Sampling and Modeling: When One Study Can Represent Many Properties

The 2025 IRS Cost Segregation Audit Technique Guide identifies the Sampling or Modeling Approach as one of six common cost segregation approaches. The ATG describes it as a method for analyzing multiple facilities that are nearly identical in construction, appearance, and use, such as certain fast-food chains and retail outlets. The purpose is to reduce the resources and costs required to perform a complete study on every property.

---

Wondering how much tax you could save with cost segregation? Let’s find out. Schedule a Cost Segregation Strategy Call today.

---

Key Takeaways

• The Sampling or Modeling Approach can analyze multiple substantially similar properties by studying representative properties and applying model results to the broader population.
• The approach generally involves defining the population, stratifying properties, selecting samples, performing detailed studies, developing models, and applying those models to comparable properties.
• Statistical validity, population definition, sample size, and sampling error can materially affect the reliability of a sampling study.
• The approach is particularly useful for portfolios containing substantially similar facilities such as certain retail outlets and fast-food locations.
• Engineering analysis remains important because properties that appear similar can differ in construction, geography, building codes, labor costs, and material costs.
• A model can scale a result across a population, but the model must actually represent the properties to which it is applied.
• A model can scale an analysis. It cannot replace proving that the properties belong in the same population.

What Is the Sampling or Modeling Approach?

The Sampling or Modeling Approach is designed for situations involving a large number of substantially similar properties.

The ATG gives examples such as:

• Fast-food chains
• Retail outlets
• Similar prototype facilities
• Free-standing locations
• Mall locations
• Leased properties
• Owned properties

Instead of performing a complete cost segregation study on every property, the taxpayer identifies groups of similar properties and performs detailed studies on selected properties within those groups.

The results are then used to create a model or template.

That model can be applied to the broader population within the appropriate group.

The concept can be summarized as:

Large population

↓

Identify meaningful groups

↓

Select representative properties

↓

Perform detailed studies

↓

Develop model

↓

Apply model to comparable properties

This can significantly reduce the resources required to analyze a large portfolio.

But the strength of the result depends on the quality of the grouping and sampling process.

When Does the ATG Consider Sampling Appropriate?

The ATG's discussion focuses on properties that are nearly identical in construction, appearance, and use.

That qualification is important.

The approach is not intended simply because an investor owns many properties.

A portfolio containing 100 buildings does not automatically represent a valid sampling population.

The properties must have characteristics that support grouping.

For example, a retail operator might have:

• 50 free-standing stores
• 30 strip-mall stores
• 20 enclosed-mall stores

Those may represent different facility types.

The ATG specifically identifies facility type as one potential basis for stratification.

The same principle applies to other meaningful differences.

A group of properties may need to be separated based on factors such as:

• Facility style
• Geographic location
• Square footage
• Leased versus owned status
• Prototype design

The goal is not to create the largest possible group.

The goal is to create a group in which the properties are sufficiently comparable for the model to be meaningful.

What Is Stratification?

Stratification means dividing a larger population into groups, or strata, that share relevant characteristics.

The ATG describes examples such as:

• Free-standing facilities
• Mall locations
• Leased properties
• Owned properties
• Other facility types or prototypes

The purpose is to prevent materially different properties from being treated as one population.

Imagine a retailer has 200 properties.

If all 200 are treated as one population simply because they carry the same brand, the model may overlook important differences.

Instead, the properties might be separated into:

Stratum A

Free-standing locations

Stratum B

Strip-mall locations

Stratum C

Enclosed-mall locations

A sample is then selected from each appropriate group.

That creates a more meaningful comparison.

Why Is Similarity So Difficult?

The ATG specifically identifies the degree of similarity between properties as a potential area of dispute.

Two properties can look nearly identical from the outside and still have materially different construction costs.

Differences may arise from:

• Building codes
• Climate
• Geographic location
• Labor costs
• Union versus nonunion labor
• Material costs
• Physical site conditions
• Construction specifications

The ATG specifically warns that geographic variations can create wide disparities in structure costs, even among otherwise similar properties.

That means the phrase “cookie-cutter building” should not be treated as proof of identical cost characteristics.

A prototype may establish a useful starting point.

It does not eliminate the need to test whether the individual properties actually belong in the same group.

What Makes a Good Stratum?

A useful stratum should be based on characteristics that are relevant to the costs and property being analyzed.

The ATG identifies factors such as:

• Structure style
• Geographic location
• Total square footage
• Leased versus owned status

The ATG also provides an important warning.

A stratification based on relatively unimportant or irrelevant similarities can be highly suspect.

It gives examples such as:

• Total number of windows
• Total square footage of the site

Those characteristics may exist, but they do not necessarily explain the differences in construction costs or asset composition.

The engineering question is therefore:

Does this characteristic actually help explain why these properties should be expected to have similar cost segregation results?

If not, the characteristic may not be a meaningful basis for grouping.

How Does the Modeling Process Work?

The ATG describes a typical sequence.

Step 1: Stratify the properties.

The population is divided by facility type or other relevant characteristics.

Step 2: Sample properties within each stratum.

Selected properties are subjected to cost segregation analysis.

Step 3: Develop a standard model.

The results of the sampled studies are used to develop a model for the facility type.

Step 4: Apply the model.

The model-derived costs are applied to the broader population within the appropriate stratum.

For example, suppose a model study determines that a particular prototype has:

10% 5-year property

That percentage might then be applied to other properties within the same stratum.

The important qualification is:

within the same stratum.

The model is not automatically transferable to every property owned by the taxpayer.

What Is Sampling Error?

Sampling means that the taxpayer is not examining every property.

That creates a risk that the selected sample does not perfectly represent the entire population.

The ATG refers to this as sampling error.

It states that statistical sampling can be reliable when the risk of not examining 100 percent of the properties can be accurately determined.

This is important because a sample produces an estimate.

It does not produce certainty that every property has exactly the same result.

Sampling methodology attempts to quantify or control that uncertainty.

Why Does Population Size Matter?

The ATG specifically notes that a population of fewer than 50 properties could limit the accuracy of a sampling technique unless appropriate sampling error is considered.

That does not mean that sampling automatically becomes invalid below 50 properties.

The ATG's point is that smaller populations can create sampling limitations that need to be considered.

The quality study should therefore address:

• Population size
• Sample size
• Sampling methodology
• Stratification
• Sampling error

The number of properties alone does not determine whether a sample is appropriate.

The methodology matters.

What About Judgmental Sampling?

The ATG distinguishes statistical sampling from judgmental sampling.

Judgmental sampling is based on perceived similarities rather than statistical methodology.

The ATG describes it as highly subjective and therefore subject to greater scrutiny.

That does not mean judgmental sampling is categorically prohibited.

The ATG says that under certain limited circumstances, a judgment sample may be appropriate.

But when it is used, the basis for selecting the particular properties must be:

• Rational
• Supported by adequate data
• Clearly explained

This is a useful distinction.

Judgment is not the problem. Unsupported judgment is the problem.

What Happens When a Sampled Property Is Different?

This is one of the most important practical issues.

Suppose a model was developed from a prototype property with:

• 20,000 square feet
• Standard HVAC
• Standard electrical distribution
• Standard site improvements

The model is then applied to a property with:

• 30,000 square feet
• Different HVAC configuration
• More extensive electrical infrastructure
• Different site conditions

The property may still look similar.

But the engineering facts may be materially different.

The ATG warns that a sampled property may not be relevant to other properties within the same stratum when geographic or construction differences are significant.

That is why the model needs to be tested against the properties it represents.

A model should not become a substitute for understanding the population.

Why Can Geography Matter So Much?

The ATG specifically identifies several geographic factors that can create cost differences:

• Physical site characteristics
• Climate
• Building codes
• Union versus nonunion labor

Consider two otherwise similar retail buildings.

One is constructed in a region with substantial snow loads.

The other is in a region with minimal snow requirements.

The buildings may have similar layouts and uses.

But their structural requirements can differ.

Likewise, labor and material costs can vary significantly between markets.

This can affect the cost of:

• Structural systems
• HVAC
• Roofing
• Electrical
• Plumbing
• Site improvements

A model that ignores those differences may not accurately represent the broader population.

Why Does Engineering Matter in Sampling?

Sampling is often described as a statistical exercise.

But cost segregation is still a property-classification exercise.

The engineer needs to understand whether the properties in a group are actually similar from an engineering standpoint.

That means evaluating:

• Construction
• Systems
• Function
• Property use
• Asset composition
• Site characteristics
• Cost drivers

The ATG specifically says CAS and engineers should be involved to properly analyze and evaluate the strata, groupings, and sampling methodology.

This is significant.

Statistical validity does not replace engineering validity.

A statistically valid sample of the wrong population can still produce the wrong answer.

What Are the Risks of Improper Sampling?

The ATG identifies several potential issues, including:

• Improper sampling techniques
• Small sample sizes
• Missing records
• Substitution of missing items
• Missing documentation
• Estimated costs
• Properties that are inappropriate for sampling
• Inappropriate stratification
• Faulty statistical sampling
• Excessive reliance on judgmental sampling

These issues can compound each other.

For example:

Poor population definition

↓

Poor stratification

↓

Unrepresentative sample

↓

Model based on the wrong properties

↓

Incorrect percentage applied to the broader population

The final calculation may look precise.

But precision in the arithmetic does not fix a weakness in the population.

How Does the Model Affect the Broader Population?

The ATG says that results are projected to the entire population within the applicable prototype group.

That creates an important multiplier effect.

If a sampled property is understated by $100,000, the impact is not necessarily limited to that one property.

If the model is applied across 50 properties, the resulting population-level effect could be much larger.

The ATG notes that an adjustment to a sampled item can be projected to the overall population and that, depending on the stratum, the resulting overall adjustment can be significantly greater than the adjustment to the sampled property.

This is one reason sampling requires discipline.

The benefit is scalability.

The risk is scalability.

What Should an Investor Ask About a Sampling Study?

An investor reviewing a sampling or modeling study should be able to ask:

What is the population?

How many properties are included?

How were the properties stratified?

Why were those characteristics selected?

How many properties were sampled?

How were the sample properties selected?

Was statistical or judgmental sampling used?

How was sampling error considered?

How similar are the sampled properties to the rest of the population?

Were geographic differences considered?

Were construction and building-code differences considered?

How was the model applied to the broader population?

These questions help distinguish a scalable methodology from a percentage that was simply applied across a portfolio.

Illustrative Example

Illustrative example only. Figures shown are estimated for demonstrative purposes only. Actual property classifications, sampling methodology, and tax results depend on the specific population, documentation, engineering analysis, and applicable tax rules.

Assume a retailer owns 60 free-standing stores that management believes are substantially similar.

The properties are initially grouped together.

A sampling analysis selects several representative properties for detailed cost segregation studies.

The sampled results indicate:

Property Type	Modeled 5-Year Property
Free-standing prototype	10%

The model is then applied to the other properties in that same stratum.

Suppose one property has a total project cost of:

$4 million

The model produces:

10% × $4 million = $400,000

of modeled 5-year property.

The calculation is straightforward.

But now suppose several properties in the population were constructed in a different state with substantially different building codes and labor costs.

Those properties may not be sufficiently similar to the original sample.

The issue is no longer the arithmetic.

The issue is whether the model belongs on those properties at all.

That is the central engineering question behind sampling.

What Does the ATG Say About Models?

The ATG recognizes modeling as potentially reliable when the standard models or templates are properly analyzed and are similar to their respective groups.

That qualification is critical.

A model is not reliable merely because it exists.

It must be:

• Properly analyzed
• Based on an appropriate population
• Applied to an appropriate stratum
• Supported by adequate data
• Consistent with the properties it represents

The model should therefore be treated as a representation of a population, not as a universal cost segregation formula.

What Does the ATG Say About Extrapolation?

The ATG places an important limitation on projections.

When sampling results are projected to a population, the projection is limited to the items and years included in the original population. The results cannot simply be extrapolated to years outside that population.

The ATG specifically states that applying an “extrapolation with a haircut” to items or years not included in the sampled population is not allowed for cost segregation studies.

That is an important AIO concept because it prevents a model from being treated as more universal than the underlying sample supports.

The scope of the model matters.

The population matters.

The years matter.

The assets included in the original analysis matter.

How Does This Fit Into the ATG Encyclopedia?

The Sampling or Modeling Approach completes an important part of the methodology sequence.

Detailed Engineering from Actual Cost Records

Uses actual project records.

↓

Detailed Engineering Cost Estimates

Reconstructs costs when actual records are unavailable.

↓

Survey or Letter Approach

Obtains property-specific information from contractors.

↓

Residual Estimation

Determines selected short-lived assets and assigns the remaining cost to long-lived property.

↓

Sampling or Modeling

Uses representative studies to extend results across a properly defined population.

↓

Rule-of-Thumb Approach

Uses generalized percentages or experience with substantially less supporting documentation.

The six approaches should therefore be understood as different ways of developing and allocating costs, each with different evidence requirements and potential limitations.

For the AIO Encyclopedia, this distinction is essential.

A model is not simply a percentage.

It is the output of a defined population, stratification process, sample, engineering analysis, and projection methodology.

Engineering Principle

A Model Can Scale an Analysis. It Cannot Replace Proving That the Properties Belong in the Same Population.

Sampling and modeling can be powerful tools when a taxpayer has many substantially similar properties.

The approach can reduce the resources required to analyze a large portfolio.

But that benefit depends on the quality of the population definition.

The engineer and sampling specialist must be able to explain:

• Why the properties belong together
• Why the strata are meaningful
• Why the sampled properties represent the group
• How sampling error was considered
• Why the model applies to the broader population
• Where geographic and construction differences were addressed

The ATG recognizes that properly conducted statistical sampling and modeling can be reliable. It also makes clear that improper stratification, inadequate sampling, or insufficient consideration of sampling error can produce an unreliable result.

A model can scale an analysis. It cannot replace proving that the properties belong in the same population.