14.1 Modeling with Functions

Modeling with Functions is about translating real-world situations into mathematical equations. You choose the right function type based on the behavior you observe in data, then use the equation's structure to make predictions and inform decisions. This skill ties together everything from earlier units (linear, quadratic, exponential, and beyond) and puts it to work on actual problems.

Modeling real-world situations with functions

more resources to help you study

practice questions

Constructing mathematical models

Building a model starts with understanding the situation, then matching it to the right mathematical structure. Here's the general process:

1. Identify your variables. Determine what quantities are changing and define them clearly (e.g., time in hours, cost in dollars).

2. Collect and analyze data. Look at the data's behavior to spot trends. Is the change constant? Accelerating? Cyclical?

3. Choose a function type. Based on the trend you see, select the best fit: linear, quadratic, exponential, logarithmic, polynomial, rational, or trigonometric.

4. Use regression to find the equation. Technology tools (graphing calculators, Desmos, etc.) can run regression analysis to produce a best-fit equation for your data.

   • The correlation coefficient ($r$ or $r^2$) tells you how well the regression equation fits the data.

5. Consider piecewise models. Some real-world situations behave differently across different intervals. For example, a shipping cost might be flat up to 5 lbs, then increase linearly after that. In cases like these, you'll need a piecewise function with separate equations for each subdomain.

Common function types used in modeling

Each function type captures a specific kind of behavior. Matching the right type to your situation is the most important modeling decision you'll make.

• Linear ($y = mx + b$): Constant rate of change. Use when the data increases or decreases by the same amount per unit. Example: A car rental that charges a $30 base fee plus $0.50 per mile.
• Quadratic ($y = ax^2 + bx + c$): Symmetrical curve, often involving acceleration or optimization. Example: The height of a ball thrown upward, where gravity pulls it back down in a parabolic arc.
• Exponential ($y = ab^x$): Constant percent growth or decay per unit. Example: A bacteria colony starting at 500 cells that doubles every hour.
• Logarithmic ($y = a\ln(x) + b$): Rapid initial change that gradually levels off. Example: The Richter scale, where each whole number represents a tenfold increase in earthquake amplitude.
• Polynomial ($y = a_nx^n + \cdots + a_0$): Varying rates of change with possible turning points. Example: The volume of an open-top box formed by cutting squares from a sheet of cardboard, expressed as a function of the cut length.
• Rational ($y = \frac{p(x)}{q(x)}$): Asymptotic behavior or inverse variation. Example: The concentration of a medication in the bloodstream over time, which spikes then gradually approaches zero.
• Trigonometric ($y = a\sin(bx) + d$): Periodic, cyclical behavior. Example: The height of a Ferris wheel passenger as a function of time, repeating every full rotation.

Interpreting function parameters

Contextual meaning of variables and parameters

Every number in a model equation means something in the real world. The input variable represents the quantity you control or observe (like time), and the output variable represents what you're measuring or predicting (like height or cost). The parameters are the constants in the equation, and they typically represent starting amounts, rates of change, or scaling factors.

Being able to translate between the math and the context is what separates plugging numbers into a formula from actually understanding the situation.

Interpreting parameters in specific function types

• Linear model $y = mx + b$: The slope $m$ is the constant rate of change per unit of input, and $b$ is the initial value (the output when $x = 0$).
  • Example: In $C(x) = 25 + 0.50x$ for a car rental, $25$ is the base fee in dollars and $0.50$ is the cost per mile driven.
• Exponential model $y = ab^x$: The parameter $a$ is the initial amount, and $b$ is the growth factor (if $b > 1$) or decay factor (if $0 < b < 1$) per unit interval.
  • Example: In $P(t) = 100(2)^t$ for bacteria growth, $100$ is the starting population and $2$ means the population doubles every hour.
• Quadratic model $y = at^2 + bt + c$: Parameters can represent physical quantities depending on context. The constant term $c$ is often an initial value, the linear coefficient $b$ relates to an initial rate, and the leading coefficient $a$ involves acceleration or curvature.
  • Example: In $h(t) = -16t^2 + 96t + 48$ for a thrown ball, $48$ ft is the launch height, $96$ ft/s is the initial upward velocity, and $-16$ comes from half the acceleration due to gravity ($\frac{1}{2} \times -32$ ft/s²).

Effectiveness of function models

Evaluating model fit and accuracy

A model is only useful if it actually represents the data well. Here's how to assess that:

• Residuals: A residual is the difference between an actual data value and the model's predicted value ($\text{residual} = \text{actual} - \text{predicted}$). Smaller residuals across the data set mean a better fit.
• Residual plots: Plot the residuals against the input variable. If they scatter randomly around zero, the model type is appropriate. If they form a pattern (like a curve), a different function type would likely fit better.
• Correlation coefficient ($r$): For linear regression, $r$ ranges from $-1$ to $1$. Values near $1$ or $-1$ indicate a strong linear relationship; values near $0$ indicate a weak one. The coefficient of determination $r^2$ tells you the proportion of variability in the data explained by the model.
• Compare predictions to actual results. If the model consistently over-predicts or under-predicts, it may be underfitting (too simple) or overfitting (too tailored to noise in the data).

Limitations of models

Every model is a simplification, and no model is perfect.

• Models are typically valid only within a certain domain. Extrapolating far beyond your observed data range can produce unreliable predictions. (An exponential growth model for bacteria won't hold forever because resources run out.)
• Real-world data contains measurement errors, variability, and outliers, all of which affect model accuracy.
• There's a fundamental tradeoff between simplicity and accuracy. A simple model is easier to interpret and communicate, but it may miss nuanced behavior. A complex model may fit the data more closely but can be harder to use and may overfit noise rather than capturing the true trend.

Modifying function models

Identifying when a model needs improvement

If your model isn't representing the situation well, look for specific clues:

• Residual plot patterns: A curved pattern in the residuals suggests you should try a nonlinear model. Residuals that fan out (get larger) as $x$ increases suggest the model's accuracy varies across the domain.
• Weak correlation coefficient: A low $r$ value for a linear regression is a signal to try quadratic, exponential, or another regression type.
• New data that doesn't match: If new observations consistently fall outside the model's predictions, the underlying situation may have changed.

Techniques for modifying models

• Change the function type. If a linear model shows curved residuals, try quadratic or exponential regression instead.
• Update with new data. Real-world situations evolve, so re-running regression with current data keeps the model relevant.
• Add parameters to capture more complex behavior, but balance this against added complexity. A model should be as simple as possible while still being accurate (this principle is called parsimony).
• Use piecewise functions to combine different function types across different intervals, matching each subdomain's behavior separately.
• Address outliers. If a few data points are clearly anomalous (measurement errors, unusual circumstances), removing them and re-fitting can produce a more representative model. Just be careful to justify why those points are outliers rather than removing data arbitrarily.
• Apply transformations. Techniques like logarithmic transformation can linearize curved data, making it easier to fit and analyze. Weighted least squares can give more importance to certain data regions where accuracy matters most.