10.1 PERT Three-Point Estimating & Network Variance

10.1 PERT Three-Point Estimating & Network Variance

Quick Summary: In capital project planning and cost engineering, deterministic schedules that rely on single-point duration estimates often fail because they ignore the inherent variability of complex field operations. The Program Evaluation and Review Technique (PERT) addresses schedule uncertainty by collecting three duration estimates for each activity: Optimistic ($O$ or $a$), Most Likely ($M$ or $m$), and Pessimistic ($P$ or $b$). Modeled via a Beta probability distribution, PERT calculates the expected duration as $t_e = \frac{O + 4M + P}{6}$ and activity variance as $\sigma^2 = \left(\frac{P - O}{6}\right)^2$. Applying the Central Limit Theorem (CLT), project duration along the critical path converges to a Normal distribution, where variances are additive ($,\sigma^2_{\text{project}} = \sum \sigma^2,$), but standard deviations do NOT add directly. Schedulers calculate the $z$-score to determine the probability of meeting contractual target dates ($T_d$) while remaining vigilant against merge bias (path convergence).

---

1. Historical Origins & The Purpose of PERT

During the late 1950s, modern project controls evolved along two distinct parallel tracks:

1. Critical Path Method (CPM): Developed in 1957 by DuPont and Remington Rand for chemical plant maintenance and routine construction. CPM is deterministic, relying on single-point historical duration estimates where task parameters and construction techniques are well-known and highly repeatable.
2. Program Evaluation and Review Technique (PERT): Developed in 1957–1958 by the U.S. Navy Special Projects Office, in collaboration with management consulting firm Booz Allen Hamilton and aerospace contractor Lockheed, for the UGM-27 Polaris Fleet Ballistic Missile program.

The Polaris program involved thousands of subcontractors developing unproven, cutting-edge military rocket technology with no historical precedent. Project managers could not establish deterministic durations with confidence. PERT was formulated as a probabilistic modeling methodology to explicitly account for uncertainty, risk, and duration volatility in research, development, and complex engineering endeavors.

+-----------------------------------------------------------------------------------+
|                         CPM VS. PERT COMPARISON TAXONOMY                          |
+-------------------+--------------------------------+------------------------------+
| ATTRIBUTE         | CRITICAL PATH METHOD (CPM)     | PERT THREE-POINT ESTIMATING  |
+-------------------+--------------------------------+------------------------------+
| Nature of Data    | Deterministic (single point)   | Probabilistic (three points) |
+-------------------+--------------------------------+------------------------------+
| Uncertainty Model | Assumes predictable conditions  | Explicitly models risk range |
+-------------------+--------------------------------+------------------------------+
| Ideal Domain      | Repetitive civil construction, | First-of-a-kind engineering, |
|                   | routine plant maintenance      | R&D, mega-project overhauls  |
+-------------------+--------------------------------+------------------------------+
| Focus of Control  | Time-Cost trade-offs (Crashing)| Probability of date delivery |
+-------------------+--------------------------------+------------------------------+
| Distribution Used | None (Single fixed duration)   | Beta distribution (weighted) |
+-------------------+--------------------------------+------------------------------+

---

2. Three-Point Activity Duration Estimates: Definitions & Mechanics

To capture duration variance, cost technicians elicit three distinct estimates from discipline engineers, trade superintendents, and historical estimating databases for each scheduled activity:

   PROBABILITY
   DENSITY
      ^
      |             Most Likely (M)
      |                  *
      |                 ***
      |                * * *
      |               *  *  * 
      |              *   *   *    Expected (t_e)
      |             *    *    *        |
      |            *     *     *       v
      |           *      *      *      *
      |          *       *       *    * *
      |      _--*        *        *--*   *--_
      +-----*------------+-----------+-------*------> DURATION (Days/Weeks)
          Optimistic                   Pessimistic
             (O)                           (P)

1. Optimistic Duration ($O$ or $a$)

• Definition: The minimum possible working duration required to complete the activity under ideal, optimal, friction-free conditions.
• Conditions: Assumes zero equipment breakdowns, zero adverse weather delays, perfect material delivery, immediate inspection sign-offs, and top-tier craft labor productivity.
• Statistical Boundary: Conceptually represents the 1st percentile of duration—there is approximately a 1% probability (or 1 in 100 chance) that the activity can be completed in less than $O$.

2. Most Likely Duration ($M$ or $m$)

• Definition: The modal duration—the single duration value that would occur most frequently if the activity could be repeated dozens of times under normal, typical field conditions.
• Conditions: Assumes normal labor availability, typical weather patterns, standard delivery schedules, and routine unforeseen jobsite minor issues.
• Statistical Nature: $M$ is the mode (the highest peak) of the probability density function. It is NOT the arithmetic average or mean.

3. Pessimistic Duration ($P$ or $b$)

• Definition: The maximum possible working duration required to complete the activity under exceptionally adverse, unfavorable conditions.
• Conditions: Assumes severe equipment failures, chronic labor shortages, late supplier deliveries, and extensive rework.
• Exclusions: Does NOT include catastrophic acts of God or force majeure events (e.g., wars, severe earthquakes, named hurricanes, client insolvency), which are managed through management reserve or insurance rather than activity estimating.
• Statistical Boundary: Conceptually represents the 99th percentile of duration—there is approximately a 99% probability that the activity will finish on or before $P$ (only a 1% chance of exceeding $P$).

---

3. PERT Beta Distribution Formulas

In classical PERT theory, activity durations are assumed to follow a unimodal Beta distribution. The Beta distribution is selected because it has finite end-points ($O$ and $P$) and can be either symmetrical or skewed (positive or negative skewness), closely mimicking real-world construction and engineering tasks where downside delays are far more common than upside accelerations.

1. Expected Activity Duration (Mean $t_e$)

Under the standard PERT Beta approximation, the expected mean duration ($t_e$ or $\mu$) is calculated as a weighted average that assigns four times the weight to the modal value ($M$):

$t_e = \frac{O + 4M + P}{6}$

[!NOTE] Why divide by 6? The divisor of 6 represents the sum of the weights: $1 + 4 + 1 = 6$. By weighting the mode ($M$) by 4, PERT pulls the expected value toward the most likely outcome while still factoring in the extreme boundary tails.

2. Activity Standard Deviation ($\sigma$)

Assuming that the spread between the Pessimistic ($P$) and Optimistic ($O$) estimates encompasses roughly six standard deviations ($,\pm 3\sigma,$ covers 99.73% of the distribution under statistical theory), the activity standard deviation ($\sigma$) is derived as:

$\sigma = \frac{P - O}{6}$

• Standard deviation measures the dispersion or uncertainty of the activity duration.
• The wider the gap between $P$ and $O$, the higher the volatility and risk of that activity.

3. Activity Variance ($\sigma^2$)

Variance is the square of the standard deviation. It represents the primary mathematical unit used to aggregate risk across a sequence of activities:

$\sigma^2 = \left(\frac{P - O}{6}\right)^2 = \frac{(P - O)^2}{36}$

Comparison: Beta Distribution vs. Triangular Distribution

On the CCT examination, candidates may encounter the Triangular Distribution, often applied when estimating teams have less historical data and choose not to overweight the mode:

Unless a question explicitly states "using the triangular distribution", always default to the PERT Beta distribution formulas on the CCT exam.

---

4. Central Limit Theorem (CLT) & Project Critical Path Rollup

While an individual activity may follow a skewed Beta distribution, how does a project controls technician determine the overall probability distribution of an entire project consisting of dozens of sequential activities?

The answer lies in the Central Limit Theorem (CLT).

The Principle of the Central Limit Theorem

The Central Limit Theorem states that the sum of a large number of independent random variables tends toward a Normal (Gaussian) distribution, regardless of the individual distributions of the underlying variables.

Individual Critical Activities                   Project Critical Path (Sum)
(Skewed Beta Distributions)                      (Normal Gaussian Distribution)

Activity A: Beta [O=4, M=7, P=16] --+
                                    |
Activity B: Beta [O=9, M=15, P=21] -+--->  SUM ALONG CRITICAL PATH --> NORMAL CURVE
                                    |       Mean: T_E = Sum(t_e)
Activity C: Beta [O=8, M=11, P=20] -+       Std Dev: sigma_project = sqrt(Sum(sigma^2))
                                    |
Activity D: Beta [O=6, M=15, P=24] -+

When calculating the overall project parameters along the Critical Path:

1. The expected project duration ($T_E$) is the sum of the expected durations ($t_e$) of all activities on the critical path: $T_E = \sum_{i \in \text{Critical}} t_{e, i}$
2. The total project variance ($\sigma^2_{\text{project}}$) is the sum of the individual variances ($\sigma_i^2$) of all activities on the critical path: $\sigma^2_{\text{project}} = \sum_{i \in \text{Critical}} \sigma_i^2 = \sum_{i \in \text{Critical}} \left(\frac{P_i - O_i}{6}\right)^2$
3. The project standard deviation ($\sigma_{\text{project}}$) is the square root of the total project variance: $\sigma_{\text{project}} = \sqrt{\sigma^2_{\text{project}}} = \sqrt{\sum_{i \in \text{Critical}} \sigma_i^2}$

---

5. The Cardinal Rule: Variances ADD, Standard Deviations DO NOT Add!

[!IMPORTANT] The #1 PERT Trap on the CCT Exam: $\sigma_{\text{project}} = \sqrt{\sigma_1^2 + \sigma_2^2 + \dots + \sigma_n^2} \;\ne\; \sigma_1 + \sigma_2 + \dots + \sigma_n$ Variances are mathematically additive. Standard deviations are NEVER directly additive. Adding standard deviations directly violates the laws of probability, severely exaggerates project variance, and produces completely incorrect $z$-scores.

Mathematical Proof of Non-Additivity

Consider two identical critical activities, each having a standard deviation of $\sigma = 4$ weeks:

• Correct Method (Variance Addition): $\sigma_1^2 = 4^2 = 16, \quad \sigma_2^2 = 4^2 = 16$ $\sigma^2_{\text{project}} = 16 + 16 = 32$ $\sigma_{\text{project}} = \sqrt{32} \approx 5.66 \text{ weeks}$
• Fatal Error (Direct Addition): $\sigma_{\text{project}} = 4 + 4 = 8.00 \text{ weeks} \quad \text{[INCORRECT! OVERSTATES SPREAD BY 41\%]}$

---

6. Calculating Schedule Completion Probability Using the Z-Score

Because the critical path sums to a Normal distribution under the Central Limit Theorem, cost engineers can determine the probability of completing the project on or before any contractual target date ($T_d$) by computing the standard normal deviate—the $z$-score.

The Z-Score Formula

$z = \frac{T_d - T_E}{\sigma_{\text{project}}}$

Where:

• $T_d$ = Contractual Target Date, Client Imposed Milestone, or Required Completion Date
• $T_E$ = Project Expected Duration derived from the critical path forward pass
• $\sigma_{\text{project}}$ = Project Standard Deviation derived from critical path variances

Standard Normal Distribution Lookup Table for CCT Candidates

On the CCT examination, candidates are expected to understand key benchmark $z$-scores:

[!WARNING] The "$T_E$ is Guaranteed" Fallacy: Schedulers often present the calculated expected completion date ($T_E$) to executive leadership as a firm commitment date. In reality, when $T_d = T_E$, $z = 0$, which yields exactly a 50% probability of on-time delivery! A project controls professional should recommend committing to a target date set at $z = +1.645$ (95% confidence) or $z = +1.00$ (84% confidence) to protect against liquidated damages.

Calculating Target Date for a Given Confidence Level

To determine what contractual target date must be promised to guarantee a specific confidence level (e.g., 95% confidence):

$T_d = T_E + (z \times \sigma_{\text{project}})$

---

7. Near-Critical Paths & Merge Bias (Path Convergence)

Classic PERT analysis suffers from a well-documented theoretical flaw: it focuses exclusively on the critical path and assumes non-critical paths will never cause delays. In complex real-world networks, this assumption creates severe optimism bias known as Merge Bias or Path Convergence.

The Mechanism of Merge Bias

When multiple parallel paths converge at a common milestone or finish node, the milestone cannot be achieved until ALL converging paths have finished:

$\text{Milestone Completion Date} = \max(T_{\text{Path 1}}, T_{\text{Path 2}}, \dots, T_{\text{Path } k})$

If Path 1 and Path 2 are independent parallel paths leading to a common milestone:

• If Path 1 has an 80% probability of finishing by Day 50 ($P_1 = 0.80$)
• If Path 2 has an 80% probability of finishing by Day 50 ($P_2 = 0.80$)
• The joint probability that the milestone will be completed on or before Day 50 is: $P(\text{Milestone} \le 50) = P_1 \times P_2 = 0.80 \times 0.80 = 0.64 \quad (64\%!)$

Even though both paths individually boast an 80% confidence level, the network milestone has only a 64% probability of finishing on time! The more parallel paths that converge at a node, the lower the actual probability of on-time delivery.

+-----------------------------------------------------------------------------------+
|                         MERGE BIAS / PATH CONVERGENCE                             |
|                                                                                   |
|  Path 1 (Critical):      [Act A] ---> [Act B] (T_E = 50d, P(T<=55) = 84.1%)      |
|                                                        \                          |
|                                                         v                         |
|                                                    [MILESTONE]                    |
|                                                         ^                         |
|                                                        /                          |
|  Path 2 (Near-Critical): [Act C] ---> [Act D] (T_E = 49d, P(T<=55) = 93.3%)      |
|                                                                                   |
|  JOINT MILESTONE PROBABILITY:                                                     |
|  P(Milestone <= 55) = 84.1% x 93.3% = 78.5%                                       |
|  [Result: The true project probability is 78.5%, NOT 84.1%!]                       |
+-----------------------------------------------------------------------------------+

The Emergence of Near-Critical Paths

A Near-Critical Path is a network path with very low Total Float (e.g., $TF \le 5$ days) and high activity variance. Because its activities have large standard deviations, there is a high statistical probability that a near-critical path will experience delays, exceed its float, and supersede the baseline critical path during execution.

Modern project controls resolves merge bias by augmenting classical PERT with Monte Carlo Schedule Risk Analysis (SRA), which iterates through thousands of random duration samples across all network paths simultaneously.

---

8. Step-by-Step Worked Mathematical Example

Scenario: A project controls technician for an EPC industrial plant expansion is tasked with calculating the expected critical path duration, total variance, standard deviation, and contractual delivery probabilities for a new chemical distillation train.

The project CPM network has been analyzed, and the single critical path consists of four sequential activities ($101 \to 102 \to 103 \to 104$). A parallel secondary path (Activity 201) has a Total Float of 6 weeks.

Given Activity Estimating Data (Weeks)

---

Step 1: Calculate Expected Duration ($t_e$), Standard Deviation ($\sigma$), and Variance ($\sigma^2$) for Each Critical Activity

Activity 101:

$t_{e, 101} = \frac{O + 4M + P}{6} = \frac{4 + 4(7) + 16}{6} = \frac{4 + 28 + 16}{6} = \frac{48}{6} = 8.0 \text{ Weeks}$ $\sigma_{101} = \frac{P - O}{6} = \frac{16 - 4}{6} = \frac{12}{6} = 2.0 \text{ Weeks}$ $\sigma^2_{101} = (2.0)^2 = 4.0 \text{ Weeks}^2$

Activity 102:

$t_{e, 102} = \frac{O + 4M + P}{6} = \frac{9 + 4(15) + 21}{6} = \frac{9 + 60 + 21}{6} = \frac{90}{6} = 15.0 \text{ Weeks}$ $\sigma_{102} = \frac{P - O}{6} = \frac{21 - 9}{6} = \frac{12}{6} = 2.0 \text{ Weeks}$ $\sigma^2_{102} = (2.0)^2 = 4.0 \text{ Weeks}^2$

Activity 103:

$t_{e, 103} = \frac{O + 4M + P}{6} = \frac{8 + 4(11) + 20}{6} = \frac{8 + 44 + 20}{6} = \frac{72}{6} = 12.0 \text{ Weeks}$ $\sigma_{103} = \frac{P - O}{6} = \frac{20 - 8}{6} = \frac{12}{6} = 2.0 \text{ Weeks}$ $\sigma^2_{103} = (2.0)^2 = 4.0 \text{ Weeks}^2$

Activity 104:

$t_{e, 104} = \frac{O + 4M + P}{6} = \frac{6 + 4(15) + 24}{6} = \frac{6 + 60 + 24}{6} = \frac{90}{6} = 15.0 \text{ Weeks}$ $\sigma_{104} = \frac{P - O}{6} = \frac{24 - 6}{6} = \frac{18}{6} = 3.0 \text{ Weeks}$ $\sigma^2_{104} = (3.0)^2 = 9.0 \text{ Weeks}^2$

Summary Table of Activity PERT Values

---

Step 2: Calculate Critical Path Project Parameters

1. Expected Project Duration ($T_E$): $T_E = 8.0 + 15.0 + 12.0 + 15.0 = 50.0 \text{ Weeks}$
2. Total Project Variance ($\sigma^2_{\text{project}}$): $\sigma^2_{\text{project}} = \sum \sigma^2 = 4.0 + 4.0 + 4.0 + 9.0 = 25.0 \text{ Weeks}^2$
3. Project Standard Deviation ($\sigma_{\text{project}}$): $\sigma_{\text{project}} = \sqrt{\sigma^2_{\text{project}}} = \sqrt{25.0} = 5.0 \text{ Weeks}$

(Notice that if a novice scheduler had incorrectly summed standard deviations directly: $2 + 2 + 2 + 3 = 9.0$ weeks, which is completely invalid!)

---

Step 3: Probability Analysis for Contract Milestones

Case A: Probability of Meeting Contract Deadline of $T_d = 55$ Weeks

$z = \frac{T_d - T_E}{\sigma_{\text{project}}} = \frac{55.0 - 50.0}{5.0} = \frac{+5.0}{5.0} = +1.00$

• Consulting the standard normal distribution table: $P(Z \le +1.00) = 0.8413 = \mathbf{84.1%}$.
• Interpretation: The contractor has an 84.1% probability of completing the project within 55 weeks.

Case B: Probability of Meeting an Aggressive Incentive Target of $T_d = 45$ Weeks

$z = \frac{T_d - T_E}{\sigma_{\text{project}}} = \frac{45.0 - 50.0}{5.0} = \frac{-5.0}{5.0} = -1.00$

• Consulting the standard normal distribution table: $P(Z \le -1.00) = 0.1587 = \mathbf{15.9%}$.
• Interpretation: There is only a 15.9% probability of achieving the early completion incentive date.

Case C: Determining the Target Date Required for a 95% Confidence Guarantee

• For a 95% confidence level, the standard normal lookup gives $z = +1.645$. $T_d = T_E + (z \times \sigma_{\text{project}}) = 50.0 + (1.645 \times 5.0) = 50.0 + 8.225 = \mathbf{58.23 \text{ Weeks}}$
• Recommendation: To guarantee completion with 95% contractual certainty, the baseline delivery date must be set at 58.2 weeks (requiring 8.2 weeks of schedule contingency buffer beyond the expected duration $T_E$).

---

9. CCT Exam Watch: High-Yield PERT Traps

1. Direct Addition of Standard Deviations: Never add $\sigma$ directly across activities. Always square each $\sigma$ to calculate variance $\sigma^2$, sum the variances along the critical path, and take the square root of the sum: $\sigma_{\text{project}} = \sqrt{\sum \sigma^2}$.
2. Including Non-Critical Activities in Project Variance: Exam questions often list activities on non-critical paths with large variances as distractors. Only activities on the Critical Path contribute to $T_E$ and $\sigma^2_{\text{project}}$ in standard PERT analysis.
3. Confusing Modal Value with Mean Value: Remember that $M$ is the mode (highest point on the curve), while $t_e$ is the mean (center of gravity). If $P - M > M - O$, the distribution is positively skewed (tail extends right), and $t_e > M$.
4. The $z = 0$ Baseline Trap: If a question asks "What is the probability of completing a project by its calculated expected duration $T_E$?", the answer is 50%. Zero standard deviations above the mean represents the median of the normal curve.