Dashboards and data visualisation for M&E

Two artefacts, two jobs

An M&E dashboard is a recurring decision aid: the same viewer returns to the same layout on a cadence, asking “where do we stand, and what needs attention?” A report figure is an argument: it appears once, inside a narrative, to carry a specific claim. The design rules differ — a dashboard optimises for scan speed, comparability across periods and stable layout; a figure optimises for a single message with everything else stripped away — but both serve a decision, and Kusek and Rist’s rule for performance reporting governs both: tailor the information to the user and the decision, and deliver it continuously enough to matter [3]. A chart nobody acts on is decoration with a data pipeline attached.

The shared obligation is stricter than usefulness, though. A chart is a factual claim rendered in geometry: bar lengths assert magnitudes, slopes assert rates of change, areas assert proportions. Readers extract those claims perceptually, in under a second, without checking the numbers — which is precisely why visualisation is an ethics topic. A chart that exaggerates a difference misinforms its reader as effectively as a fabricated number, with the added protection that almost nobody audits a picture.

The ethics of representation

The distortions below recur in M&E reporting not because designers are dishonest but because each one, applied “innocently”, makes a programme look better. That is exactly why they must be treated as bright-line rules rather than stylistic preferences. The evaluation field’s standard grading instrument — the Evergreen and Emery Data Visualization Checklist — encodes the core requirement as checklist items: the graph’s proportions must accurately represent the data’s proportions [1].

The recurring distortions and the honest alternative

Distortion	How it misleads	Honest alternative
Truncated bar axis	Bars encode value as length; starting the axis at 60% makes a 62%→65% change look like a doubling	Bars start at zero, always; if the change matters, show the change itself as its own chart
Zoomed line axis, unmarked	A clipped y-axis turns noise into drama; readers read slope as rate of change	Zooming a line chart is defensible for trend inspection — but only with the axis range unmistakable and, ideally, a zero-based companion
Dual axes	Two independent scales let the designer choose the crossing point and manufacture visual correlation	Two stacked panels sharing an x-axis; or index both series to a common base period
Area and 3D scaling	Doubling a circle's radius quadruples its area; 3D pies distort slice sizes by perspective	Encode value by length or position, not area or volume; retire 3D charts entirely
Cumulative-only curves	A cumulative line can never fall, so delivery that has stalled still looks like progress	Pair cumulative totals with the per-period series, where a stall is visible as a drop
Unlabelled denominators	"87% coverage" means nothing without the population it divides by — and denominator switches move the number silently	State the denominator and its source on the chart face; flag any denominator change on the trend
Cherry-picked windows	A time axis starting just after the bad year tells a true lie	Show the full available series, or state why the window starts where it does

One principle underlies the whole table: the reader’s perceptual inference should equal the statistical inference. If a viewer’s eye concludes “roughly double”, the numbers should say roughly double. Where a defensible design choice (a zoomed line axis) weakens that guarantee, the chart must announce it loudly enough that the inference is corrected.

Choosing the chart for the M&E question

Most M&E reporting needs a small set of shapes, and the field’s checklist-driven practice has converged on a recommended form for each [1] [3]:

• Progress against target — a horizontal bar with an explicit target marker (or a bullet-style bar), not a gauge: gauges spend enormous space encoding one number and hide the target-versus-actual comparison that matters.
• Trend against baseline — a line chart with the baseline value and target band drawn as reference lines, so the viewer reads distance-to-target, not just direction.
• Disaggregation comparison — sorted horizontal bars (sorted by value, not alphabet), which make the laggard group visible instantly; this is where disaggregated indicators either reveal who is left behind or bury it.
• Composition — a single stacked bar or simple table; pies survive only with few categories and no 3D.
• Geographic coverage — a map coloured by rate (never by raw count, which mostly redraws population density), with the denominator caveat attached.

Dashboard architecture: overview, drill-down, caveat

A working dashboard is layered: an overview answering “what needs attention” in one screen; drill-down views one click deep for the questions the overview raises; and thresholds tied to targets, so that attention is directed by pre-agreed rules rather than by whoever built the display. Alert colouring should encode distance from target as defined in the indicator’s metadata — which is one reason a dashboard is only as good as the information system beneath it.

The most neglected element is the caveat on the face of the chart. Routine data arrive incomplete and improve over weeks; a dashboard that displays last month’s figure without its reporting completeness invites false conclusions the WHO’s data-quality metrics exist to prevent [4]. The minimum honest furniture on any routine-data dashboard: a data-freshness stamp (“data to 31 July”), the reporting completeness for the period shown, and a visible marker on any figure still expected to move.

Accessibility and low-bandwidth reality

A dashboard read only at headquarters on a large screen is a dashboard most of its data suppliers never see. Practical floors: colour-vision safety (a meaningful share of any audience cannot distinguish the default red–green pairing — never encode meaning by colour alone; pair colour with position, labels or icons, as the checklist’s colour items require [1]); mobile rendering, since field and county staff will meet the dashboard on a phone; and a printable, static fallback, because the quarterly review meeting in a low-connectivity county happens whether the live dashboard loads or not. The same logic argues for restraint in interactivity: every filter is a way for two viewers to see different numbers and believe they saw the same dashboard.

Where figures meet reports

Charts inside an evaluation report inherit the report’s evidence standards: a figure asserts a finding, so it must survive the same traceability and honesty tests as the prose around it — with its data source, period and n on the face [2]. The full argumentative machinery of report figures is covered under evaluation reports; the underlying question of what the data can honestly carry belongs to data quality dimensions.