Rounded Measurements and Apparent Piles

When Many Measurements Share One Value

A dotplot of heights may show a tall stack at 170 cm, or a histogram may show an unexpectedly high bar around that value. The stack is real as a feature of the recorded data: several entries have the same recorded height. But it does not automatically mean that the people’s unrounded heights were concentrated at exactly 170 cm. If heights were rounded before being recorded, a range of different measurements can all receive that same value.

The important distinction is between the quantity of interest, such as a person’s height, and the number entered in the data set. Graphs display recorded values. To decide what an apparent pile means, ask how those values were obtained: to what precision were heights measured, and did people or recorders tend to choose particular rounded numbers?

For example, a height recorded as 170 cm to the nearest centimetre can represent an unrounded height from 169.5 cm up to, but not including, 170.5 cm. The record does not reveal where in that interval the height lies. A dotplot stacks all records of 170 cm at one position; it cannot show the variation hidden inside the rounding interval.

A stack at each recorded integer is an expected consequence of recording to whole centimetres. A different pattern may be more revealing: if many people report heights ending in zero or five, there may be extra heaping at those preferred values. That pattern could result from estimating or rounding to the nearest five or ten rather than measuring to the stated precision. It is a clue about the recording process, not by itself proof of what the unrounded heights were.

Why Rounding Creates Repeated Values

Suppose an underlying height \(H\) varies continuously and is recorded to the nearest \(q\) units. The recorded value \(R\) can only take steps of size \(q\), even though \(H\) could lie anywhere in a continuous range. All heights that fall into the rounding interval for one recorded value receive the same entry. Ties are usually not important for a genuinely continuous measurement, because the probability of landing on any one exact boundary is zero.

A useful justification comes from considering the probability of a recorded value. If the underlying heights have a probability density \(f\), then the probability of recording \(r\) is the probability that the height falls in its rounding interval:

If \(f\) changes gradually over this short interval, its value is approximately \(f(r)\) throughout the interval. The probability is then approximately the interval width times the density near \(r\):

Thus, when the underlying distribution is locally smooth, adjacent rounding intervals of equal width tend to receive similar probabilities, unless the underlying density itself changes substantially across them. A repeated value is therefore not surprising simply because the unrounded measurements were continuous. However, a striking excess at selected endings may indicate that values were rounded more coarsely or estimated to convenient numbers. A graph alone cannot determine which explanation is correct.

Ordinary Rounding and Extra Heaping

It helps to distinguish two patterns that can look similar at first. Ordinary rounding creates repeated values at every allowed recording step. For example, measurements rounded to whole centimetres can repeat at 167, 168, 169, and so on. Extra heaping occurs when some allowed values appear far more often than nearby ones, often because people report estimates using familiar endings such as zero or five.

The word “far” matters. A value can be common because many observations genuinely lie near it, because the underlying distribution is denser there, or because the recording process favors it. To assess the pattern, examine the precision and source of the data. A measuring instrument with a stated resolution suggests one mechanism; self-reports or estimates suggest another. Preferential endings are evidence to investigate, not a definitive diagnosis.

Use Recording Details to Assess an Apparent Pile

A graph shows the values that were recorded; it does not disclose the precision of the original measurement unless that information is supplied separately. When a dotplot or histogram suggests a pile, the following questions help separate what the graph establishes from what remains uncertain:

• What was the recording precision? Whole centimetres, half-centimetres, and multiples of five centimetres represent different levels of resolution. Repetition at the recorded step size is expected.
• Were the values measured or estimated? A person estimating a height may prefer a convenient number. An instrument or protocol can also limit the precision, so “measured” does not automatically mean “unrounded.”
• Are certain endings especially common? Look for a pattern at preferred values, such as multiples of five, relative to nearby recorded values. Treat this as a diagnostic description, not a formal test or a conclusive explanation.
• Are more precise records available? Comparing the original measurements with rounded reports, if both exist, can show which differences were hidden by rounding.

The display itself can also affect how prominent a pile looks. In a dotplot, identical recorded values are stacked together, making their repetition visible. In a histogram, a bar combines all observations falling within its bin; depending on the bin boundaries and width, a preferred value may be combined with neighboring values or separated from them. Changing the display can help reveal whether a pattern depends strongly on how values are grouped, but it cannot recover detail that rounding removed.

Common Mistake: Treating Every Stack as a Real Cluster

A common mistake is to see a tall stack at a rounded value and conclude, “Many people had exactly that height, so there must be a genuine cluster there.” That conclusion confuses a recorded value with the range of measurements it represents. If heights were rounded to the nearest centimetre, everyone in a half-centimetre interval receives the same entry; if heights were reported to the nearest five centimetres, the interval is five times as wide.

A careful account separates observation from explanation. “Eight reports ended in zero” is a statement about the recorded data. “People rounded their estimates to the nearest ten centimetres” is a possible explanation that requires information about how reports were collected. “The underlying heights were concentrated at those values” is a claim about unrounded measurements that the rounded graph alone cannot establish.

This distinction matters whenever repeated measurements appear in a graph. Rounding can hide variation within intervals, create ties, and make a dotplot look more concentrated than the underlying values are. At the same time, a pile should not be dismissed automatically: it may reflect a real concentration, a reporting habit, or a combination. The sensible interpretation begins with what is known about measurement and recording, and states clearly what the graph can and cannot show.