Why Goodness of fit evaluation in SciPy? - Purpose & Use Cases

The Big Idea

What if you could instantly know if your data truly fits your expectations without guessing?

The Scenario

Imagine you have collected data on daily sales for a store and want to check if it follows a normal pattern. You try to eyeball the data or calculate some averages by hand to see if your guess fits.

The Problem

Doing this manually is slow and tricky. You might miss subtle differences, make calculation mistakes, or spend hours comparing numbers without a clear answer.

The Solution

Goodness of fit evaluation uses statistical tests to quickly and accurately measure how well your data matches a chosen model. It gives clear numbers and confidence levels, removing guesswork.

Before vs After

✗ Before

mean = sum(data)/len(data)
# eyeball histogram and guess fit

✓ After

from scipy import stats
stat, p = stats.kstest(data, 'norm')
print(f'p-value: {p}')

What It Enables

It enables you to trust your data models and make decisions based on solid evidence, not just intuition.

Real Life Example

A health researcher uses goodness of fit tests to check if patient recovery times follow an expected distribution, helping to validate treatment effects.

Key Takeaways

Manual checks are slow and error-prone.

Goodness of fit tests provide fast, reliable evaluation.

This builds confidence in data-driven decisions.

Practice

1. What does the chi-square goodness of fit test in scipy.stats.chisquare primarily evaluate?

easy

A. The mean difference between two samples

B. How well observed data matches expected frequencies

C. The correlation between two variables

D. The variance within a single dataset

Solution

1. Step 1: Understand the purpose of chi-square test

   The chi-square goodness of fit test compares observed data frequencies to expected frequencies to check if they match.

2. Step 2: Identify what scipy.stats.chisquare does

   This function calculates the chi-square statistic and p-value to evaluate the fit between observed and expected counts.

3. Final Answer:

   How well observed data matches expected frequencies -> Option B

4. Quick Check:

   Goodness of fit = observed vs expected match [OK]

Hint: Chi-square tests observed vs expected frequencies [OK]

Common Mistakes:

• Confusing goodness of fit with correlation
• Thinking it measures mean differences
• Mixing variance analysis with goodness of fit

2. Which of the following is the correct way to import the chi-square goodness of fit test function from scipy?

easy

A. import scipy.chisquare

B. from scipy import chisquare

C. from scipy.stats import chisquare

D. import scipy.stats.chisquare as cs

Solution

1. Step 1: Recall the module structure of scipy

   The chi-square test function is inside the stats submodule of scipy.

2. Step 2: Identify correct import syntax

   The correct import is from scipy.stats import chisquare to directly access the function.

3. Final Answer:

   from scipy.stats import chisquare -> Option C

4. Quick Check:

   Correct import = from scipy.stats import chisquare [OK]

Hint: Import from scipy.stats for statistical tests [OK]

Common Mistakes:

• Trying to import chisquare directly from scipy
• Using incorrect module paths
• Using alias without import

3. What will be the output of the following code?

from scipy.stats import chisquare
observed = [20, 30, 50]
expected = [25, 25, 50]
result = chisquare(f_obs=observed, f_exp=expected)
print(round(result.statistic, 2), round(result.pvalue, 3))

medium

A. 2.0 0.368

B. 3.0 0.223

C. 0.5 0.778

D. 1.0 0.607

Solution

1. Step 1: Calculate chi-square statistic manually

   Chi-square = sum((observed - expected)^2 / expected) = ((20-25)^2/25) + ((30-25)^2/25) + ((50-50)^2/50) = (25/25)+(25/25)+0 = 1+1+0 = 2.0

2. Step 2: Interpret p-value from scipy output

   Using scipy.stats.chisquare with these values gives a p-value around 0.368, indicating moderate fit.

3. Final Answer:

   2.0 0.368 -> Option A

4. Quick Check:

   Chi-square stat = 2.0, p-value ≈ 0.368 [OK]

Hint: Calculate chi-square stat then check p-value [OK]

Common Mistakes:

• Forgetting to square differences
• Dividing by wrong expected values
• Mixing up statistic and p-value

4. Identify the error in this code snippet for performing a chi-square goodness of fit test:

from scipy.stats import chisquare
observed = [15, 25, 35]
expected = [20, 20]
result = chisquare(f_obs=observed, f_exp=expected)
print(result)

medium

A. Observed and expected arrays have different lengths

B. chisquare function is not imported correctly

C. Expected frequencies must be integers

D. Missing p-value extraction from result

Solution

1. Step 1: Check input array lengths

   The observed array has 3 elements, but expected has only 2 elements, which is invalid for chi-square test.

2. Step 2: Understand scipy requirement

   Both observed and expected arrays must be the same length to compare frequencies correctly.

3. Final Answer:

   Observed and expected arrays have different lengths -> Option A

4. Quick Check:

   Array length mismatch causes error [OK]

Hint: Observed and expected must be same length [OK]

Common Mistakes:

• Ignoring length mismatch
• Assuming expected must be integers
• Thinking import or print is the error

5. You have observed counts of [40, 35, 25] for three categories. You expect them to be equally likely. Using scipy.stats.chisquare, what is the p-value indicating if the observed data fits the equal distribution? (Hint: expected counts are equal for all categories.)

hard

A. 0.789

B. 0.223

C. 0.456

D. 0.174

Solution

1. Step 1: Calculate expected counts for equal distribution

   Total counts = 40+35+25 = 100. Expected counts = [100/3, 100/3, 100/3] ≈ [33.33, 33.33, 33.33].

2. Step 2: Perform chi-square test with scipy

   Using chisquare(f_obs=[40,35,25], f_exp=[33.33,33.33,33.33]) gives a chi-square statistic ≈ 3.5 and p-value ≈ 0.174.

3. Final Answer:

   0.174 -> Option D

4. Quick Check:

   Unequal counts vs equal expected gives p-value ≈ 0.174 [OK]

Hint: Equal expected counts = total/number categories [OK]

Common Mistakes:

• Using observed counts as expected
• Not dividing total counts equally
• Misinterpreting p-value significance