BIOSTATISTICS • DESCRIPTIVE STATISTICS & VISUALIZATION

Time Series Plots — Interpret time series plots (intro)

Learn to read temporal patterns in health and biological data through the foundational language of time series visualization.

Historical Context & Motivation

The desire to visualize how quantities change over time is one of the oldest impulses in data analysis. Long before formal statistical methods existed, physicians tracked the course of fevers, astronomers charted planetary positions night after night, and economists recorded commodity prices across trading seasons. The time series plot — a graph that places time on the horizontal axis and a measured variable on the vertical axis — emerged from this practical need to see temporal patterns at a glance. In the biomedical sciences, understanding how a variable evolves over hours, days, or years is often more informative than any single snapshot measurement, making time series visualization an indispensable tool in the biostatistician's repertoire.

1786
Playfair's Line Charts
William Playfair published The Commercial and Political Atlas, introducing the line chart to display England's trade data over time — the first systematic use of time series plots in publication.
1854
Snow's Cholera Investigation
John Snow plotted cholera deaths by date during the Broad Street outbreak in London. His temporal plots helped identify the epidemic curve and pinpoint the contaminated water pump as the source.
1900s
Clinical Fever Charts
Hospitals widely adopted bedside fever charts, plotting patient temperature over consecutive days. These simple time series plots became a standard clinical tool for monitoring disease progression and treatment response.
1970
Box–Jenkins Methodology
George Box and Gwilym Jenkins formalized ARIMA modeling for time series data, but emphasized that visual inspection of the time series plot should always precede model fitting — a principle still central to biostatistical practice today.
2020
COVID-19 Epidemic Curves
The global pandemic brought time series plots — daily case counts, hospitalization rates, and mortality curves — into mainstream public discourse. Millions of people learned to interpret trends, peaks, and plateaus from these visualizations.

Across these milestones, a consistent theme emerges: before any sophisticated modeling takes place, the analyst must first look at the data. A time series plot is the most natural way to do so when observations are ordered by time. The central question this lesson addresses is deceptively simple — how do we read and interpret the visual features of a time series plot so that we can extract meaningful biological or clinical insights from temporal data?

Core Principles & Definitions

A time series is a sequence of data points collected or recorded at successive, typically equally spaced, points in time. When we plot these observations — with time on the x-axis and the variable of interest on the y-axis, connecting successive points with line segments — we obtain a time series plot. The visual form reveals several fundamental features that are the building blocks of interpretation. Mastering these features allows a biostatistician to describe a dataset's temporal behavior before resorting to formal tests or models.

1

Trend

A trend is the long-term increase or decrease in the data. In biostatistics, this might represent a gradual rise in antimicrobial resistance rates over a decade or a steady decline in infant mortality. Trends can be linear, exponential, or follow more complex trajectories.
2

Seasonality

Seasonality refers to regular, repeating patterns tied to a fixed period — for example, influenza case counts peaking every winter or pollen counts rising every spring. The key distinction is that the pattern recurs with a predictable period.
3

Cyclical Variation

Unlike seasonality, cyclical variation involves fluctuations that do not have a fixed period. Disease outbreaks, for instance, may recur at irregular intervals driven by population immunity dynamics or environmental perturbations.
4

Irregular / Random Variation

The residual fluctuation remaining after trend, seasonal, and cyclical components are accounted for is called irregular (or random) variation. This noise reflects unpredictable factors — measurement error, stochastic biological processes, or rare events.
5

Structural Breaks

A structural break is an abrupt change in the level or behavior of the series — such as a sudden spike in emergency department visits after a policy change or a natural disaster. Identifying these on a plot is essential for valid interpretation.
KEY TAKEAWAY
Think of a time series plot like an ECG (electrocardiogram) for your data. Just as a cardiologist reads rhythm, rate, and anomalies from the tracing on the monitor, a biostatistician reads trend, seasonality, and anomalies from a time series plot. The 'heartbeat' of your data tells you whether things are stable, worsening, improving, or oscillating — and whether any sudden events have disrupted the normal pattern.

Visual Explanation — Anatomy of a Time Series Plot

The diagram below illustrates a hypothetical time series of monthly hospital-acquired infection (HAI) rates over a three-year period. Study the labeled features carefully: each annotation corresponds to one of the core components introduced in Section 2. The ability to visually decompose a plot into these components is the foundation of all time series interpretation.

A simulated 36-month time series of hospital-acquired infection rates. The dashed purple line shows the downward trend; the red shaded zones highlight seasonal peaks (winter months); and the yellow dashed line marks a possible structural break where a new infection control protocol was introduced.

Several features are immediately apparent from the plot. First, the overall trajectory of the data slopes downward from left to right, which the dashed trend line makes explicit — the hospital's infection rate is declining over the three-year observation window. Second, within each year there are recurring rises and falls that correspond to seasonal variation; the peaks align with winter months, consistent with known patterns of nosocomial respiratory infections. Third, around the midpoint of Year 2, the series appears to shift to a consistently lower level, suggesting a structural break — perhaps the effect of a new hand-hygiene protocol or a change in surveillance definitions. Finally, the irregular jitter around the trend and seasonal pattern represents random variation, the noise inherent in any biological measurement process. The ability to identify and articulate these four features — trend, seasonality, structural breaks, and noise — constitutes the essential skill of time series plot interpretation.

Mathematical Framework — Decomposition of a Time Series

While interpreting a time series plot is fundamentally a visual exercise, the underlying concepts are formalized through the classical decomposition model. This model states that any observed time series value can be expressed as a combination of systematic components and random noise. Two standard formulations exist: the additive model and the multiplicative model. At the introductory level, the additive model is most intuitive and is appropriate when the seasonal fluctuations remain roughly constant in magnitude over time.

ADDITIVE DECOMPOSITION
Yₜ = Tₜ + Sₜ + Eₜ
Where Yₜ is the observed value at time t, Tₜ is the trend-cycle component, Sₜ is the seasonal component, and Eₜ is the irregular (error) component. Each component is expressed in the same units as the original data.
MULTIPLICATIVE DECOMPOSITION
Yₜ = Tₜ × Sₜ × Eₜ
In the multiplicative model, the seasonal component Sₜ is expressed as a ratio (e.g., 1.15 means 15% above the trend). This model is appropriate when the amplitude of seasonal fluctuations grows proportionally with the level of the series.
MOVING AVERAGE (TREND ESTIMATION)
T̂ₜ = (1/k) × Σᵢ₌₋ₘᵐ Yₜ₊ᵢ where k = 2m + 1
A simple moving average of order k smooths the observed series by averaging k consecutive observations centered at time t. This removes short-term fluctuations and reveals the underlying trend. For monthly data with annual seasonality, a 12-point moving average (or a 2 × 12 centered average) is commonly used.

When you examine a time series plot, you are implicitly performing a qualitative decomposition. You assess whether there is a discernible trend (is the series drifting upward or downward?), whether there are periodic oscillations (seasonality), and how much scatter remains around those systematic patterns (noise). The formal equations above give mathematical precision to what your eyes do naturally when reading the plot. Understanding that Yₜ = Tₜ + Sₜ + Eₜ reinforces a critical interpretive principle: any single data point on the plot is a composite of multiple underlying processes, and isolating them requires looking at the overall shape rather than fixating on individual observations.

Detailed Breakdown — Recognizing Components on a Plot

Identifying the individual components of a time series on a plot requires systematic observation. The diagram below presents four idealized patterns — each illustrating one component in isolation. When analyzing real biostatistical data, these components appear superimposed; your task as an analyst is to mentally (and later, computationally) separate them.

Panel A shows a clear upward trend with no oscillation. Panel B shows pure seasonality — a repeating wave of fixed period. Panel C shows cyclical variation with no fixed period. Panel D shows irregular random fluctuation around a stable mean.

In practice, real biostatistical data will display a mixture of these patterns. Consider daily COVID-19 case counts in a given region: there was a strong upward trend during each wave, a weekly seasonality driven by reporting delays (fewer cases reported on weekends), cyclical patterns corresponding to successive epidemic waves separated by troughs of variable duration, and day-to-day noise from testing variability. When interpreting a time series plot, a useful strategy is to ask yourself four questions in sequence. First, is the overall level of the series rising, falling, or stable? Second, are there repeating oscillations of a consistent period? Third, are there longer-term fluctuations that do not follow a fixed schedule? Fourth, how much scatter or noise remains after accounting for these systematic patterns? Answering these four questions provides a structured interpretation of any time series plot.

⚠️ Common Pitfall
Do not confuse seasonality with cyclical variation. Seasonality has a fixed, known period (e.g., 12 months, 7 days), while cyclical patterns have variable period lengths. On a time series plot, seasonal peaks will appear at evenly spaced intervals; cyclical fluctuations will not. Mislabeling one as the other can lead to incorrect modeling decisions downstream.

Worked Example — Interpreting a Clinical Time Series

Suppose you are given a time series plot showing the monthly count of new tuberculosis (TB) diagnoses in a metropolitan health district over five years (60 months). Your task is to provide a structured interpretation of the plot. Below, we walk through the systematic approach step by step.

Interpreting a TB Incidence Time Series Plot
1
Step 1 — Identify the Axes and UnitsBegin by reading the axis labels. The x-axis displays months from January of Year 1 through December of Year 5. The y-axis shows the number of new TB cases diagnosed per month. The units are counts (integers), and the range spans from 0 to approximately 45 cases per month.
Variable: monthly TB case count. Time span: 60 months.
2
Step 2 — Assess the Overall TrendVisually trace the general direction of the data from left to right. In this example, the data starts near 30 cases/month in Year 1 and ends near 20 cases/month in Year 5. Imagining (or overlaying) a straight line through the data confirms a downward trend. This suggests that TB incidence is declining in this district, possibly due to improved screening or treatment programs.
Downward linear trend: approximately −2 cases/month per year.
3
Step 3 — Look for Seasonal PatternsExamine whether the data oscillates with a regular period. In many temperate regions, TB diagnoses show a spring–summer peak. In our plot, local maxima occur around April–May each year, and local minima occur around October–November. The distance between successive peaks is approximately 12 months, confirming a seasonal pattern with a 12-month period.
Seasonal pattern present with a 12-month period, peaking in spring.
4
Step 4 — Identify Any Structural Breaks or AnomaliesScan the plot for abrupt shifts in level or unusual spikes/dips. In this example, there is a notable spike in Month 27 (March of Year 3) where the count reaches 44 — well above the seasonal maximum for that period. Investigating this anomaly might reveal an outbreak in a congregate setting such as a shelter or prison. No other structural breaks are apparent; the downward trend appears smooth and continuous.
One anomalous spike at Month 27 — potential outbreak event requiring further investigation.
5
Step 5 — Characterize the Noise LevelAssess the scatter of individual points around the trend and seasonal pattern. If most observations cluster tightly around the expected value (within ±3 cases), the noise is low and the systematic components dominate. If scatter is wide (±10 cases), the signal-to-noise ratio is low, and any trend or seasonal interpretation should be made cautiously. In our example, the scatter is moderate (±4 cases), giving us reasonable confidence in the identified patterns.
Moderate noise (±4 cases around systematic pattern). Signal-to-noise ratio adequate for confident interpretation.
📋 INTERPRETATION CHECKLIST
When interpreting any time series plot, follow this order: (1) axes and units, (2) overall trend, (3) seasonal patterns, (4) anomalies or structural breaks, (5) noise level. Think of it like a clinical exam: you take vitals (axes), assess overall health (trend), look for recurring symptoms (seasonality), flag acute episodes (anomalies), and note baseline variability (noise).

Strengths, Limitations & Comparison with Other Plot Types

Time series plots are remarkably effective for their intended purpose, but like any visualization tool, they have both strengths and limitations. Understanding these trade-offs helps you decide when a time series plot is the right choice and when supplementary or alternative visualizations may be needed.

Strengths and limitations of time series plots in biostatistical practice.
FeatureStrengthLimitation
Temporal patternsExcels at revealing trends, seasonality, and structural breaks in chronological data.Cannot show relationships between two non-temporal variables.
Ease of interpretationIntuitive for broad audiences; the left-to-right flow mirrors how humans naturally perceive time.Can be misleading if the y-axis is truncated, compressed, or uses a non-linear scale without clear labeling.
Multiple seriesTwo or three series can be overlaid for comparison (e.g., two treatment arms over time).Overlaying more than three series creates visual clutter; consider small multiples (facets) instead.
Sample sizeWorks well with moderate to large numbers of time points (20–500+).With very few time points (< 10), a table or bar chart may be clearer; connecting sparse points can imply continuity that does not exist.
Distributional detailShows central tendency over time; can add confidence bands or ranges.Does not inherently show the distribution of observations at each time point — box plots or violin plots may be needed for that purpose.
🔄 WHEN TO USE ALTERNATIVES
A time series plot is optimal when your primary question is 'How does this variable change over time?' If instead you are asking 'How does this variable relate to another variable?' use a scatter plot. If you need to compare distributions at specific time points, use box plots arranged along a time axis. If you are comparing cumulative quantities, consider a stacked area chart. The right visualization depends on the question, not the data type.

Connection to Advanced Time Series Methods

The interpretive skills you develop by reading time series plots form the qualitative foundation for more advanced analytic techniques. In subsequent coursework and research, you will encounter formal methods that automate and refine what you currently do by eye. The table below maps each visual feature you have learned to the corresponding advanced method, providing a roadmap for further study.

From visual interpretation to formal analysis: a roadmap for advanced study.
Visual Feature (This Lesson)Advanced MethodApplication in Biostatistics
Overall trend (upward/downward)Linear regression, LOESS smoothing, segmented regressionEstimating annual change in disease incidence rates
Seasonal peaks and troughsSeasonal ARIMA (SARIMA), Fourier terms, STL decompositionForecasting influenza peaks for hospital capacity planning
Structural breaksInterrupted time series (ITS) analysis, change-point detection algorithmsEvaluating the impact of a policy intervention on vaccination rates
Noise / random variationAutocorrelation function (ACF), partial ACF, white noise testsChecking model adequacy by examining residual patterns
Comparing multiple seriesCross-correlation, Granger causality, multivariate time series models (VAR)Assessing whether air pollution levels predict asthma ED visits with a lag

A recurring theme in the biostatistics workflow is that visual inspection always precedes formal modeling. Even the most sophisticated SARIMA or interrupted time series analysis begins with plotting the data and interpreting what you see. The features you learn to identify in an introductory time series plot — trend, seasonality, breaks, and noise — are precisely the features that guide model specification in advanced coursework. Developing strong visual literacy now will make you a more effective and more cautious analyst later, because you will know what to expect before a model tells you what it found.

Practice Problems

PROBLEM 1CONCEPTUAL
A researcher presents a time series plot of weekly emergency department visits for asthma over four years. She claims the data show a clear seasonal pattern. What specific visual feature on the plot would support her claim, and how would you distinguish it from cyclical variation?
PROBLEM 2BASIC CALCULATION
A time series plot of monthly malaria cases shows the following annual averages: Year 1 = 120 cases/month, Year 2 = 108 cases/month, Year 3 = 97 cases/month. Estimate the average annual decline in cases per month. If this linear trend continues, what would you predict for the Year 4 average?
PROBLEM 3INTERMEDIATE
You are examining a time series plot of daily blood glucose readings (mg/dL) for a Type 2 diabetes patient over 90 days. The plot shows readings fluctuating between 130 and 200 mg/dL for the first 60 days, then abruptly shifting to a range of 100–160 mg/dL for the remaining 30 days. The patient started a new medication at Day 60. Using the additive decomposition model Yₜ = Tₜ + Sₜ + Eₜ, describe what happened to each component at the transition point.
PROBLEM 4APPLIED
A public health department publishes a time series plot of weekly COVID-19 hospitalizations from March 2020 through December 2022 (approximately 145 weeks). You observe: (a) three distinct waves of increasing then decreasing hospitalizations, (b) the third wave has a much higher peak than the first two, (c) within each wave there are small dips every 7 days, and (d) the baseline between waves is higher after each successive wave than the baseline before it. Write a structured interpretation report identifying all four time series components.
PROBLEM 5CRITICAL THINKING
Two researchers each produce a time series plot of the same dataset — monthly suicide rates in a country over 10 years. Researcher A's plot has the y-axis starting at 0 and extending to 20 per 100,000. Researcher B's plot has the y-axis starting at 10 and extending to 14 per 100,000. Both plots are technically accurate. Discuss how the choice of y-axis range affects the visual interpretation of trend and variability. What ethical considerations arise in the context of communicating public health data?

Summary & Review

A time series plot displays observations in chronological order with time on the x-axis and the measured variable on the y-axis, making it the most fundamental visualization for temporal data in biostatistics. Interpreting these plots requires systematically identifying four core components: the trend (long-term direction), seasonality (fixed-period recurring patterns), cyclical variation (irregular-period fluctuations), and irregular noise (random scatter). The additive decomposition model (Yₜ = Tₜ + Sₜ + Eₜ) provides the mathematical formalization of this visual decomposition.

Effective interpretation also involves watching for structural breaks — abrupt level shifts that may signal interventions, policy changes, or outbreak events — and critically evaluating how axis scaling choices influence visual perception. These foundational skills in visual time series interpretation are prerequisite to advanced techniques including ARIMA modeling, interrupted time series analysis, and seasonal decomposition. Remember: in biostatistical practice, always plot your data before fitting a model — the time series plot is both the first step and the most important diagnostic tool.

Varsity Tutors • Biostatistics • Time Series Plots — Interpret time series plots (intro)