2026-03-24 17:03 Tags:
1. Start from the core equation
Focus on one variable (x_j):
2. What does (\beta_j) mean?
Think: increase (x_j) by 1
What happens?
So:
A 1-unit increase in (x_j) changes log-odds by (\beta_j)
But log-odds is not intuitive.
So we exponentiate.
3. Convert to something interpretable
This is the key.
4. Final interpretation (memorize this)
A 1-unit increase in (x_j) multiplies the odds by (e^{\beta_j})
5. Concrete example
Suppose:
Then:
Interpretation:
Each 1-year increase in age → odds are multiplied by 0.5
→ odds are cut in half
Another example:
Interpretation:
Each +1 in score → odds double
6. Important: this is NOT probability
This is where most people get confused.
You are NOT saying:
probability increases by 0.2
You are saying:
odds are multiplied by 2
7. Why odds instead of probability?
Because the relationship is nonlinear.
Example:
If baseline probability = 0.1
odds = 0.11
Multiply odds by 2 → 0.22
new probability ≈ 0.18
If baseline probability = 0.8
odds = 4
Multiply odds by 2 → 8
new probability ≈ 0.89
Same coefficient, very different probability change.
That’s why we use odds.
8. Direction vs magnitude
Sign of coefficient
-
(\beta > 0): increases probability
-
(\beta < 0): decreases probability
Magnitude
-
larger (|\beta|) → stronger effect
-
but better to compare using (e^{\beta})
9. Special case: binary variable
If:
Then:
means:
odds for group 1 vs group 0
Example:
-
gender (female = 1, male = 0)
-
(e^{\beta} = 1.5)
Interpretation:
females have 1.5× the odds compared to males
10. Intercept (\beta_0)
Often not meaningful unless variables are centered.
12. Common mistakes
Mistake 1
“β = 0.7 means probability increases by 0.7”
Wrong.
Mistake 2
Comparing coefficients directly across scaled vs unscaled data
Mistake 3
Ignoring that effect depends on baseline probability
13. Mental model
Think of logistic regression as:
-
linear model → log-odds
-
exponent → multiplicative effect on odds
14. One-line summary
Example: Interpreting Logistic Regression Coefficients
Model
Where:
-
tachycardia = 1→ patient has tachycardia -
tachycardia = 0→ no tachycardia
Step 1: Baseline (no tachycardia)
Convert to odds:
Convert to probability:
Interpretation
Without tachycardia, the probability of hospitalization is about 12%
Step 2: With tachycardia
Interpretation
With tachycardia, the probability of hospitalization is about 21%
Step 3: What does the coefficient mean?
We focus on:
Exponentiate:
Key Interpretation
Having tachycardia doubles the odds of hospitalization
Step 4: Verify numerically
-
Original odds = 0.135
-
New odds = 0.27
So the interpretation is exact.
Step 5: Important insight
The coefficient affects:
Odds (multiplicative change)
Not:
Probability (additive change)
Step 6: Why probability change is not fixed
From the example:
- 12% → 21% (increase of 9%)
But if baseline were different:
Suppose baseline = 50%
-
odds = 1
-
multiply by 2 → odds = 2
Now:
- 50% → 67% (increase of 17%)
Key Insight
Same coefficient:
-
always multiplies odds by 2
-
but probability change depends on baseline
Step 7: Proper professional wording
You can write:
The presence of tachycardia is associated with a 2-fold increase in the odds of hospitalization.