Good hypothesis testing is surprising!

Timothy Fitz
Timothy FitzIndependent Contractor at Timothy Fitz
GOOD

HYPOTHESIS

TESTING
IS SURPRISING

Timothy Fitz (.com)
Timothy Fitz
Coined Continuous Deployment
Technical Lead at IMVU
CTO at Canvas
Software Consultant
TimothyFitz.com
SystemsLive.com
@TimothyFitz
WHICH SHADE OF BLUE?
Buy Buy Buy
Buy Buy Buy
Buy Buy Buy
Buy Buy Buy
MULTI-ARMED BANDIT
GOOD HYPOTHESIS
• Documented
• Testable
• Actionable
• Interesting
• Surprising
DOCUMENTED
NCSU Libraries Mobile Redesign for 2012
Objective
The redesign of the NCSU Libraries Mobile website will improve the extensibility of the service and introduce several user experience
enhancements to the existing set of mobile library services.
Requirements
● Redesign the homepage interface in a way that enables deployment of new apps with minimal effort
● Migrate mobile dev framework from forked MIT Mobile Web framework to jQuery Mobile framework
● Implement editorial recommendations on app names, taglines, and graphic identity; update the color scheme to be consistent with
the NCSU Libraries website
● Perform a comprehensive review of the usage stats; use this information to remove underutilized services
● Maintain or improve usage data collection; improve usage data reporting tools
● Retire legacy MoboiLIB service
Out of Scope
● Development of brand new apps
● Deep redesign of existing apps
● Custom iPad mobile website
● Native app deployment (iOS/Android)
Team
Project Manager: Tito Sierra
UI/UX Lead: Adam Rogers
Tech Lead: Cory Lown
Schedule
Launch: 2012.12.08
DOCUMENTED
Hypothesis
Students want to use their phones to track their
outstanding library books.
The details really matter.
Tiny code changes can change
results dramatically.
TESTABLE
• Are you collecting the data?
• Can you collect and analyze it quickly?
• Is it statistically significant?
HIGHTEMPOTESTING
3 experiments per week!
Hypothesis:
Adding Group Chat (a 12-month effort) will result in
a 25% increase in 3-month retention.
Hypothesis: Users want to chat with strangers.
Hypothesis: Users want to chat with strangers
about specific topics.



Hypothesis: Users want a directory of group chats.
Hypothesis: Users who group chat will

have higher retention.
Hypothesis: Users want to chat with strangers.
Hypothesis: Users want to chat with strangers
about specific topics.



Hypothesis: Users want a directory of group chats.
Hypothesis: Users who group chat will

have higher retention.
✓
Hypothesis: Users want to chat with strangers.
Hypothesis: Users want to chat with strangers
about specific topics.



Hypothesis: Users want a directory of group chats.
Hypothesis: Users who group chat will

have higher retention.
✓
Hypothesis: Users want to chat with strangers.
Hypothesis: Users want to chat with strangers
about specific topics.



Hypothesis: Users want a directory of group chats.
Hypothesis: Users who group chat will

have higher retention.
ACTIONABLE
The results of testing your
hypothesis cause action.
ACTIONABLE
Hypothesis: Adding a new payment method will
increase overall revenue by 5%
• If confirmed, we will keep the payment method
INTERESTING
The action you take is
important and

dependent on the outcome.
Hypothesis: Adding a new payment method will
increase overall revenue by 5%
• If confirmed, we will keep the payment method
• If denied, we will keep the payment method
INTERESTING+DOCUMENTED
GOOD PRODUCT MANAGERS
Can rationalize ANY action
from ANY outcome.

:( :( :(
Hypothesis: Adding a new payment method will
increase overall revenue by 2%
• If confirmed, we will keep the payment method
• If denied, we will DELETE the payment method
INTERESTING+DOCUMENTED
SURPRISING
Bayesian definition of surprise:





“The total number of wows experienced when
simultaneously considering all models is obtained
through the integration over the model class.”
http://ilab.usc.edu/surprise/
SURPRISING
surprise
=
change in beliefs
SURPRISING
The results of your hypothesis tests
should frequently and significantly
change your beliefs about your product!
Don’t let a single hypothesis
test invalidate

your whole vision.
GOOD HYPOTHESIS
• Documented
• Testable
• Actionable
• Interesting
• Surprising
1 of 26

Recommended

Identifying VariablesIdentifying Variables
Identifying Variablesjane1015
5K views25 slides
Kinds Of VariableKinds Of Variable
Kinds Of VariableDr. Cupid Lucid
114.4K views8 slides
Identifying variablesIdentifying variables
Identifying variableskimiworley
72K views22 slides
Types of VariablesTypes of Variables
Types of VariablesAli Mustafa
239.9K views31 slides
Research hypothesisResearch hypothesis
Research hypothesisNursing Path
187.5K views25 slides

More Related Content

More from Timothy Fitz(12)

Scaling Up Continuous DeploymentScaling Up Continuous Deployment
Scaling Up Continuous Deployment
Timothy Fitz1.5K views
Realtime web2012Realtime web2012
Realtime web2012
Timothy Fitz2K views
Continuous DeploymentContinuous Deployment
Continuous Deployment
Timothy Fitz1.9K views
Continuous DeploymentContinuous Deployment
Continuous Deployment
Timothy Fitz838 views
ShdhShdh
Shdh
Timothy Fitz10 views
Socket.ioSocket.io
Socket.io
Timothy Fitz3.1K views
ShdhShdh
Shdh
Timothy Fitz565 views
Continuous DeploymentContinuous Deployment
Continuous Deployment
Timothy Fitz2.8K views

Recently uploaded(20)

Advanced API Mocking TechniquesAdvanced API Mocking Techniques
Advanced API Mocking Techniques
Dimpy Adhikary17 views
LAVADORA ROLO.docxLAVADORA ROLO.docx
LAVADORA ROLO.docx
SamuelRamirez835247 views
HarshithAkkapelli_Presentation.pdfHarshithAkkapelli_Presentation.pdf
HarshithAkkapelli_Presentation.pdf
harshithakkapelli10 views
SAP FOR TYRE INDUSTRY.pdfSAP FOR TYRE INDUSTRY.pdf
SAP FOR TYRE INDUSTRY.pdf
Virendra Rai, PMP14 views
Unleash The MonkeysUnleash The Monkeys
Unleash The Monkeys
Jacob Duijzer7 views

Good hypothesis testing is surprising!

  • 2. Timothy Fitz Coined Continuous Deployment Technical Lead at IMVU CTO at Canvas Software Consultant TimothyFitz.com SystemsLive.com @TimothyFitz
  • 3. WHICH SHADE OF BLUE? Buy Buy Buy Buy Buy Buy Buy Buy Buy Buy Buy Buy
  • 5. GOOD HYPOTHESIS • Documented • Testable • Actionable • Interesting • Surprising
  • 6. DOCUMENTED NCSU Libraries Mobile Redesign for 2012 Objective The redesign of the NCSU Libraries Mobile website will improve the extensibility of the service and introduce several user experience enhancements to the existing set of mobile library services. Requirements ● Redesign the homepage interface in a way that enables deployment of new apps with minimal effort ● Migrate mobile dev framework from forked MIT Mobile Web framework to jQuery Mobile framework ● Implement editorial recommendations on app names, taglines, and graphic identity; update the color scheme to be consistent with the NCSU Libraries website ● Perform a comprehensive review of the usage stats; use this information to remove underutilized services ● Maintain or improve usage data collection; improve usage data reporting tools ● Retire legacy MoboiLIB service Out of Scope ● Development of brand new apps ● Deep redesign of existing apps ● Custom iPad mobile website ● Native app deployment (iOS/Android) Team Project Manager: Tito Sierra UI/UX Lead: Adam Rogers Tech Lead: Cory Lown Schedule Launch: 2012.12.08
  • 7. DOCUMENTED Hypothesis Students want to use their phones to track their outstanding library books.
  • 8. The details really matter. Tiny code changes can change results dramatically.
  • 9. TESTABLE • Are you collecting the data? • Can you collect and analyze it quickly? • Is it statistically significant?
  • 11. Hypothesis: Adding Group Chat (a 12-month effort) will result in a 25% increase in 3-month retention.
  • 12. Hypothesis: Users want to chat with strangers. Hypothesis: Users want to chat with strangers about specific topics.
 
 Hypothesis: Users want a directory of group chats. Hypothesis: Users who group chat will
 have higher retention.
  • 13. Hypothesis: Users want to chat with strangers. Hypothesis: Users want to chat with strangers about specific topics.
 
 Hypothesis: Users want a directory of group chats. Hypothesis: Users who group chat will
 have higher retention. ✓
  • 14. Hypothesis: Users want to chat with strangers. Hypothesis: Users want to chat with strangers about specific topics.
 
 Hypothesis: Users want a directory of group chats. Hypothesis: Users who group chat will
 have higher retention. ✓
  • 15. Hypothesis: Users want to chat with strangers. Hypothesis: Users want to chat with strangers about specific topics.
 
 Hypothesis: Users want a directory of group chats. Hypothesis: Users who group chat will
 have higher retention.
  • 16. ACTIONABLE The results of testing your hypothesis cause action.
  • 17. ACTIONABLE Hypothesis: Adding a new payment method will increase overall revenue by 5% • If confirmed, we will keep the payment method
  • 18. INTERESTING The action you take is important and
 dependent on the outcome.
  • 19. Hypothesis: Adding a new payment method will increase overall revenue by 5% • If confirmed, we will keep the payment method • If denied, we will keep the payment method INTERESTING+DOCUMENTED
  • 20. GOOD PRODUCT MANAGERS Can rationalize ANY action from ANY outcome.
 :( :( :(
  • 21. Hypothesis: Adding a new payment method will increase overall revenue by 2% • If confirmed, we will keep the payment method • If denied, we will DELETE the payment method INTERESTING+DOCUMENTED
  • 22. SURPRISING Bayesian definition of surprise:
 
 
 “The total number of wows experienced when simultaneously considering all models is obtained through the integration over the model class.” http://ilab.usc.edu/surprise/
  • 24. SURPRISING The results of your hypothesis tests should frequently and significantly change your beliefs about your product!
  • 25. Don’t let a single hypothesis test invalidate
 your whole vision.
  • 26. GOOD HYPOTHESIS • Documented • Testable • Actionable • Interesting • Surprising