Measuring health outcomes: A new approach

A- A A+

Measuring people’s health status is vital for designing effective policies that support different age groups. But economists working in this field face a difficult challenge when it comes to integrating accurate health measures into their economic models. This then makes it difficult for policy-makers to make straightforward model-based decisions. This column offers a new metric and seeks to solve the issue of measuring health statuses and their associated economic outcomes. The new approach makes integrating important health data into economic models easier, helping yield more accurate results.

Household surveys offer extensive data on various aspects of people’s health. This is particularly important when designing policies relating to older people, with surveys such as HRS, SHARE and ELSA used extensively by researchers to explore a range of key questions. But economists often face the challenge of selecting a single health measure that summarises the wealth of available information. There is a difficult trade-off to be made between granularity and ease of use.

A catch-all measure is useful because it means researchers can incorporate health as a key state variable in their structural models. But using a singular metric runs the risk of dimensionality problems – if researchers try to incorporate numerous health variables, the model takes too long to be solved numerically. So, a parsimonious measure (the simplest metric with the least assumptions but with the greatest explanatory power) is useful because it allows researchers to solve the model fast, without losing clarity.

In this column, we present a novel data-driven approach that addresses these challenges and provides a parsimonious health measure. By leveraging panel-data information on Activities of Daily Living (ADLs) and Instrumental Activities of Daily Living (IADLs), we are able to develop a new metric that overcomes and summarises health measures observed in the data. We also explore how this approach can benefit economists by tackling the curse of dimensionality and providing insights into the economic consequences of ageing and health disparities.


How does the econometric model help us estimate health status?

Our approach involves estimating something called a ‘dynamic latent variable model’. This approach uses unobserved (or latent) variables to explain the relationships between a larger set of observed variables. By using this method, we can identify individuals’ health groups based on reported difficulties with ADLs and IADLs. 

We then determine a finite number of latent health groups, each characterised by specific health limitations. In our analysis, we find that four groups effectively capture all individuals’ health statuses: the healthy, the physically frail, the mentally frail, and the impaired. These groups differentiate individuals based on the presence or absence of physical and cognitive limitations, providing a comprehensive representation of health status across the population.


How might this new model help economists?

For economists studying the economic implications of ageing and health-related policies, incorporating a rich health measure is crucial. Our data-driven approach offers a more nuanced understanding of health status while addressing the curse of dimensionality. 

By focusing on the most informative ADLs and IADLs, economists can reduce the dimension of the state space of their models without sacrificing the richness of health representation. This not only improves the accuracy of health classifications but also enables more precise analyses of how health influences survival, savings, insurance choice, as well as other economic outcomes.

Our results are consistent with previous studies and reaffirm well-documented findings regarding aging and health. As people get older, health tends to deteriorate, and the probability of mortality increases. We also observe the well-known gender disparity in life expectancy, with females generally outliving males. Educational attainment plays a significant role in health outcomes, with individuals with higher education levels more likely to belong to healthier groups. These insights emphasise the importance of considering age, gender, and education within economic models – especially those that aim to understand health disparities and develop targeted interventions.


What does the new model reveal about healthcare?

People’s needs also vary. Our new model sheds light on the differing ways that different groups use healthcare services and long-term care. Individuals in the impaired group tend to have higher out-of-pocket medical expenditures, while mentally frail people have a greater likelihood of residing in nursing homes. By incorporating these findings into economic models, policy-makers can gain a better understanding of the economic consequences of such health statuses and develop effective strategies to allocate resources and improve access to healthcare services for different health groups. As with most policies, there is no one-size-fits-all solution for healthcare. Understanding the needs of each sub-group is a crucial starting point.


How accurate is the model? How useful is it?

A model’s usefulness is bound tightly to its accuracy. To assess the predictive power of our design, we compare it with commonly used health classifications found in past research. Our approach generates more differentiated health groups, leading to higher explanatory and predictive power for health-related spending variables. Compared with self-reported health or frailty indices, our measure explains a larger proportion of the variance in medical expenses, nursing home residency, and care receipts. Our measure also outperforms alternative classifications in predicting mortality, enhancing its relevance for economic analysis. 

This then has implications for economists designing larger ‘life-cycle’ models. As part of our research, we explore the implications of our health classification in a life-cycle model by solving an existing model used in previous studies. The results indicate substantial differences compared with previous predictions. Unhealthy individuals, as classified by our approach, are found to spend more than they earn (known as ‘dissaving’) during retirement. This is reflected by the tighter correlation between health and medical expenses, and the differential survival probabilities across health groups. Differential health and economic outcomes are found to interact, and this is seen clearly during retirement.



Our data-driven approach provides economists with a powerful new tool. Those working with structural models can now incorporate a rich health measure that addresses the curse of dimensionality. By leveraging panel data information on ADLs and IADLs, we generate a new health measure that captures the complexity of health status. 

This approach enhances our understanding of ageing and health disparities, as well as their economic implications. It also enables more accurate modeling of survival, savings, insurance choices, and healthcare, leading to improved policy interventions and targeted strategies for people with varying health profiles. Without good data and good models, it is very difficult for policy-makers to make evidence-based decisions. Our new approach is a step in the right direction.


Author: Jesus Bueren