Continuous Variable
A continuous variable is a numerical variable that can take any value within an interval, so between any two possible values another one always lies — height, time, mass, temperature, blood pressure, the voltage on a wire. In principle such a quantity has unlimited precision: a person is not exactly 172 centimeters tall but some value like 172.0381 and onward, and the number reported depends on how finely the instrument can read.
The mathematical signature of a continuous variable is that probability attaches to intervals through a probability density function, and the probability of any single exact value is zero. Probability is the area under the density curve between two points, the total area is 1, and the height of the curve is a density, not a probability, so it can legitimately exceed 1 for a narrow distribution. This explains a result that surprises beginners: the chance that someone is exactly 170 centimeters tall is zero, yet people of that height clearly exist, because the practical question is always about a range such as "between 169.5 and 170.5." The cumulative distribution function is a smooth curve, and quantiles, percentiles, and medians come from inverting it. Moving from sums over a mass function to integrals over a density is the main technical change from the discrete case.
Familiar continuous distributions include the normal, uniform, exponential, gamma, and beta, and each arises from a physical or statistical mechanism. The normal distribution, developed by Carl Friedrich Gauss in the early 1800s in work on astronomical measurement error, describes sums of many small independent effects. The exponential distribution describes waiting time between events in a Poisson process and is memoryless, the uniform describes complete ignorance across a range, and the beta distribution describes proportions bounded between 0 and 1. In practice no instrument records true continuity: measurements are rounded or binned, so a continuous variable is observed through a discrete lens, and treating it as continuous is a modeling choice that works when the grid is fine relative to the spread.
Continuous variables dominate measurement-based science and many machine learning tasks. Regression predicts a continuous target; neural networks operate on continuous activations and are trained by gradient descent, which depends on differentiability; kernel density estimation, introduced by Murray Rosenblatt in 1956 and Emanuel Parzen in 1962, estimates an unknown density from data. In physics, engineering, and medicine, quantities such as concentration, strain, and reaction time are modeled as continuous, and the relevant uncertainty is expressed with standard deviations and intervals. Continuous outcomes also allow powerful tools such as the t-test and analysis of variance, usually with checks on distribution shape.
Typical pitfalls are confusing density with probability, forgetting that recorded data are rounded, and imposing artificial cut points. Reading a density value of 3.2 as a 320% probability is an error, since densities have units of one over the variable's unit. Heavy rounding creates heaping (ages reported as multiples of five) and ties that distort methods assuming smoothness. Cutting a continuous variable into high and low groups discards information and manufactures sharp boundaries that the underlying quantity lacks, so a blood-pressure threshold is a decision convention rather than a natural divide. Finally, strictly bounded quantities such as proportions or positive times are poorly described by a normal curve that ignores the bounds.
Read a continuous variable as a quantity that varies smoothly with no gaps: probability lives in the area over ranges, not on any single value, and the numbers in a spreadsheet are only a rounded, finite sampling of an underlying quantity that never actually stops at the last decimal place.