Does a long list of permitted choices make a field multivalued?

No. A field can offer many choices while accepting only one at a time.

Must multiple memberships be stored in one column?

No. Separate relationship rows are another representation.

Are no values and unknown values equivalent?

Not necessarily. Their meaning must be defined for the application.

AI Glossary / Data engineering and architecture / Advanced

Multivalued Domain

Definition

A multivalued domain, in this glossary's database-modeling sense, permits an attribute to represent several values together for one entity. For example, a resource may have several assigned topic labels. This differs from a domain offering many possible choices while allowing only one choice per record. The label describes a modeling idea; the precise representation and collection rules depend on the data model and implementation.

What it means

Having several allowed values is normal for almost any useful domain. A status field might allow draft, reviewed or published while storing only one status at a time. A multivalued attribute instead holds or represents a collection, such as both writing and analytics as a resource's topics. The supplied Dr Mike database glossary uses multivalued domain for this multiple-values-at-once idea. Modern systems expose different mechanisms: Microsoft Access provides multivalued fields, while PostgreSQL supports array-valued columns. Those features should not be treated as identical merely because each can represent multiple elements.

How it works

First define the allowed elements and what a collection means. Decide whether order matters, whether duplicates are allowed, and how to represent an empty collection or unknown information. Then choose an appropriate representation. Access documents a multivalued field that presents multiple selections while managing their storage behind the interface. PostgreSQL arrays have their own element and dimensional semantics; its documentation also discusses separate rows as an alternative for searchable collections. In an original topic-label design, a ResourceTopic relation with one resource-topic pair per row could represent the same intended memberships without placing a displayed list in one ordinary text field.

Three practical examples

A learning resource is associated with both writing and analytics, so its topic membership cannot be represented by a single-choice field without losing information.
A project records several assigned reviewers, while separately deciding whether reviewer order has any meaning.
An AI extraction workflow returns several proposed labels and validates each label against an approved vocabulary before saving the accepted collection.

Think of it this way

Imagine a form that says choose one preferred session and another that says select every session you can attend. Both present several possible answers, but only the second permits several selections together. The distinction concerns the meaning of the response, not how many boxes are printed on the form. A system must also decide whether an unanswered form means no available sessions or information that has not yet been supplied.

Worked example

Consider an original fictional label vocabulary containing writing, analytics and automation. Resource R1 has the set {writing, analytics}; R2 has {analytics}; R3 has the empty set. There are three resources, three resource-label memberships and two distinct labels actually used. A separate ResourceTopic relation would contain R1-writing, R1-analytics and R2-analytics, while the Resource relation still retains R3. Under these set semantics, adding analytics to R1 again does not create a new membership. If a storage mechanism allows repeated elements, the application must enforce or otherwise handle the intended uniqueness. An inner join from Resource to ResourceTopic would omit R3, so it cannot by itself count all resources. This example explicitly allows an empty known collection. It does not treat unknown labels as equivalent to having no labels, and it does not assign significance to label order.

Illustrative case study

In a fictional AI content directory, an extraction service produces topic lists for learning resources. An early report counts every list element, including repeated labels, and also loses unlabeled resources after expanding the lists into rows. The team defines topic membership as a set, validates labels against the approved vocabulary and keeps the complete resource inventory separate from membership records. Reviewers check R1, R2 and R3 from the worked example and reconcile both resource and membership totals. They also label pending extraction separately from a reviewed empty result. The scenario illustrates specifying collection meaning before choosing storage. It does not claim that an array, a multiselect interface or an AI-generated list automatically supplies these business rules.

Common misconceptions and mistakes

A domain containing many possible scalar values is not necessarily multivalued in this sense. Conversely, a comma-separated display does not reveal the underlying storage model. Ordinary text containing commas may simply be a string that requires parsing. Do not confuse a multivalued attribute with a multivalued dependency, which is a particular constraint in relational theory. Another mistake is assuming sets and arrays are interchangeable: ordering and repetition can change their meaning. Finally, dropping empty collections during expansion can silently change the population represented in a report.

Business use

For an AI tagging feature, define what users need to ask before selecting the representation. In the fictional directory, questions about the number of resources differ from questions about topic memberships. State whether a resource can have no labels, how unprocessed resources are marked, and whether a repeated label is invalid or merely redundant. Check how the chosen platform searches, updates and exports collections. Test round-trip import and export with empty and multiple-value examples. This connects the modeling decision to reporting and maintenance requirements rather than assuming a convenient input widget settles the design.

Check your understanding

How many resources exist in the toy example?

Answer: Three.

How many resource-label memberships exist?

Answer: Three.

Why are only two distinct labels used?

Answer: Automation is allowed but assigned to none of the resources.

Why must R3 remain in the resource inventory?

Answer: Its known empty label set does not mean the resource does not exist.

Would label order matter under the example's set semantics?

Answer: No.

Related terms

Array
Set
Multivalued dependency
Attribute domain
multivalued attribute
associative table

Learn next

Design a tiny label collection with one empty result, one multiple-label result and one unknown result. Write separate questions that count entities, memberships and distinct labels. Explain how your chosen representation preserves all three meanings when exported or expanded into rows.

Sources

Supplied terminology source, Dr Mike, Glossary of Database Terms: https://dr-mikes-maths.com/database-glossary.html Microsoft Support, Create or delete a multivalued field: https://support.microsoft.com/en-us/access/create-or-delete-a-multivalued-field PostgreSQL documentation, Arrays: https://www.postgresql.org/docs/current/arrays.html The resource-label arithmetic, form analogy and AI-directory scenario are original educational illustrations. Multivalued domain is scoped to the supplied glossary's modeling usage; the cited products illustrate distinct implementation choices. The glossary directly establishes its own usage; its cited Blaha textbook passage was not independently consulted.

Previous: Multivalued dependency Next: Multivalued Fact