Choosing Chart Types for Distributions, Cross-Tabs, Flows and Maps

1.How values are spread: histogram, box and heatmap

Some questions are not about totals or trends. They are about shape: are order values bunched together or spread out, and does one category vary more than another? The dashboard concepts page groups three types for this: box plots, histograms and heatmaps, listed together "for distribution analysis."

A histogram takes one numeric column and shows whether its values cluster in a few ranges or spread widely. It is drawn as bars, but you choose the number of bins. The documentation's example uses o_totalprice with 20 bins. Bin count matters because it changes how coarse or fine the distribution appears.

A box chart summarizes a distribution with quartiles, optionally grouped by category. That lets you "quickly compare the value ranges across categories" and see locality, spread and skew. The example places return flag on the x-axis and extended price on the y-axis, so each return flag gets its own box.

A heatmap uses color to show numbers in a grid of two categorical axes. It "blend[s] features of bar charts, stacked charts, and bubble charts." The example puts order priority on the x-axis and ship mode on the y-axis, and color intensity shows the summed order count. Heatmaps can display up to 64K rows or 10MB. The dataset behind that example is pre-aggregated in SQL:

SELECT
    o.o_orderpriority AS priority,
    l.l_shipmode AS ship_mode,
    COUNT(*) AS order_count,
    o.o_orderdate
FROM
    samples.tpch.orders AS o
JOIN
    samples.tpch.lineitem AS l
ON
    o.o_orderkey = l.l_orderkey
GROUP BY
    o.o_orderpriority,
    l.l_shipmode,
    o.o_orderdate
ORDER BY
    priority,
    ship_mode;

Sources12

2.Cross-tabs, retention and flows: pivot, cohort and sankey

When readers need exact numbers broken down by two dimensions, a chart can be harder to read than a grid. A pivot visualization "aggregates records from a query result into a tabular display," much like PIVOT or GROUP BY in SQL. You place fields in Rows, Columns and Values, and you can turn on totals for each dimension. The example sums l_quantity, with return flag as rows and ship mode as columns. Pivot tables support rendering up to 1000 columns by 1000 rows.

Retention is a special case of a cross-tab. A cohort chart groups users by a shared starting point, such as sign-up date, and tracks how many remain active in later periods. In AI/BI dashboards there is no separate cohort type: you use a pivot. Cohort goes in Rows, active period goes in Columns, and the retention cell uses a Color Scale style, so darker cells show higher retention. The notebook and SQL editor visualizations do have their own cohort type, which only aggregates over dates. Any other aggregation must happen in the query.

For movement between states, such as where taxi trips start, where they end and which fare band they fall in, a sankey diagram "visualizes the flow from one set of values to another." You configure it with ordered stages plus a value, such as SUM(value).

Sources234

3.Geography: choropleth, point map or path map

Map choice depends on the geographic data the query returns. A choropleth colors regions, such as countries, states or counties, by an aggregated value. That needs locations by name, or a GEOMETRY or GEOGRAPHY column. The documentation's example sums total_sales by U.S. state name. A point map places a marker at each coordinate, so the result must include latitude and longitude pairs (or a geometry column). A path map draws lines, either from a line geometry or by connecting points in order. It fits routes, transit lines, rivers and GPS tracks.

In short, use a choropleth for totals by region, a point map for individual locations, and a path map for routes and trajectories.

Sources52

4.Putting it together: insight to visualization

Each type covered here answers a question that trend, proportion and KPI charts cannot. The table maps each kind of question to the type built for it.

Sources2

Exam traps

Each one states something that sounds right. Open it to see what is actually true.

1. 1.A histogram is just a bar chart of a measure by category.Why is that wrong?

   A histogram plots how often values occur, grouped into bins you control. It shows a distribution, not a total per category.

   Covered in How values are spread: histogram, box and heatmap

2. 2.A choropleth can be drawn from raw latitude/longitude points.Why is that wrong?

   A choropleth colors regions and needs locations by name or a GEOMETRY/GEOGRAPHY column. Latitude and longitude pairs go to a point map.

   Covered in Geography: choropleth, point map or path map

3. 3.AI/BI dashboards have a dedicated cohort visualization type, as in notebooks.Why is that wrong?

   In AI/BI dashboards, a cohort chart is built as a pivot visualization over retention data, styled with a color scale.

   Covered in Cross-tabs, retention and flows: pivot, cohort and sankey