Analyzing Model Reasonableness — I can analyze a model's reasonableness and revise assumptions when needed. - Math 1

SECTION 1

Historical Context & Motivation

Throughout history, humans have tried to predict the world around them using simplified descriptions of reality—what we now call mathematical models. From ancient astronomers sketching planetary orbits to modern engineers simulating bridge loads, every model involves choices about what to include and what to leave out. Those choices are called assumptions, and they determine whether a model's predictions are reasonable or wildly off-target.

When a model's output doesn't match the real-world data it's supposed to describe, mathematicians and scientists don't just throw it away. Instead, they analyze which assumptions caused the mismatch and revise the model to make it more accurate. This cycle of building, testing, and revising models is one of the most powerful tools in all of mathematics and science.

~150 CE • Ptolemy's Geocentric Model

Ptolemy modeled planetary motion with Earth at the center. Predictions were reasonable for centuries, but growing errors revealed the assumption of a geocentric universe was flawed.

1687 • Newton's Laws of Motion

Isaac Newton created models for gravity and motion that assumed constant gravitational fields and frictionless surfaces. These simplifying assumptions worked remarkably well for everyday objects but broke down at extreme speeds.

1900s • Statistical Modeling Emerges

Statisticians developed regression models and residual analysis, giving us formal tools to measure how well a model fits data and to identify where assumptions fail.

2020s • Data-Driven Model Revision

Today, scientists and analysts use computers to rapidly test millions of model variations, but the core question remains the same: does this model make sense, and if not, what assumption should we change?

The central question this lesson addresses is deceptively simple: How do you tell if a model is reasonable, and what do you do when it isn't? Answering that question requires a clear process for evaluating model predictions, identifying problematic assumptions, and making targeted revisions.

SECTION 2

Core Principles of Model Reasonableness

Before you can judge a model, you need to understand what makes one "reasonable." A reasonable model doesn't have to be perfect—it just has to produce outputs that are close enough to reality to be useful for a specific purpose. The following principles form the foundation of model analysis.

Purpose Alignment

A model is reasonable only relative to the question it's designed to answer. A model that predicts a car's fuel cost doesn't need to account for tire wear—unless your question is about tire wear.

Assumption Transparency

Every model rests on assumptions—statements you treat as true to simplify the problem. Listing your assumptions explicitly is the first step to analyzing reasonableness.

Output Validation

Compare model predictions to known data, common sense, or expected ranges. If a model says a person walks 60 miles per hour, something is wrong.

Residual Analysis

The residual is the difference between the predicted value and the actual value. Large or patterned residuals signal that an assumption needs revision.

Iterative Revision

Model building is a cycle: build, test, revise, repeat. Each revision should target a specific assumption that was identified as problematic during validation.

✦

KEY TAKEAWAY

Think of a mathematical model like a GPS route. The GPS assumes roads are open, your speed is consistent, and there's no construction. If your estimated arrival is way off, you don't throw away the GPS—you update the assumption that caused the error (maybe there was heavy traffic) and let it recalculate. Analyzing model reasonableness works exactly the same way.

SECTION 3

The Model Analysis Cycle

The diagram below illustrates the complete cycle of analyzing and revising a mathematical model. Notice that the process is cyclical, not linear. You don't finish after one pass—you continue refining until the model's predictions are reasonable for your purpose.

The five-stage cycle begins with defining your question, then building a model with explicit assumptions. After testing predictions against real data, you analyze the residuals (prediction errors). If the model is unreasonable, you revise your assumptions and loop back to rebuild. This cycle continues until the model is fit for its purpose.

Notice the central decision point labeled "Reasonable?" This is where your judgment comes in. You decide whether the model's errors are small enough for your purpose. A weather model that's off by 2°F might be perfectly acceptable for planning a picnic, but unacceptable for launching a spacecraft. Reasonableness is always relative to the context of the problem.

SECTION 4

Mathematical Framework for Evaluating Models

While much of model analysis involves judgment and common sense, there are concrete mathematical tools you can use to measure how well a model fits. The most fundamental tool is the residual—the difference between what the model predicts and what actually happens.

RESIDUAL

Residual = Actual Value − Predicted Value

A positive residual means the model underestimated (actual was higher). A negative residual means the model overestimated (actual was lower). Residuals near zero indicate a good fit.

PERCENT ERROR

Percent Error = |Actual − Predicted| ÷ Actual × 100%

Percent error expresses the residual as a percentage of the actual value, making it easier to compare model accuracy across different scales. A model predicting a $50 cost that's off by $5 has a 10% error.

LINEAR MODEL

y = mx + b

Many first models are linear: m is the rate of change (slope), b is the initial value (y-intercept), x is the input variable, and y is the predicted output. If the residuals show a curved pattern, you may need a nonlinear model instead.

⚠

When to Revise Your Model

Revise your model when: (1) residuals are consistently large, (2) residuals show a clear pattern (like curving up over time), or (3) the model gives outputs that violate common sense. Each of these signals points to a specific assumption that needs to change.

SECTION 5

Residual Plots & Identifying Flawed Assumptions

One of the most powerful ways to evaluate a model is to create a residual plot. In a residual plot, the x-axis represents the input variable and the y-axis represents each residual (actual − predicted). If your model is a good fit, the residuals should be randomly scattered around zero with no visible pattern. A pattern in the residuals tells you that your model is systematically missing something.

The left panel shows a good fit: residuals bounce randomly above and below zero with no clear pattern. The right panel shows a poor fit: the U-shaped pattern indicates the data has a curve that a linear model cannot capture. This pattern tells you to revise the assumption of linearity.

In the right panel, the curved residual pattern is a clear signal. The model assumed a constant rate of change (linear relationship), but the data actually curves. The revision? Switch from a linear model (y = mx + b) to a quadratic or exponential model that can capture the curvature. Each type of residual pattern points to a specific kind of assumption failure, which makes the revision process targeted rather than random.

Residual Pattern	What It Tells You	Likely Revision
Random scatter around zero	Model fits well; assumptions are reasonable	No revision needed
U-shaped or inverted-U curve	Relationship is nonlinear, not linear	Try a quadratic or exponential model
Residuals grow larger as x increases	Variability increases with x (fan shape)	Consider a multiplicative or percentage-based model
One or two extreme outliers	Unusual data points or errors in data collection	Investigate outliers; remove if justified

Common residual patterns and their corresponding model revisions

SECTION 6

Worked Example — School Fundraiser

A school club is selling candles to raise money. They create a linear model to predict total revenue based on the number of candles sold. Their model assumes every candle sells for $8 and there are no up-front costs. After the event, they compare their predictions to the actual data and analyze whether the model was reasonable.

Fundraiser Revenue Model — Analyze & Revise

1
Step 1 — State the Model and Assumptions

The club's model is R = 8n, where R is total revenue in dollars and n is the number of candles sold. Assumptions: (1) every candle sells for exactly $8, (2) there are no other costs or revenues, and (3) every candle produced gets sold.
2
Step 2 — Compare Predictions to Actual Data

The club tracked actual revenue at several points during the sale. At 10 candles, predicted = $80, actual = $65. At 25 candles, predicted = $200, actual = $170. At 50 candles, predicted = $400, actual = $325. At 75 candles, predicted = $600, actual = $490.
3
Step 3 — Calculate Residuals

Residual = Actual − Predicted. At n = 10: 65 − 80 = −15. At n = 25: 170 − 200 = −30. At n = 50: 325 − 400 = −75. At n = 75: 490 − 600 = −110. All residuals are negative and growing in magnitude, indicating the model consistently overestimates revenue, and the overestimation gets worse as sales increase.
4
Step 4 — Identify the Problematic Assumption

The growing negative residuals suggest the $8 flat price assumption is wrong. After investigating, the club discovers they offered a bulk discount: candles were $8 each for the first 20, but $6 each after that. Additionally, they had a $15 supply cost they didn't include.
5
Step 5 — Revise the Model

The revised model is a piecewise function. For n ≤ 20: R = 8n − 15. For n > 20: R = 8(20) + 6(n − 20) − 15 = 6n + 25. Let's test: at n = 50, R = 6(50) + 25 = 325. This matches the actual data exactly.

Revised model: R = 6n + 25 (for n > 20), which accounts for the bulk discount and supply cost. This passes the reasonableness check.

💡

LESSON FROM THE EXAMPLE

The original model wasn't useless—it gave the club a starting point. But by checking residuals systematically, they found two hidden assumptions (constant price and zero cost) that made the model unreasonable. Revising those assumptions produced a model that matched reality.

SECTION 7

Strengths & Limitations of Common Model Types

Different types of models are suited to different situations. Understanding the strengths and limitations of each type helps you choose the right starting model and anticipate where it might fail. The table below compares three model types you'll encounter frequently in Math 1.

Model Type	Strengths	Limitations	Common Assumption Failures
Linear (y = mx + b)	Simple, easy to interpret, good for constant rates of change	Cannot capture curves, acceleration, or saturation effects	Assumes rate of change never varies; ignores diminishing returns
Exponential (y = a × bˣ)	Models growth/decay that compounds; good for populations and finance	Predicts unlimited growth; unrealistic for large x values	Assumes unlimited resources; ignores carrying capacity
Quadratic (y = ax² + bx + c)	Captures symmetric curves; good for projectiles and area problems	Assumes symmetry; may not fit data that curves in only one direction	Assumes effects are symmetric; may overpredict in one direction

Comparison of common model types used in Math 1

✦

KEY TAKEAWAY

No single model type works for every situation. The key is to start with the simplest model that might work, then let the residuals guide you toward a better model type when the simple one fails. A curved residual pattern from a linear model doesn't mean you made a mistake—it means you've learned something about the data that helps you improve.

SECTION 8

Connection to Advanced Modeling

The model-analyze-revise cycle you've learned in this lesson is the foundation of all advanced mathematical modeling. In later courses, the tools become more sophisticated, but the fundamental question remains the same: Is this model reasonable, and what should I change if it isn't?

What You Learn Now	Where It Leads
Calculate residuals by hand	In statistics, you'll compute R² (coefficient of determination) to measure overall model fit
Identify curved residual patterns	In Algebra 2, you'll use regression analysis to find the best-fit curve automatically
Revise one assumption at a time	In calculus and physics, you'll use sensitivity analysis to measure how each assumption affects the output
Choose between linear, exponential, and quadratic	In data science, you'll compare dozens of model types and use cross-validation to select the best one

How today's skills connect to future coursework

Whether you go into engineering, medicine, business, or social science, you'll use models daily. The ability to evaluate whether a model's assumptions are justified—and to revise them when they're not—is one of the most transferable skills you'll develop in math class.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

A student builds a model that predicts the temperature inside a parked car based on the outside temperature. The model says the inside temperature equals the outside temperature. Explain why this model might be unreasonable. What assumption is the student making, and how would you revise it?

Show answer

The student is assuming that the car's interior temperature equals the outside temperature, which ignores the greenhouse effect inside a closed car. Sunlight passes through the windows and heats the interior, but the heat gets trapped inside. In reality, a parked car in sunlight can be 20–40°F hotter than outside. A revised model might be: Inside Temp = Outside Temp + H, where H is a heating factor that depends on sun exposure, window tinting, and time. This revision acknowledges that a car's interior absorbs and traps heat differently than open air.

Question 2: BASIC CALCULATION

A model predicts that a plant grows 2 cm per week. After 4 weeks, the actual heights are: Week 1 = 2.3 cm, Week 2 = 4.1 cm, Week 3 = 5.5 cm, Week 4 = 6.8 cm. Calculate the residual for each week and determine whether the model is reasonable.

Show answer

Predicted heights: Week 1 = 2, Week 2 = 4, Week 3 = 6, Week 4 = 8. Residuals (Actual − Predicted): Week 1 = 2.3 − 2 = +0.3, Week 2 = 4.1 − 4 = +0.1, Week 3 = 5.5 − 6 = −0.5, Week 4 = 6.8 − 8 = −1.2. The residuals start positive and become increasingly negative, showing a pattern. The model overestimates more and more each week, suggesting the plant's growth rate slows over time. The assumption of a constant 2 cm/week growth rate is unreasonable for later weeks.

Question 3: INTERMEDIATE

A ride-share company models the cost of a trip as C = 2.50 + 1.25d, where C is the cost in dollars and d is the distance in miles. A customer takes a 12-mile trip during rush hour and pays $22.50. Calculate the percent error and explain which assumption likely caused the discrepancy. Suggest a revised model.

Show answer

Predicted cost: C = 2.50 + 1.25(12) = 2.50 + 15.00 = $17.50. Percent error = |22.50 − 17.50| ÷ 22.50 × 100% = 5.00 ÷ 22.50 × 100% ≈ 22.2%. This is a substantial error. The model assumes a constant rate per mile, but ride-share prices often include surge pricing during rush hour and per-minute charges for time spent in traffic. A revised model might be: C = 2.50 + 1.25d + 0.30t, where t is the trip duration in minutes. This accounts for time-based charges that increase during rush hour when traffic is heavy.

Question 4: APPLIED

A small business models its monthly profit as P = 15n − 800, where P is profit in dollars and n is the number of units sold. After three months, actual data is: Month 1 (n = 60, P = $80), Month 2 (n = 90, P = $420), Month 3 (n = 120, P = $650). Analyze the residuals, identify the problematic assumption, and propose a revised model with justification.

Show answer

Predicted profits: Month 1 = 15(60) − 800 = $100, Month 2 = 15(90) − 800 = $550, Month 3 = 15(120) − 800 = $1000. Residuals: Month 1 = 80 − 100 = −20, Month 2 = 420 − 550 = −130, Month 3 = 650 − 1000 = −350. The residuals are all negative and growing in magnitude, meaning the model increasingly overestimates profit. The assumption that each unit generates exactly $15 in revenue is likely wrong—perhaps the business lowers prices for bulk orders or faces increasing costs as production rises. A revised model might be P = 12n − 640, which uses a lower per-unit profit to reflect the actual revenue pattern. Alternatively, a model with decreasing marginal profit like P = −0.03n² + 18n − 800 could capture the diminishing returns.

Question 5: CRITICAL THINKING

Two students model the number of students using the school library after school. Student A uses a linear model: U = 5w + 10, where w is the week number. Student B uses an exponential model: U = 10 × 1.4ʷ. In Week 1 the actual count is 15, Week 2 is 19, Week 3 is 22, Week 4 is 24, and Week 5 is 25. Which model is more reasonable through Week 5? What would happen to each model's predictions by Week 20, and what does that reveal about long-term model reasonableness? Explain which assumptions would need to be revised for each model.

Show answer

Student A predictions: W1 = 15, W2 = 20, W3 = 25, W4 = 30, W5 = 35. Residuals: +0, −1, −3, −6, −10. Student B predictions: W1 = 14, W2 ≈ 19.6, W3 ≈ 27.4, W4 ≈ 38.4, W5 ≈ 53.8. Residuals: +1, −0.6, −5.4, −14.4, −28.8. Through Week 5, Student A's linear model is initially better, but both models overestimate increasingly—Student B's exponential model diverges much faster. By Week 20, Student A predicts 110 users and Student B predicts over 8,000, while the actual number is likely capped by the school's population (perhaps 200–300 students). The actual data shows growth that is slowing down, suggesting a logistic or saturation model. Student A needs to revise the assumption of constant weekly growth; Student B needs to revise the assumption of unlimited compounding. A better model might be U = 26 − 12 × 0.7ʷ, which approaches a ceiling. This problem shows that model reasonableness depends on the prediction range—a model that's reasonable for short-term forecasts can be wildly unreasonable for long-term ones.

SUMMARY

Lesson Summary

Summary

Analyzing model reasonableness means checking whether a model's predictions align with real data and common sense. Every model is built on assumptions—simplifications you make to create the model. You evaluate a model by computing residuals (Actual − Predicted) and percent error, then examining whether the residuals are small, random, and pattern-free. If they show a systematic pattern—like a curve or a fan shape—it signals that a specific assumption is wrong.

The model analysis cycle (define → build → test → analyze → revise) is iterative: you continue revising assumptions until the model is fit for its purpose. Common revisions include switching from a linear model to a nonlinear one, adding costs or factors that were initially ignored, or accounting for limits on growth. Remember that reasonableness is always relative to your purpose—the same model can be reasonable for one question and unreasonable for another.