Skip to content
Back to search results

Indirect Standardization

Introduced by
Classical demographic method (formalized by Preston, Heuveline & Guillot)
Year
2001
Type
Rate adjustment using a standard schedule of group-specific rates
Family
Standardization & decomposition
Font size

What does it do?

Indirect standardization is a demographic technique for comparing summary rates when a study population's own group-specific rates are too sparse to be reliable.

Instead of reweighting the study population's rates, it applies a trusted standard schedule of group-specific rates to the study population's own structure to compute the number of events that would be expected. The ratio of observed to expected events — the standardized mortality ratio (SMR) — measures how the study population's risk compares with the standard, adjusted for its composition.

The intuition behind the method

Imagine a small town with only a handful of deaths each year. Its age-specific death rates bounce around wildly because each age band has so few people, so direct standardization would be unstable. Indirect standardization sidesteps this: it borrows a well-measured set of age-specific rates from a large reference population and asks, 'If our town's people died at those standard rates, how many deaths would we expect?' Comparing the deaths we actually observed with that expected number tells us, in a single robust ratio, whether the town is dying faster or slower than the standard predicts — even though the town's own rates were too noisy to trust.

History

Indirect standardization arose in 19th- and early-20th-century vital statistics and life-insurance practice, where analysts needed to compare mortality in small districts, occupations, or insured subgroups whose own age-specific rates were too sparse to trust. The standardized mortality ratio became the workhorse index of occupational epidemiology, used to flag elevated mortality in specific trades. Alongside the direct method, it was codified as a core demographic technique in the standard textbook treatment by Preston, Heuveline, and Guillot (2001), which frames both methods as complementary responses to the confounding of composition.

When to use it

Use indirect standardization when the study population is small or events are rare, so its own group-specific rates would be unstable, but you still need a composition-adjusted comparison against a reference. It is the standard tool for occupational and small-area mortality studies, where the standardized mortality ratio is the usual output. Assumptions: a credible external standard schedule of group-specific rates exists, the total observed event count is known, and the study population's group structure is available. Do NOT use indirect standardization to compare two study populations with one another unless they share very similar age structures — SMRs computed against a common standard are only strictly comparable to the standard, not freely to each other, because each SMR uses a different (its own) composition. When the study population's own rates are reliable, direct standardization is preferable.

Limitations and considerations

  • SMRs from different study populations are not strictly comparable to one another, since each uses its own (different) composition as the weighting.
  • Results depend on the choice of standard rate schedule; an inappropriate or outdated standard biases the expected count and the SMR.
  • It adjusts for composition only along the standardized dimension, leaving residual confounding from other factors.
  • It assumes the standard's age pattern of risk is a reasonable proxy for the study population; strong effect modification by age violates this.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Robust when events are rare or the study population is small, because it relies on the total observed count rather than noisy group-specific rates.
  • Requires less detailed data: only the study population's group structure and total events, plus an external standard rate schedule.
  • Yields the standardized mortality ratio, an intuitive single index of relative risk versus the standard, with a simple Poisson-based variance.
  • Widely used and well understood in occupational epidemiology and small-area analysis, with established reporting conventions.

Applications

  • Occupational epidemiology, where SMRs compare mortality among workers in an industry against the general population.
  • Small-area public-health surveillance, computing SMRs for towns or districts with too few events for stable direct rates.
  • Insurance and actuarial work, assessing whether an insured cohort's mortality exceeds a standard table.
  • Hospital and clinical studies, comparing observed deaths in a patient cohort against expected deaths from population rates.
  • Cohort studies of exposed groups, summarizing relative mortality risk versus a reference population in a single ratio.

How it works

How it works

1

Start by obtaining a standard schedule of group-specific rates rᵢˢ — for example age-specific death rates from a large national or reference population, computed as standard events Dᵢˢ over standard population Nᵢˢ in each group. Because the standard comes from a large population, these rates are stable, which is exactly what the sparse study population lacks. The same standard schedule is used for every study population being compared.

Formula
r_i^{s} = \frac{D_i^{s}}{N_i^{s}}
2

Apply the standard rates to the study population's own group sizes Nᵢ and sum across groups to obtain the expected number of events E. This is the number of events the study population would experience if its members were subject to the standard's risks while keeping the study population's actual composition. Crucially, this step needs only the study population's group structure and the standard rates, not the study population's own (unreliable) group-specific rates.

Formula
E = \sum_i r_i^{s}\, N_i
3

Form the standardized mortality ratio as the total observed events O divided by the expected events E. An SMR of 1 (or 100 when expressed as a percentage) means the study population experiences exactly the events predicted by the standard; values above 1 indicate higher-than-standard risk and below 1 indicate lower. The SMR depends only on the total observed count, so it remains stable even when individual group counts are tiny.

Formula
\text{SMR} = \frac{O}{E} = \frac{\sum_i D_i}{\sum_i r_i^{s} N_i}
4

Multiplying the SMR by the crude rate of the standard population Rˢ converts the ratio into an indirectly standardized rate on the original scale. An approximate variance for the SMR follows from treating the observed count as Poisson, giving Var(SMR) ≈ O / E², from which confidence intervals are constructed. This lets analysts test whether the study population's risk differs significantly from the standard.

Formula
\text{ISR} = \text{SMR}\times R^{s},\qquad \widehat{\mathrm{Var}}(\text{SMR}) \approx \frac{O}{E^2}

Common pitfalls

  • Directly comparing two SMRs as though they were on a common scale, when each is anchored to its own population structure.
  • Choosing a standard schedule that does not resemble the study population's risk pattern, distorting expected counts.
  • Reporting an SMR without its confidence interval, hiding the substantial uncertainty when observed counts are small.
  • Confusing the indirectly standardized rate (SMR × standard crude rate) with a directly standardized rate; the two methods answer different questions.

Frequently asked questions

Why use indirect standardization instead of the direct method?

Because the direct method needs the study population's own group-specific rates, which become unstable when group counts are small or events are rare. Indirect standardization avoids this by borrowing stable rates from a large standard population and relying only on the study population's total observed events, which is far less noisy. It is therefore the method of choice for small populations and rare outcomes.

What exactly does an SMR of 120 mean?

An SMR of 120 (or 1.20) means the study population experienced 20 percent more events than would be expected if it had been subject to the standard population's group-specific rates, given its own composition. An SMR of 100 means observed equals expected, and below 100 means fewer events than expected. It is a composition-adjusted measure of relative risk against the chosen standard.

Can I compare SMRs between two different occupational groups?

Only with caution. Each SMR is weighted by its own group's age structure, so two SMRs computed against the same standard are not on a strictly common scale unless the two groups have very similar structures. For rigorous comparison of two study populations with one another, direct standardization to a common standard is the cleaner approach when the data permit it.

Related methods

Sources

  1. Preston, S. H., Heuveline, P., & Guillot, M. (2001). Demography: Measuring and Modeling Population Processes. Blackwell.

    ISBN: 9781557864512
    Copied.