Log transformation: Applies the natural logarithm to the dependent variable to linearize exponential relationships.
- Plot original data to identify non-linear pattern
- Apply log transformation to appropriate variable
- Plot transformed data to verify linearity
- Fit regression line to transformed data
- Interpret results in context
Time: 1, 2, 3, 4, 5
Population: 100, 250, 600, 1400, 3200
This appears to be exponential growth (non-linear)
ln(Population): ln(100), ln(250), ln(600), ln(1400), ln(3200)
ln(Population): 4.605, 5.521, 6.397, 7.244, 8.071
Time (x): 1, 2, 3, 4, 5
ln(Population) (y): 4.605, 5.521, 6.397, 7.244, 8.071
Calculate slope and intercept using least squares
Slope (b) ≈ 0.866
Intercept (a) ≈ 3.741
Regression equation: ln(ŷ) = 3.741 + 0.866x
Exponentiate both sides: ŷ = e^(3.741 + 0.866x)
ŷ = e^3.741 × e^(0.866x) = 42.1 × e^(0.866x)
The log transformation linearizes the exponential growth. The transformed regression equation is ln(ŷ) = 3.741 + 0.866x, indicating exponential growth in the original scale.
• Log transformation: Used for exponential relationships
• Back-transformation: Exponentiate to return to original scale
• Pattern recognition: Exponential growth suggests log transformation
• Original: Exponential growth pattern
• Transformed: Linear relationship
• Growth rate: Approximately 86.6% per hour
Square root transformation: Applies the square root function to reduce the impact of large values and linearize quadratic relationships.
Day: 1, 3, 5, 7, 9
Area: 1, 9, 25, 49, 81
Notice: Area = Day², which is a quadratic relationship
√Area: √1, √9, √25, √49, √81
√Area: 1, 3, 5, 7, 9
Day (x): 1, 3, 5, 7, 9
√Area (y): 1, 3, 5, 7, 9
Perfect linear relationship: y = x
Slope (b) = 1
Intercept (a) = 0
Regression equation: √ŷ = 0 + 1×x = x
Square both sides: ŷ = x²
Original: R² = 1.0 (perfect quadratic)
Transformed: R² = 1.0 (perfect linear)
The square root transformation successfully linearizes the quadratic relationship. The transformed regression equation is √ŷ = x, confirming the original quadratic pattern.
• Square root transformation: Used for quadratic relationships
• Pattern recognition: Perfect squares suggest square root transformation
• Linearity assessment: Check if transformed data is linear
• Original: Quadratic relationship (area grows as day squared)
• Transformed: Linear relationship (square root of area grows linearly)
• Perfect fit achieved through transformation
Reciprocal transformation: Applies the reciprocal function (1/x) to linearize inverse relationships, such as those described by Boyle's Law.
Pressure: 1, 2, 3, 4, 5
Volume: 100, 50, 33.3, 25, 20
Product PV ≈ 100 (constant), suggesting inverse relationship
1/Volume: 1/100, 1/50, 1/33.3, 1/25, 1/20
1/Volume: 0.01, 0.02, 0.03, 0.04, 0.05
Pressure (x): 1, 2, 3, 4, 5
1/Volume (y): 0.01, 0.02, 0.03, 0.04, 0.05
Linear relationship appears: y ≈ 0.01x
Slope (b) ≈ 0.01
Intercept (a) ≈ 0
Regression equation: 1/ŷ = 0.01x
1/ŷ = 0.01x → ŷ = 1/(0.01x) = 100/x
This confirms Boyle's Law: V = k/P where k = 100
PV = P × (100/P) = 100 (constant)
This confirms the inverse relationship
The reciprocal transformation linearizes the inverse relationship between pressure and volume. The transformed regression confirms Boyle's Law with PV = 100.
• Reciprocal transformation: Used for inverse relationships
• Physical laws: Boyle's Law suggests reciprocal transformation
• Constant product: PV = constant indicates inverse relationship
• Original: Inverse relationship (volume decreases as pressure increases)
• Transformed: Linear relationship (reciprocal of volume increases linearly with pressure)
• Physical law confirmed through transformation
Linearization: The process of applying mathematical transformations to convert a non-linear relationship into a linear one.
Log transformation: Converts exponential relationships to linear ones by taking the natural logarithm of variables.
Square root transformation: Reduces the impact of large values and linearizes quadratic relationships.
Power transformation: A family of transformations that includes log, square root, and reciprocal as special cases.
Box-Cox transformation: A more general family of power transformations that includes the optimal power parameter.
- Visual inspection: Plot original data to identify non-linear patterns
- Pattern recognition: Identify the type of non-linearity (exponential, quadratic, inverse, etc.)
- Select transformation: Choose appropriate transformation based on pattern
- Apply transformation: Transform variables using selected method
- Fit regression: Perform linear regression on transformed variables
- Assess linearity: Check if transformed data shows linear relationship
- Validate assumptions: Verify regression assumptions on transformed scale
- Interpret results: Understand the relationship in original scale
• Log transformation: Exponential growth/decay, multiplicative effects, wide range of values
• Square root: Quadratic relationships, count data, variance stabilization
• Reciprocal: Inverse relationships, rates, physical laws (Boyle's, Ohm's)
• Power transformation: General curved relationships, Box-Cox optimization
• All values must be positive for log transformation
• Square root can handle zeros but not negative values
• Reciprocal fails when original value is zero
• Transformations may affect interpretation of coefficients
Transformation comparison: Evaluating multiple transformations to determine which best linearizes the relationship.
x: 1, 2, 3, 4, 5
y: 2, 8, 18, 32, 50
Pattern: y appears to grow quadratically (y ≈ 2x²)
ln(y): 0.693, 2.079, 2.890, 3.466, 3.912
Correlation with x: r ≈ 0.95
√y: 1.414, 2.828, 4.243, 5.657, 7.071
Correlation with x: r ≈ 1.0 (perfect)
No transformation: r ≈ 0.98
Log transformation: r ≈ 0.95
Square root: r ≈ 1.0
Square root transformation: √ŷ = 1.414x
Back-transform: ŷ = (1.414x)² = 2x²
Original data fits ŷ = 2x² perfectly
Square root transformation achieves perfect linearity
The square root transformation achieves the best linearity with perfect correlation (r = 1.0). The original relationship is quadratic: ŷ = 2x².
• Transformation selection: Choose based on highest correlation
• Quadratic pattern: Square root transformation is most effective
• Validation: Verify the identified relationship
Pharmacokinetic modeling: Using transformations to understand drug elimination patterns and half-life.
Hours: 1, 2, 4, 8, 12
Concentration: 50, 25, 12.5, 6.25, 3.125
Pattern: Concentration halves approximately every 2 hours
ln(Concentration): 3.912, 3.219, 2.526, 1.833, 1.139
This appears linear with respect to time
Time (x): 1, 2, 4, 8, 12
ln(Concentration) (y): 3.912, 3.219, 2.526, 1.833, 1.139
Regression: ln(ŷ) = 4.265 - 0.259x
ŷ = e^(4.265 - 0.259x) = e^4.265 × e^(-0.259x)
ŷ = 71.1 × e^(-0.259x)
The rate constant k = 0.259 hr⁻¹
Half-life = ln(2)/k = 0.693/0.259 ≈ 2.68 hours
The drug follows first-order elimination kinetics
Dosing intervals should consider the ~2.7-hour half-life
The log transformation reveals first-order elimination kinetics. The drug has a half-life of approximately 2.7 hours, which is crucial for dosing recommendations.
• Exponential decay: Log transformation linearizes first-order kinetics
• Medical interpretation: Half-life is clinically relevant
• Real-world application: Transformations enable practical insights
Regression analysis: A statistical method that examines the relationship between dependent and independent variables.
Linearity assumption: The fundamental assumption in linear regression that the relationship between variables is linear.
Residual analysis: Examining the differences between observed and predicted values to validate model assumptions.
Model specification: The process of selecting the appropriate functional form for the regression model.
Scale invariance: Properties that remain unchanged under transformations of measurement scale.
- Exploratory data analysis: Plot raw data and calculate basic statistics
- Pattern identification: Recognize common non-linear patterns
- Transformation selection: Choose appropriate transformation based on pattern
- Model fitting: Apply regression to transformed variables
- Diagnostic checking: Validate assumptions on transformed scale
- Model comparison: Compare different transformations using criteria
- Interpretation: Translate results back to original scale
- Validation: Check model performance with new data
• Preservation of relationships: Transformations maintain the underlying relationship while changing its representation
• Reversibility: Most transformations are invertible, allowing interpretation in original units
• Assumption validation: Check all regression assumptions on the transformed scale
• Model selection: Use criteria like R², AIC, or BIC to compare transformations
• Domain expertise: Consider subject matter knowledge when selecting transformations
• Transformations may affect the interpretation of standard errors and confidence intervals
• Prediction intervals should be computed on the transformed scale and then back-transformed
• Some transformations compress the range of values, others expand it
• Consider robust regression methods when transformations don't fully address issues