Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
how_to_ab_test_with_confidence_railsconf.pdf
Search
Frederick Cheung
April 13, 2021
Programming
0
43
how_to_ab_test_with_confidence_railsconf.pdf
Frederick Cheung
April 13, 2021
Tweet
Share
More Decks by Frederick Cheung
See All by Frederick Cheung
Fixing Performance and Memory Problems (RubyWine)
fcheung
0
47
Fixing Performance and Memory Problems
fcheung
2
410
Asking questions
fcheung
0
46
Extending Ruby
fcheung
1
400
Introduction to Version Control
fcheung
0
62
Other Decks in Programming
See All in Programming
Blue/Greenデプロイの導入による 運用フローの改善
kudoas
1
380
冗長なエラーログを削減し、スタックトレースを手に入れる / Reducing Verbose Error Logs and Obtaining Stack Traces
upamune
0
790
GitHub Copilotのススメ
marcy731
1
200
Rethinking UI building strategies @ SFI 2024
letelete
0
270
Rubyでたのしむクリエイティブコーディング/Enjoy Creative coding with Ruby
chobishiba
1
180
From Spring Boot 2 to Spring Boot 3 with Java 21 and Jakarta EE
ivargrimstad
0
100
ゆるい個人開発のススメ
kuroppe1819
10
990
MicrosoftのPlatform Engineeringガイドを読んで実際になにかやってみた
ymd65536
1
340
GraphQLサーバの構成要素を整理する #ハッカー鮨 #tsukijigraphql / graphql server technology selection
izumin5210
4
840
R言語の環境構築と基礎 Tokyo.R 112
bob3bob3
0
270
単体テストを書かない技術 #phpcon_odawara
o0h
PRO
27
8.3k
PostmanでAPIの動作確認が楽になった話
h455h1
0
170
Featured
See All Featured
Rebuilding a faster, lazier Slack
samanthasiow
73
8.2k
The World Runs on Bad Software
bkeepers
PRO
61
6.7k
4 Signs Your Business is Dying
shpigford
175
21k
Visualizing Your Data: Incorporating Mongo into Loggly Infrastructure
mongodb
34
8.9k
Testing 201, or: Great Expectations
jmmastey
28
6.4k
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
6
1.5k
A designer walks into a library…
pauljervisheath
200
23k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
30
6k
Music & Morning Musume
bryan
41
5.6k
The Language of Interfaces
destraynor
151
23k
CoffeeScript is Beautiful & I Never Want to Write Plain JavaScript Again
sstephenson
155
14k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
25
2.3k
Transcript
How to A/B Test with con fi dence @fglc2 Photo
by Ivan Aleksic on Unsplash
None
The Plan • Intro: What's an A/B Test? • Test
setup errors • Errors during the test • Test analysis errors • Best practices Photo by Javier Allegue Barros on Unsplash
What is an A/B test?
Buy Now Order Or
🧛🙋🙋🙋🧕🧑✈👨🌾👩💼💁🧑🎨 🧑🎤👩💼🙋👷🙋👩🏭🕵🙋🧑🚀🧝 👨🎓💁👨🏭💂👩🌾🧛🧑✈💁🧝💁 🙋🕵👩🏭👨🚀🙋🧕👨🦱👰👨🎓🕵 👩🔧🧑🚒👩🚀🧝👨🎓🥷🧑🏭🧕🧑✈🧟
💁👨🏭🙋🙋🧕🧕🧝 👩🏭👨🚀🧛👩💼💁👰👨🎓 🕵🧟💁🧑🎨🧑🎤🧕👨🎓 🙋💂👨🌾👩🏭 🕵👩🚀🧝👨🎓👨🦱🧑✈👩🔧 🕵🥷🧑🏭🧑✈👩🌾👩💼👷 🙋🙋🧑🚒🙋🧑🚀🧑✈💁 🧝🧛🙋🙋 Buy Now
Order 49 orders 56 orders
Is the difference real?
• Layouts / designs / fl ows • Algorithms (eg
recommendation engines) • Anything where you can measure a di ff erence Not just buttons!
Jargon
Signi fi cance • Is the observed di ff erence
is just noise? • p value of 0.05 = 5% chance it’s a fl uke • The statistical test depends on the type of metric • No guarantees on the magnitude of the di ff erence
Test power Photo by Michael Longmire on Unsplash Test power
Test power • How small a change do I want
to detect? • 10% to 20% is much easier to measure than 0.1% to 0.2%
Sample size • Check this is feasible! • Ideally you
don’t look / change anything until sample size reached • Be wary of very short experiments
Bayesian A/B testing
Bayesian A/B testing
Bayesian A/B testing • Allows you to model your existing
knowledge & uncertainties • Can be better at with low base rates • The underlying maths are a bit more complicated
Test setup errors
Group Randomisation Photo by Macau Photo Agency on Unsplash
class User < ActiveRecord::Base def ab_group if id % 2
== 0 'experiment' else 'control' end end end
class User < ActiveRecord::Base def ab_group(experiment) hash = Digest::SHA1.hexdigest( “#{experiment}-#{id}"
).to_i(16) if hash % 2 == 0 'experiment' else 'control' end end end
Non random split • Newer users in other group •
Older users in one group • New users were less loyal!
Starting too early
Home Page 50,000 Users Home Page 50,000 Users
30,000 Users 30,000 Users Home Page 50,000 Users Home Page
50,000 Users
15,000 Users 15,000 Users 30,000 Users 30,000 Users Home Page
50,000 Users Home Page 50,000 Users
Checkout Page A Checkout Page B 5,000 Users 5,000 Users
15,000 Users 15,000 Users 30,000 Users 30,000 Users Home Page 50,000 Users Home Page 50,000 Users
2600 conversions 2500 conversions Checkout Page A Checkout Page B
5,000 Users 5,000 Users 15,000 Users 15,000 Users 30,000 Users 30,000 Users Home Page 50,000 Users Home Page 50,000 Users
2600 conversions 2500 conversions Home Page 100,000 Users 60,000 Users
30,000 Users Checkout Page A Checkout Page B 5,000 Users 5,000 Users
Not agreeing setup • Scope of the test (what pages,
users, countries ...) • What is the goal? How do we measure it? • Agree *one* metric
Errors during the test Photo by Sarah Kilian on Unsplash
A test measures the impact of all differences
Ecommerce Service Recommendation Service
Ecommerce Service Recommendation Service 10x more crashes
Repeated signi fi cance testing • Invalidates signi fi cance
calculation • Di ffi cult to resist! • Stick to your Sample Size • This is fi ne with Bayesian A/B testing
Test analysis errors Photo by Isaac Smith on Unsplash
Do the maths • Use the appropriate statistical test •
Signi fi cance on one metric does not imply signi fi cance on another
Outliers Photo by Ministerie van Buitenlandse Zaken
Photo by Ministerie van Buitenlandse Zaken
Photo by Ministerie van Buitenlandse Zaken
Understanding the domain
-4 -3 -2 -1 0 week 1 week 2 week
3
-4 -2 0 2 4 6 8 week 1 week
2 week 3 week 4 week 5 week 6 week 7
Results splitting
💰
💰
We aren't neutral
If the result is 'right' 🎉
If the result is 'wrong' • Start looking at result
splits • Start digging for potential errors • Hey what about this other metric • Well documented test can help
Best practices Photo by SpaceX on Unsplash
Don't reinvent the wheel • Split, Vanity gems do a
good job • Consider platforms (Optimizely, Google Optimize) • But understand your tool, drawbacks
Resist the urge to check/tinker • Repeated signi fi cance
testing • Changing the test while it is running (restart the test if necessary)
A/A tests • Do the full process but with no
di ff erence between the variants • Allows you to practise
Be wary of overtesting • Let's test everything! • Can
be paralysing/time consuming • Not a substitute for vision / talking to your users
Document your test • Metric (inc. outliers etc.) • Success
criteria • Scope • Sample size / test power • Signi fi cance calculation/process • Meaningful variant names
Thank you! @fglc2
Further Reading • https://www.evanmiller.org/how-not-to-run-an-ab-test.html • https://making.lyst.com/bayesian-calculator/ • https://www.chrisstucchio.com/blog/2014/ bayesian_ab_decision_rule.html @fglc2