Delta of niet?

Bij het analyseren van ratiostatistieken komt het voor dat de analyseeenheid in de metriek die je onderzoekt gedetailleerder is dan de randomisatie-eenheid. Zo kunnen we randomiseren op user_id en toch het effect van onze wijzigingen willen meten op per-sessie- of pageview-metrieken. Dit schendt enkele aannames rond de berekening van de variantie die we gebruiken in significantietests. De Delta-methode kan helpen om met dit verschil om te gaan.

Deze oefening maakt deel uit van de cursus

A/B-testen in Python

Cursus bekijken

Praktische interactieve oefening

Zet theorie om in actie met een van onze interactieve oefeningen.

Begin met trainen

Deze oefening maakt deel uit van de cursus

A/B-testen in Python

SkillTag.level.intermediateSkillTag.label

4.8+

Begin de cursus gratis

In this chapter, you’ll learn the foundations of A/B testing. You’ll explore clear steps and use cases, learn the reasons and value of designing and running A/B tests, and discover the most commonly used metrics design and estimation frameworks.

Exercise 1: What is A/B testing?Exercise 2: When an A/B test is not best Exercise 3: A/B testing steps Exercise 4: Randomization effects Exercise 5: Why run experiments?Exercise 6: Correlation visualization Exercise 7: Correlation or causation?Exercise 8: Metrics design and estimation Exercise 9: Means and proportions Exercise 10: Ad impressions metrics

In Chapter 2, you’ll cover the experiment design process. Starting with learning how to formulate strong A/B testing hypotheses, you’ll also cover statistical concepts such as power, error rates, and minimum detectable effects. You’ll finish the chapter by learning to estimate the appropriate sample size needed to yield conclusive results and tackle scenarios with multiple comparisons.

Exercise 1: Hypothesis formulation and distributions Exercise 2: Strong hypothesis formulation Exercise 3: Plotting distributions Exercise 4: Central limit theorem for means Exercise 5: Experimental design: setting up testing parameters Exercise 6: Interpreting p-values Exercise 7: Error rates in the wild Exercise 8: Experimental design: power analysis Exercise 9: Plotting power curves Exercise 10: Sample size for means Exercise 11: Sample size for proportions Exercise 12: Multiple comparisons tests Exercise 13: Is a multiple comparisons correction needed?Exercise 14: Corrected p-values

Here, you’ll discover a concrete workflow for cleaning, preprocessing, and exploring AB testing data, as well as learn the necessary sanity checks we need to follow to ensure valid results. You’ll explore a detailed explanation and example of analyzing difference in proportions A/B tests.

Exercise 1: Data cleaning and exploratory analysis Exercise 2: Proportions EDA Exercise 3: A/B test data cleaning Exercise 4: Sanity checks: Internal validity Exercise 5: SRM Exercise 6: Distributions balance Exercise 7: Sanity checks: external validity Exercise 8: Novelty effects detection Exercise 9: Simpson's paradox in action Exercise 10: Analyzing difference in proportions A/B tests Exercise 11: Difference in proportions A/B test Exercise 12: Interpretation of confidence intervals Exercise 13: Confidence intervals for proportions

In the final chapter, you’ll develop frameworks for analyzing differences in means and leveraging non-parametric tests when several assumptions aren't met. You’ll also learn how to apply the Delta method when analyzing ratio metrics and discover the best practices and some advanced topics to continue the A/B testing mastery journey.

Exercise 1: Verschillen in gemiddelden in A/B-tests analyseren Exercise 2: T-test voor verschil in gemiddelden Exercise 3: Paarsgewijze t-toetsen Exercise 4: Niet-parametrische statistische toetsen Exercise 5: Parametrisch of niet-parametrisch?Exercise 6: Mann-Whitney U-toets Exercise 7: Chi-kwadraattoets voor onafhankelijkheid Exercise 8: Ratiometrieken en de delta-methode Exercise 9: Delta of niet?

Huidige oefening

Exercise 10: Delta-methode Exercise 11: Intro: best practices en geavanceerde onderwerpen voor A/B-testen Exercise 12: Best practices Exercise 13: Dag-van-de-week-effect Exercise 14: Afronding: A/B-testen in Python