Your schedule says the substation is energized in week 35. Each of the three teams feeding into that date is 80% sure of making week 40. That sounds like week 40 is safe. In fact the chance of making it's about one in two.
Merge bias is the delay that builds up where parallel paths meet. A milestone that waits for several paths starts only when the last of them finishes. Each path can look comfortable on its own while the milestone is anything but. A schedule built on single durations can't show this, which is why it underestimates how late a project with many parallel paths will finish.
Why parallel paths make you late
Take two paths that each have an 80% chance of finishing by week 40. For the milestone to be on time, both must make it. If they run independently, that chance is 80% times 80%: 64%. Add a third path and it drops to about 51%. The milestone carries the bad luck of every path, while the good luck of one path is wasted waiting for the others.
The US Government Accountability Office describes this in its Schedule Assessment Guide: the risk at a merge point depends on the risk on each path, the float before the milestone and the number of paths that merge there. More paths means more chances that one of them runs late.
Worked example: the Northfield substation
Illustrative case. The project and figures are fictional, built to show the effect.
A grid operator plans a new substation. Energization can only happen once three parallel paths are complete. Each path gets a minimum, most likely and maximum duration in weeks:
| Path to energization | Min | Most likely | Max | Chance of week 40 |
|---|---|---|---|---|
| Civil works and building | 30 | 34 | 46 | 81% |
| Transformer: manufacture, delivery, installation | 29 | 33 | 47 | 81% |
| Protection, control and grid connection | 32 | 35 | 45 | 81% |
The deterministic schedule uses the most likely durations. The protection and grid connection path is critical, and energization lands in week 35. The other two paths have one and two weeks of float. On paper, the project looks healthy.
A Monte Carlo schedule risk analysis tells a different story (200,000 iterations):
| Energization | Week | Chance |
|---|---|---|
| Deterministic date | 35 | 4% |
| Week each path is 80% sure of | 40 | 53% |
| P50 | 39.8 | 50% |
| P80 | 42.3 | 80% |
| P90 | 43.4 | 90% |
The planned week 35 has a 4% chance. Week 40, which every team on its own would sign up to, is a coin toss. And to be 80% sure of energization, the project needs to plan for week 42 to 43, more than two weeks beyond what any single path suggests.
The more paths, the worse it gets
The effect grows with every path that merges. Here, each path is the same civil works range (30, 34 and 46 weeks), and on its own has an 81% chance of finishing by week 40:
| Paths merging | Chance of week 40 | P80 of the milestone (week) |
|---|---|---|
| 1 | 81% | 39.8 |
| 2 | 66% | 41.5 |
| 3 | 54% | 42.3 |
| 4 | 44% | 42.8 |
| 5 | 35% | 43.1 |
Energy and infrastructure projects are full of these points: energization, commissioning, a track possession, a road closure, a handover. Each one waits for design, permits, long-lead equipment, civil works and third parties at the same time.
What about correlation?
Paths that share a cause tend to run late together: the same contractor, the same permit, the same winter. That actually softens merge bias a little, because the paths are less likely to be late at different times. With a correlation of 0.5 between the three Northfield paths, the chance of week 40 rises from 53% to 63%. The P80 barely moves: 41.9 instead of 42.3 weeks. Correlation doesn't make merge bias go away; it changes which risks you should look at.
What to do about it
- Run a schedule risk analysis on the real network. Merge bias only shows up when the logic is simulated, not when durations are added up. A schedule risk analysis reads the date straight from the simulated finish dates.
- Count the merge points. List the milestones that wait for three or more paths. Those are where the schedule is most optimistic.
- Watch the near-critical paths. A path with a week of float isn't safe if its range is wide. Track paths by criticality, not only by float.
- Reduce the merge. Where you can, take paths off the merge point: deliver equipment earlier, split a commissioning milestone, or pre-test what can be tested before the others are ready.
- Put the margin where it belongs. A schedule margin in front of the merge milestone protects the date better than padding each path, for the same reason a pooled cost reserve beats contingency spread over work packages.
Frequently asked questions
Does PERT account for merge bias?
No. Classic PERT calculates the expected duration of the critical path only and ignores the parallel paths that can overtake it. That's why it tends to be optimistic for schedules with many merge points. A Monte Carlo simulation of the full network does capture it.
Is merge bias the same as the critical path moving?
They are two sides of the same effect. Because each path can be the slowest in a given run, the critical path changes from run to run, and the milestone finishes later than any single path would predict.
Does this only apply to schedules?
Mostly. Costs add up, so they don't merge in the same way. But delay costs money: when a merge milestone slips, crews, site overheads and equipment wait with it. See how to calculate contingency with Monte Carlo for the cost side.
Quantil's Schedule Risk Analyzer Pro simulates your real CPM network in Excel, so merge bias is in the result: P50, P80 and P90 finish dates, with risks linked to activities and a Primavera P6 import. Try the free demo.
References
- US Government Accountability Office (2015). GAO-16-89G, Schedule Assessment Guide: Best Practices for Project Schedules, section "Merge Bias and Schedule Underestimation".
- AACE International. Recommended Practice 57R-09: Integrated Cost and Schedule Risk Analysis Using Monte Carlo Simulation of a CPM Model.