Halstead's Software Science

Overview

Halstead's Software Science (Maurice Halstead, 1977) was one of the first attempts to treat software quantitatively, measuring the "size"/complexity of a program based purely on the number and types of operators and operands in the code.

Basic measures (defined from source code)

• n1 = number of distinct operators (e.g., +, −, =, if, while, function calls, keywords).
• n2 = number of distinct operands (variables, constants).
• N1 = total number of operator occurrences.
• N2 = total number of operand occurrences.

Derived measures

1. Vocabulary: $n = n_1 + n_2$ (total distinct operators + operands).
2. Program length: $N = N_1 + N_2$ (total operators + operands occurrences).
3. Estimated length: $\hat N = n_1 \log_2 n_1 + n_2 \log_2 n_2$.
4. Volume: $V = N \log_2 n$ — bits needed to represent the program (measures size).
5. Difficulty: $D = \dfrac{n_1}{2} \times \dfrac{N_2}{n_2}$ — reflects how error-prone/hard to understand the program is.
6. Effort: $E = D \times V$ — effort to implement/understand (measured in elementary mental discriminations).
7. Time (seconds): $T = \dfrac{E}{18}$ (with the Stroud number, an empirically derived constant).
8. Bugs (estimated): $B = \dfrac{V}{3000}$ (estimated number of defects, if desired).

How it helps assess software quality

• Size measure independent of language — volume $V$ can compare programs written in different languages.
• Complexity indicator — difficulty $D$ and effort $E$ flag modules that are hard to understand/maintain and more likely to contain defects.
• Estimation — length and effort estimates help predict implementation effort and testing effort.
• Maintainability — high $V$/$D$/$E$ ⇒ a module should be simplified, split, or given extra review/testing.

Limitations

• Measures are implementation-based (need code), not requirements-based.
• The empirical constants (18, 3000) are approximate and not universally accepted.
• Interpreters/tools differ on what counts as an operator, reducing consistency; it ignores control-flow structure quality to some degree.

Summary table

Symbol	Measure	Meaning
n1, n2	Distinct operators/operands	Vocabulary
N1, N2	Operator/operand occurrences	Length
V = N log₂ n	Volume	Size in bits
D = (n1/2)(N2/n2)	Difficulty	Error-proneness
E = D·V	Effort	Implementation/understanding effort
T = E/18	Time	Estimated implementation time

Summary: Halstead's Software Science quantifies program size, difficulty, effort, and (estimated) defects from operator/operand counts, giving objective metrics to gauge complexity and quality/comparability of software.