5.1 Line Graphs, Time Series & Trend Analysis

Time Series Dynamics in Public Administration

A time-series line chart plots a quantity against time. This section uses hypothetical administrative data to practise reading levels, interval changes, annual growth and recurring patterns. Use the format and timing specified for your competition rather than a universal seconds-per-item assumption.

---

Dual-Axis Graph Architecture & Scale Alignment

A dual-axis graph, also called a secondary-axis chart, requires careful matching of series and scales. These displays feature two independent vertical axes sharing a common chronological base:

1. Primary Vertical Axis (Left $y$-axis): Typically calibrated to financial or macroeconomic volume (e.g., Gross Expenditure in € millions).
2. Secondary Vertical Axis (Right $y$-axis): Often calibrated to operational activity, client caseloads, or percentage indices (e.g., Applications Processed in thousands, or Staff Full-Time Equivalents).

For a hypothetical display with linear axes 0–50 on the left and 0–200 on the right, the same physical height maps to different values:

The legend determines which scale belongs to a series. The shared vertical position does not make euros and inquiries interchangeable or prove a relationship between them.

The Scale Alignment Rule

A point at the right-axis reading 140 means 140,000 inquiries for the inquiry series. The corresponding left-axis position may read 35, but that does not mean 35 inquiries or €140 million. Match each series to the labelled scale and unit before calculating.

Truncated Axes and Zero-Suppression Traps

A line chart need not have a zero origin. A narrow vertical scale can make a small change look steep, so inspect the labelled range and time spacing rather than judging a percentage change by the line’s angle. Always calculate differences using absolute coordinates rather than visual slopes.

---

Evaluating Trends: Rate of Change vs. Absolute Level

A critical conceptual requirement in public sector quantitative evaluation is separating the absolute volume level from the rate of change (the slope):

• Absolute Level: The precise vertical coordinate of a data point at time $t$.
• Slope (Rate of Change): The steepness and direction of the line segment connecting two points, representing marginal velocity: $\text{Slope } (m) = \frac{\Delta y}{\Delta x} = \frac{y_2 - y_1}{x_2 - x_1}$

The "Flattening Curve" Fallacy

When an upward-sloping curve flattens—transitioning from a steep incline to a gentler positive slope—candidates often conclude that the metric has "declined." This is false. As long as the slope remains positive ($m > 0$), the absolute volume continues to expand. A flattening curve signifies decelerating growth, meaning the metric is still increasing, but at a reduced pace.

Inflection Points

For a continuous curve, an inflection point is where concavity changes: from bending upward to bending downward, or conversely. A zero second derivative alone does not establish such a change. Annual observations connected by line segments do not determine a unique smooth curve or exact inflection point.

For sampled yearly data, describe what is known: which interval has the greatest increase, and whether subsequent increments become smaller. A peak in the observed annual increment is not proof of a calculus inflection at the labelled year, nor proof that total adoption has peaked.

---

Mathematical Formulations: Slope, AAGR, and CAGR

Two different summaries of multi-year change are:

1. Average Annual Growth Rate (AAGR)

AAGR is the arithmetic mean of individual annual percentage growth rates across $n$ consecutive intervals: $\text{AAGR} = \frac{1}{n} \sum_{t=1}^n g_t = \frac{g_1 + g_2 + \dots + g_n}{n}$ For positive preceding values, each single-year percentage growth rate $g_t$ is: $g_t = \left(\frac{y_t - y_{t-1}}{y_{t-1}}\right) \times 100$

2. Compound Annual Growth Rate (CAGR)

For positive endpoints and $n>0$ annual intervals, CAGR is the constant compounded annual rate joining the endpoints. Four annual observations contain three intervals. The formula gives a decimal rate; multiply by 100 to express it as a percentage: $\text{CAGR} = \left(\frac{y_n}{y_0}\right)^{\frac{1}{n}} - 1$

Worked Administrative Example: Regulatory Agency Spend

Consider an environmental monitoring agency whose gross annual expenditure evolved as follows:

• 2021: €20.0 million
• 2022: €24.0 million ($g_1 = +20.0\%$)
• 2023: €27.6 million ($g_2 = +15.0\%$)
• 2024: €30.36 million ($g_3 = +10.0\%$)

Calculating the performance indicators:

• Linear Change: $y_3 - y_0 = \text{EUR }30.36\text{m} - \text{EUR }20.0\text{m} = +\text{EUR }10.36\text{m}$.
• Total Multi-Year Percentage Growth: $\left(\frac{30.36 - 20.0}{20.0}\right) \times 100 = +51.8\%$.
• AAGR: $\frac{20.0\% + 15.0\% + 10.0\%}{3} = \frac{45.0\%}{3} = 15.0\%$.
• CAGR: $(30.36 / 20.0)^{1/3} - 1 = (1.518)^{1/3} - 1 \approx 14.93\%$.

The two summaries answer different questions. The CAGR reproduces the ending value when compounded for three years; the arithmetic average of the varying annual rates need not do so.

---

Operational Seasonality vs. Secular Trends

A seasonal pattern repeats around similar parts of the calendar. A hypothetical applications series may rise before summer and fall later in the year. A trend concerns a longer-term direction, while one-off events can also affect observations.

Compare like quarters across years to reduce the effect of a repeated calendar pattern. This is not itself a formal seasonal-adjustment method, and one positive year-on-year comparison does not prove a lasting structural trend. In the quiz dataset, every quarter and the annual total are 15% higher in the second year, although Q3 is below Q2 in both years. Those facts undermine a conclusion of general contraction from that quarter-to-quarter drop alone.

Formal seasonal adjustment estimates and removes recurring calendar effects. Use adjusted data or a supplied model when the question requires that analysis; do not invent an adjustment from two observations.

---

Avoiding the Extrapolation Trap

Psychometric test designers test whether candidates recognize the boundaries of provided data. If a graph displays historic operational volumes from 2020 through 2024, questions may ask: "Will agency expenditure exceed €45 million in 2026?"

For a task requiring conclusions from supplied historical data alone, a future value is not established without an applicable forecast, rule or other sufficient constraint. If a model is supplied, calculate its conditional prediction and identify it as a forecast rather than an observed fact. A question may also explicitly ask you to estimate; follow that instruction without treating the estimate as certain.

---

Trend Indicators & Interpretation Reference