Measurement Scales

Understanding Measurement Scales

In measurement theory, a scale is the system used to assign numbers or symbols to the attributes of entities we are measuring. The type of scale determines the mathematical operations that are meaningful for the data and influences the statistical analyses that can be applied. Understanding scales is crucial for correctly interpreting software metrics.

There are five primary types of measurement scales:

• Nominal Scale
• Ordinal Scale
• Interval Scale
• Ratio Scale
• Absolute Scale

Nominal Scale

This is the simplest scale. It categorizes items without any order or ranking. Numbers or symbols are used merely as labels or identifiers.

• Characteristics: Classifies data into distinct, unordered categories.
• Empirical Relation: Equivalence (items are either in the same category or different categories).
• Meaningful Operations: Counting frequencies (e.g., number of defects of type ‘UI’, ‘Logic’, ‘Performance’), finding the mode.
• Meaningless Operations: Arithmetic operations (addition, average), comparisons like ‘greater than’.
• Software Example: Classifying programming languages used in a project (e.g., 1=Java, 2=Python, 3=C#), defect types (e.g., ‘Security’, ‘Functional’, ‘Performance’), module names.

Ordinal Scale

This scale ranks items in a specific order, but the differences (intervals) between the ranks are not necessarily equal or meaningful.

• Characteristics: Classifies data into ordered categories.
• Empirical Relation: Equivalence and Order (‘greater than’ or ‘less than’).
• Meaningful Operations: Counting frequencies, mode, median, non-parametric statistical tests based on rank.
• Meaningless Operations: Arithmetic operations (average, standard deviation) because the intervals are not defined.
• Software Example: Ranking defect severity (e.g., 1=Critical, 2=High, 3=Medium, 4=Low), user satisfaction ratings (e.g., ‘Very Satisfied’, ‘Satisfied’, ‘Neutral’, ‘Dissatisfied’), complexity ratings assigned by developers (e.g., ‘Simple’, ‘Moderate’, ‘Complex’). Note that while we know ‘Critical’ is worse than ‘High’, we don’t know how much worse.

Interval Scale

This scale orders items, and the intervals between consecutive values are equal and meaningful. However, it lacks a true zero point (a point indicating complete absence of the attribute).

• Characteristics: Ordered data with equal, meaningful intervals.
• Empirical Relation: Equivalence, Order, and known differences between values.
• Meaningful Operations: Counting, mode, median, mean, standard deviation, addition, subtraction.
• Meaningless Operations: Multiplication, division, ratios (because there’s no true zero). Example: 20°C is not ‘twice as hot’ as 10°C.
• Software Example: Calendar dates (e.g., project start date, completion date - the interval between Jan 1 and Jan 10 is the same as between Feb 1 and Feb 10). Temperature in Celsius or Fahrenheit. Sometimes, standardized test scores are treated as interval scale.
• Note: Few common software metrics naturally fall purely into the interval scale.

Ratio Scale

This is the most informative scale. It has all the properties of an interval scale, plus a true zero point, indicating the complete absence of the attribute being measured. Ratios between measurements are meaningful.

• Characteristics: Ordered data, equal intervals, and a true zero point.
• Empirical Relation: Equivalence, Order, known differences, and known ratios.
• Meaningful Operations: All arithmetic operations (addition, subtraction, multiplication, division), calculation of ratios, mean, median, mode, standard deviation, geometric mean.
• Software Example: Lines of Code (LOC) (0 LOC means no code), execution time (0 seconds means no time elapsed), memory usage (0 bytes means no memory used), number of defects (0 defects means none found), effort in person-hours (0 hours means no effort), cost in currency (0 means no cost).

Absolute Scale

This scale is used when the measurement involves simply counting the number of elements in a set. The measurement value is unique and corresponds directly to the count.

• Characteristics: Measurement is a count of elements.
• Empirical Relation: Based on the structure of the elements being counted.
• Meaningful Operations: All arithmetic operations applicable to counts.
• Software Example: Number of modules in a system, number of developers on a team, count of test cases executed. It’s essentially a ratio scale where the unit of measure is inherently fixed (the count itself).

Choosing the correct scale is vital. Applying statistical methods designed for ratio scales (like calculating the average) to ordinal data (like severity rankings) can lead to nonsensical results (e.g., an ‘average severity’ of 2.5 doesn’t clearly map back to the categories). Always consider the scale type when analyzing and reporting metrics.