# The “Flaw of Averages” in Cost Estimating

{% hint style="info" %}
This is a living document. It is maintained and updated without notice. It's best to refresh pages to ensure delivery of the latest iteration from the CDN Content Delivery Network in your area
{% endhint %}

**Square-foot estimating**—quoting a job by a single cost per square foot—is a convenient early-stage method, but research shows it can be misleading. A key issue is **economies of scale**: larger projects often have lower unit costs than smaller ones. For example, an NAHB analysis found that the median cost **per square foot** of new homes *declines systematically as home size increases*. Homes under 1,200 sq. ft. had a median cost around $200/sq.ft., whereas homes 5,000 sq. ft. or larger averaged only about $132/sq.ft. This trend occurs because certain fixed costs (design, setup, etc.) are spread over more square footage in bigger projects, lowering the unit price. **If one were to apply a single average $/sq.ft. rate across all sizes, the errors could be dramatic**—using the cost per sq.ft. of a 3,000 ft² home to estimate a 1,500 ft² home would *“drastically underestimate”* the smaller home’s total cost. In short, a one-size-fits-all square-foot price fails to account for scale differences, leading to **systematic underestimation on small projects and overestimation on large ones**.

*Median sale price per square foot vs. home size (NAHB data). Smaller homes cost significantly more per sq.ft. than larger homes, illustrating economies of scale. Basing estimates on a single average $/sq.ft. for all sizes would skew high for big houses and dangerously low for small ones.*

Beyond size, **design and scope variables** undermine square-foot rates. An electrical contracting study noted that *“bidding by square foot…fails to account for critical variables”*—such as differing numbers of fixtures, wiring complexity, or commercial vs. residential requirements—making the method inherently flawed. In practice, two projects of equal area can have vastly different components and difficulty. **Using an average $/ft² ignores these specifics**, so important cost drivers (specialty systems, custom features, etc.) get averaged out. The result is often **erroneous bids**—contractors who rely solely on square-foot pricing *“risk underbidding… or worse—going out of business”* due to the inaccuracies. Industry guidance, therefore, pegs rough **square-foot estimates as only ±20% accurate** at best, suitable for preliminary budgets but too variable for final pricing. In summary, studies and industry experts agree that square-foot pricing **must be used with caution**. It should be limited to very concept-stage estimates or projects that truly mirror the average prototype. Otherwise, averaging cost per area across diverse conditions will yield unreliable numbers.

### Time-Based Pricing and Production Rate Variability

Estimating based on **average labor hours or production rates** faces similar reliability problems. Contractors often use historical productivity rates (e.g., square feet painted per hour, linear feet painted per hour, or per day) as a baseline. The **flaw is in assuming that those averages hold constant** regardless of project-specific nuances. Research published by The Estimating Edge observes that *estimators often plug in “historical data or rough averages, resulting in inaccurate labor projections.”* Labor requirements can **“fluctuate significantly”** with factors like job complexity, crew skill, or site conditions. In other words, the actual hours required may deviate significantly from a simplistic average, especially if the current project differs from the historical average.

Real-world data underscores the variability. In one field study of new-home painting (where conditions were kept fairly consistent), **a small 1,336 sq.ft. house took 54.5 labor hours** to paint, while a **45% larger house (1,937 sq.ft.) took only 50 hours**. The larger home was painted *faster* despite its size—likely due to simpler layout or efficiencies of scale—whereas the small home had more labor-intensive detail per square foot. **Using a single average production rate (sq. ft./hour) for both** would badly miss the mark. If an estimator averaged these cases to roughly **34 sq.ft. per hour**, that rate would be *“worthless”*—it would **underestimate the small home’s hours by \~27% and overestimate the large home’s by 13–16%**. Indeed, calculating the painting hours using the small house’s slower rate led to a **58% overestimation** for the big house, whereas using the big house’s faster rate caused a **37% underestimation** on the small one . This example demonstrates how averaging production rates across different conditions yields large errors in *both* directions. Each scenario had its own productivity curve, and the average of all three fell in a no-man’s-land that **matched none of them**.

Such variability is not unique to painting. Construction productivity studies show that labor output rates can be highly inconsistent and even **non-normally distributed**. Radosavljevic and Horner’s analysis of 12 productivity datasets revealed “complex variability”—productivity did *not* cluster tightly around a single mean, and the variance was so large as to defy normal statistical assumptions. In practical terms, this means using a simplistic average productivity can be misleading; there may be multiple modes (e.g., “fast-track” vs. “slow/challenging” jobs) or heavy-tail effects where occasional outlier projects skew the mean. The data suggests **averages often conceal the true complexity** of labor performance. Without adjusting for specific project conditions—crew experience, site layout, weather, and so on—an average production rate gives a false sense of certainty. It might be right “on average” but very wrong for any given case. This is why experienced estimators incorporate *range estimates, contingency, or adjustments* for difficulty rather than relying blindly on a single historical average. As one industry article put it, focusing only on a past average rate **“ignores job complexity”** and local factors, whereas a detailed breakdown by task and crew often proves far more accurate.

### The “Flaw of Averages” in Cost Estimating

The problems above exemplify a broader principle known in risk analysis as the **“flaw of averages.”** Simply stated, *“plans based on assumptions about average conditions usually go wrong.”* When you use average inputs, you implicitly assume an average scenario—but real projects vary, and the average scenario itself is rare. In construction estimating, this flaw shows up in many ways. One vivid example is the use of **average unit prices** (e.g., a statewide average cost per unit of work) without accounting for context. Civil engineers Gransberg et al. note that departments of transportation often base estimates on the **average low bid prices** from past projects, but *“the problem with using statewide average unit prices is that the mean value is influenced by outliers, and prices are not correlated to project location.”* Location differences alone can swing costs dramatically—for instance, one state’s data showed certain regions consistently had nearly double the unit cost for asphalt paving compared to others due to terrain and short work seasons. In that case, relying on a single state-wide average price would wildly underprice work in high-cost regions (and overprice it in low-cost areas). The study highlights how incorporating specific variables (like geospatial cost data) leads to much more reliable estimates. In general, whenever estimators **average across unlike conditions—whether project size, location, design complexity, or market timing—they introduce a systematic error.** The average blurs the distinctions that drive actual cost or duration outcomes.

The **“flaw of averages”** reminds us that averaging is not forecasting. A classic anecdote is the statistician who drowned in a river *“that was on average 3 feet deep.”* In construction terms, an estimate based on average assumptions will be too low half the time and too high the other half—and rarely dead-on. Moreover, when things go wrong, they tend to go **really** wrong (for example, underestimating a complex project by 30% can erase profit or incur a loss, whereas overestimating by 30% might mean losing the bid entirely). Thus, **formal studies urge caution in using simple averages.** Instead of one-number simplifications, estimators are encouraged to use range estimates, adjust for known factors, or employ data-driven models that capture variability. For instance, rather than a blanket productivity rate, one can maintain **tables of production rates** by project type or difficulty level or use **simulation techniques** to account for uncertainty. Modern cost models (such as those by RSMeans or DOTs with big data) often segment data by key drivers (size, quality level, region, etc.) specifically to avoid the pitfalls of a broad average. The goal is to **condition the estimate on the project’s particulars** rather than a generic mean. By doing so, estimators significantly improve reliability and reduce the risk of over/underestimation that stems from “averaging across different conditions.” In summary, extensive research and industry experience confirm that averaging methods—be it square-foot pricing, time-per-task rates, or unit costs—**carry inherent variability and often systematic bias.** They are quick budgeting tools, not precision instruments. Without careful adjustments, the use of averages can mislead estimators into consistent errors, proving the adage that *plans built on averages are, on average, wrong*.

I’ve found it striking how many insights came from painting thousands of new-construction homes over the last forty years. New construction teaches lessons the residential repaint market never could. When the substrate, products, methods, and standards are consistent, you can finally **see what’s really driving the results**. What remains are the **variables that actually govern production**. Such an environment reveals mathematical and production truths that the complexities of residential repaint work, along with its numerous condition variables, often obscure. This is why **so many contractors end up with mystery margins**. They’re the natural outcome of estimating systems that average away the variables that actually control time. I’ve raised these issues directly with others who promote the use of averages in the painting industry. The problem isn’t that they’re unaware of the limitations of averages—**it’s that they continue to promote them after those limitations have been clearly demonstrated.** That isn’t oversight. **It’s willful ignorance**. Observing field data over forty years in any trade *is* a controlled study. Not because variables are eliminated, but because they’re allowed to exist naturally. **Jobs are permitted to be what they actually are**, with all of their variation intact. That’s as close to reality as trade work gets. You don’t simulate conditions. You don’t flatten outcomes. You observe what repeatedly happens when real work meets real constraints, year after year, across thousands of jobs.

### References

* Paul Emrath, *“Economies of Scale in Single-family Home Construction,”* NAHB Eye on Housing (Sept. 2024)—Analysis of cost per square foot vs. home size.
* Douglas D. Gransberg *et al.*, *“How to Visualize the Cost Impact of Design Decisions,”* Civil Engineering (ASCE Magazine, Mar. 2024)—Discusses the flaw of using statewide average unit prices and the importance of context-specific data.
* Jack Pauhl, *"The Square Footage Lie,"* (2025)—Field data from new-home painting exposes errors from averaging production rates.
* Dave Chapman, *“Five Biggest Errors Construction Estimators Make,”* Estimating Edge (June 2025) – Notes that relying on rough average labor rates leads to inaccurate estimates.
* Sam L. Savage, *“The Flaw of Averages,”* Harvard Business Review (Nov. 2002)—Explains why plans based on average assumptions often fail.
* *“The Square Foot Myth: Debunking a Common Pricing Practice,”* Best Bid Estimating (2023) warns that square-foot pricing ignores key variables, causing underbids.

### **Related Reading**

If this sounds familiar, it’s because the same flaw shows up in square-foot pricing. [*The Square Footage Lie*](https://jackpauhl.gitbook.io/archive/field-notes/business-strategy/the-square-footage-lie) walks through field data from new-construction painting and shows how estimating systems built on linear assumptions break down even under controlled conditions.