Quantitative Data Types in Data Mining and Data Warehousing

Quantitative data is the backbone of analytical processing in both data mining and data warehousing. It consists of numerical values that represent measurable properties of data objects — quantities, counts, measurements, and metrics that can be subjected to arithmetic operations like addition, subtraction, multiplication, and division. Unlike qualitative data, which describes qualities or categories, quantitative data tells us "how much," "how many," or "how often" something occurred 2.

In the context of data mining, quantitative data drives pattern discovery through statistical analysis, predictive modeling, and association rule mining. In data warehousing, quantitative data forms the core of fact tables, which store the measurable business metrics that organizations analyze, aggregate, and visualize for decision-making 2.

The classification of quantitative data originates from the work of psychologist Stanley Smith Stevens, who in 1946 published a landmark paper in Science introducing a hierarchy of four measurement scales: nominal, ordinal, interval, and ratio. This framework, often called Stevens' typology, gave researchers a shared vocabulary for discussing the precision and mathematical properties of data, and it remains the near-universal standard in research methodology and statistical reporting today . Understanding these levels is not optional for anyone who collects, mines, or warehouses data — the measurement scale determines which statistical analyses are valid, which data mining algorithms are applicable, and how data should be structured in warehouse schemas.

Footnotes

1. Australian Bureau of Statistics — Quantitative and Qualitative Data — Official definitions of quantitative vs. qualitative data. ↩

2. FullStory — What is Quantitative Data? — Characteristics and types of quantitative data. ↩

3. Streamkap — Best Practices in Data Warehousing — Dimensional modeling and star schema design with quantitative measures. ↩

4. Redgate — Facts and Dimensions in a Data Warehouse — How quantitative attributes function as measures in fact tables. ↩

5. Wikipedia — Level of Measurement — Stevens' typology and the four scales of measurement. ↩

The Four Levels of Measurement — Deep Dive

Stevens' typology classifies data into a cumulative hierarchy: each successive level retains the properties of the previous one and adds new capabilities. The first two levels (nominal and ordinal) are qualitative, while the last two (interval and ratio) are quantitative.

1. Nominal Scale (Qualitative)

The simplest level. Categories are labels only with no inherent ordering. The only permitted measure of central tendency is the mode, and the only arithmetic operation is counting (frequency) . Examples: gender, nationality, product category, blood type.

2. Ordinal Scale (Qualitative)

Categories possess a meaningful rank or order, but the intervals between ranks are not necessarily equal. The median can be calculated in addition to the mode. Examples: education level (high school, bachelor's, master's), customer satisfaction ratings (poor, fair, good, excellent) .

3. Interval Scale (Quantitative)

Interval data features equal intervals between consecutive values, but lacks a true zero point — meaning ratios are not meaningful. A temperature of $40°C$ is not "twice as hot" as $20°C$. The mean and standard deviation become valid measures. Examples: temperature in Celsius/Fahrenheit, IQ scores, calendar dates 2.

4. Ratio Scale (Quantitative)

Ratio data possesses all properties of interval data plus a true zero point, which means the absence of the measured quantity. This makes ratios meaningful: $10$ minutes is twice as long as $5$ minutes, and $0$ kg means no weight. All statistical operations are valid, including the coefficient of variation. Examples: height, weight, age, revenue, count of items 2.

Footnotes

1. Researcher.Life — Levels of Measurement — Detailed explanation of Stevens' four measurement scales with examples. ↩ ↩² ↩³ ↩⁴

2. Wikipedia — Level of Measurement — Stevens' typology and the four scales of measurement. ↩ ↩²

Discrete vs. Continuous Quantitative Data

Within the interval and ratio levels, quantitative data is further classified as discrete or continuous:

• Discrete data: Represents countable values, usually whole numbers. There are gaps between possible values. The number of items in a shopping cart, the number of students in a class, or the count of website clicks are all discrete. Discrete data answers the question "how many?" 2.

• Continuous data: Can take any value within a given range, constrained only by the precision of the measurement instrument. Height, weight, temperature, and time are continuous — between any two values, there are infinitely many other possible values. Continuous data answers the question "how much?" 2.

The distinction between discrete and continuous data influences the choice of statistical tests, probability distributions (e.g., Poisson vs. Normal), and data mining algorithms. For instance, classification trees may split discrete data on specific values, while continuous data is split on thresholds.

Footnotes

1. FullStory — What is Quantitative Data? — Characteristics and types of quantitative data. ↩ ↩²

2. GeeksforGeeks — Understanding Data Attribute Types — Discrete vs. continuous attribute differences in data mining. ↩ ↩²

Quantitative Data in Data Warehousing

In a data warehouse, quantitative data plays a central role through dimensional modeling, a design technique championed by Ralph Kimball. The fundamental building blocks are fact tables and dimension tables 2.

Fact tables contain two types of attributes:

1. Qualitative attributes — These are foreign keys referencing dimension tables (e.g., ProductID, CustomerID, DateID). They provide context but are not themselves measures .
2. Quantitative attributes — These are the measures: amounts, quantities, durations, or any other measurable numerical values upon which statistical calculations like SUM, AVERAGE, and VARIANCE can be applied .

The most common warehouse schema, the Star Schema, places a large fact table at the center containing quantitative measurements (like sales figures or quantities), surrounded by smaller dimension tables holding descriptive attributes 2.

In this schema, Total_Revenue, Quantity_Sold, and Discount_Amount are quantitative measures stored in the fact table. Analysts can rapidly aggregate revenue by product category or region by joining the fact table to only the relevant dimensions — the denormalized structure minimizes join complexity and maximizes query speed .

Types of quantitative measures in fact tables include:

• Additive measures: Can be summed across all dimensions (e.g., revenue, quantity sold)
• Semi-additive measures: Can be summed across some dimensions but not others (e.g., account balance — additive across customers but not across time)
• Non-additive measures: Cannot be summed across any dimension (e.g., ratios, percentages, averages)

Footnotes

1. Streamkap — Best Practices in Data Warehousing — Dimensional modeling and star schema design with quantitative measures. ↩

2. Exasol — Data Warehouse Schemas: Star, Snowflake, and Galaxy Explained — Comprehensive guide to warehouse schema types and their components. ↩ ↩²

3. Redgate — Facts and Dimensions in a Data Warehouse — How quantitative attributes function as measures in fact tables. ↩ ↩²

4. Snowflake — What Is a Star Schema? Complete Guide — Star schema design principles and retail sales example. ↩ ↩²

Quantitative Association Rule Mining in Detail

[Quantitative association rules]{def "Association rules where at least one attribute is numerical, requiring discretization or optimization techniques to mine meaningful patterns"} (QAR) are a special type of association rule in which at least one attribute — on either the left-hand or right-hand side — must involve a numerical (quantitative) attribute. This extends traditional association rule mining, which was originally limited to binary or categorical attributes 2.

The general form of a QAR is:

$X \rightarrow Y, \quad \text{where at least one of } X, Y \text{ contains a numerical attribute}$

The two main approaches to handling quantitative attributes in association rule mining are:

1. Discretization-Based Approaches: Partition numeric attributes into intervals, converting them to categorical items. The seminal Srikant-Agrawal algorithm uses equi-depth discretization, where each interval contains roughly the same number of tuples. Other discretization strategies include equi-width (equal-sized intervals), K-means clustering-based partitioning, and entropy-based methods 2.

2. Optimization-Based Approaches: Avoid discretization entirely by finding optimal intervals through mathematical optimization. The half-space approach by Ruckert et al. (2004) represents rules based on half-spaces in the attribute space, eliminating the information loss inherent in discretization. However, the resulting rules can be harder to interpret .

Footnotes

1. arXiv — Numerical Association Rule Mining: A Systematic Literature Review — Comprehensive review of QAR mining algorithms and discretization strategies. ↩ ↩² ↩³

2. Springer — Quantitative Association Rules — Formal definition and classification of quantitative association rules. ↩

3. GeeksforGeeks — Types of Association Rules in Data Mining — Overview of association rule types including QAR. ↩