Interactive comparisons: Linear vs. Binary Logistic vs. Multinomial vs. Ordinal Analysis
Notice how Linear Regression (blue line) shoots past 1.0 (100%) and below 0.0 (0%), while Binary Logistic Regression (green curve) forms an S-shape bounded between 0 and 1.
| Feature | Linear Regression | Binary Logistic Regression |
|---|---|---|
| Continuous (e.g., Exam Score 0-100) | Binary / Dichotomous (0 = Fail, 1 = Pass) | |
| $Y = \beta_0 + \beta_1 X$ | $\ln(\frac{P}{1-P}) = \beta_0 + \beta_1 X$ | |
| $-\infty$ to $+\infty$ | Bounded Probability $[0, 1]$ | |
Analyze > Regression > Linear |
Analyze > Regression > Binary Logistic |
Multinomial Logistic Regression is used when the outcome variable has 3 or more unordered nominal categories (e.g., Career Choice: Tech, Business, Arts).
Predicted Probabilities:
Instead of comparing 1 vs. 0, Multinomial Logistic Regression designates one group as a Reference Category and estimates separate baseline logit models for each remaining category.
Ordinal Logistic Regression (PLUM in SPSS) handles outcome variables with an inherent order (e.g., Low, Medium, High satisfaction), assuming proportional odds.
Cumulative Likelihood:
Use Ordinal Logistic Regression when response levels have a natural rank, but distance between categories is not equal (e.g., Likert scales, disease severity stages).
Key Assumption: Parallel Lines (Proportional Odds)
Assumes the relationship between covariates and log-odds is constant across all cut-points.
| Analysis Method | Dependent Variable Type | Example Outcome |
|---|---|---|
| Linear Regression | Continuous (Interval/Ratio) | Salary ($), Blood Pressure, Test Score |
| Binary Logistic | Binary / Dichotomous (2 outcomes) | Disease (Yes/No), Pass/Fail, Churn (0/1) |
| Multinomial Logistic | Nominal (> 2 unordered categories) | Transportation choice (Car/Bus/Train) |
| Ordinal Logistic | Ordinal (> 2 ordered categories) | Pain level (Mild/Moderate/Severe) |