The Three Knowledge Gaps Behind Retirement Planning
The IZA study is not only asking whether people know how long similar people live. It is testing whether the knowledge needed for retirement planning exists as a connected system: an understanding of longevity, public pension rules, and basic financial mechanics. The useful judgment is therefore not whether respondents passed or failed a quiz. It is whether the survey’s measures show an accurate and integrated planning toolkit—and what the study’s local design allows us to infer from the answer.
That distinction matters because an average can conceal a distribution, and a score can conceal a missing connection. A group may appear reasonably informed about longevity while many individuals are far from the actuarial benchmark. A respondent may understand compound interest yet misunderstand the pension income that will determine how much saving is needed. The study makes those differences visible by measuring the pieces separately before asking how they relate.
What the survey could measure
The individual survey took place in February 2024 among 1,000 people aged 25 to 60 in Trentino–South Tyrol, a region in northeastern Italy. Interviews used computer-assisted telephone interviewing. The sample was designed around the reference population’s age, gender, county, employment status, and primary language. This gives the comparisons a defined population and institutional setting. It does not make the results a global estimate of what working-age adults know about retirement.
The measurement sequence was important. Respondents were asked to estimate the average age at death for people with the same relevant age, gender, and county characteristics. That question concerns a reference group, not the respondent’s own medical prognosis. The researchers compared the answer with the official actuarial estimate for that same group. Later, respondents were asked about their own expected age at death. Separating those questions allowed the study to measure population longevity literacy without forcing the first answer to carry the respondent’s private health beliefs.
The researchers used point estimates rather than asking respondents for a probability of surviving to several target ages. That choice reduces the burden of translating a vague chance into a numerical probability and avoids some focal-point problems that can arise when people round uncertain probabilities. It also makes the comparison straightforward: how far is the reported age at death from the actuarial benchmark? But a point estimate remains a response to a particular question format. It is not a complete probability distribution and should not be treated as one.
The survey then tested two other knowledge domains. Pension questions asked how state benefits are calculated and how the public system is financed. Financial questions tested compound interest, inflation, and diversification. These are not interchangeable versions of the same ability. Pension literacy concerns the rules and income stream that a person expects to receive; financial literacy concerns how savings and risks behave over time. Longevity literacy concerns the length of the period those resources may need to cover.
The information experiment was placed within this broader sequence. A randomly selected half received the official actuarial life expectancy for their age, gender, and county. The control group was told that the national statistical office calculates such estimates but did not receive the numerical figure. Because the survey had already measured the knowledge domains, the researchers could examine both the level of each input and the possibility that an official statistic would enter the respondent’s personal planning process.
This design gives each number a proper denominator and comparator. The longevity estimate is compared with an actuarial value for a defined reference group. The pension and financial results are proportions of this 1,000-person sample answering particular questions correctly. The treatment comparison is between randomly assigned groups within the same survey, not between this region and a different country. Those distinctions are easy to lose when results are compressed into a headline.
It also separates two kinds of evidence that are often blended together. The survey’s descriptive results tell us how far beliefs and answers were from the benchmarks. The randomized treatment can support a causal comparison of what happened when a numerical actuarial figure was supplied. Random assignment does not make every descriptive feature causal, and it does not remove the geographic limits of the sample. For example, the fact that a respondent was inaccurate about the lifespan of similar people does not by itself show why they were inaccurate or what they would do with better information.
The benchmark is doing specific work here. It is not a claim that an actuarial average is a person’s destiny. It gives the researchers a common reference point for judging population longevity literacy. A respondent’s error can then be interpreted as a difference between a stated estimate and the official estimate for the same age, gender, and county. That makes the first question less emotionally loaded than asking someone to forecast their own death, while preserving the possibility of comparing it later with the personal forecast.
The point-estimate format has a related advantage and a related limit. Saying “age 82” is cognitively simpler than assigning probabilities to several survival ages, and it avoids treating a rounded probability as a precise belief. But a single age does not reveal how confident the respondent is, how they imagine the distribution around that age, or how they would respond to different risks. The measure is well suited to the study’s comparison, but it cannot carry more psychological meaning than the question elicits.
Why the averages are misleading
Respondents’ average estimate for the lifespan of similar people was about 82 years and four months. The corresponding actuarial benchmark was about 84 years and nine months, so the average bias was pessimistic by roughly 2.4 years. If the mean were the only information available, one might describe the group as somewhat miscalibrated but broadly close.
The distribution tells a more consequential story. More than 32% underestimated the benchmark by over five years, while about 8% overestimated it by over five years. Together, roughly 40% of respondents were more than five years away from the actuarial estimate in one direction or the other. Only 7% gave an estimate within one year of the official figure. The average is therefore not a typical individual. Large errors on opposite sides can partly offset one another mathematically while leaving many people with a materially different planning horizon.
The direction of the errors also varies across groups. Female respondents underestimated the longevity of comparable women by more than three years. The paper suggests that such an error could contribute to undersaving, especially in private, non-annuitized retirement assets, but the survey did not directly observe a savings shortfall caused by the belief. The careful interpretation is that the measured error creates a plausible exposure to risk; it is not proof that the proposed financial consequence occurred.
The same average-versus-distribution problem appears in pension knowledge. More than 35% of respondents said they did not know how state pension benefits were calculated. About 32% correctly identified the contribution-based calculation system, and about 20% correctly identified pay-as-you-go financing. Only 6% answered both pension questions correctly. These figures do not show that respondents lacked every kind of pension knowledge. They show that understanding the specific institutional rules tested by the survey was uncommon.
Financial knowledge was uneven as well. About 34% correctly answered the compound-interest question, 36% correctly answered the inflation question, and 53% correctly answered the risk-diversification question. Only 14% answered all three correctly. The pattern is not a single ladder on which a person is simply “more” or “less” financially literate. Someone may understand that a diversified fund is safer than a single stock while still misunderstanding how inflation erodes purchasing power or how interest compounds.
These answers correspond to different failure points in the life-cycle budget. Misunderstanding compound interest can distort how a balance grows over time. Misunderstanding inflation can make a future nominal payment look more generous than its purchasing power will be. Misunderstanding diversification can leave a person exposed to the outcome of one company or one concentrated asset rather than the broader risk they intended to take. None of these questions determines a retirement choice on its own, but each can alter the resources available at the transition from work to retirement.
The pension questions create a different kind of uncertainty. A person may know that they need income after leaving work without knowing how the public benefit is calculated or whether current contributions finance current retirees. That uncertainty affects the amount of private wealth the person believes they must build. It also means that a longevity estimate cannot be evaluated in a vacuum: the same expected age at death can imply a different private saving requirement under a different pension rule. This is why the study treats pension literacy as a separate input rather than as a footnote to financial knowledge.
That fragmentation is the central result of the measurement exercise. Longevity literacy was essentially uncorrelated with pension literacy, with a correlation of 0.00, and nearly uncorrelated with financial literacy, at −0.05. Pension and financial literacy had a low positive correlation of 0.29. In other words, the two financial domains had some shared variation, but knowledge about how long similar people live did not line up with either one.
A correlation is not a causal explanation. A near-zero association does not prove that longevity knowledge can never affect a financial choice, nor does it prove that pension knowledge is irrelevant to saving. It says that, in these measures and this sample, respondents who were more accurate about the reference group’s lifespan were not systematically the same respondents who scored higher on the pension or financial questions. The standard planning model requires the domains to work together; the data show that they are usually held apart.
The paper also constructs standardized indices so that performance across the three domains can be compared on a common scale. That is useful for describing an overall pattern, but it does not turn separate facts into a single mental account. An index can summarize low or high knowledge; it cannot show that a respondent uses longevity information when interpreting pension benefits, or that a pension expectation changes how they evaluate a savings return. Integration is a relationship among inputs, not merely an average of scores.
This distinction matters for both diagnosis and intervention. If a person answers one financial question correctly, a planner should not assume that the person understands the pension replacement they will receive or the number of years it must cover. Conversely, a low score on one question does not prove that the person is incapable of making a sound decision with advice or a clear statement. The survey identifies where the measured knowledge is thin and disconnected; it does not observe every reasoning step between a correct answer and an actual retirement choice.
The survey’s measurement choices also establish boundaries around the interpretation. The pension questions were tied to the Italian system, so their difficulty partly reflects country-specific institutions. The financial module contained three standard concepts, not every skill involved in managing a retirement portfolio. The longevity question concerned a group average and used a point estimate, not a personal survival curve. These are sensible measures for the study’s purpose, but they should not be inflated into a complete assessment of retirement competence.
The value of measuring the domains separately is that it prevents a false remedy. If longevity knowledge were the only weak input, sending a more accurate life-expectancy number might plausibly address the problem. But the survey describes a planning environment in which the inputs are both incomplete and disconnected. Improving one answer does not automatically teach the pension rules, clarify the real purchasing power of future income, or show how a longer horizon changes a household’s balance between work, saving, and consumption.
It also explains why the population question must remain separate from the personal question. The survey can show that a respondent’s estimate for similar people is far from the actuarial benchmark. It cannot, from that fact alone, tell us whether the respondent will revise their own expected age at death or change their retirement plan. The next difficulty lies in the bridge between those beliefs: whether people treat the lifespan of a reference group as evidence about themselves.
The evidence thus supports a restrained conclusion. Retirement planning is not just a matter of finding one correct life-expectancy number. It requires several kinds of knowledge to be accurate enough and connected enough to guide a choice. The IZA survey shows why an apparently modest average error can coexist with large individual mistakes, and why even a corrected population estimate may enter a system whose other parts remain weak. That leaves the most personal link still unresolved: how respondents move—or fail to move—from beliefs about similar others to beliefs about their own future.