A defensible labour-market dashboard
14 min read
Lesson objectives
- Build a concise dashboard that states population, period, denominator and decision-relevant caveats.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
A dashboard is not decoration beside a decision already made. Its value is preventing a reader from assigning a meaning to a valid rate that the rate cannot support. Consider a fictional training agency with two places in a job-search programme, choosing districts for further needs assessment rather than making a final allocation. Its dashboard states the period and population covered, then employment, unemployment and participation. It adds hours or earnings when the question concerns job quality. Numbers alone do not prove a cause or settle the allocation, but a consistent display shows where deeper evidence is needed.
Give each indicator row five fields: clear name, numerator, denominator, period, and source or coverage note. For example: 'A unemployment rate, first quarter, household survey: 60 unemployed ÷ 660 in the labour force ≈ 9.1%.' Before comparing the next quarter, check district boundaries, working-age definition, survey weights, and search and availability questions. An apparent change can result from measurement rather than the market. If the local sample is small, say the district estimate is uncertain instead of presenting one decimal place as certainty. A sound dashboard needs auditable definitions more than many colours.
Choose indicators for the decision. An entry-to-work programme needs participation, employment and a breakdown of those outside the labour force. An additional-hours initiative needs actual hours, desire and availability. Purchasing-power analysis needs real earnings with worker coverage stated. Do not compare the mean earnings of large firms in A with the median for all workers in B, or put twenty employed workers with insufficient hours into unemployment. That corrupts both status and denominator. If an item is unavailable, label it 'not measured', not zero. Zero is a measured result; absence is a data state. The distinction can prevent a confident recommendation built on an empty cell.
An employment team has two hours before choosing its next pilot. It owns a large table, but the immediate question is which need deserves further investigation first. Put that question above the dashboard. Identify the reader and the decisions available to them. A small centre manager can change service timing or request follow-up interviews; a national report serves broader purposes. Let the intended use determine dashboard scope. Choose measures that can change the ordering of alternatives and add what helps interpret them. Place denominator definitions beside results and keep period order consistent. A colleague should be able to reproduce the result, then distinguish observations from proposals. This turns the display into a shared reasoning tool instead of an image from which everyone selects a preferred number.
Start the data card with the observation unit. A centre may count visits while a survey counts people. One person visiting four times creates four visits and one person. Dividing hires by visits and labelling the result a jobseeker placement rate changes the denominator's meaning. Define vacancies too: actual positions, advertisements, or employer requests? An advertisement appearing in two channels may still concern one position. Use a clear period for flows such as monthly hires, and distinguish these from the number of jobseekers present on a specific day. A point-in-time stock and a period flow describe different things. A short line specifying unit, period and coverage below each indicator catches errors before drawing, particularly when the display combines service records, household surveys and advertisements.
Before interpreting colours across two areas, inspect their populations. One area may contain more students and the other more retired people, affecting participation through composition. Show comparable age groups or display overall results beside population composition. Choose detail that serves the decision while retaining adequate sample sizes. Splitting a small sample into many cells can make comparisons fragile. Ask the survey producer for uncertainty measures and publication guidance, and retain them beside the table. An honest rounded value is preferable to an extra decimal implying unwarranted precision. If search definitions or area boundaries change, mark the series break. Present the periods with an explicit comparability limit, or construct a consistent series when data permit. Readers can then distinguish measurement changes from market changes.
Chart design shapes the message. For bars comparing magnitudes, start at zero or clearly explain an unavoidable truncation. Name percentage units in the title. A rise from 10% to 12% is two percentage points and 20% relative growth. Choose wording for the question, while showing both levels makes the comparison easy. Keep colour meanings consistent and make labels sufficient without colour. Before ranking areas, ask whether differences exceed measurement uncertainty. Similar areas can swap ranks because of sampling. Grouping close results may be more defensible than a strict league table, provided the grouping rule is stated. The result should be a display readers can question, rather than one that dictates interpretation through colour and type size.
Keep a short route from every published number back to its source. In your project, record the table name, source version, period and processing rule. If you remove duplicate advertisements, record the count and the rule identifying one underlying position. When a source changes after a meeting, retain the version used for the decision and note the update. Reproducibility helps when a stakeholder challenges a particular figure. Identify the cell, record and definition, then review the affected result. Sometimes the number is correct but its headline exceeds its coverage; wording then needs repair. With individual data, publish aggregates that protect privacy and use only the detail needed for the question. This combines auditability with responsible record use without turning the dashboard into a store of personal details.
Differences between sites may reflect seasonality before policy. A centre serving new graduates can receive more visits around graduation than one serving continuing workers. Comparing the same month across years may reduce some seasonal influence but still needs stable coverage and context. With short data, show months separately and state that an annual trend is unavailable. Avoid selecting two convenient points. Distinguish registration from placement dates: this month's hires may concern people registered months earlier. A cohort defined by entry month and followed over a fixed duration answers the beneficiary-journey question more clearly than a ratio of simultaneous activities involving different people.
Lesson summary
Build a concise dashboard that states population, period, denominator and decision-relevant caveats.