Response Model Lab

Model Fitting

Name the shape, pick the form, then turn the parameters by hand until the error stops falling. This is the exercise regression automates two weeks from now.

πŸ‘¨β€πŸ« Professor Mode: Guided Learning Experience

New to response models? Enable Professor Mode for step-by-step guidance through naming a shape, choosing a functional form, and calibrating it by hand!

OVERVIEW & WHAT THIS TOOL SOLVES

Before any optimization algorithm can fit a marketing response model, the analyst must make four distinct choices: what shape the customer response follows, which mathematical equation expresses that shape, how parameter sliders move the curve, and which loss function judges a "good" fit. The Response Model Lab lets you fit Linear, Diminishing-Returns, S-Curve, and Threshold response models by hand and compare them against automated least-squares/numerical optimization.

🎯 Questions It Answers

  • Does our marketing outcome grow at a constant rate (Linear), taper off as spend rises (Diminishing Returns), or need a minimum spend to wake up before saturating (S-Curve)?
  • How does each parameter (intercept, slope, curvature, ceiling, inflection point) physically bend or shift the response curve?
  • Why do Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) disagree when a dataset contains an outlier promotional week?

πŸ“‚ What Data You Need

  • Bivariate Marketing Dataset: Built-in marketing scenarios (ad spend, pricing, sales calls, threshold promotions) or custom CSV upload
  • Functional Form Choice: Select Linear, Diminishing Returns (Log/Power), S-Curve (Logistic/Hill), or Piecewise Threshold
  • Loss Criterion: Choose RMSE/SSE (squared error) or MAE (absolute error) to guide manual and automated fitting

πŸ’‘ What You'll Discover

  • Interactive Curve & Residual Sticks: Real-time visual feedback showing vertical errors as you drag parameter sliders
  • Manual vs. Optimal Fit Comparison: How close your hand-tuned parameters come to the mathematical optimum
  • Misspecification Diagnostics: Why the best-fitting straight line still leaves systematic residual patterns on a curved market
πŸ” Key Nuances & Common Traps to Watch For
  • Optimization Cannot Fix the Wrong Functional Form: Least squares finds the best straight line effortlessly, but if ad spend truly saturates at $50K, a linear model will tell you to spend infinity. Choosing the functional form is a strategic judgment, not arithmetic.
  • Diminishing Returns vs. S-Curve: In a pure diminishing-returns curve, the first dollar spent has the highest marginal return; in an S-curve, tiny budgets barely register below the threshold, marginal returns peak at the inflection point, and then saturation sets in.
  • RMSE Chases Outliers, MAE Ignores Them: Squaring residuals makes a single 10-unit outlier count 100Γ— as much as a 1-unit errorβ€”pulling the whole fitted curve toward extreme promotional spikes.
πŸ“ Under the Hood: Model Equations & Variable Definitions

1. Canonical Marketing Response Functional Forms

\[ \hat{y}_{\text{lin}} = a + b x, \qquad \hat{y}_{\text{dim}} = a + b \ln(1 + c x), \qquad \hat{y}_{\text{S-curve}} = a + \frac{L}{1 + \exp\!\left(-k(x - x_0)\right)} \]

πŸ’¬ Read it in plain English: Linear assumes constant marginal return b per unit of x; Diminishing Returns bends downward as spend x grows; and the Logistic S-Curve rises from baseline a toward ceiling a + L with steepest slope at inflection spend xβ‚€.

x, Ε·
Marketing Input & Predicted ResponseYOUR DATA β“˜
Marketing effort input x (e.g., ad spend, price) and model-predicted outcome Ε·.
a, b
Baseline Intercept & Scale SlopeESTIMATED β“˜
Outcome level when x = 0 and marginal sensitivity parameter scaling the response.
L, k, x0
S-Curve Ceiling, Steepness & InflectionESTIMATED β“˜
Maximum incremental lift L, steepness k, and midpoint spend xβ‚€ where marginal return peaks.

2. Residual Error & Loss Functions (RMSE vs. MAE)

\[ e_i = y_i - \hat{y}_i, \qquad \text{RMSE} = \sqrt{\frac{1}{n}\sum_{i=1}^{n} e_i^2}, \qquad \text{MAE} = \frac{1}{n}\sum_{i=1}^{n} |e_i| \]

πŸ’¬ Read it in plain English: Each observation's residual e_i is the vertical gap between actual outcome y_i and the curve Ε·_i; RMSE penalizes large errors quadratically while MAE weights every dollar of error equally.

yi, ei
Observed Outcome & ResidualYOUR DATA β“˜
Actual outcome y_i for observation i and vertical prediction error e_i = y_i βˆ’ Ε·_i.
RMSE
Root Mean Squared ErrorINFERENCE β“˜
Square root of mean squared residuals (heavily sensitive to outliers).
MAE
Mean Absolute ErrorINFERENCE β“˜
Average absolute vertical distance between data points and the response curve.

DATA SOURCE

Start by loading a real-world marketing scenario or uploading your own dataset.

Use a Case Study

Load a preset scenario to explore response shapes, functional forms, and parameter calibration.

Upload Your Own Data

Two numeric columns: the marketing input first, the outcome second. A header row is expected and is used to label the axes.

Drag & Drop CSV File

Expects two numeric columns (predictor X, response Y)

Step 1Name the shape you expect

Read the scenario above and commit to a shape before you look at the scatter. Choosing the form because it happened to fit best is not a theoretical justification, and it is how overfitting starts.

Steps 2 & 3Pick a form, then calibrate it

Parameters

The automatic search minimises whichever loss is selected above. Run it once under each loss and compare the parameters it lands on.

MAE
β€”
average miss
MSE
β€”
squared, punishes big misses
Largest miss
β€”
single worst point

Where the model is wrong

A loss value is one number for the whole curve. The residual plot shows where the misses are. Structure here β€” a run of points all above the line, or a U shape β€” means the form itself is wrong, and no amount of parameter tuning will fix it. The plot shows the misses of your current parameters, but the verdict below judges the best-fitting (least-squares) curve of the chosen form, so moving the sliders does not change it.

Compare the three forms

Each form is fitted to the current scenario under the selected loss. Fit is not the only criterion: a form that fits marginally better but cannot respect a ceiling, or that bends violently outside the observed range, is usually the worse choice.

Exploratory Questions

  1. Fit the TV reach scenario with a linear model, then read the prediction at 2,000 GRPs. What does it claim, and why is that impossible?
  2. On the email frequency scenario, fit a linear model and then a power series. What does the linear model recommend about send frequency, and what does it miss?
  3. Pick any scenario. Fit it once minimising MAE and once minimising MSE. Where do the two curves differ most, and which points explain the difference?
  4. Add an extreme point to a scenario by uploading a modified CSV. Which loss function moves further, and what does that tell you about using MSE when your data contains genuine outliers?
  5. The trade promotion scenario keeps rising across the observed range. Would you still choose a form with a ceiling? What would you need to know to decide?