I've been tinkering with what it would take to accelerate GenAIBro's growth, and have been reflecting quite a bit on experimentation.
Granted, GenAIBro is still in its infancy, but with the trajectory it's on, I'd be remiss if I werent planning next steps already.
Now, to the Article! When conducting Experiments, especially in Marketing, it's crucial to:
- Determine the appropriate sample size to achieve desired precision and confidence.
- Understand how to calculate p-values to assess the statistical significance of your results.
- Know when to use different statistical tests and how to interpret them.
I created this guide as a refresher, and it should walk you (read: us) through these concepts using practical examples, focusing on the use case of estimating click-through rates (CTR) on platforms.
Determining Sample Size for Estimating a Proportion
Step-by-Step Method
To determine the minimum sample size (N) required to estimate a population proportion (P) within a margin of error (δ) at a specific confidence level:
- Choose the Confidence Level and Z-value:
Common Z-values for confidence levels are:
- 90%: Z ≈ 1.645
- 95%: Z ≈ 1.96
- 99%: Z ≈ 2.58
- Decide on the Margin of Error (δ):
This represents the maximum acceptable difference between your sample estimate and the true population proportion.
- Estimate the Population Proportion (P):
Use prior data or P = 0.5 if unknown for a conservative estimate.
- Apply the Sample Size Formula:
N = (Z * √(P(1-P)) / δ)^2
Examples
Example 1: Estimated Proportion Known
- Estimated CTR (P): 10%
- Margin of Error (δ): 2%
- Confidence Level: 95% (Z = 2)
Calculations yield a minimum sample size of 900.
Example 2: Estimated Proportion Unknown
- P = 0.5
- δ = 2%
- Confidence Level: 95% (Z = 2)
Calculations yield a sample size of 2,500.
Calculating P-Values and Hypothesis Testing
The framework for hypothesis testing includes:
- Formulate Hypotheses:
Null Hypothesis (H₀) assumes no effect, while the Alternative Hypothesis (H₁) tests for an effect or difference.
- Choose Significance Level (α):
Common choices are 0.05, 0.01, or 0.10.
- Calculate Test Statistics:
For large sample sizes, use the z-statistic:
Z = (P̂ - P₀) / √(P₀(1-P₀) / N)
When to Use t-Statistic vs. z-Statistic
- Z-Statistic: Used when population standard deviation is known and sample size is large (N ≥ 30).
- T-Statistic: Used when population standard deviation is unknown, especially with small samples (N < 30).
One-Tailed vs. Two-Tailed Tests
- One-Tailed Test: Tests for an effect in one direction.
- Two-Tailed Test: Tests for any significant difference, regardless of direction.
Choosing between these depends on whether you are testing for a specific direction or any deviation.
Comprehensive Example
Scenario: Testing if a new ad design increases CTR from 5%.
- Determine Sample Size:
With a desired margin of error of ±2%, a confidence level of 95%, and an estimated CTR of 5%, you calculate a sample size of approximately 475.
- Collect Data:
Sample size: 500
Observed clicks: 35
Observed CTR: 7%
- Conduct Hypothesis Test:
Null Hypothesis (H₀): P = 0.05
Alternative Hypothesis (H₁): P > 0.05
- Calculate Test Statistic:
Using the observed data, calculate Z ≈ 2.051.
- Calculate P-Value:
One-tailed p-value ≈ 0.0202, which is less than α = 0.05, leading to the rejection of the null hypothesis.
Conclusion
By following this guide, you can:
- Determine sample sizes for statistical studies.
- Calculate p-values for statistical significance.
- Decide between t-statistics and z-statistics.
- Choose the appropriate hypothesis testing method for your analysis.
In summary, these statistical principles ensure the reliability of experimental findings and help make informed, data-driven decisions.