9.3 PERT Three-Point Estimating & Probabilistic Duration Variance

9.3 PERT Three-Point Estimating & Probabilistic Duration Variance

Deterministic Critical Path Method (CPM) scheduling assumes that activity durations can be estimated as single, fixed point values. However, in complex engineering projects, research and development (R&D), first-of-a-kind technological deployments, and severe weather environments, activity durations are inherently uncertain.

To account for this variability, the Program Evaluation and Review Technique (PERT) was developed by the U.S. Navy Special Projects Office in 1958 for the Polaris missile program. PERT utilizes a probabilistic three-point estimating model to calculate expected project completion dates, standard deviations, and the statistical probability of meeting contractual delivery milestones.

For Certified Cost Professional (CCP) candidates, understanding PERT mathematical formulations, Central Limit Theorem aggregation, Z-score transformations, and the structural limitations of PERT (such as Merge Bias) is vital.

---

1. The PERT Three-Point Estimating Model

PERT models the probability distribution of an activity's duration using a unimodal Beta distribution. Estimators provide three distinct duration estimates for each activity:

+-----------------------------------------------------------------------------+
|                        THE PERT BETA PROBABILITY DISTRIBUTION               |
|                                                                             |
|   Probability Density f(t)                                                  |
|          ^                                                                  |
|          |                 Mode (Most Likely = M)                           |
|          |                    *                                             |
|          |                   / .                                            |
|          |                  /   .     Expected Mean (te)                    |
|          |                 /     .    |                                     |
|          |                /       .   v                                     |
|          |               /         .                                        |
|          |              /           .                                       |
|          |   *---------/             .--------------------*                 |
|          +---+------------------------+-------------------+------------> t  |
|             Optimistic                                Pessimistic           |
|                (O)                                        (P)               |
|                                                                             |
|          |<---------------- Total Range = P - O ------------------>|        |
|          |                 (Approximates 6 Sigmas)                 |
+-----------------------------------------------------------------------------+

The Three Time Estimates:

1. Optimistic Duration (O or a): The minimum possible duration required to complete the activity assuming ideal operating conditions, zero unforeseen delays, and maximum labor/equipment productivity (typically representing the 1st or 5th percentile of the probability distribution).
2. Most Likely Duration (M or m): The modal duration representing the highest point of the probability density function—the duration that would occur most frequently under normal, standard operating conditions.
3. Pessimistic Duration (P or b): The maximum possible duration required to complete the activity assuming extremely adverse conditions, severe delays, and worst-case execution (excluding major force majeure or cataclysmic events, typically representing the 95th or 99th percentile).

---

2. Core PERT Mathematical Formulations

PERT assumes that the total spread (P - O) encompasses approximately 6 standard deviations (6σ, corresponding to 99.73% coverage under a normal distribution). The expected mean is weighted toward the mode (M) with a weight of 4:

+-----------------------------------------------------------------------------+
|                       PERT BETA DISTRIBUTION FORMULAS                       |
|                                                                             |
|   1. Expected Activity Duration:    te = (O + 4M + P) / 6                   |
|                                                                             |
|   2. Activity Standard Deviation:   σ = (P - O) / 6                         |
|                                                                             |
|   3. Activity Variance:             σ² = [ (P - O) / 6 ]² = (P - O)² / 36   |
+-----------------------------------------------------------------------------+

Comparison: PERT Beta vs. Triangular Distribution

In conceptual cost estimating and Monte Carlo simulation software, cost engineers frequently encounter the Triangular Distribution as an alternative three-point model:

---

3. Project-Level Statistical Aggregation (Central Limit Theorem)

To determine the overall project duration distribution, PERT relies on the Central Limit Theorem (CLT) from probability theory:

The Central Limit Theorem in PERT: Regardless of the individual probability distributions of individual activities (whether Beta, Triangular, or skewed), the sum of a sufficiently large number of independent random variables (n >= 4) along the critical path converges toward a Normal (Gaussian) Distribution.

+-----------------------------------------------------------------------------+
|                    PROJECT-LEVEL PERT STATISTICAL FORMULAS                  |
|                                                                             |
|   1. Project Expected Mean Duration:                                        |
|      μ_proj = E[T_proj] = Σ te,i  (Summed over ALL critical path activities)|
|                                                                             |
|   2. Project Critical Path Variance:                                        |
|      σ²_proj = Σ σ²_i             (Summed over ALL critical path activities)|
|                                                                             |
|   3. Project Standard Deviation:                                            |
|      σ_proj = √(σ²_proj) = √( Σ σ²_i )                                      |
+-----------------------------------------------------------------------------+

[!IMPORTANT] Cardinal Exam Rule: Variances are Additive; Standard Deviations are NOT!

• Correct: Compute the variance σ_i² for each critical activity, add all variances together to find σ_proj², and then take the square root: σ_proj = √(Σ σ_i²).
• Fatal Exam Mistake: Never sum the individual standard deviations directly (σ_proj != Σ σ_i). Summing standard deviations violates probability theory and grossly overstates project variance.

---

4. Probability of On-Time Completion & Z-Score Calculations

Once the project mean (μ) and standard deviation (σ_proj) are established, the cost engineer can calculate the probability of completing the project on or before any scheduled contractual target date (Ts).

+-----------------------------------------------------------------------------+
|                           THE Z-SCORE TRANSFORMATION                        |
|                                                                             |
|   Z = (Target Scheduled Date - Project Expected Mean) / Project Std Dev     |
|                                                                             |
|   Z = (Ts - μ_proj) / σ_proj                                                |
|                                                                             |
|   Probability of On-Time Completion = P(T <= Ts) = Φ(Z)                     |
+-----------------------------------------------------------------------------+

Standard Normal Distribution Table for Cost Engineers:

Calculating the Target Date for a Desired Confidence Level:

To determine what contractual delivery date (Ts) guarantees a specific confidence level (e.g., 95% confidence where Z = 1.645): Ts = μ_proj + Z * σ_proj

---

5. The Merge Bias Phenomenon in Complex Networks

While PERT provides a powerful statistical foundation, it contains a significant structural limitation known as Merge Bias.

+-----------------------------------------------------------------------------+
|                          THE MERGE BIAS MECHANISM                           |
|                                                                             |
|   Path 1:  (Mean = 40 days, P(T1 <= 42) = 90%) ────────┐                   |
|                                                        ├──> [ Merge Node ]  |
|   Path 2:  (Mean = 40 days, P(T2 <= 42) = 90%) ────────┘                   |
|                                                                             |
|   Deterministic PERT Evaluation:                                            |
|   Evaluates only Path 1 -> Reports 90% Probability at 42 Days.              |
|                                                                             |
|   Actual Joint Probability (Merge Bias):                                    |
|   P(Project <= 42) = P(T1 <= 42) * P(T2 <= 42) = 0.90 * 0.90 = 81.0%        |
+-----------------------------------------------------------------------------+

Why PERT Underestimates Project Risk:

1. Focus on a Single Critical Path: Classical PERT evaluates variance along only the single deterministic critical path, completely ignoring non-critical paths.
2. Parallel Converging Paths: In reality, when multiple near-critical paths converge at a common milestone (merge node), the project milestone cannot be reached until all converging paths finish:
   T_merge = max(Path 1, Path 2, ..., Path k)
3. Joint Probability Multiplication: Assuming independent paths, the actual probability of completing the project on or before Ts is the product of all path probabilities: P(T_proj <= Ts) = Product( P(Path_j <= Ts) ) Even if two parallel paths each have a 90% chance of on-time completion, the joint probability that both finish on time is 0.90 * 0.90 = 81.0%. If four parallel paths converge, 0.90^4 = 65.6%.

Modern Solution: Monte Carlo Schedule Risk Analysis

To overcome Merge Bias, modern cost engineering employs Monte Carlo Simulation (Quantitative Schedule Risk Analysis / QSRA). Monte Carlo samples all activity distributions simultaneously across thousands of iterations, computing the Criticality Index (the percentage of iterations in which a specific activity falls on the critical path).

---

6. Comprehensive Step-by-Step Worked PERT Case Study

Case Study Scenario: An engineering team prepares a probabilistic schedule for an offshore gas processing skid. The critical path consists of four sequential activities (A → B → C → D):

Step 1: Calculate Individual Activity Expected Durations & Variances

1. Activity A:

   • t_e,A = (4 + 4(7) + 16) / 6 = (4 + 28 + 16) / 6 = 48 / 6 = 8.0 days
   • σ_A = (16 - 4) / 6 = 12 / 6 = 2.0 days → σ_A² = 2.0² = 4.0 days²

2. Activity B:

   • t_e,B = (9 + 4(12) + 21) / 6 = (9 + 48 + 21) / 6 = 78 / 6 = 13.0 days
   • σ_B = (21 - 9) / 6 = 12 / 6 = 2.0 days → σ_B² = 2.0² = 4.0 days²

3. Activity C:

   • t_e,C = (5 + 4(11) + 17) / 6 = (5 + 44 + 17) / 6 = 66 / 6 = 11.0 days
   • σ_C = (17 - 5) / 6 = 12 / 6 = 2.0 days → σ_C² = 2.0² = 4.0 days²

4. Activity D:

   • t_e,D = (8 + 4(14) + 20) / 6 = (8 + 56 + 20) / 6 = 84 / 6 = 14.0 days
   • σ_D = (20 - 8) / 6 = 12 / 6 = 2.0 days → σ_D² = 2.0² = 4.0 days²

---

Step 2: Calculate Overall Project Statistical Parameters

• Project Expected Mean Duration (μ_proj): μ_proj = Σ t_e = 8.0 + 13.0 + 11.0 + 14.0 = 46.0 days
• Project Critical Path Variance (σ_proj²): σ_proj² = Σ σ² = 4.0 + 4.0 + 4.0 + 4.0 = 16.0 days²
• Project Standard Deviation (σ_proj): σ_proj = √(16.0) = 4.0 days

---

Step 3: Probabilistic Evaluation of Contractual Milestones

1. Question 1: What is the probability of completing the project within 50 days?

   • Calculate Z-score: Z = (Ts - μ) / σ_proj = (50 - 46) / 4.0 = 4.0 / 4.0 = +1.00
   • From the standard normal table, Φ(+1.00) = 0.8413 or 84.13%.

2. Question 2: What is the probability of completing the project within 42 days?

   • Calculate Z-score: Z = (42 - 46) / 4.0 = -4.0 / 4.0 = -1.00
   • From symmetry: Φ(-1.00) = 1 - Φ(+1.00) = 1 - 0.8413 = 0.1587 or 15.87%.

3. Question 3: What completion date (Ts) guarantees a 95% probability of on-time completion?

   • For a 95% confidence level, Z = +1.645. Ts = μ + Z * σ_proj = 46.0 + (1.645 * 4.0) = 46.0 + 6.58 = 52.58 days ≈ 52.6 days