THERP (Technique for Human Error Rate Prediction)

Description

Decomposes a task into an event tree of human actions, assigns tabled error probabilities with performance shaping factors and dependence, and computes overall task failure probability.

THERP is the original full-scale human reliability analysis. Each task is broken into steps represented as binary success/failure branches in an HRA event tree. Basic human error probabilities come from data tables in the Swain and Guttmann handbook, adjusted for performance shaping factors such as stress, experience and interface quality. Crucially, THERP models dependence between steps (zero, low, moderate, high, complete), recognising that a person who misreads one display is likely to misread the next. Recovery factors such as independent checks are included. The method yields an overall failure probability and, through sensitivity analysis, shows where a design change buys the most reliability. It is laborious but uniquely explicit about dependence and recovery.

Attributes

Loss type
Unmigrated
Effort
Weeks

Steps

  1. Conduct a task analysis and identify the human actions relevant to the failure of interest.
  2. Build an HRA event tree with success and failure branches for each action, in sequence.
  3. Assign nominal error probabilities from THERP tables or empirical data, with uncertainty bounds.
  4. Adjust for performance shaping factors and apply the dependence model between successive actions.
  5. Add recovery branches (self-check, second-person check, system alerts) and compute the end-state failure probability.
  6. Run sensitivity analysis to identify which steps and factors dominate risk, and redesign accordingly.

Sources

  • Swain & Guttmann, Handbook of Human Reliability Analysis with Emphasis on Nuclear Power Plant Applications (NUREG/CR-1278, 1983)
  • Kirwan, A Guide to Practical Human Reliability Assessment (1994)

Also known as

  • THERP
  • Swain and Guttmann method

Categories

  • Accessibility
  • Human Factors & Error