Reference Class Forecasting
An estimation method that bases predictions on actual outcomes of similar past projects rather than the specifics of the current plan.
Also known as: Outside View, Base Rate Forecasting, RCF
Category: Decision Science
Tags: estimation, planning, decision-making, cognitive-biases, project-management, statistics
Explanation
Reference Class Forecasting (RCF) is a method for making predictions by looking at what actually happened in a class of similar past projects, rather than building an estimate from the inside out based on the specifics of the current plan.
## The problem it solves
Most project estimates are built 'inside out' - decomposing the project into tasks, estimating each task, and summing them up. This approach systematically produces overoptimistic results because:
- Planners focus on the unique features of their project
- They underweight the base rate of how similar projects actually performed
- Optimism bias and anchoring skew individual task estimates
- Unforeseen problems are, by definition, not included in the plan
Daniel Kahneman calls this the **inside view** versus the **outside view**.
## How it works
1. **Identify a reference class** - Find a set of past projects that are relevantly similar to the current one
2. **Collect outcome data** - Gather actual durations, costs, or other metrics from these past projects
3. **Establish a distribution** - Determine the statistical distribution of outcomes in the reference class
4. **Position the current project** - Place the current project within this distribution, adjusting for any genuinely distinctive features
## Example
If you're estimating a new e-commerce platform:
- Look at 20 similar e-commerce builds from the past 5 years
- Find that the average took 14 months (not the planned 8-10)
- Find that 90% completed within 20 months
- Use 14 months as your baseline estimate rather than your bottom-up calculation of 9 months
## Key insight
People consistently believe their project is special and will avoid the problems that plagued others. Reference class forecasting forces confrontation with the base rate of actual outcomes.
## Applications beyond software
RCF was developed by Bent Flyvbjerg for large infrastructure projects (bridges, tunnels, railways) where cost overruns of 50-300% are common. It has since been adopted in:
- Government project planning (mandated in the UK since 2004)
- IT project estimation
- Pharmaceutical drug development timelines
- Construction budgeting
## Limitations
- Requires access to historical data from comparable projects
- Defining the right reference class can be subjective
- Truly novel projects may have no valid reference class
- Does not replace detailed planning, but rather serves as a reality check on it
Related Concepts
← Back to all concepts