Wide and Long Data

Wide and long describe how a chosen measurement dimension is arranged. Population by country and year can use one country row with year columns, or one country-year row with a population column. Long does not mean exactly three columns: population and average age can remain separate measures.

Reshaping can preserve the observations when identifiers, units, and missing values remain explicit. Returning to wide form needs one value per destination cell. If a country-year has two records, choose an additional key or an explicit aggregation; summing them is a modeling decision, not a lossless rearrangement. An absent cell and an observed unknown value can also need different treatment.

A prejoined wide table is a different idea. It repeats descriptive attributes alongside facts and need not pivot any measurement dimension. Long form is useful when adding periods should add rows, while wide form can suit comparisons or model inputs. Neither shape alone guarantees speed, storage efficiency, or normalization. pandas melt reshapes columns into rows; pivot requires unique destination cells, whereas pivot_table can aggregate.

Reference: pandas reshaping. See Data Modeling for Operational and Analytical Systems for the worked example.