Understanding Variability in Data: Range, Standard Deviation, and More

Two classrooms can have the exact same average marks and still be completely different places. In one, almost every student scores close to the average. In the other, half the class tops the paper while the other half struggles to pass. The mean hides this difference entirely. To see it, you need a second set of tools: measures of variability. These tell you how spread out your data really is, and in fields like informetrics and scientometrics, where citation counts and publication outputs are anything but evenly distributed, they often matter more than the average itself.

Why variability matters in data analysis 🔗

Measures of central tendency such as the mean, median, and mode tell you where the centre of your data lies. But they say nothing about how the values scatter around that centre. Variability describes how far apart data points lie from each other and from the centre of a distribution, and it is also called spread, scatter, or dispersion. Without it, your picture of the data is half-complete.

Consider research evaluation. If two departments both average 12 citations per paper, that single number suggests they perform identically. Yet one department might have steady, consistent output while the other rides on one or two spectacularly cited papers. Variability is what separates a reliable, predictable dataset from a volatile one, and it is central to judging the consistency of any measurement.

Range: the simplest measure of variability 🔗

The range is the most basic measure of spread. You calculate it by subtracting the smallest value in your dataset from the largest. If a journal’s papers receive between 0 and 240 citations, the range is 240. It is quick, intuitive, and easy to explain to anyone.

The problem is that the range relies on only two values, the maximum and the minimum, and ignores everything in between. This makes it extremely sensitive to outliers. A single highly cited “blockbuster” paper can inflate the range dramatically, even if every other paper received roughly the same number of citations. The range also tends to grow with sample size, since larger datasets are more likely to contain an extreme value.

That said, the range is not useless. In quality-control settings where samples of a fixed size are collected regularly, an unusually large range on a given day can act as a signal to investigate that day’s unusual variability. For a first, rough sense of spread, it does its job. For anything more serious, you need a measure that accounts for every data point.

Standard deviation and variance: measuring spread around the mean 🔗

Variance and standard deviation solve the range’s biggest weakness by using all the data, not just the extremes. Both work by measuring how far each value sits from the mean.

Variance 🔗

Variance is the average of the squared distances between each data point and the mean. You find the difference between each value and the mean, square each difference so that positive and negative deviations do not cancel out, add them all up, and then divide. For a sample, you divide by the number of values minus one; for an entire population, you divide by the total count. The squaring step is what makes variance mathematically convenient, which is why it sits at the heart of advanced techniques like analysis of variance (ANOVA) and regression.

The catch is interpretation. Because the deviations are squared, variance is expressed in squared units. If your data is in citations, the variance is in “citations squared,” which has no real-world meaning. That is exactly the gap the standard deviation fills.

Standard deviation 🔗

The standard deviation is simply the square root of the variance. Taking the square root brings the measure back into the original units, so it is once again expressed in citations, rupees, or marks rather than their squares. This is why the standard deviation is meaningful: it represents the standard difference between the observed values and the mean in units you can actually understand.

A small standard deviation means the data points cluster tightly around the mean and the dataset is consistent. A large standard deviation means the values are widely scattered. For data that follows a roughly bell-shaped normal distribution, reporting the mean alongside the standard deviation is by far the most common and informative summary you can give.

Why these two matter together 🔗

Variance and standard deviation are closely related but serve different roles. Variance is the engine used inside statistical theory and inferential methods because squared terms are easy to work with algebraically. Standard deviation is the reporting tool, the number you actually communicate to readers because it is interpretable. Both are absolute measures of variability: they carry the units of the original data. That strength becomes a limitation the moment you try to compare two datasets measured in different units, which is where the next tool comes in.

Coefficient of variation: comparing variability across datasets 🔗

Suppose you want to compare the spread of citation counts in a physics journal with the spread of word counts in a humanities journal. The raw standard deviations are not comparable, because one is measured in citations and the other in words, and their averages differ wildly. You need a measure that strips away the units. That is the coefficient of variation (CV).

The CV is a relative measure of dispersion. You calculate it by dividing the standard deviation by the mean, usually multiplying by 100 to express the result as a percentage. Because both the numerator and denominator share the same units, those units cancel, leaving a pure, dimensionless number. Introduced by the statistician Karl Pearson, it is sometimes also called the relative standard deviation.

This makes the CV ideal for cross-dataset comparison. A dataset with a CV of 14% shows less relative variability than one with a CV of 20%, regardless of what either one measures. A low CV signals data clustered tightly around its mean and therefore greater consistency; a high CV signals greater relative spread. Researchers use it everywhere from finance, where it captures the risk-to-reward ratio of an investment, to epidemiology, where it compares variability in health indicators across populations with very different baseline rates.

The CV does have one important blind spot. Because the mean sits in the denominator, the measure becomes unreliable, and even mathematically undefined, when the mean is close to zero. A tiny shift in a near-zero mean can swing the CV wildly, so it should be used with caution for data that hovers around zero.

Skewness and kurtosis: the shape of the distribution 🔗

Range, variance, standard deviation, and the CV all describe how much the data spreads. They say nothing about its shape. Two datasets can have identical means and standard deviations yet look entirely different when plotted. Skewness and kurtosis capture this shape, going beyond centre and spread to describe symmetry and the behaviour of the tails.

Skewness: symmetry and asymmetry 🔗

Skewness is a measure of symmetry, or more precisely, the lack of it. A perfectly symmetric distribution, like the ideal bell curve, looks the same on both sides of its centre, and its skewness is zero. When the values stretch out into a long tail on the right, the distribution is positively skewed (right-skewed); when the tail extends to the left, it is negatively skewed. A useful rule of thumb: when the mean is pulled above the median, the data is right-skewed, and when the median dominates the mean, it is left-skewed.

Kurtosis: peaks and tails 🔗

Kurtosis describes the tails. It indicates how the tails of a distribution differ from those of a normal distribution. High kurtosis means heavy tails and a sharp peak, signalling a greater frequency of extreme values or outliers. Low kurtosis means light tails and a flatter shape, with fewer extremes. In short, skewness tells you which way the data leans, while kurtosis tells you how prone it is to extreme values.

Why these measures matter for citation and publication data 🔗

This is where variability becomes essential reading for anyone working with bibliometric data. Nearly all distributions in bibliometrics are skewed, and citation counts in particular are heavily right-skewed: a small minority of papers attract the overwhelming share of citations, while most receive very few or none. This pattern echoes the classic regularities of Lotka, Bradford, and Zipf, and Derek de Solla Price described it as a process of cumulative advantage, often called the Matthew effect, where the already-cited get cited still more.

The practical consequence is serious. When a distribution is this skewed, the plain mean can be misleading because a handful of blockbuster papers drag it upward, away from what a typical paper actually achieves. Many individual bibliometric indicators are well represented by a log-normal function, which appears as a highly skewed curve with a long tail on the high-productivity side. Knowing the skewness and kurtosis of your dataset warns you not to trust the average blindly, and it explains why researchers often turn to medians, percentile shares, or adjusted means when ranking journals, authors, or research units. Understanding variability, in other words, is not an academic nicety here. It is what keeps a citation analysis honest.

What do you think? If a department’s average citation count is propped up by one or two exceptional papers, should funding decisions rely on the mean at all, or would a measure like the median or coefficient of variation give a fairer picture? And when you next read a “top-performing” ranking, how would knowing the skewness of the underlying data change the way you interpret it?

References

1. https://www.scribbr.com/statistics/variability/
2. https://statisticsbyjim.com/basics/variability-range-interquartile-variance-standard-deviation/
3. https://www.itl.nist.gov/div898/handbook/eda/section3/eda35.htm
4. https://corporatefinanceinstitute.com/resources/data-science/coefficient-of-variation/
5. https://www.itl.nist.gov/div898/handbook/eda/section3/eda35b.htm
6. https://support.minitab.com/en-us/minitab/help-and-how-to/statistics/basic-statistics/supporting-topics/data-concepts/how-skewness-and-kurtosis-affect-your-distribution/
7. https://link.springer.com/article/10.1007/s11192-020-03476-8
8. https://www.nature.com/articles/palcomms201764