5.1 Descriptive Statistics & Measures of Dispersion

5.1 Descriptive Statistics & Measures of Dispersion

Quick Summary: Descriptive statistics provide cost technicians with the quantitative foundation required to summarize historical project performance, benchmark labor productivity rates, evaluate contractor bid spread, and isolate anomalies in cost tracking. While measures of central tendency (arithmetic mean, weighted mean, median, mode) define the expected midpoint of a dataset, measures of dispersion (range, variance, standard deviation, coefficient of variation) quantify operational uncertainty and volatility. Understanding the mathematical distinction between sample statistics (using $n - 1$ degrees of freedom) and population parameters is a core competency tested on the AACE CCT examination.

---

1. The Role of Statistics in Total Cost Management (TCM)

In capital project execution, cost engineers rarely deal with single deterministic numbers. An estimate of concrete placement labor is not a fixed certainty; it represents a distribution of historical productivities subject to weather, site congestion, craft skill, and managerial efficiency. In accordance with AACE International's Total Cost Management (TCM) Framework, descriptive statistics serve several critical functions:

• Historical Normalization: Converting raw accounting records from past projects into standardized unit rates (e.g., labor hours per linear foot of pipe or cost per square meter of gross floor area).
• Risk and Contingency Analysis: Quantifying historical cost variance to determine appropriate probabilistic contingency reserves rather than relying on arbitrary percentage markups.
• Procurement & Bid Analysis: Evaluating competitive tender submissions against historical benchmarks to detect irregular bidding patterns, unbalancing, or commercial errors.
• Quality & Productivity Control: Tracking craft productivity and unit cost trends across reporting periods to identify negative variances before project budgets are breached.

---

2. Measures of Central Tendency

Measures of central tendency identify the single central value that best describes the center or typical value of a probability distribution or dataset.

+-----------------------------------------------------------------------------------+
|                         MEASURES OF CENTRAL TENDENCY                             |
+-------------------+---------------------------------------------------------------+
| Arithmetic Mean   | Sum of all observations divided by the total number of items. | 
|                   | Heavily distorted by extreme values or unrepresentative data. |
+-------------------+---------------------------------------------------------------+
| Median            | The physical middle observation in an ordered array (50th     |
|                   | percentile). Highly resistant (robust) to extreme outliers.   |
+-------------------+---------------------------------------------------------------+
| Mode              | The most frequently occurring value in the dataset. The only  |
|                   | measure of central tendency applicable to nominal/categorical.|
+-------------------+---------------------------------------------------------------+
| Weighted Mean     | Averages data where individual observations carry unequal     |
|                   | relative importance, hours, volume, or statistical weights.   |
+-------------------+---------------------------------------------------------------+

The Arithmetic Mean: Sample vs. Population

The arithmetic mean represents the mathematical center of gravity of a dataset. In cost engineering, practitioners must strictly differentiate between a population parameter (when every unit in an entire universe is measured) and a sample statistic (when an estimator analyzes a subset of historical projects to infer properties of future projects):

• Population Mean ($\mu$): $\mu = \frac{\sum_{i=1}^{N} X_i}{N}$ Where $N$ is the total number of items in the entire population.

• Sample Mean ($\bar{x}$): $\bar{x} = \frac{\sum_{i=1}^{n} X_i}{n}$ Where $n$ is the number of observations in the sample.

The Median

The median is the value that splits an ordered data array into two equal halves. To calculate the median, observations must first be arranged in ascending numerical order:

• If the sample size $n$ is odd, the median is the value at position $\frac{n + 1}{2}$.
• If the sample size $n$ is even, the median is the arithmetic mean of the two central values located at positions $\frac{n}{2}$ and $\frac{n}{2} + 1$.

Cost Engineering Significance: Project cost data frequently exhibits strong positive skewness due to occasional catastrophic cost overruns, supplier bankruptcies, or massive legal claims. In such datasets, the arithmetic mean is pulled upward toward the tail, presenting an inflated estimate of typical performance. The median remains unaffected by extreme outliers and therefore often reflects a more realistic operational baseline.

The Mode

The mode is the data value that appears with the greatest frequency. A distribution may have no mode (all values occur equally), a single mode (unimodal), two modes (bimodal), or multiple modes (multimodal). In estimating, a bimodal distribution often signals mixed underlying populations—for instance, combining open-shop and union labor rates, or conflating domestic and imported material supply costs into a single unsegregated dataset.

The Weighted Mean (Composite Rates)

When historical observations contribute unequally to the total cost object, a simple arithmetic average introduces severe mathematical bias. The weighted mean ($\bar{x}_w$) weights each value by its corresponding volume, hours, or relative quantity: $\bar{x}_w = \frac{\sum_{i=1}^{k} (w_i \cdot X_i)}{\sum_{i=1}^{k} w_i}$ Where $w_i$ represents the weight or volume associated with value $X_i$.

Worked Practical Example (Composite Crew Wage Rate): A mechanical contractor constructs a specialized piping crew consisting of:

• 1 Working Piping Foreman earning $58.00/hr
• 4 Journeyman Pipefitters earning $46.00/hr
• 2 Certified Pipe Welders earning $52.00/hr
• 1 Apprentice Pipefitter earning $28.00/hr

To compute the composite hourly crew labor rate ($8$ total labor hours per crew hour): $\bar{x}_w = \frac{(1 \times 58.00) + (4 \times 46.00) + (2 \times 52.00) + (1 \times 28.00)}{1 + 4 + 2 + 1} = \frac{58.00 + 184.00 + 104.00 + 28.00}{8} = \frac{374.00}{8} = \$46.75 \text{ per craft hour}$

If the technician had improperly calculated an unweighted arithmetic average of the four wage classifications, the result would have been $(58 + 46 + 52 + 28) / 4 = $46.00$, understating true hourly labor cost.

---

3. Measures of Dispersion and Variability

Central tendency describes where data concentrates, but dispersion describes how spread out the values are around that center. Two cost distributions can have identical means of $100,000, yet one may range from $95,000 to $105,000 (low risk, predictable) while the other ranges from $40,000 to $220,000 (extreme uncertainty, high risk).

The Range

The range ($R$) is the simplest measure of dispersion, defined as the algebraic difference between the highest and lowest observed values: $R = X_{\text{max}} - X_{\text{min}}$ While intuitive, the range utilizes only two extreme values from the dataset, making it acutely sensitive to anomalies and providing zero information regarding the distribution of values between the extremes.

Variance

Variance measures the average squared deviation of individual data points from their mean. By squaring the deviations $(X_i - \bar{x})^2$, negative differences do not cancel out positive differences, and larger deviations are penalized exponentially.

• Population Variance ($\sigma^2$): $\sigma^2 = \frac{\sum_{i=1}^{N} (X_i - \mu)^2}{N}$

• Sample Variance ($s^2$): $s^2 = \frac{\sum_{i=1}^{n} (X_i - \bar{x})^2}{n - 1}$

Degrees of Freedom and Bessel's Correction ($n - 1$)

A critical concept on the CCT exam is why the sample variance formula divides by $n - 1$ rather than $n$. When using the sample mean $\bar{x}$ as a proxy for the unknown true population mean $\mu$, the sample data points are naturally closer to $\bar{x}$ than they are to $\mu$. Dividing by $n$ produces a systematically downward-biased estimate of the population variance. Dividing by $n - 1$ (Bessel's correction) mathematically corrects this bias, rendering $s^2$ an unbiased estimator of $\sigma^2$. The denominator $n - 1$ represents the degrees of freedom ($df$)—the number of independent observations remaining after calculating the sample mean.

Standard Deviation

Because variance is expressed in squared units (e.g., dollars squared, $$^2$), it cannot be directly compared to original project metrics. The standard deviation is the positive square root of the variance, restoring the metric to the original units of measure (dollars, hours, tons):

• Population Standard Deviation ($\sigma$): $\sigma = \sqrt{\sigma^2} = \sqrt{\frac{\sum_{i=1}^{N} (X_i - \mu)^2}{N}}$

• Sample Standard Deviation ($s$): $s = \sqrt{s^2} = \sqrt{\frac{\sum_{i=1}^{n} (X_i - \bar{x})^2}{n - 1}}$

Coefficient of Variation (CV)

When comparing the dispersion of two datasets with drastically different monetary magnitudes (e.g., comparing a $50,000 painting package against a $20,000,000 structural turbine package), standard deviation alone is misleading. The turbine package will naturally have a much larger standard deviation in absolute dollars, but the painting package may exhibit greater percentage volatility.

The Coefficient of Variation (CV), also called relative dispersion, normalizes the standard deviation relative to the mean, yielding a unitless ratio or percentage: $CV = \frac{s}{\bar{x}} \quad (\text{or } CV\% = \frac{s}{\bar{x}} \times 100\%)$

Analytical Takeaway: Site dewatering carries six times the relative cost uncertainty of structural steel ($30%$ vs. $5%$), highlighting where estimating contingency percentages must be concentrated.

---

4. Visualizing Cost Data & Outlier Detection

Descriptive statistics rely heavily on graphical representations to validate data distribution before running parametric estimating models or cost regression.

Histograms & Frequency Polygons

• Histogram: A graphical bar chart displaying the frequency distribution of continuous cost data grouped into discrete, non-overlapping intervals (bins). The vertical axis represents absolute frequency or relative frequency (percentage), while the horizontal axis represents the cost or productivity intervals.
• Frequency Polygon: A line graph created by connecting the midpoints of the histogram bin tops. It is particularly useful for overlaying multiple historical distributions to compare crew performance across consecutive project years.

Box-and-Whisker Plots & The Five-Number Summary

A box-and-whisker plot summarizes a dataset through five key boundary points known as the Five-Number Summary:

1. Minimum ($X_{\text{min}}$): The lowest non-outlier value.
2. First Quartile ($Q_1$): The 25th percentile (25% of data falls below this point).
3. Median ($Q_2$): The 50th percentile.
4. Third Quartile ($Q_3$): The 75th percentile (75% of data falls below this point).
5. Maximum ($X_{\text{max}}$): The highest non-outlier value.

               OUTLIER          Q1       MEDIAN      Q3                OUTLIER
                  *         |---------|----------|---------|              *
              |<------------|         |    |     |         |------------>|
             MIN            |---------|----------|---------|            MAX
                            <-------------IQR------------->

The Interquartile Range (IQR) and Outlier Identification Rule

The Interquartile Range (IQR) represents the spread of the middle 50% of the observations: $IQR = Q_3 - Q_1$ Under standard AACE and statistical auditing conventions, data points are flagged as suspected outliers if they fall beyond the inner fences: $\text{Lower Fence} = Q_1 - 1.5 \times IQR$ $\text{Upper Fence} = Q_3 + 1.5 \times IQR$

Values beyond these thresholds indicate atypical occurrences—such as catastrophic equipment failure, severe design rework, or accounting entry errors—and should generally be excluded from normalized historical cost estimating databases.

---

5. Comprehensive Worked Mathematical Example: Unit Concrete Placement Costs

Scenario: A cost technician gathers historical direct installation costs per cubic yard for ready-mix structural concrete across five recently completed bridge footing projects: $\$115, \quad \$120, \quad \$130, \quad \$135, \quad \$150$

Calculate the sample mean, median, range, sample variance, sample standard deviation, and coefficient of variation.

Step 1: Calculate the Sample Mean ($\bar{x}$)

$\bar{x} = \frac{\sum X_i}{n} = \frac{115 + 120 + 130 + 135 + 150}{5} = \frac{650}{5} = \$130.00 \text{ per cu yd}$

Step 2: Determine Median and Range

• The data is already ordered: $115, 120, 130, 135, 150$.
• Since $n = 5$ is odd, the median is the value at position $(5 + 1)/2 = 3$, which is $130.00/cu yd.
• Range: $R = X_{\text{max}} - X_{\text{min}} = $150 - $115 = $35.00/cu yd$.

Step 3: Compute Deviations and Sum of Squared Deviations

Check: The sum of deviations from the mean must always equal exactly zero.

Step 4: Calculate Sample Variance ($s^2$)

Applying Bessel's correction with $n - 1 = 5 - 1 = 4$ degrees of freedom: $s^2 = \frac{\sum (X_i - \bar{x})^2}{n - 1} = \frac{750}{4} = 187.50 \text{ (in } \$^2/\text{yd}^6)$ (Notice: If this were improperly calculated as a population variance, the denominator would be $5$, yielding an understated variance of $150.00$).

Step 5: Calculate Sample Standard Deviation ($s$)

$s = \sqrt{s^2} = \sqrt{187.50} \approx \$13.693 \text{ per cu yd}$

Step 6: Calculate Coefficient of Variation ($CV$)

$CV = \frac{s}{\bar{x}} = \frac{13.693}{130.00} \approx 0.1053 \quad (10.53\%$

---

6. Exam Watch: High-Yield Traps & Rules of Thumb

[!WARNING] The "$n$ vs. $n - 1$" Denominator Trap: The single most common computational error on the CCT exam is dividing by $n$ when asked for sample variance or sample standard deviation. If a test question mentions "a sample of 8 projects" or provides a small table of recorded tasks, you must divide the sum of squared deviations by $n - 1 = 7$. Only divide by $N$ if the question explicitly states that the data represents the entire population of all historical records ever executed.

[!CAUTION] The Skewness Direction Trap: When cost distributions are positively skewed (skewed to the right by large overruns), remember the mathematical sequence: $\text{Mode} < \text{Median} < \text{Mean}$. The mean is pulled farthest into the right tail. Never assume that mean, median, and mode are interchangeable in skewed cost environments.

[!TIP] Sum of Deviations Check: In any manual table calculation of variance, immediately sum the raw deviations $(X_i - \bar{x})$. If this sum does not equal exactly zero, you have made an arithmetic error in calculating your mean—stop and correct it before squaring the numbers!