How to Calculate the Infant Mortality Rate

The infant mortality rate is calculated by dividing the number of deaths among infants under one year of age by the total number of live births in the same period, then multiplying by 1,000. The result is expressed as deaths per 1,000 live births. That formula sounds simple enough, but every element in it raises questions that affect the final number: what counts as a live birth, where the data come from, how deaths get classified, and what statistical adjustments are needed when the raw data are incomplete or inconsistent.

The Basic Formula

At its core, the infant mortality rate (IMR) uses one fraction: infant deaths divided by live births, multiplied by 1,000. If a country recorded 5,000 deaths of infants under age one and 500,000 live births during the same calendar year, the IMR would be 10 per 1,000 live births. The multiplication by 1,000 simply rescales the number so you are not dealing with tiny decimals. Most countries and international agencies report IMR on this per-1,000 scale, so any figure you see in a news headline or policy report is already in these units.

This formula is technically a rate in demographic convention, though statisticians sometimes call it a “probability of dying before age one.” For practical purposes, the distinction rarely matters unless you are comparing it with a true person-time rate used in epidemiology, where the denominator is person-years of exposure rather than a simple birth count. For nearly all public health work, the live-birth-based version is the standard.

What Counts as a Live Birth

The denominator matters just as much as the numerator, and it is less straightforward than it looks. The World Health Organization defines a live birth as any product of conception that shows signs of life at birth, regardless of birthweight or how long the pregnancy lasted. But countries apply that definition very differently. In Canada, England and Wales, and the United States, registration is required for all live births meeting the WHO definition. Other countries set minimum thresholds: the Czech Republic, France, and the Netherlands, for instance, impose some combination of gestational age, birthweight, or survival criteria before a birth must be registered.¹

These differences have a direct effect on cross-country comparisons. Very premature infants who are born alive but die within hours will be counted as both a live birth and an infant death in countries with definition-based registration. In countries with pragmatic thresholds, that same infant might never enter the system at all, lowering both the numerator and the denominator but disproportionately shrinking the death count. A population-based study of this problem found that rankings of countries shifted substantially depending on whether very small births were included. When crude neonatal mortality rates were used, the United States ranked 22nd; after excluding births under 1,000 grams, the US rank improved to 11th, while Canada jumped from 18th to 12th.² That shift is entirely an artifact of registration rules, not actual differences in care.

Neonatal and Postneonatal Components

Demographers routinely split the infant mortality rate into two sub-rates because the causes of death change dramatically across the first year. The neonatal mortality rate covers deaths in the first 28 days, calculated the same way: neonatal deaths divided by live births, times 1,000. The postneonatal mortality rate covers deaths from day 28 through day 364, but its denominator is typically the number of neonatal survivors rather than total live births. This distinction matters because an infant who died in the first week never entered the postneonatal risk pool.

These sub-rates reveal patterns that the overall IMR can obscure. In the United States, for example, data from linked birth and death records have shown that the neonatal mortality risk for Black infants was 1.6 times higher than for white infants, while the postneonatal mortality risk was 2.09 times higher.³ The postneonatal gap being wider suggests that the causes of later infant death, things like infections, injuries, and sleep-related deaths, are amplified by social and environmental factors beyond what drives mortality in the first few weeks.

Where the Numbers Come From

How a country gets its birth and death counts depends on the strength of its civil registration system. In high-income countries with comprehensive vital statistics, birth and death certificates are the primary data source. In the United States, the National Center for Health Statistics links each infant death certificate back to its birth certificate, creating a paired record that allows researchers to calculate birthweight-specific and gestational-age-specific mortality rates.⁴ This linkage process itself can introduce small errors: one study in Cleveland found that roughly 4.6% of deceased infants had unregistered births, and these tended to be very premature infants from socially disadvantaged backgrounds.⁵

In countries where civil registration is weak or incomplete, household surveys fill the gap. The Demographic and Health Surveys (DHS) program is the most widely used, collecting birth histories from women of reproductive age in dozens of low- and middle-income countries. In a “full birth history,” each woman reports the date of birth and, if applicable, the date of death for every child she has ever had. From those dates, analysts can directly calculate age-specific mortality rates. A “summary birth history” is simpler: a woman reports only how many children she has had and how many are still alive. That stripped-down version requires mathematical models to estimate the timing of deaths, since the data do not contain the exact age at death.⁶

A massive validation effort analyzed 243 DHS surveys from 76 countries, covering 8.5 million children, to develop improved methods for extracting mortality trends from summary birth histories.⁷ In settings where these surveys are the only source of data, even imperfect estimates are far more useful than no estimate at all. An assessment of a civil registration and vital statistics system in one Indonesian district demonstrated how strengthening local registration can complement survey data, producing district-level indicators like an IMR of 15 per 1,000 live births directly from administrative records.⁸

Why Raw Numbers Often Need Adjustment

Even when surveys collect birth histories, the data are not always clean. One well-documented problem is “age heaping,” where mothers round the age at death to convenient numbers. A child who died at 11 months might be reported as dying at 12 months, shifting a death from the infant category into the child mortality category and artificially lowering the IMR. Research into this pattern found that direct estimates from DHS birth histories are more robust than previously feared, but individual surveys still need case-by-case evaluation, especially when error margins sit near critical thresholds.⁹

A more serious problem is the misclassification of neonatal deaths as stillbirths. If a baby who briefly showed signs of life is recorded as stillborn, neither the numerator (infant deaths) nor the denominator (live births) registers the event, and the IMR drops without any real change in survival. A study in Malawi compared classifications from a household survey with those from a follow-up verbal and social autopsy and found that about one in five cases identified as neonatal deaths in the survey had been classified as stillbirths in the follow-up assessment.¹⁰ In low- and middle-income settings more broadly, research across 17 countries has confirmed that survey-based estimates are frequently affected by both underreporting and misclassification between stillbirths and early neonatal deaths, requiring statistical adjustments to produce reliable trend estimates.¹¹

Political and institutional factors can warp the numbers too. After the breakup of the Soviet Union, official vital registration statistics in Kyrgyzstan showed a continued decline in infant mortality. But independent analysis revealed that the decline had actually stalled, and that infants of Central Asian ethnicity and those born in rural areas were at considerably higher risk than official numbers suggested.¹² Incomplete reporting was the culprit, not improving health conditions.

How International Agencies Produce Country-Level Estimates

No single data source tells the whole story for most countries. The UN Inter-agency Group for Child Mortality Estimation (UN IGME) compiles every nationally representative data point it can find for a given country, adjusts for known biases, and then fits a statistical model to produce smoothed time series. The approach fits a country-specific regression model to one indicator, such as under-five mortality, and then uses the relationship between age-specific mortality rates to derive the IMR.¹³ The Institute for Health Metrics and Evaluation (IHME) uses a different modeling approach. The two groups have sometimes produced somewhat different estimates for the same country-year. A formal comparison found that the key differences lay in how each group chose smoothing parameters and handled uncertainty, not in the underlying data.¹⁴

In its 2013 update, UN IGME moved from a local regression smoother to a penalized B-spline regression model to better capture recent trends, particularly rapid declines. Uncertainty was assessed through a Bayesian framework that accounts for noise in the data and in the model parameters.¹⁵ For anyone using published IMR figures in policy documents or academic work, understanding which modeling group produced the estimate, and for which year, is important because the numbers can differ at the margin.

Breaking the Rate Down by Subgroups

A national IMR is an average, and averages can hide enormous disparities. One of the most powerful uses of the rate is disaggregation, calculating separate IMRs for different populations defined by geography, ethnicity, income, or maternal education. In the United States, linked birth and death data from 2007 to 2010 revealed that infants of college-educated Black women experienced 3.1 more deaths per 1,000 live births than infants of white women with only a high school degree. The gap was fully accounted for by shorter gestational lengths among Black mothers at every education level.¹⁶ That finding illustrates why calculating IMR separately for subgroups is not just a statistical exercise; it points directly at where interventions are needed.

Similar disaggregation work in Brazil used regression models to assess how maternal race and skin color related to child mortality after controlling for education, region of residence, and year of birth.¹⁷ In both cases, the overall national IMR gave a misleading picture of who was actually dying.

Mapping Mortality at the Local Level

National or even regional IMR figures can mask “hot spots” of high mortality that are small enough to vanish in aggregate statistics. Spatial modeling techniques address this by borrowing information from neighboring areas and adjacent time periods to stabilize estimates for areas with few births. A study spanning 122 DHS surveys across 262 subregions in 35 African countries used a Bayesian space-time model to produce more precise subnational estimates of under-five mortality that were consistent with each other in both space and time.¹⁸ Similar methods have combined survey data with demographic surveillance system data to produce “small area estimates” of child mortality in individual countries where vital statistics coverage is thin.¹⁹

For a district health officer or a non-governmental organization, these localized IMR estimates are far more actionable than a national figure. They can guide where to deploy community health workers, where to build neonatal care facilities, and where surveillance needs strengthening.

Capture-Recapture as a Correction Tool

When you suspect that deaths are being missed entirely, a technique borrowed from wildlife ecology can help. Capture-recapture compares two or more independent lists of deaths, each of which catches some cases the other missed, to estimate how many deaths neither list caught. A study in the Philippines applied capture-recapture to civil registration data and found that infant mortality was underreported by 62%.²⁰ In India, the same approach was used in Rajasthan to estimate institutional perinatal mortality at about 36 per 1,000 births, revealing substantial under-reporting by the government’s passive reporting system.²¹

Capture-recapture has also been used for cause-specific infant mortality estimation. In Korea, three independent data sources were compared to estimate the total number of sudden infant death syndrome (SIDS) cases, with a log-linear model suggesting that the true count of SIDS deaths was roughly three times what any single source had recorded.²² The technique requires that the data sources be reasonably independent of each other; if the same reporting gap affects all lists, the correction will still undercount.

Identifying Causes of Death Where Doctors Are Scarce

Knowing how many infants died is not enough for designing interventions; you also need to know why they died. In settings without routine medical certification, verbal autopsy is the primary tool. A trained interviewer visits the family of the deceased and asks structured questions about the symptoms and circumstances preceding death. Algorithms then assign a probable cause of death. In Mozambique, researchers compared verbal autopsy diagnoses with tissue sampling results and found meaningful misclassification across all the standard verbal autopsy algorithms, which required a calibration step to produce reliable cause-specific mortality fractions.²³

In India, the Sample Registration System uses its own verbal autopsy tool, with two independent physicians assigning the cause of death and a senior physician adjudicating disagreements.²⁴ These cause-specific rates are calculated the same way as the overall IMR, just restricted to deaths from a single cause. If 200 out of 5,000 infant deaths in a region were attributed to pneumonia and there were 500,000 live births, the pneumonia-specific infant mortality rate would be 0.4 per 1,000 live births. Those cause-specific figures are what health planners use to decide between investing in vaccines, clean water, or neonatal intensive care.

Handling Uncertainty and Confidence Intervals

An IMR calculated from a complete national vital registration system is close to a census figure, with minimal statistical uncertainty. But when the number comes from a household survey that sampled only a fraction of the population, it carries sampling error. The standard approach is to calculate a confidence interval that reflects the survey’s design, including its clustering and stratification. Software tools like the DHS.rates R package automate this, producing not just the point estimate of mortality but also the standard error, design effect, and confidence interval.²⁵

In humanitarian emergencies, surveys are often conducted under time pressure with complex sampling designs. A review of confidence interval methods for mortality rates in those settings recommended using a specific “ratio command” approach available in standard statistical software, rather than simpler approximations that can understate the true uncertainty.²⁶ If you see an IMR figure from a crisis zone presented without a confidence interval, treat it as a rough estimate rather than a precise measurement.

Tracking Progress Against Global Targets

IMR is one of the indicators used to monitor Sustainable Development Goal 3, which calls for all countries to reduce neonatal mortality to no more than 12 per 1,000 live births and under-five mortality to no more than 25 per 1,000 live births by 2030. IMR does not have its own explicit SDG target, but it sits between the neonatal and under-five rates and is tracked alongside them. A review of the WHO South-East Asia Region found that while the region as a whole has made substantial progress, the pace varies enormously between countries.²⁷

Ghana illustrates both the progress and the gap. Modeling of historical IMR data from 1960 to 2022, with projections to 2030, showed a consistent downward trend, from about 31 per 1,000 live births in 2023 to an estimated 25 by 2030. The researchers concluded that while the decline is real, the current pace is not fast enough to meet the most ambitious targets, calling for stronger equity-focused health strategies.²⁸ The ability to calculate, project, and disaggregate IMR is ultimately what turns an abstract number into a planning tool. Every methodological refinement described above exists because policymakers need IMR estimates they can trust well enough to stake budgets on.