Interpreting Trends: Real Change vs Random Variation
Learn how to distinguish genuine trends from random fluctuation in longitudinal data, and how to describe trends correctly in your results section.
You'll learn
How to distinguish a real trend from random variation, and how to describe trend direction and magnitude correctly.
Use this when
You have longitudinal data and need to report whether a quantity is increasing, decreasing, or stable.
What Is a Trend?
A trend is a sustained, directional change in a quantity over time. Not every upward or downward movement is a trend — short-term fluctuations, seasonal cycles, and random noise can all create the appearance of a trend in a small window of data.
📖 Trend vs noise
Trend is the long-run direction of a series after removing seasonal and irregular components. A single year of declining rates does not confirm a downward trend; it may be within the normal random variation of a stable system. Reliable trend detection requires at least 5–7 time points, and ideally 10+.
Statistical Tests for Trend
| Test | When to use | Output |
|---|---|---|
| Cochran-Armitage test | Testing for trend in proportions across ordered categories | p-value for linear trend |
| Jonckheere-Terpstra test | Non-parametric test for trend across ordered groups | p-value for monotonic trend |
| Linear regression on time | Continuous outcome, regularly spaced time points | Slope ± SE, R², p-value |
| Joinpoint regression | Identifying when a trend changes direction | Annual percent change (APC) per segment |
| Mann-Kendall test | Non-parametric monotonic trend in time series | τ statistic and p-value |
For annual trend reporting in epidemiology and public health, joinpoint regression with Annual Percent Change (APC) is the standard. The NCI Joinpoint Trend Analysis Software is free and widely used.
How to Describe Trends in Results
Describe the direction, magnitude, and statistical significance of the trend. Never report a trend p-value without the effect size (slope or APC).
- ●"The incidence rate increased from 12.4 to 18.6 per 100,000 between 2015 and 2023 (annual percent change +4.2%, 95% CI: 2.8–5.7%, p < 0.001)."
- ●"No significant trend was observed (APC: +0.8%, 95% CI: −1.2 to +2.9%, p = 0.41)."
- ●"A significant downward trend was observed from 2010 to 2016 (APC: −3.1%, p = 0.008), after which the trend stabilised (APC: +0.4%, p = 0.72)."
⚠️ Saying a trend is "increasing" without a test
Saying "the rate increased over the study period" without a p-value or CI is an unsupported claim. The rate may have fluctuated randomly. Always test whether the apparent trend is statistically significant.
Practice with your own dataset
Visualise and test a trend in your longitudinal dataset.
- 1.Upload a dataset with a year variable and a numeric or categorical outcome
- 2.Go to Analyze → Trend Report
- 3.Select the year variable and outcome variable
- 4.Choose the chart type (line, stacked area)
- 5.Download the trend table or figure for your paper