Data Structure Definition

From Fusion Registry Wiki
Jump to navigation Jump to search


An SDMX Data Structure Definition (DSD) provides a template which describes the dimensionality of related datasets in terms of their dimensions, attributes and measures.

Structure Properties

Structure Type Standard SDMX Structural Metadata Artefact
Maintainable Yes
Identifiable Yes
Item Scheme No
SDMX Information Model Versions 1.0, 2.0, 2.1
Concept ID DSD

Context within the SDMX 2.1 Information Model

SDMX Information Model - Core Artefacts - DSD.png

The schematic illustrates the Data Structure Definition artefact within the SDMX 2.1 Information Model


Data Structure Definitions (DSDs) are used to describe the structure of datasets by specifying their constituent Components:

and optionally the Representation for each Component.

Each Dataflow references a single DSD which describes the structure of the dataset that the Dataflow represents.


DSD IDs are conventionally uppercase using underscores '_' as separators if required. Examples:

Agency DSD ID Description SDMX-ML
Eurostat NA_MAIN European National Accounts Main Aggregated Statistical Indicators SDMX-ML
IMF BPO Balance of Payments and International Investment Position SDMX-ML
IMF ALT_FISCAL_DSD Alternate Fiscal Data Structure Definition SDMX-ML
World Bank WDI World Development Indicators SDMX-ML

The SDMX standard does not preclude using lowercase or mixed case for structure IDs. However IDs are case sensitive meaning that a DSD with ID 'NATIONAL_ACCOUNTS' is distinct from another named 'National_Accounts'.
For Time Series DSDs, the Time Dimension Component is conventionally given the ID 'TIME_PERIOD'.

Data Structure Components

The Role of Concepts In Defining a DSD's Components
Every Dimension, Attribute and Measure is described by a predefined Concept. Concepts have their own default Representation which can be overridden by defining a Local Representation for the Component in the DSD. That's particularly helpful when using some standard Concepts like the SDMX Cross Domain Concepts where the default Representation is 'String', but the Component needs to be Enumerated or have some use case specific restriction on what values are allowable.

A DSDs Dimensions are the minimal set of statistical concepts capable of uniquely identifying a specific series. For Time Series, the Dimensions in combination with the Time Dimension, uniquely identify an Observation.
In this sense, the Dimensions of a dataset together form its primary key.

Ordering of Dimensions in a DSD
The Dimensions in a DSD have a defined order and together form the dataset's Series Key.
Below is a simple example DSD:

Position Component Type Component ID Description
1 Dimension INDICATOR Indicator
2 Dimension REF_AREA Reference Area
3 Dimension FREQUENCY Data Frequency
n/a Time Dimension TIME_PERIOD Observation Time
n/a Attribute UNIT_MULT Unit Multiplier e.g. tens, thousands, millions
n/a Attribute Observation Status Observation Status e.g. Estimated, Final
n/a Primary Measure Observation Value The observation value

The Series Key is the concatenation of the Dimensions in the order specified in the DSD. In this example, the Series Key is:


Attributes do not form part of the Series Key so have no explicit or implied ordering.

Attributes allow extra concepts to be added to the dataset to provide additional information about the variable being measured such as the unit multipler or observation status.
Attributes are unique in that they must be attached to specific levels in the dataset at DSD design time.

Primary Measure
All DSDs must have a Primary Measure Component, which is used for the observation value of the main variable being measured. Like all components, the Primary Measure must reference a Concept. For many series, the measure is numeric, but does not need to be so.

Time Dimension
A Time Dimension is required for DSDs representing Time Series datasets. Again, the Time Dimension must reference a Concept which should have an appropriate time representation - typically Observational Time Period.

Attribute Attachment Levels

In designing a DSD, attributes must be attached to specific levels in the dataset.

Attachment Level Description
Dataset A single value for the attribute is set for the complete dataset
Series A different value for the attribute can be set for each series
Observation A different value for the attribute can be set for each individual observation in a time series
Group A different value for the attribute can be set for a Group of series

Example 'Demography' DSD with Attributes attached at the Series and Observation level:

DSD With Attribute Attachments.PNG

Time Series

DSDs for Time Series are characterised by having an explicit Time Dimension.
In combination with the DSD's other Dimensions, the Time Dimension uniquely identifies an individual Observation within a Dataset.

Non Time Series

DSDs can be designed for non Time Series datasets by excluding the Time Dimension. This supports use cases like census statistics which, although the observations are from a fixed point in time, there's no sequence of observations over a period of time.

Data Structure Definitions with Multiple Measures