LOD & Filters — Combine LOD expressions with filters responsibly (context filters conceptually)

Before the advent of modern visual analytics platforms, analysts working with SQL-based reporting tools frequently encountered a frustrating limitation: any filter applied to a query would uniformly restrict every aggregation in the result set. If you filtered to show only the year 2023, every sum, average, and count was scoped exclusively to 2023 data — there was no straightforward way to display a 2023-filtered metric alongside a grand-total metric computed across all years. Achieving such mixed-granularity results required subqueries, common table expressions, or window functions that many business users found inaccessible. Tableau set out to democratize analytics, but as its user base grew more sophisticated, the demand for calculations that could transcend or restrict the visible level of detail became impossible to ignore.

The central question that LOD expressions address is deceptively simple: how do you compute a measure at a granularity that differs from what the current view shows, and how do filters interact with that computation? Understanding this interaction is not optional — misusing it produces silently incorrect numbers. The remainder of this lesson dissects Tableau's query pipeline, clarifies the role of context filters, and equips you to combine LOD expressions with filters responsibly.

Before diving into the mechanics, it is essential to formalize the vocabulary. A Level of Detail (LOD) expression in Tableau is a calculated field whose aggregation granularity is explicitly declared in the formula itself, rather than being inferred from the dimensions present on the visualization shelves. The three LOD keywords — FIXED, INCLUDE, and EXCLUDE — each establish a different relationship between the formula's grain and the view's grain. Meanwhile, Tableau processes filters in a strict sequence known as the order of operations (or query pipeline), and the position of an LOD expression within that pipeline determines which filters can affect it.

The diagram above is the conceptual backbone of every decision you will make when combining LOD expressions with filters. When you drag a dimension to the Filters shelf and set it to show only a subset of values, Tableau by default places that filter at stage 5 — after FIXED LOD computations have already been resolved. This means a { FIXED [Region] : SUM([Sales]) } calculation will sum sales across all categories even if a dimension filter restricts the view to a single category. To make the FIXED expression respect that filter, you must promote the filter to a context filter (stage 3), effectively shrinking the dataset before the FIXED computation fires. INCLUDE and EXCLUDE expressions, in contrast, sit at stage 6, meaning they naturally respect dimension filters without any promotion.

Formal Semantics of LOD Expressions

While Tableau does not expose SQL directly to the user, each LOD expression translates to a subquery or a window function under the hood. Understanding this translation clarifies why filter placement matters. Consider a view with [Category] on Rows and SUM([Sales]) on Columns. Adding a FIXED expression { FIXED [Region] : SUM([Sales]) } effectively generates a correlated subquery that groups by Region and joins the result back to every row in the main query. Because this subquery is logically evaluated before dimension filters, the subquery's WHERE clause includes only extract-level, data-source-level, and context-level filter predicates.

Context Filters — Conceptual Model

When you add a dimension filter to context, Tableau conceptually creates a temporary materialized table containing only the rows that pass the context filter predicate. All subsequent stages — including FIXED LOD — operate on this reduced table rather than the full data source. This is analogous to creating a CTE (Common Table Expression) in SQL: WITH ctx AS (SELECT * FROM DataSource WHERE [context_predicate]) and then running every other query against ctx. The performance implication is worth noting: because Tableau materializes this intermediate result, context filters can introduce latency on large datasets, but they guarantee semantic correctness when FIXED expressions must honor certain filter conditions.

Tableau supports several filter types, each occupying a specific slot in the query pipeline. Misidentifying which filter type you are using is a common source of incorrect LOD results. The table below provides a comprehensive classification, including each filter's pipeline stage, its interaction with FIXED LOD, and practical use cases.

The second diagram makes the practical implications concrete. In Scenario A, the FIXED expression calculates total sales per region across all categories because the Category filter has not yet executed when FIXED fires. The view shows only Furniture rows, but the FIXED value of $500K for East includes Technology and Office Supplies as well — a classic gotcha. In Scenario B, promoting Category to a context filter restricts the underlying data to Furniture rows before FIXED executes, yielding the correct $180K. The decision flowchart codifies a simple heuristic: whenever you combine FIXED LOD with a dimension filter and you intend the filter to constrain the FIXED computation, add the filter to context.

Suppose you are analyzing the Superstore dataset. Your goal is to build a bar chart showing the number of customers whose first purchase fell in each year, but only for customers who have purchased items in the Technology category. The first-purchase year is computed via a FIXED LOD expression on [Customer ID].

The interplay between LOD expressions and filters extends into more advanced Tableau patterns. Nested LOD expressions — where one LOD wraps another — follow the same pipeline rules, but the innermost expression is evaluated first, and its result is treated as a row-level value for the outer expression. Set-based filtering introduces another nuance: when a set is placed on the Filters shelf, it behaves as a dimension filter (stage 5), meaning it does not affect FIXED LOD unless the set filter is added to context. Similarly, parameter-driven filters implemented via calculated fields on the Filters shelf follow dimension filter semantics unless explicitly promoted.

As you progress into Tableau Prep, Tableau Server (now Tableau Cloud), and embedded analytics, the LOD-filter interaction remains unchanged — the query pipeline is a constant across all Tableau products. Building a solid mental model of this pipeline now will pay dividends when you encounter row-level security filters (which operate at the data source level, stage 2) or initial SQL (which runs before even extract filters). Each mechanism is simply another station on the same assembly line.