions, the negative coefficients on female lump sum estimates were often statistically significant, and we certainly did not find any evidence that women were estimating higher healthcare costs than men, as experts suggest they should. On balance, our results suggest the possibility that women may well fail to appreciate that they face higher healthcare costs in retirement and, indeed, may be systemically underestimating costs as compared to men, at least with respect to their lump sum estimates.


In the final module of our survey, we divided all respondents into two separate groups and asked each group a series of questions designed to elicit their assessment of three sources of potential risk for out-of-pocket healthcare expenses in retirement: variation in individual health experience, unanticipated medical cost growth, and changes in government policies with respect to Medicare and other government programs. As discussed earlier in our literature review, all three of the risks could be material. Variations in individual health experience could double to triple individual out-of-pocket costs above median levels, and government policy changes could as much as double them. Respondents across the board did not identify individual health and policy changes as the most salient risks, nor did they appreciate the extent to which these risks could increase their out-of-pocket spending for healthcare in retirement. Thus, we found that while respondents' understanding of typical out-of-pocket spending is variable, their understanding of uncertainty is pervasively low.

1. Group One: Assessments of Concern and Severity

For half of our respondents, we asked them to make a qualitative assessment of the risk that each form of uncertainty poses to future spending. First, we asked respondents in this group to evaluate on a four-point scale how concerned they were about each of the risks. Second, we asked them if the risk should materialize, how much more they would need to budget for out-of-pocket healthcare costs if they wanted to be "highly confident" of having sufficient resources to cover the costs.

In this formulation, respondents seemed to identify policy changes and medical inflation as being the greater sources of risk, but they underestimated the potential magnitude of both. In terms of level of concern, summarized in Figure Six, these two risks dominated across age cohorts. Again, this response is inconsistent with expert perceptions of the relative risk, which would clearly rank variations in individual health experience, and probably also policy uncertainty, as a more significant risk than unanticipated medical inflation, especially for those at or near retirement for whom any inflation will have limited impact.


On the issue of how large of a financial impact respondents estimated that the risks could pose to their budgets, nearly all respondents underestimate the magnitude of these risks, especially with regard to individual health experience risk. Figure Seven reports responses for Individual Health Experience (on the left) and Policy Changes (on the right). For example, only a fifth of all respondents estimated that adverse individual health experience could lead to a more than 50% increase in out-of-pocket costs, although expert opinion suggests those who end up in the seventy-fifth or ninetieth percentile of out-of-pocket costs are likely to spend double to triple that of someone at the median. (195) Similarly, less than a third of respondents reported that they would need at least 50% more in financial resources to compensate for adverse changes in government policy, even though expert views are that some current reform proposals for Medicare could have a much larger effect. (196)

2. Group Two: Willingness to Pay

To gain an alternative perspective on the topic of risk, we posed questions to the other half in terms of their willingness to pay to be flee of each of these specific risks. (197) As reported in Table Ten, the median respondent was willing to pay a monthly insurance premium of about $150 to be relieved from the risk of higher out-of-pocket costs from individual health experience. While it is difficult to know if this specific estimate is actuarially sound, what is most interesting is that the willingness to pay responses for each of these three questions were roughly similar. While the medians for responses on willingness to pay for protection against medical inflation and willingness to pay for protection against policy changes were a bit lower ($125 and $120, respectively) than the health experience analog ($150), the distributions were roughly comparable. Certainly, there is no indication in this data that respondents overall were especially concerned about individual health experience or policy changes; indeed the latter had the lowest median and distribution ranges of the three. 198 Nor was there strong evidence in our results that younger workers were particularly wary about risking healthcare costs. Responses to the other two willingness to pay questions suggest that younger respondents placed a higher value on protection against bad individual health experience and policy changes than did older respondents, but even there the trends were not especially strong. …

Log in to your account to read this article – and millions more.