The Machinery of Invisibility: Why Older People Go Uncounted, and What Measurement Science Can Do About It

When WHO published its report on unmet health and social care needs in the Eastern Mediterranean Region, its first recommendation was to build robust national data systems that include older populations. In the previous blog, Unseen, Unheard: The Unmet Needs of Older People in the Eastern Mediterranean, Mohamed Ismail and Dr. Shereen Hussein described how many older people in the region have care needs that go unmet. In many places nobody is yet measuring those needs at all. I want to pick up that first recommendation and ask a practical question. What does it take to measure older people’s needs well, and what can measurement science offer while stronger data systems are being built?

It’s easy to think of measurement as a technical step that happens before the real work of policy and care begins. In practice, it decides who gets counted. Surveys, records, and statistical models all have blind spots, and those blind spots tend to fall on the people with the greatest need. Often that means the very old and frail, people living in care homes, and people in areas affected by conflict. If we can see where the blind spots come from, we have a much better chance of closing them.

The Layers of Invisibility 

I find it helpful to think of being uncounted as three separate problems rather than one.

The first is never being measured at all. Someone can be missed because surveys don’t sample their household, because the care home they live in is excluded, or because their region is too insecure for fieldwork. They may also be too frail to answer. If no survey reaches them, they’re simply absent from the data and no amount of analysis can bring back someone who was never in the sample.

The second is being measured poorly. A survey exists, but its design misses people who are hard to reach. Think of someone who is housebound or lives in an institution, someone without a phone, or someone who just doesn’t respond. People who don’t respond are rarely a random slice of the population. They tend to be poorer, sicker, older and more isolated. When the hardest to reach are also the people with the highest need, a survey can look complete and still be systematically incorrect about the people who matter most.

The third is biased analysis. Even with good data, a model or measure can work less well for some groups than for others. It can end up reporting lower need for exactly the populations that were under-represented to begin with.

Separating the layers matters because each one needs a different fix. You can’t analyse your way out of a sampling gap, and you can’t sample your way out of an analytic bias.

A Response for Each Layer

The report is clear that investment and political commitment are needed. Alongside them, measurement science has something specific to offer at each layer.

For people a survey never reaches, the answer lies mostly in design. Sampling frames can be built to deliberately include groups that are usually left out, such as older people in care homes or confined to their own homes. Groups known to be hard to reach can be oversampled. Routine and administrative records can help find people that household surveys miss. This is the hardest layer to fix because the work has to happen before any data are collected. Once someone falls outside the frame, nothing later in the process will recover them.

For people measured poorly, there’s a well established toolkit. Survey weighting and post-stratification go some way toward correcting for who is over or under represented. Non-response adjustment deals with the systematic absence of those who didn’t answer. Validation studies compare what people report with what records show, which tells us how much a given instrument misses and for whom. These are the everyday methods of survey science. They matter a great deal for unmet need, though. Unmet need is usually measured by asking people whether they went without care they needed, and the answer can shift depending on who responds and how the question is worded.

Sometimes the survey data exist, but there are too few responses to rely on. A district might have only a handful of respondents, or a subgroup might be too small to estimate. This is where small-area estimation helps. Instead of treating each area in isolation, these models borrow strength from elsewhere. They combine the limited survey data with information that’s available everywhere, such as age structure and deprivation. They also rely on the reasonable assumption that similar places tend to resemble each other. The result is an estimate for every area, including places the survey barely touched, with an honest measure of uncertainty attached. In the United States, the CDC’s PLACES project uses this approach to publish health estimates for neighbourhoods across the country, even though no survey samples every neighbourhood directly. Similar methods are now being adapted for lower-resource settings. They use data from large international household surveys and are packaged so that analysts without specialist programming skills can use them.

Biased analysis is the layer most often linked with artificial intelligence, and there’s a growing body of fairness methods to address it. These methods ask a simple question. Is the model just as accurate for the oldest, frailest and most disadvantaged people as it is for everyone else? If not, it can be retrained or adjusted to close the gap. This becomes more important as predictive models are used to decide where attention and resources go. A model learns from whatever data it’s given. If those data already under-represent the hardest to reach, the model can make them even less visible while looking entirely objective. 

One caution applies to all of this. Every one of these methods produces estimates rather than certainties, and each is only as good as the data underneath it. Measurement science can extend what we see, but it can’t conjure data where none exist. Technical fixes can’t reach the people who were never measured at all, which is why design and investment have to come first.

What is Already Being Tried in the Region

The WHO report found that evidence on unmet need among older people in the Eastern Mediterranean Region is limited. Only three settings (Egypt, Lebanon and the occupied Palestinian territory) were able to contribute additional evidence of ongoing work. The published estimates that do exist vary widely and are rarely comparable from one country to the next. Two new national cohorts will help, A Longitudinal Study of Egyptian Healthy Aging (AL-SEHA) and the Lebanon Study on Aging and Health (LSAHA). For now, each speaks only for its own country, and both will take years to mature.

Rather than waiting for better surveys, the report called for alternative ways of estimating need where direct data are missing. That call is already being acted on. Researchers in the MENARAH network have published a Long-Term Care Needs Index that estimates care dependency across data-scarce MENA countries. It draws on information that exists almost everywhere, namely population age structure, disability prevalence and the burden of major non-communicable diseases. From these it produces a comparable figure for each country. It shows that useful estimates can be built even where dedicated surveys don’t exist.

Just as important, the Index gives countries a common yardstick, which the region has lacked. The Index works at the national level. A natural progression would be methods that also look within countries, telling urban areas apart from rural ones and richer regions from poorer ones. Those methods would put an honest range around each estimate, so decision makers know how far to trust the number. Small-area estimation and hierarchical models are well suited to that work, and they would build directly on the foundation the Index has laid.

Measurement is necessary, but is it sufficient?

Better measurement is only one part of what the report calls for. Political commitment matters too, and so do investment and the decision to treat the care of older people as a public responsibility. Each depends on the others. Money spent without good measurement can’t be aimed at the places where need is greatest. Measurement without investment just produces detailed maps of needs that nobody has the resources to meet.

Whether older people are counted comes down to a series of choices about who we sample, how we measure, which models we trust and what we decide is worth making visible. None of those choices is fixed. Each one is a chance to count people who have been missed.