Data Preparation
From connection to clean data source
Data Preparation questions ask whether you can turn a raw connection into a trustworthy, analysis-ready source before you build a single chart. Know the layers: the logical layer (relationships, added in 2020.2, drawn as a "noodle") and the physical layer (joins and unions inside each logical table). Relationships do not duplicate rows - Tableau aggregates at the level of detail of whatever is in the view - while a join can multiply rows. A union stacks rows of the same shape; a join adds columns.
Reshaping and cleaning
Expect questions on the Data Interpreter (cleans up Excel-style headers, removes titles and notes), Pivot (turns crosstab columns into rows), Split (needs a delimiter and creates new fields), changing data types, aliases, default properties, hiding unused fields, and renaming. Know where filters live: a data source filter applies to every sheet using that source; an extract filter is applied when the extract is built and limits what ships to the extract; a context filter is independent and changes the order of operations for LOD expressions.
Traps to watch
- Confusing join with union, or relationships with joins.
- Thinking a filter on a worksheet narrows an extract.
- Pivot vs Split: Pivot goes wide-to-tall, Split goes one-column-to-two.
- Data Interpreter only makes sense for spreadsheet-style sources.
Worked example
A sales sheet has columns Region, Jan, Feb, Mar. Select the three month columns, choose Pivot, and you get Pivot Field Names (Month) and Pivot Field Values (Sales). Row count triples; Month is now a usable dimension for trend lines. Rule of thumb: if a field names a time period inside a column header, pivot it; if a field is a single measure at the right grain, leave it alone.