RISK

For demographers, the risk of a demographically significant event–such as birth, death, the onset of illness, marriage, migration, or labor force entry or exit–is the probability that the event will occur. Demographically significant events define entry or exit from demographically significant conditions, such as life, death, residence in a politically defined region, various marital statuses, employment, and school enrollment.

The demographic definition of risk ignores the desirability and impact of risked events. For example, sexually-active women of childbearing age are "at risk" of pregnancy, but the demographer's calculation of that risk does not consider if women regard pregnancy with delight or dread. Nor does the demographer's risk evaluation consider differences between pregnancy that is unwanted due to minor timing inconveniences and pregnancy that is unwanted because it would precipitate the mother's death. In common language, negative consequences of events are losses and positive consequences are benefits. The demographic approach is technical. The technical analysis is sometimes simplified by calling the consequence of a risked event a loss; a benefit then is a negative loss.

Demographic Rates

Individuals are at risk of an event if and only if their risk exceeds zero. A demographic rate is a time-related measure of exposure to risk. The rate is measured by the number of occurrences of an event per at-risk person per unit time. If events are non-recurring (e.g., death), and the time interval for the measurement is one unit (e.g., a year), then the rate is the proportion of at-risk persons who experience the event per time period. In demography, rates at which an event occurs are distinguished from proportions of population segments who experience the event. The denominator of the proportion, but not the denominator of the rate, may include persons who are not at risk (e.g., men are not at risk of giving birth).

Rates are used to calculate or estimate important time-related measures, including the extent to which members of a population who enter a demographically significant state remain in it over time, the probabilities that an individual who enters that state will remain in it for various numbers of consecutive time periods, and the expected or mean future time remaining in a state for persons who already have been in the state a particular length of time. These measures include the so-called life-table quantities: age-specific death rates, age-specific expected length of remaining life, and proportion of the population surviving at each specific age.

Age-specific rates for a population are often applied to hypothetical or standard age distributions to compute standardized or adjusted rates, life expectancies, and other quantities for the entire population. Alternatively, hypothetical or standard rates are applied to the observed age distribution of a population to produce adjusted rates and expectancies for population aggregates.

Methods of Analysis

Risks can be simple or competing. For example, employed persons are at risk of job loss from mortality, retirement, layoff, mandatory military service, incarceration, and voluntary job termination; employed persons who leave their jobs by dying cannot also leave by retirement, layoff, or any other means. Competing risks are used to produce multiple decrement life tables in which members of a population can exit a demographic condition via several specified, mutually exclusive routes (e.g., one can exit the civilian non-institutionalized population by mortality, emigration, or institutionalization). Demographic risk analysis often focuses on socioeconomic differentials in exposure to risk of death and other demographically significant events, implicitly examining the effects on mortality of socioeconomic factors such as schooling, occupation, and race.

Because of practical limitations on the size of available datasets, empirical analysis of many socioeconomic differentials in risk requires multivariate statistical methods. Methods such as logit and probit analysis can be applied in some situations involving a risked event that can occur only once. Poisson regression methods are useful in those situations when the event can occur more than once. Multinomial logit and multinomial probit methods are useful in those situations when there are competing risks, only one of which can occur, and only once. For data that gives the duration of spells (uninterrupted periods spent in a demographic state of interest) various types of survival analysis methods are useful, including those based on exponential, Weibull, lognormal, and loglogistic distributions. Cox's proportional hazard method is frequently useful. Appropriate methods also appear in the literature on event history analysis.

Risk and Loss

Effective design of government policy and business strategy often requires prognostications of (a) future demographic risks, rates and proportions, and (b) the exposure to losses (i.e., costs and benefits) that would be associated with these risks, rates, and proportions, if they occurred as projected. The future or past size and age distribution of a population in a demographically significant condition can be projected or estimated by application of a set of age-specific survival rates to the current age-distribution of that population. In practice, all estimates and projections necessarily are based on a combination of information and conjecture about past, current, and sometimes future risks and other factors. Data limitations and methodological disputes add uncertainty. A common but incomplete response to this uncertainty is to make demographic projections in sets, each element of which is based on different assumptions about unknown information. But demography offers no standard procedures for choosing among the members of a set of projections, and the choice is inescapably subject to dispute. Production of a set of projections saves the demographer from the need to defend intrinsically-subjective speculation about the unknown, and it pushes disputes about demographic projections outside of demography.

The loss distribution. If it is possible to evaluate the losses associated with demographic events, then it is possible and often useful to evaluate the general level of exposure to loss from a set of risks, or from different subsets of those risks. Common descriptive statistics in addition to the mean and variance are informative but not routinely used. The expected loss is the first moment of the loss distribution, otherwise known as its mean or expectation. If outcomes x are continuously differentiable and occur with probability Pr(x) and loss L), then the expected loss, E), is given by E(L) = ∫x)Pr(x)dx. If outcomes are discrete, then E(L) = Σ_i L(x)Pr(x_i). The variance of) describes the accuracy with which loss can be anticipated without additional predictive information. The higher the variance, the less informative is the mean about the loss that one is likely to experience. The worst case loss is the maximum of the loss distribution.

In the absence of concrete knowledge about the future, insurance provides a defense against disruptively large losses and, more generally, a hedge against variance in the distribution of losses. Insurance permits individuals to experience some present loss with certainty (in the form of payment of premiums) in exchange for protection against uncertain future losses that exceed a threshold (the insurance deductible). Insurance commonly is available for only some risked events; for those that cannot be insured, the analysis of risk and loss exposure, and planning on the basis of that analysis, is particularly useful.

Valuing losses. Demography itself is seldom, if ever, informative about how to compare different types of losses. Comparison of dissimilar losses requires a theory of value, or at least some principles about how to compare dissimilar demographic states and the events, such as birth, death, employment, and migration that cause them to change. For example, how is one to compare the losses associated with 100 deaths from workplace injuries to job loss by 60,000 employed persons? Numerous and conflicting economic, legal, aesthetic, emotional, political, religious, and other analyses of value exist. Thus, disputes are endemic to considerations of the losses associated with demographic projections. Policies are often evaluated on their actual or projected effects on mortality and other demographically significant events. These disputes are especially severe when they concern social policies that involve tradeoffs between risks of different types, such as increased unemployment risk and increased mortality risk.

Conflicts also often focus on risk (probability) estimation and worst case analysis. The worst imaginable event in any situation is likely to be the demographic tragedy of massive loss of human life. Imaginable events are not necessarily possible. Because the demographic framework examines risk only for those who are at risk, the first question is whether or not the risk of the worst imaginable event is zero or so close to zero that it should be treated as such. If this risk is distinguishable from zero, then this loss is the worst case loss. But if this risk is not distinguishable from zero, then this loss passes out of consideration. Heated debate over the risk of the worst imaginable event has been a prominent feature of public policy discussion concerning nuclear power, genetically modified plants and animals, environmental pollution, workplace safety, and other matters.

Expert Versus Popular Views of Risk

Much of the disagreement between experts and the lay public appears to stem from, or to be exacerbated by, the following:

Differences in probability estimation. Lacking technical training and often distrustful of expert pronouncements, substantial proportions of the lay public seem to prefer their own subjective estimates of risk probabilities to the data-based estimates of technical experts. A substantial segment of the population appears to lack intuitive understanding of very small decimal fractions, with consequent difficulty understanding the frequency of occurrence of low-probability events.

Differences in valuation of risked events. Experts tend to focus on quantitative loss measures and tend to use generally accepted estimation methods. In contrast, large segments of the general public rely on subjective evaluations that are quite dissimilar to expert evaluations.

Differences in attention. There are differences in attention given to the worst imaginable loss versus the average, expected, or most-likely loss. Attentive to the accuracy of their predictions over the long run, technical experts tend to give the greatest weight to scenarios that are most likely, and no weight to scenarios that have no probability of occurrence. Substantial segments of the general public focus on the worst imaginable case, perhaps because it inspires the greatest emotional response.

Differences in the conceptualization of losses. At their best, risk experts apply methods that let them make finely-graded comparisons of the losses associated with the occurrence of a risked event to the losses associated with its nonoccurrence. For example, technologies periodically fail disastrously, and disasters take lives (e.g., airplanes crash, bridges collapse, and physician errors kill patients). A simple and popular measure of the impact of technology failure is the number of lives lost from it. But technologies that fail periodically also can prolong and improve the quality of lives. At a minimum, one should compare the number of lives lost from a technology failure to the number of lives that would have been lost if the technology had not been deployed at all. And since everyone dies eventually, regardless of what technology is or is not deployed, the relevant measure is even better approximated by the number of person-years of life lost by the failure of a technology, compared to the number of person-years of life that would be lost by not deploying that same technology. Technical experts can apply life tables or analogous methods to calculate the loss of person-years of remaining life. Analysis and comparison of age-specific death rates (rather than numbers of deaths) is yet more complicated and directs attention to the societal rather than the individual consequences of deadly events. Quality-of-life issues are important too.

Differences in considerations of "spillover" effects. Experts tend to confine their analyses to variables that they can measure; substantial segments of the general public appear to consider the consequences of a risked event on their entire way of life. Losing a job can be seen as a simple loss of income, or it can be seen as the unraveling of everything supported by that income in the family of the employed person.

Differences in treatment of losses associated with unfamiliar risks. Substantial portions of the general public appear to respond to danger from an unfamiliar event (e.g., anthrax infection by contaminated mail, real or imagined illness from radiation-sterilized food) by increasing their estimate of the risk (probability) of experiencing the event, increasing their estimate of the loss that would result from the experience, or both, sometimes with anxiety, hostility, and the growth of social movements and collective action added.

Differences between technical and lay approaches to risk and loss exposure lead to questions about when it is useful to apply the technical analysis to public policy debates, and how to present it to those who combine high emotional interest in the subject with low exposure to the technical issues. Answers to these questions are external to demography, and they rest on judgments and strategic decisions about what is worth studying in detail, and what social choice inferences should be emphasized, stated, or left implicit. Widely-felt emotions and subjective impressions are social facts that cannot be ignored, but they are poor tools for analysis of risk and loss exposure. Those who claim that risk and loss exposure are equivalent to the general public's perception of them risk seriously flawed results.

Equally unstable analyses may result from the so-called rival rationalities view that conceives of experts as focused narrowly on statistical analysis of that which can be quantified, and an equally rational general public focused on a wide range of qualitative aspects of risk, including voluntariness and fairness of risk and loss exposure, and the dread with which a possible loss is perceived. The rival rationality view of risk assessment is not subject to any requirement for empirical evidence on risk magnitudes. This method is likely to be particularly troublesome when the risks of advanced technologies are considered. When science is misunderstood, as it often is, then popular misconceptions can and do lead to perceptions of imagined risks involving horrible but imaginary future losses. Finally, there appears to be confusion regarding the perceptions that are the basis for lay assessments of risk and loss exposure: It has been argued that anxiety about a risked event makes exposure to that event seem less voluntary, thereby raising the perceived risk of the event and exposure to loss from it, regardless of any actual difference in risk or exposure.

Conclusion

In summary, demography offers a particular conceptual and methodological framework for the measurement and analysis of risk; for predicting, forecasting, and comparing risks in different times and places; and for understanding how a given risk structure affects a population. The demographic approach to risk emphasizes the explicit connection between the structure of risk experienced by a population and the structure of age–the time spent in a demographically significant condition–that the population develops over time.

Demographic techniques emphasize the proper technical calculation of demographic risk measures. Demographic methods permit and even encourage analysis based on hypothetical values of demographic risks. These calculations are an important step in the analysis of exposure to loss. But demography offers no guidance about how to value the losses (and benefits) associated with the occurrence of risked events. Thus, demographic analysis of loss exposure requires combining demographic risk calculation with loss evaluations provided by other disciplines, engendering all the difficulties described above.

RISK

Demographic Rates

Methods of Analysis

Risk and Loss

Expert Versus Popular Views of Risk

Conclusion

BIBLIOGRAPHY