1. **Create a Contingency Table:**
Organize your data into a table where rows represent gender and columns represent color choices. For example:
| | Green | Pink | Blue | Total |
|--------------|-------|------|------|-------|
| Boys | O1 | O2 | O3 | B |
| Girls | O4 | O5 | O6 | G |
| **Total** | T1 | T2 | T3 | N |
- O1, O2, O3, O4, O5, O6: Observed frequencies
- B: Total number of boys
- G: Total number of girls
- T1, T2, T3: Totals for each color
- N: Total number of participants
2. **Calculate Expected Frequencies:**
For each cell in the table, calculate the expected frequency using the formula:
E_ij = (Row Total_i * Column Total_j) / Grand Total
For instance, the expected frequency for boys choosing green would be:
E_B_Green = (B * T1) / N
3. **Compute the Chi-square Statistic:**
Use the formula:
χ^2 = Σ ((O_ij - E_ij)^2 / E_ij)
where `O_ij` is the observed frequency and `E_ij` is the expected frequency for each cell.
4. **Determine Degrees of Freedom:**
Degrees of freedom `df` are calculated as:
df = (Number of Rows - 1) * (Number of Columns - 1)
For this table:
df = (2 - 1) * (3 - 1) = 2
5. **Compare with Critical Value or Compute p-value:**
Compare the Chi-square statistic with the critical value from the Chi-square distribution table for `df` at your chosen significance level (usually 0.05). Alternatively, compute the p-value.
6. **Interpret Results:**
- **If the Chi-square statistic is higher than the critical value** or if the p-value is less than 0.05, **reject the null hypothesis**. This indicates a significant association between gender and color preference.
- **If not**, there’s insufficient evidence to suggest a significant association.
In summary, the Chi-square test helps assess whether the observed differences in color preferences between boys and girls are statistically significant or if they could have occurred by chance.
### **Degrees of Freedom Calculation:**
1. **Goodness-of-Fit Test:**
- **Purpose:** Tests if the observed frequencies match an expected distribution.
- **Degrees of Freedom Formula:**
df = Number of Categories - 1
- **Example:** If you’re testing preferences among 3 colors (Green, Pink, Blue),
df = 3 - 1 = 2
2. **Test of Independence (Contingency Table):**
- **Purpose:** Tests if two categorical variables are independent of each other.
- **Degrees of Freedom Formula:**
df = (Number of Rows - 1) * (Number of Columns - 1)
- **Example:** For a table with 2 rows (Boys, Girls) and 3 columns (Green, Pink, Blue),
df = (2 - 1) * (3 - 1) = 2
In summary, the degrees of freedom calculation method changes based on whether you're testing a single categorical variable (Goodness-of-Fit) or the relationship between two categorical variables (Independence). Each method uses degrees of freedom to determine the appropriate Chi-square distribution for assessing statistical significance.
### **Understanding the Chi-Square Statistic in Tobit Models**
The Chi-Square test in a Tobit regression typically refers to:
1. **Wald Test**: Tests whether individual coefficients are significantly different from zero.
- Formula:
- Chi-Square = (Beta / Standard Error)²
- This is the squared z-statistic, used to check the significance of each coefficient.
2. **Likelihood Ratio Test (LRT)**: Compares the log-likelihood of the full model versus a restricted model (e.g., only intercept).
- Formula:
- Chi-Square = -2 * (Log-Likelihood of Restricted Model - Log-Likelihood of Full Model)
3. **p-value Calculation**:
- p-value = 1 - chi2.cdf(Chi-Square, Degrees of Freedom)
- Degrees of Freedom = Number of parameters tested.
---
### **Recovering Chi-Square in Python**
Assuming you are using the `tobit` package, here’s how to compute both the **Wald test** and **Likelihood Ratio Test**.
#### **1. Wald Test for Each Coefficient**
import numpy as np
import scipy.stats as stats
from tobit import TobitModel
# Fit Tobit Model
model = TobitModel(y, X, left=0) # Assuming left-censored at 0
results = model.fit()
# Extract coefficients and standard errors
betas = results.params_
se = results.bse_
# Compute Wald Chi-Square statistic
wald_chi2 = (betas / se) ** 2
# Compute p-values from Chi-Square distribution (df=1 for each test)
p_values = 1 - stats.chi2.cdf(wald_chi2, df=1)
# Display results
for i, (beta, chi2, p) in enumerate(zip(betas, wald_chi2, p_values)):
print(f"Variable {i}: Chi-Square = {chi2:.4f}, p-value = {p:.4f}")
#### **2. Likelihood Ratio Test (LRT)**
This compares the full model to a restricted model (intercept only).
# Fit a restricted model (only intercept)
X_restricted = np.ones((X.shape[0], 1)) # Only intercept
restricted_model = TobitModel(y, X_restricted, left=0)
restricted_results = restricted_model.fit()
# Compute Likelihood Ratio Test statistic
ll_full = results.llf_ # Log-likelihood of full model
ll_restricted = restricted_results.llf_
lrt_chi2 = -2 * (ll_restricted - ll_full)
df = X.shape[1] - 1 # Degrees of freedom (difference in parameters)
# Compute p-value
lrt_p_value = 1 - stats.chi2.cdf(lrt_chi2, df=df)
print(f"Likelihood Ratio Test: Chi-Square = {lrt_chi2:.4f}, p-value = {lrt_p_value:.4f}")
---
### **Interpreting Results**
- **Wald Test Chi-Square**: If a variable has a large Chi-Square, its coefficient is significantly different from zero.
- **Pr > Chi-Square (p-value)**: If p < 0.05, the coefficient is statistically significant.
- **Likelihood Ratio Test**: If the p-value is small, the full model is significantly better than the restricted model.
