Linear Modeling

The Central Idea

Suppose a delivery service charges a fixed fee of \$8 plus \$2.50 per mile.

Let

and

The fixed fee gives the intercept:

The per-mile rate gives the slope:

So the model is

This equation is useful because each parameter has a practical meaning:

• \(2.50\) dollars per mile is the rate of change;
• \$8 is the cost when distance is zero.

A good linear model turns a real situation into a mathematical relationship while preserving the meaning of its quantities.

What Is a Mathematical Model?

A mathematical model is a mathematical representation of a real or conceptual system.

A model may use:

• equations;
• functions;
• graphs;
• tables;
• algorithms;
• probability distributions;
• geometric or numerical structures.

A model simplifies reality so that important relationships can be studied more clearly.

What Is a Linear Model?

A linear or affine model in one input variable commonly has the form

In strict Malone Math terminology:

• \(y=mx\) is linear;
• \(y=mx+b\) with \(b\ne0\) is affine.

In modeling practice, both are often grouped under the familiar phrase linear model.

The Meaning of the Parameters

For

the slope

represents the constant change in output per unit change in input.

The intercept

represents the model's output when

These are not merely algebraic symbols. They should be interpreted in the context of the problem.

Modeling Begins with Variables

Before writing an equation, define the quantities.

For example:

Then a model such as

has a clear interpretation.

Without variable definitions and units, the same equation is much less informative.

Independent and Dependent Variables

In a model

the input \(x\) is usually treated as the independent variable.

The output \(y\) depends on it.

For a cost model, for example:

Then cost depends on quantity.

Units Are Part of the Model

Suppose

where distance is in miles and time is in hours.

Then the slope must have units

The intercept has units

Dimensional consistency helps verify that the model has been constructed correctly.

The Constant-Rate Assumption

The defining assumption of an affine model is that the rate of change is constant:

Equivalently,

If the real process does not maintain approximately constant rate over the relevant domain, a single linear model may be inappropriate.

Exact Models Versus Approximate Models

Some relationships are exactly affine by definition or design.

For example, a service may literally charge

Then

may be exact.

Other relationships are only approximately linear.

For noisy data, write

or describe the equation as a trend model.

Start with the Story, Not the Formula

Before choosing

ask:

• what does the input represent?
• what does the output represent?
• what causes or corresponds to change?
• is there a fixed starting amount?
• does a constant-rate assumption make sense?
• what values of the variables are realistic?

The equation should emerge from the structure of the situation.

Model from a Rate and Initial Value

A tank contains \(500\) liters and drains at \(40\) liters per minute.

Let

and

The slope is

The initial value is

Therefore,

Interpret the Tank Model

For

the slope

means the volume decreases by \(40\) liters each minute.

The intercept

is the volume at time zero.

Find the Natural Domain

The tank reaches zero when

So

A natural physical domain is

The algebraic line extends farther, but the physical model need not.

Model from One Observation and a Rate

Suppose a vehicle is \(140\) miles from a reference point at

hours and continues away at \(60\) miles per hour.

The observed point is

Use point-slope form:

Thus,

Why Point-Slope Form May Be Better for a Model

The form

keeps the known observation visible.

Expanding gives

Both are equivalent, but the first emphasizes measured information while the second exposes the intercept.

The best representation depends on what we want to communicate.

Model from Two Observations

Suppose a machine produces \(120\) units after \(3\) hours and \(240\) units after \(7\) hours.

Assume production is affine over that interval.

The slope is

Using \((3,120)\):

So

Interpret the Implied Intercept Carefully

The model

implies

That may mean \(30\) units had already been completed at time zero.

But if the observed data began later, the intercept may simply be an extrapolated feature of the affine model.

Do not assign practical meaning to an intercept automatically.

Model from a Table

Suppose:

The rate is

and

The constant rate supports an affine model.

Using \((1,45)\):

So

Check the Model Against Every Data Point

For

check:

All observations are reproduced exactly.

This supports the model for the data provided.

Model from Noisy Data

Suppose observations are:

The changes are close to \(2\) per input unit, but not exact.

A reasonable informal trend model might be

This equation summarizes the pattern rather than passing through every observation.

Residuals Measure Model Error

If the model predicts

and the observed value is

then the residual is

Residuals measure how far observations lie from the model.

Example: Residual Check

Using

at \(x=3\) the model predicts

If the observed value is

then

The observed point lies \(0.2\) units above the model.

Residual Pattern Matters More Than One Residual

One small residual does not prove the model is good.

Inspect residuals across the data.

If they remain small and show no obvious structure, a linear model may be reasonable.

If they form a curved pattern, the model may be missing nonlinear behavior.

Graph the Data and the Model Together

A graph helps reveal whether the proposed line captures the data pattern.

Look for:

• points scattered close to the line;
• systematic curvature;
• outliers;
• changes in rate;
• thresholds or discontinuities.

Graphical checking complements algebraic checking.

Modeling with Standard Form

Not every useful linear model needs to isolate one variable immediately.

Suppose product \(x\) requires \(2\) labor hours and product \(y\) requires \(5\) labor hours.

If exactly \(100\) labor hours are used, then

This standard-form relation models a resource constraint.

Interpret the Constraint Intercepts

For

if

then

If

then

The intercepts describe extreme production combinations under the resource limit.

Equality Versus Inequality Constraints

The equation

means exactly \(100\) labor hours are used.

The inequality

means no more than \(100\) labor hours may be used.

The first describes a boundary line; the second describes a feasible region.

Modeling Thresholds

Suppose a subscription costs

where \(m\) is months.

If the budget is at most \$150, solve

Then

The linear model becomes a decision tool.

Break-Even Modeling

Suppose revenue is

and cost is

Break-even occurs when

So

Profit Region

Profit is positive when

Thus,

The equality locates the threshold; the inequality identifies the favorable region.

Comparing Competing Linear Models

Suppose two service plans are

and

Plan \(A\) starts higher but grows more slowly.

Plan \(B\) starts lower but grows faster.

Set them equal to find the crossover:

Then

Interpret the Crossover

At

both plans have the same cost.

For smaller \(x\), compare one test value.

At \(x=0\):

So \(B\) is cheaper initially.

For large \(x\), the larger slope of \(B\) eventually makes \(A\) cheaper.

Interpolation in a Linear Model

If a linear model is valid over an observed interval, it can estimate values between known observations.

For example, suppose

fits observed costs for

Estimating at

is interpolation.

The model gives

Extrapolation in a Linear Model

Using the same observed range

predicting at

is extrapolation.

The formula still produces a number:

But the reliability depends on whether the same cost structure continues that far.

Short-Term Versus Long-Term Models

A model may be useful over one interval and poor over another.

Examples:

• constant-speed motion may work over a few minutes but not an entire trip;
• a fixed production rate may change after equipment reaches capacity;
• a temporary sales trend may not persist for years;
• a temperature trend may change after sunset.

Model validity is usually tied to a domain.

Model Assumptions Should Be Stated

A useful model report may say:

• assume constant speed;
• assume no additional fixed charges;
• assume production rate remains constant;
• assume the observed trend continues;
• assume no capacity limit is reached.

Stating assumptions makes the model easier to evaluate and revise.

Model Assumptions Are Testable

If a model assumes constant rate, new data can test that assumption.

If observed slopes begin changing systematically, the model may need to be updated.

A model is not a permanent declaration. It is a working mathematical representation that can be compared with evidence.

Piecewise-Linear Models

If the rate changes at known thresholds, one affine equation may be too simple.

A piecewise-linear model can use a different rate on each interval.

For example, overtime pay may use one hourly rate before \(40\) hours and another afterward.

Example: Overtime Model

Suppose a worker earns \$20 per hour for the first \(40\) hours and \$30 per hour afterward.

Then

A piecewise-linear model preserves the changing rate without abandoning linear structure within each region.

When a Nonlinear Model Is Better

A single linear model may be poor when:

• rates increase or decrease continuously;
• growth is percentage-based;
• the graph curves strongly;
• physical saturation occurs;
• periodic behavior dominates;
• residuals show systematic structure.

In such cases, another function family may better represent the system.

Constant Percentage Change Is Not Linear

If a quantity grows by \(5\%\) each period, the change depends on the current amount.

The process is multiplicative, not additive.

Such behavior is naturally modeled by an exponential function rather than by

Curvature Is Evidence Against One Global Line

If data rise slowly at first and then increasingly rapidly, one constant slope cannot capture the whole pattern.

A local line may still be useful over a small interval, but a global nonlinear model may be more appropriate.

Outliers and Model Choice

An outlier may strongly affect a fitted line.

Before changing or discarding the model, ask:

• was the observation recorded correctly?
• did an unusual event occur?
• did the system change?
• is the outlier actually evidence that the simple model is incomplete?

Do Not Remove Data Just to Improve the Fit

A model should represent the system, not merely look neat.

Removing inconvenient observations without justification can hide important information.

Data quality decisions should be explained.

Graph Scale Can Mislead

A line may appear nearly flat or extremely steep depending on axis scaling.

Always interpret slope numerically using the variable units.

Do not judge rate only by the visual angle of the graph.

Zero May Be Outside the Relevant Domain

The intercept \(b\) describes the model at

But if the meaningful domain is

then \(b\) is extrapolated far outside the observed region.

It may still be mathematically useful without having a realistic direct interpretation.

Physical Bounds Can Limit a Model

A line may predict impossible values if extended too far.

Examples include:

• negative inventory;
• more than \(100\%\) capacity;
• negative population;
• unbounded growth in a finite system.

Known bounds should constrain the valid domain.

Discrete Inputs

Some models use inputs that can take only whole-number values.

For example,

may model the cost of \(n\) items.

If \(n\) counts items, then

The formula is affine, but the practical graph consists of discrete points rather than every point on the continuous line.

Continuous Inputs

Other variables may vary continuously.

Examples include:

• time;
• distance;
• temperature;
• mass;
• volume.

For continuous domains, the model may reasonably include all real inputs in an interval.

Model Precision Should Match Data Precision

If measurements are only accurate to the nearest kilogram, reporting a model prediction to twelve decimal places creates false precision.

Round results at a level appropriate to the quality of the data and the purpose of the model.

Example: Reasonable Rounding

Suppose a model predicts

minutes, but measured times are recorded only to the nearest minute.

A practical report might state

or perhaps \(47.4\) minutes if tenths are justified.

Sensitivity to the Slope

Small slope errors become more important over large input changes.

If two models differ in slope by

then after an input change of \(100\), their predictions differ by

This is one reason distant extrapolation can amplify small modeling errors.

Sensitivity to the Intercept

If two models have the same slope but intercepts differing by \(5\), their predictions differ by exactly \(5\) everywhere.

The intercept shifts the entire line without changing its rate.

Model Comparison Through Residuals

If two candidate models are available, compare how well each predicts the observed data.

A model with consistently smaller, less structured residuals may be preferable.

But mathematical fit is only one criterion.

Simplicity, interpretability, physical meaning, and domain knowledge also matter.

Simple Models Have Value

A linear model may be useful even when reality is more complicated.

Its advantages include:

• easy interpretation;
• fast calculation;
• clear graphical meaning;
• simple threshold analysis;
• easy comparison with other models.

A model need not be perfect to be useful.

But Simplicity Has a Cost

Every simplification ignores some details.

A good modeling decision balances:

The right balance depends on the task.

Prediction Is Not Explanation

A line may predict data well without explaining why the pattern occurs.

For example, two variables may rise together because both are influenced by a third factor.

Mathematical association does not automatically establish causation.

Modeling and Causation

To make causal claims, additional evidence may be needed, such as:

• controlled experiments;
• mechanistic understanding;
• careful study design;
• control of confounding variables.

A linear equation alone cannot supply that evidence.

Communicating a Linear Model

A complete model statement should usually include:

• the equation;
• definitions of the variables;
• units;
• interpretation of slope;
• interpretation of intercept when meaningful;
• domain or range restrictions;
• important assumptions;
• whether the relationship is exact or approximate.

Example of a Complete Model Statement

Suppose

A complete interpretation might be:

“For production levels between \(20\) and \(50\) units, total cost is modeled as \$340 plus approximately \$8 per unit produced. The slope represents marginal cost per additional unit. The intercept is the affine model's implied cost at zero production, but zero production lies outside the observed range and should not automatically be interpreted as an observed fixed cost.”

Model Validation with New Data

A strong test of a model is how well it performs on data not used to construct it.

If a line was built from earlier observations, compare future observations with its predictions.

Large systematic errors suggest the model should be revised.

Updating a Model

Suppose a machine originally produced at

but newer observations show a rate closer to

The old model may no longer represent current operation.

Models should be updated when the system or evidence changes.

Modeling as an Iterative Process

A practical modeling cycle is:

Modeling is not a one-time calculation.

It is a process of using mathematics to represent, test, and improve our understanding of a system.

A Reliable Linear-Modeling Workflow

1. Identify the quantities.
2. Define variables and units.
3. Decide whether constant-rate behavior is plausible.
4. Identify the slope from a rate or data.
5. Identify an intercept or known point.
6. Choose a useful equation form.
7. State the valid domain.
8. Interpret the parameters.
9. Check the model against data and context.
10. Revise or replace the model if the assumptions fail.

A Reliable Data-to-Model Workflow

1. Plot or inspect the data.
2. Check whether the rate is constant or approximately constant.
3. Distinguish exact data from noisy observations.
4. Construct an exact line or approximate trend line as appropriate.
5. Compute residuals or deviations.
6. Check for curvature, thresholds, or outliers.
7. Use interpolation more confidently than distant extrapolation.

A Reliable Applied-Decision Workflow

1. Construct the relevant linear models.
2. Translate the decision into an equation or inequality.
3. Solve for the threshold or feasible interval.
4. Apply domain restrictions such as nonnegativity or integrality.
5. Interpret the result in units and context.
6. Check whether the model remains valid near the decision boundary.

A Reliable Model-Communication Workflow

1. State the equation.
2. Define every variable.
3. Give units.
4. Interpret slope and intercept.
5. State assumptions and domain.
6. Label predictions as predictions.
7. Describe limitations rather than hiding them.

Common Errors

Using a Linear Model Without Checking Constant Rate

A line should reflect the structure of the data or system, not merely be chosen because it is familiar.

Treating an Approximate Trend as an Exact Law

Noisy data generally support approximate statements, not exact equality.

Ignoring Units

Slope must carry output units per input unit.

Interpreting Every Intercept Literally

If \(x=0\) lies outside the meaningful or observed domain, the intercept may not have direct physical meaning.

Extrapolating Too Far

A locally useful line may fail outside the region where it was tested.

Ignoring Physical Bounds

Negative inventories, impossible percentages, or unlimited growth may signal model failure.

Ignoring Discrete Domains

If the variable counts objects, only whole-number inputs may be meaningful.

Confusing Correlation with Causation

A strong linear association does not by itself establish a causal mechanism.

Reporting False Precision

Model outputs should not imply more precision than the data justify.

Keeping a Model After the System Changes

Models should be revised when new evidence shows that their assumptions no longer hold.

Practice

1. Rate and Initial Value

A service charges \$12 initially plus \$7 per hour. Construct a cost model \(C(h)\).

2. Observation and Rate

At \(t=3\) hours, a vehicle is \(210\) miles from a reference point. It continues away at \(50\) miles per hour. Construct a model.

3. Two Observations

A company's cost is \$420 at \(10\) units and \$660 at \(30\) units. Assume an affine relationship. Construct \(C(q)\).

4. Interpret the Slope

For

interpret \(12\).

5. Intercept Caution

If the model above was based only on observations for \(10\le q\le30\), should the intercept \(300\) automatically be treated as an observed fixed cost?

6. Natural Domain

A tank model is

Assuming time begins at \(0\) and the model ends when the tank is empty, find the natural domain.

7. Break-Even

Revenue is

and cost is

Find the break-even quantity.

8. Profit Region

Using the same models, for what quantities is revenue greater than cost?

9. Compare Two Plans

Plan A costs

and Plan B costs

Find where they cost the same.

10. Detect Model Failure

A population model is

What warning should be given before using it for very large \(t\)?

11. Exact or Approximate?

Observed data do not lie exactly on one line but cluster close to

How should the relationship be written?

12. Choose a Better Model

A process grows by \(10\%\) each month. Is an exact affine model the natural choice?

Key Ideas

• A linear model represents a system using constant-rate structure.
• Modeling begins by defining variables, units, domain, and assumptions.
• In \(y=mx+b\), \(m\) is the modeled rate of change and \(b\) is the modeled output at \(x=0\).
• Exact relationships use equality; approximate trends should be described as approximations.
• A model can be constructed from rates, intercepts, observations, tables, constraints, or fitted trends.
• Point-slope form is often useful when the known observation does not occur at \(x=0\).
• Standard form is useful for resource and constraint models involving two variables.
• Equations identify thresholds and intersections; inequalities identify feasible or favorable regions.
• Residuals help measure how well an approximate model matches observed data.
• Interpolation is generally safer than distant extrapolation.
• Domain restrictions, physical bounds, and discrete inputs must be respected.
• Piecewise-linear models are useful when rates change at known thresholds.
• Curvature, changing rates, or percentage growth may require nonlinear models.
• Simple models are valuable when they are accurate enough for the task and their limitations are understood.
• Modeling is iterative: observe, construct, predict, check, and revise.

Where This Leads

Linear modeling completes the study of Linear and Affine Relationships by bringing the section's tools together into one practical process. Slope describes rate, intercepts establish position, equation forms organize information, graphs reveal structure, inequalities identify decision regions, and interpolation and piecewise functions extend the model to realistic situations.

These ideas now support the next layers of Algebraic Relationships: systems of equations and matrix methods, polynomial and rational models, exponential and logarithmic change, sequences, graphical methods, numerical algorithms, and introductory analysis. The line remains one of the simplest mathematical models, but it becomes a reference point against which more complicated relationships can be understood.