Exploring Measures of Dispersion in Statistical Analysis

When researchers in population and family health studies look at numbers, the average often gets the spotlight. But two villages with the exact same average income, the same mean haemoglobin level, or the same mean age at marriage can still tell very different stories. That difference lies in how spread out the values are. This is where measures of dispersion step in. They reveal whether the data clusters tightly around the average or scatters across a wide range, and that distinction can change how policies, programmes, and clinical decisions are made.

What is dispersion in data? 🔗

Dispersion, also called variability or spread, describes how much individual values in a dataset differ from each other and from a central value like the mean or median. While measures of central tendency tell us about the typical observation, measures of dispersion describe how much variation exists around that centre. A dataset with low dispersion has values packed close together, while one with high dispersion has values stretched far apart.

Consider two districts where the average birth weight of newborns is 2.8 kg. In the first district, most babies weigh between 2.6 and 3.0 kg. In the second, weights range from 1.5 to 4.2 kg. The averages are identical, but the second district clearly has more low-birth-weight infants who need urgent attention. Without measuring dispersion, this critical public health insight would be invisible.

Why dispersion matters in health research 🔗

In population and family health studies, variability is not just noise. It is a signal. Researchers analysing data from the National Family Health Survey use standard deviation as a key quality-check indicator for anthropometric measures like child height-for-age and weight-for-height. When the standard deviation of these z-scores falls outside the expected range, it can point to measurement errors, interviewer bias, or genuine inequality across regions. Dispersion, in this sense, doubles as both a descriptive statistic and a diagnostic tool.

Key measures of dispersion 🔗

Statisticians have developed several ways to capture variability. Each has its own logic, strengths, and blind spots. The four most commonly used in health research are the range, percentiles (and the related quartile-based measures), mean deviation, and standard deviation.

Range 🔗

The range is the simplest measure of dispersion. It is calculated by subtracting the smallest value in a dataset from the largest. If the systolic blood pressure readings in a clinic sample run from 95 mmHg to 175 mmHg, the range is 80 mmHg. Range is easy to compute, but it relies only on the two outermost values, which means two datasets with the same range could still have drastically different distributions.

Range is most useful when researchers want a quick sense of how wide the data is, or when they are checking for extreme values. Its main weakness is sensitivity to outliers. A single unusually high or low value can stretch the range and give a misleading picture of variability.

Percentiles and quartiles 🔗

Percentiles divide a dataset into 100 equal parts after the values are arranged in order. The 50th percentile is the median, the 25th percentile (Q1) is the lower quartile, and the 75th percentile (Q3) is the upper quartile. The interquartile range (IQR), which is Q3 minus Q1, captures the spread of the middle 50% of the data and ignores the extremes.

This makes percentile-based measures especially useful for skewed data, which is common in public health. Income, household expenditure, age at first birth, and length of hospital stay rarely follow a neat bell curve. The IQR is best used when data are skewed or asymmetric, while the standard deviation is best suited for approximately symmetric data. Percentiles also drive growth charts. WHO child growth standards plot height and weight at the 3rd, 50th, and 97th percentiles, helping clinicians flag children who fall outside the expected band for their age.

Mean deviation 🔗

Mean deviation, sometimes called average absolute deviation, takes a more democratic approach. It calculates the absolute difference between each value and the mean (or median), then averages those differences. The formula in simple terms is the sum of |x − mean| divided by the number of observations. Unlike the range, mean deviation uses every value in the dataset. Unlike variance, it avoids squaring, which keeps the units intuitive.

Despite this elegance, mean deviation is used less often in advanced statistical work. The reason is mathematical: absolute values are awkward to handle in calculus and algebra, so analysts usually prefer measures based on squared differences. Mean deviation is defined as the arithmetic mean of the differences between values and their mean, and it remains a valuable teaching tool because it makes the idea of “average distance from the centre” easy to grasp.

Standard deviation 🔗

Standard deviation is the workhorse of statistical analysis. It is calculated by taking the squared difference between each value and the mean, averaging those squared differences (which gives the variance), and then taking the square root to return to the original units. The result is, roughly, the typical distance any observation lies from the mean.

A small standard deviation means the data is tightly clustered around the mean. A large one means it is widely scattered. A larger standard deviation indicates that the data values are very spread out, while a smaller standard deviation means the data are quite concentrated around the mean. Because it uses every data point and is mathematically tractable, standard deviation forms the basis for confidence intervals, hypothesis tests, regression models, and most of the inferential statistics used in epidemiology and demography.

Interpreting dispersion in real research 🔗

Calculating these numbers is only half the job. Interpretation is what turns statistics into evidence. When standard deviation is paired with the mean and the assumption of a roughly normal distribution, it unlocks the empirical rule: about 68% of observations fall within one standard deviation of the mean, 95% within two standard deviations, and 99.7% within three. This rule is the foundation of reference ranges in clinical medicine. When a laboratory reports that “normal” fasting blood glucose lies between 70 and 110 mg/dL, that band is often built from the mean plus or minus two standard deviations of a healthy reference population.

Spotting data quality problems 🔗

Dispersion also helps researchers judge whether their data can be trusted. International guidelines suggest that the standard deviation of height-for-age and weight-for-height z-scores should typically fall between 0.8 and 1.2 in high-quality data, and substantial deviations from this range usually indicate problems with measurement accuracy. When the standard deviation of these z-scores in NFHS-5 went up from earlier rounds, analysts read it as a sign of persistent measurement variability, not just a real change in child nutrition. This is a powerful illustration of how dispersion functions as a quality control mechanism in large surveys.

Comparing groups and exposing inequality 🔗

Two states might report the same average institutional delivery rate, but the district-level standard deviation can reveal whether progress is even or patchy. A state with a low standard deviation has spread its gains widely. One with a high standard deviation hides pockets of deprivation behind a respectable average. Studies of child undernutrition across India’s 640 districts have shown that examining the relationship between the mean and standard deviation is essential when defining areas of high burden, because small-area inequalities can be invisible at the state or national level.

Choosing the right measure 🔗

No single measure of dispersion suits every situation. The range gives a fast snapshot but is fragile against outliers. The IQR is robust for skewed data but ignores values outside the middle 50%. Mean deviation is intuitive but mathematically inconvenient. Standard deviation is statistically powerful but assumes a roughly symmetric distribution and is influenced by extreme values. Skilled researchers often report more than one measure to give a complete picture, especially when the underlying distribution is unclear.

Dispersion as the silent storyteller 🔗

Averages are useful, but they can flatten reality. Two communities, two clinics, or two states with the same mean indicator can have radically different lived experiences. Measures of dispersion bring those experiences back into view. They tell us whether progress is shared, whether measurements are trustworthy, and whether the “average” person in our dataset actually exists or is just a mathematical illusion. For anyone working with population and family health data, fluency in dispersion is not optional. It is the difference between describing a population and understanding it.

What do you think? If you were comparing two states with identical average child stunting rates, which measure of dispersion would you reach for first, and what would a very high value tell you about the data or the population? Have you ever come across a statistic where the average painted a very different picture from what was actually happening on the ground?

References

1. https://www.betterevaluation.org/methods-approaches/methods/measures-dispersion
2. https://pmc.ncbi.nlm.nih.gov/articles/PMC9576578/
3. https://libguides.lib.miamioh.edu/data_analysis/dispersion
4. https://bookdown.org/pkaldunn/Book/computing-the-variation.html
5. https://www.geeksforgeeks.org/maths/measures-of-dispersion/
6. https://libguides.leedsbeckett.ac.uk/skills-for-learning/maths-and-stats/statistics/measurements-of-dispersion
7. https://www.sciencing.com/relationship-between-standard-deviations-percentiles-8768703/
8. https://pmc.ncbi.nlm.nih.gov/articles/PMC12943685/
9. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7907088/