Competing Function Model Validation - AP Precalculus

1. What transformation makes an exponential model $y=ab^x$ linear for regression?

   Take logs: $ln(y)=\ln(a)+x\ln(b)$. This linearizes to $Y = \ln(a) + X\ln(b)$ where $Y=\ln(y)$, $X=x$.

2. Which residual plot pattern indicates a model is likely appropriate: random scatter or a curved pattern?

   Random scatter around $0$. Random residuals suggest the model captures the data's pattern well.

3. Which model is preferred if two models have similar context fit but one has larger $R^2$?

   The model with larger $R^2$. Higher $R^2$ means the model explains more variance in the data.

4. What is $SST$ in an $R^2$ calculation for data $y_1,\dots,y_n$?

   $SST=\sum (y_i-\bar{y})^2$. Total sum of squares measures total variance from the mean.

5. Which model is appropriate for a constant additive rate of change: linear or exponential?

   Linear. Linear models add the same amount each unit increase.

6. What does a funnel shape (increasing spread) in a residual plot indicate?

   Non-constant variance; errors grow with $x$ or $y$. Widening spread shows prediction accuracy decreases as values increase.

7. What does a larger value of $R^2$ indicate when comparing two regression models on the same data?

   Larger $R^2$ indicates more variance explained by the model. R² ranges from 0 to 1; higher values mean better predictive power.

8. What is the primary goal when validating competing function models for the same data set?

   Select the model that best fits and makes sense for the context. Good models balance statistical fit with real-world applicability.

9. Which option is valid: extrapolating far beyond the data range or staying within the observed $x$-values?

   Staying within the observed $x$-values (interpolation). Extrapolation beyond data is risky; interpolation is more reliable.

10. Which validation warning applies when a model is based on too few data points?

    Overfitting risk; the model may not generalize. Small samples can lead to models that fit noise rather than true patterns.

11. What does a curved pattern in a residual plot usually indicate about the chosen model?

    The model form is likely incorrect (systematic error remains). Curves in residuals mean the model misses a systematic pattern.

12. Identify the better fit if Model A has $SSE=120$ and Model B has $SSE=95$ on the same data.

    Model B. Lower SSE means Model B has smaller prediction errors.

13. What does a curved pattern in a residual plot usually indicate?

    The model is missing curvature (wrong functional form). Curves in residuals suggest a linear model can't capture nonlinear data patterns.

14. Identify the better fit if Model A has $R^2=0.93$ and Model B has $R^2=0.88$ on the same data.

    Model A. Higher $R^2$ means Model A explains more variance in the data.

15. What is an outlier in a regression context?

    A point with an unusually large residual compared with others. Outliers deviate significantly from the model's predictions.

16. What transformation makes a power model $y=ax^b$ linear for regression?

    Take logs: $ln(y)=\ln(a)+b\ln(x)$. This linearizes to $Y = \ln(a) + bX$ where $Y=\ln(y)$, $X=\ln(x)$.

17. Find the residual if the observed value is $y=7$ and the model predicts $\hat{y}=9$.

    $r=-2$. Using $r = y - \hat{y} = 7 - 9 = -2$.

18. Which validation warning applies when using a model to predict beyond the data range?

    Extrapolation; predictions may be unreliable. Models trained on limited data may fail outside that range.

19. Find the percent growth rate if an exponential model is $y=a(1+r)^x$ with base $1.08$.

    $r=0.08$ (an $8\%$ increase per unit $x$). Since $1.08 = 1 + 0.08$, the growth rate $r = 0.08 = 8\%$.

20. Find the residual if the observed value is $y=12$ and the model predicts $\hat{y}=10.5$.

    $r=1.5$. Using $r = y - \hat{y} = 12 - 10.5 = 1.5$.

21. Identify the better model by $SSE$: Model A has $SSE=18$, Model B has $SSE=25$.

    Model A. Lower $SSE$ means smaller prediction errors and better fit.

22. What is the key domain restriction for a logarithmic model $y=a+b\ln(x)$?

    The input must satisfy $x>0$. Natural log is undefined for non-positive values.

23. Choose the better model if residuals for Model 1 are random, but Model 2 residuals show a U-shape.

    Model 1. Random residuals indicate appropriate model form; patterns suggest misfit.

24. What is the coefficient of determination definition in terms of $SSE$ and $SST$?

    $R^2=1-\frac{SSE}{SST}$. Measures proportion of variance explained by the model.

25. What is the key domain restriction for a power model $y=ax^b$ when $b$ is not an integer?

    Typically require $x>0$ for real-valued outputs. Non-integer powers can produce complex numbers for negative inputs.

26. Which model is appropriate for a constant percent rate of change: linear or exponential?

    Exponential. Exponential models multiply by a constant factor each period.

27. Which model is more reasonable for data that levels off toward a horizontal asymptote: exponential growth or logistic?

    Logistic. Logistic models have carrying capacity; exponentials grow unbounded.

28. Compute $R^2$ if $SSE=9$ and $SST=36$.

    $R^2=0.75$. Using $R^2 = 1 - \frac{9}{36} = 1 - 0.25 = 0.75$.

29. Identify the issue if residuals are mostly positive for small $x$ and mostly negative for large $x$.

    Systematic bias; the model consistently over/underestimates. Sign change in residuals reveals the model doesn't match data's true trend.

30. Compute $R^2$ if $SSE=20$ and $SST=80$.

    $R^2=0.75$. Using $R^2 = 1 - \frac{20}{80} = 1 - 0.25 = 0.75$.

31. What does a residual plot that is randomly scattered about $0$ indicate?

    The model form is appropriate (no systematic pattern). Random scatter means the model captures the data's trend without systematic bias.

32. What does a smaller value of $SSE$ indicate when comparing two regression models on the same data?

    Smaller $SSE$ indicates a better fit to the data. SSE measures total squared deviations; lower means less error.

33. Identify the issue if a model predicts negative values for a quantity that cannot be negative (for example, population).

    The model is not contextually valid over that interval. Models must respect physical constraints of the modeled quantity.

34. What does it mean to validate a function model against data?

    Check that predictions match data within an acceptable error. Models are valid when their outputs closely approximate actual observed values.

35. What is $SSE$ in an $R^2$ calculation using predictions $\hat{y}_i$?

    $SSE=\sum (y_i-\hat{y}_i)^2$. Sum of squared errors measures total prediction error.

36. Which option indicates a better model: $R^2=0.98$ with clear residual pattern, or $R^2=0.95$ with random residuals?

    $R^2=0.95$ with random residuals. Random residuals indicate proper model form despite slightly lower $R^2$.

37. Choose the better fit by $R^2$: Model A has $R^2=0.91$, Model B has $R^2=0.86$.

    Model A. $R^2 = 0.91$ explains 91% of variance vs 86% for Model B.

38. What is an influential point in regression?

    A point whose removal changes the regression model noticeably. These points have high leverage on the regression line's position.

Study these flashcards