GIS guides

ArcGIS Dashboards: Why Two Correct Numbers Can Tell Different Stories

Before fixing a dashboard number, check its population, date rule and counting unit. A correct calculation can still answer the wrong question.

Two analysts compare paper records with a calculator during an office review.

Before fixing a dashboard number, check its population, date rule and counting unit. A correct calculation can still answer the wrong question.

The disagreement: 70 percent or 87.5 percent?

In a synthetic public-service register, 100 requests were created during a reporting period. Seventy are resolved, ten remain open and twenty were cancelled. One dashboard shows 70% resolved. Another shows 87.5%. Both calculations can be arithmetically correct.

MeasureCalculationMeaning
Resolved share of all created requests70 / 100 = 70%Includes cancelled requests in the population
Resolved share excluding cancellations70 / 80 = 87.5%Excludes twenty cancelled requests
Outstanding share excluding cancellations10 / 80 = 12.5%Describes remaining eligible work

The first repair is a definition, not a new chart color. Label the population and decide which question the audience actually needs answered.

Audit the date rule next

The table concerns requests created in the period and their status at the stated review time. It does not measure all resolutions completed in that period. Some older requests may have been resolved during the same dates; some newly created requests may be resolved later.

Write a measure contract: “Unique requests created from the agreed start through the agreed end, excluding cancelled requests, with resolution status as of the review timestamp.” State the time zone and boundary convention. Do not imply a date filter automatically chooses the business definition.

ArcGIS Dashboards supports interactive elements and location-based visualizations. That flexibility makes consistency between filters, elements and labels a review task.

One request can become three rows

Suppose each request can have several inspection records. Counting joined rows may count inspections while the label says requests. In a small audit sample, give one request three child records and check whether the displayed request count changes. The intended counting unit should survive the relationship.

This is a proposed diagnostic test, not a statement that every dashboard uses a join or suffers this defect. Keep it small enough that a reviewer can calculate the expected answer by hand.

A release note that helps the next person

Record the numerator, denominator, distinct identifier, date field, exclusions, time zone and update time. Include the worked 100-request example as a reference case. When someone changes a filter or data source, repeat it.

The benefit is an explainable number. A reader who understands what is excluded can make a better comparison than one looking at a large percentage with no definition. Keep this public-service measure separate from maintenance completion KPIs, whose units and lifecycle may differ.

For related preparation, read testing records before collection. Discuss this workflow with InfoGraph or review relevant GIS training.