STAT 311 | Data Analysis & Model Selection | Tampa, FL Market Study
[cite_start]This case study investigates a critical question in real estate equity: Do property appraisers apply consistent valuation criteria across diverse neighborhoods? Using a dataset of 1,264 residential transactions in Tampa, Florida, I built a series of hierarchical regression models to determine if location influences not just baseline prices, but the actual "valuation rate" of land and physical improvements[cite: 65, 82].
The core objective was to identify if "blind spots" exist in standard appraisal methodologies. [cite_start]If the relationship between appraised values and sale prices differs significantly by neighborhood, it suggests inconsistent practices with profound implications for tax fairness and market efficiency[cite: 72].
[cite_start]I utilized JMP Statistical Software to develop and compare five increasingly complex models[cite: 17, 30]:
- [cite_start]Model 1: Baseline first-order model (Land + Improvements)[cite: 31].
- [cite_start]Model 2: Main effects with neighborhood dummy variables[cite: 33].
- [cite_start]Models 3-4: Interaction models testing if Land or Improvement effects vary by location[cite: 35, 36].
- [cite_start]Model 5 (Selected): Full interaction model (
$SALES \sim LAND + IMP + NBHD + LAND \times NBHD + IMP \times NBHD$ )[cite: 37, 196].
To find the most parsimonious yet accurate model, I performed:
- [cite_start]Global F-Tests for overall significance[cite: 39].
- [cite_start]Partial F-Tests for nested model comparisons (Crucial for justifying complexity)[cite: 40, 150].
- [cite_start]Adjusted
$R^2$ and MSE Comparison to evaluate goodness-of-fit[cite: 42, 160].
- [cite_start]Statistical Justification: Model 5 achieved the highest Adjusted
$R^2$ (0.9583) and lowest RMSE (81,403.59). - [cite_start]Proven Inconsistency: Highly significant Partial F-tests (
$p < 0.0001$ ) confirmed that both land and improvement valuation practices vary systematically across neighborhoods[cite: 199, 210]. - [cite_start]Conclusion: The analysis revealed that location affects both the level and the slope of property value, indicating neighborhood-specific factors that standard assessments may overlook[cite: 208].
- [cite_start]Analysis Tool: JMP [cite: 17]
- [cite_start]Methodology: Hierarchical Regression, Interaction Modeling, Hypothesis Testing (F-tests, T-tests) [cite: 30, 105]