my-markdown-library 0.1.0

data/F24LS_md/Lecture 10 - Development_.md

@@ -0,0 +1,631 @@
+---
+title: "Lecture 10 - Development\_"
+type: slides
+week: 10
+source_path: "/Users/ericvandusen/Documents/Data88E-ForTraining/F24LS/Lecture 10 - Development_.pptx"
+---
+
+## Slide 1: Data 88E: Economic Models
+
+- Lecture 10:  Development - Randomized Controlled Trials
+
+## Slide 2: Announcements
+
+- Lab 9 released today
+- Project 3 is due next week
+
+## Slide 3: Meme?
+
+- Meme?
+
+## Slide 4: Lecture 8 Outline
+
+- A/B testing
+- Facebook Ads
+- Development Economics
+- Correlation vs Causality
+- Randomized Controlled Trials
+- RCTs in Development Economics
+- Project in Kenya on Clean Drinking Water
+- I am not covering all or most of Development Economics
+- - the recent rise of Randomized Controlled Trials in Development Economics as relates to Data Science
+
+## Slide 5: A/B testing -
+
+## Slide 6: A/B Testing in Data 8
+
+## Slide 7: A/B Testing is built into the internet !
+
+- https://marketingplatform.google.com/about/optimize/
+
+## Slide 8: A/B testing is the bread and butter of data science in Web Development
+
+## Slide 9: A/B testing in Web Design
+
+## Slide 10: The Social Dilemma - we are all constantly being AB tested
+
+## Slide 11: (untitled)
+
+- https://youtu.be/3FmX8SBIeco?feature=shared
+
+## Slide 12: https://www.miamiherald.com/news/politics-government/article246429000.html
+
+## Slide 13: Dark Facebook ads + Cambridge Analytica targeting
+	Used to discourage ( black ) voters from voting
+
+## Slide 14: And at the same time - Russia was doing the same thing!
+
+- https://www.businessinsider.com/some-of-the-russian-social-media-accounts-that-could-have-influenced-the-2016-election-2018-12\#heart-of-texas-4
+
+## Slide 15: Facebook ad library
+
+- Facebook does not curate content
+- But after 2016 - make a library of the all ads shown
+- Doesn’t say exactly who the demographic profiles are
+- Other affiliated groups are advertising at the same time
+
+## Slide 16: Sampling of ads
+
+- Trump ads - many attack ads
+- Some are meant to discourage voters
+
+## Slide 17: This looks like a deterrence ad
+
+## Slide 18: This is the ad being tested - October
+
+- Different subgroups?
+- Data on engagement?
+- How many people watch the video?
+
+## Slide 19: “I see a way out” - 7m views on Trump’s Youtube page
+
+## Slide 20: (untitled)
+
+## Slide 21: (untitled)
+
+## Slide 22: Biden ad getting tested
+
+- https://www.facebook.com/ads/library/?id=358668698526428
+
+## Slide 23: Trump ads - October 2024
+
+## Slide 24: About this ad - ~60 variants
+
+## Slide 25: Who targets me
+
+- Meta ad spends
+
+## Slide 26: Total Ad spend per target group
+
+- https://favstats.github.io/us24/targeting.html
+
+## Slide 27: Who targets me
+
+- https://favstats.github.io/us24/detailed.html
+
+## Slide 28: Contested Ad Groups
+
+## Slide 29: What is this a picture of
+
+## Slide 30: What is this a picture of?
+
+## Slide 31: (untitled)
+
+## Slide 32: End of Facebook ->  A/B testing -> Experimental Design
+
+- How do we test different populations?
+- Test within a population
+- What is the population
+- If we do randomization right we can make generalizations about the entire population
+- How do we split up populations into comparable subsets
+- Internet  - easy
+- Surveys / Drug Trials  - harder
+- Need a map of the domain to do randomization from
+- What list ,  how to check the balance
+
+## Slide 33: Data 8 Chapter 2
+
+- John Snow and the Broad Street Pump
+- 1854
+- Water source caused Cholera
+- (People didn’t believe it)
+- Comparison of similar population
+- Different water company - but similar otherwise
+- This was a “natural experiment”
+- Hygiene
+  - Microbiology
+  - Disinfection
+  - Wastewater treatment
+- http://blog.rtwilson.com/john-snows-cholera-data-in-more-formats/
+
+## Slide 34: Observational Studies ~ Confounding Variables
+
+- Inferential Thinking
+- “ what if some people worked in a factory and some did not”
+- Or ….  Does Coffee cause lung cancer?
+- -- but … smoking
+- Observational Studies
+- You can measure correlation
+- But you can’t measure causation unless you can control for all confounding variables
+- Experimental Design ( Project 3)
+- Eighty-four pairs of undergraduate volunteers between the ages of 18 and 26 from North Dakota State University participated for course credit and were randomly assigned to either an asymmetrical or symmetrical camera condition.
+
+## Slide 35: A/B Testing in Project  3
+
+## Slide 36: Clinical Trials  - Randomization of population
+
+- Phase 1 - ( 20-50 people)
+- Is the Treatments Safe?
+- Measure and Observe Side Effects
+- Phase 2 - ( 50-100 people)
+- Does the treatment works
+- Recruit people with the disease
+- Phase 3 - ( 1000s of people)
+- Randomized population
+- Placebo given to Control population
+- Double Blinded - participant and researcher
+
+## Slide 37: Covid 19 Clinical Trials - 2021
+
+- 446 in Phase 3
+- Hydroxychloroquine
+- Remdesivir
+- Mononoclonal Antibodies
+- Vaccines
+- https://clinicaltrials.gov/ct2/results?cond=COVID-19&age\_v=&gndr=&type=&rslt=&phase=2&Search=Apply
+
+## Slide 38: A study taken at Random - Argentina Vaccine study
+
+- 3000 people
+- Triple blinded
+- Participant,
+- Care Provider
+- Researcher
+- 2 shots 21 days apart
+- Completed Dec 2021
+- Primary outcome -
+- Incidence of cases
+- Secondary outcome -
+- Antibodies measured
+
+## Slide 39: Cholera in Clinical Trials
+
+## Slide 40: Cholera
+
+- 3-5 million cases per year
+- 20,000-100,000 deaths per year
+- Most deaths are in children under 5
+- Iraq, Yemen, Haiti, Zimbabwe
+- Treatment
+- Oral Rehydration Therapy
+- Antibiotics
+- Vaccine -
+- Drops to 50% effective after 1 year
+
+## Slide 41: Long term solution is clean water and sanitation
+
+- https://www.who.int/news-room/fact-sheets/detail/cholera
+
+## Slide 42: Effectiveness of Government Programs - Impact Evaluation
+
+- What is the most effective way to :
+- Save lives  (e.g for children under 5)
+- Change behavior  ( sanitation & hygiene)
+- Improve people’s welfare
+- For governments and policymakers - National or Local Governments
+- For implementers  - Local or International NGOS
+- For International Donors  ( Gates Foundation)
+
+## Slide 43: Is it the domain of Economics?
+
+- Public Health – address health issues
+- Epidemiology – measure sickness
+- Microbiology – measure contamination
+- Behavioral Science – nudges
+- Environmental Engineering – who often takes on these issues
+- Economics - costs of different approaches
+  - Technology adoption
+  - Behavior changes around technology
+- Objective outcome measures
+
+## Slide 44: Randomized Controlled Trials (RCTs)  in Economics
+
+- Also called Impact Evaluation
+- Effectiveness of Aid – interest to measure impacts of
+- international projects, health interventions, schooling, welfare
+- Impact of a program, device, innovation, system, behavior change, education
+- Causality cannot be determined in observational studies
+- Compare to a control group - a counterfactual - a comparison group
+- Endogeneity of economic choices - Remove Selection Bias!
+- In an observational study you cannot separate correlation from causality – because of factors involving  selection into the treatment. Those who sign up are different from those who do not
+
+## Slide 45: 2019 Nobel Prize in Economics
+
+- Jpal - Poverty Action Lab
+- MIT and Harvard Faculty
+- Leaders in RCT movement
+- Michael Kremer
+- Esther Duflo
+- Abhijit Banerjee
+
+## Slide 46: CEGA - UC Berkeley based  Evaluation Network
+
+- http://cega.berkeley.edu/
+- Ted Miguel
+- Josh Blumenstock
+
+## Slide 47: Michael Kremer and Ted Miguel
+
+- Worked at a school in Busia Kenya
+- There was a lot of student absenteeism
+- Intestinal parasites that could be easily treated
+  - Work with NGO that was going to pilot mass de-worming
+  - Randomize which schools would get the interventions
+- Water borne diarrheal diseases
+  - Work with NGO to improve local water sources
+  - Randomize communities getting improved springs
+  - Also have water related hygiene behaviors
+
+## Slide 48: Deworming RCT
+
+- Deworming RCT
+- Simple single pill for worm ( helminths)
+- Mass Deworming - giving pill to entire sectors of population, using existing infrastructure like schools
+
+<details><summary>Speaker notes</summary>
+
+Line up the students]
+Give them a pill and 
+
+What about handwashing, and open defacation, and shoes, and concrete floors?
+
+</details>
+
+## Slide 49: https://www.povertyactionlab.org/sites/default/files/publications/2012.3.22-Deworming.pdf
+
+- https://www.povertyactionlab.org/sites/default/files/publications/2012.3.22-Deworming.pdf
+- Data for
+- Students with treatment in treatment school
+- Students without treatment in treatment school
+- Students in other schools
+
+<details><summary>Speaker notes</summary>
+
+Outcome varialbe – 14.6 more days of school
+
+Untreated also affected – this is usually a problem
+Usually you are trying to not have spillovers if you are measuring treatment – control
+But they did some more advanced techniques to establish a counterfactual for spillovers as well.
+Contaminated the control!  But it was an awesome outcome so measure and show it
+
+</details>
+
+## Slide 50: https://www.povertyactionlab.org/sites/default/files/publications/2012.3.22-Deworming.pdf
+
+- https://www.povertyactionlab.org/sites/default/files/publications/2012.3.22-Deworming.pdf
+- Ten Years Later !!!
+- Long term panel dataset - KLPS
+- Follow original deworming panel
+- Many other aspects of panel research
+
+<details><summary>Speaker notes</summary>
+
+Kept tracking people
+Some moved all over Kenya
+
+</details>
+
+## Slide 51: Reduction in disease = 7.5% increase in schooling
+Spillover effects - Reduction by other students not treated
+Health improvements by infants and siblings – cognitive improvements due to reduced disease burden
+10 year follow-up – worked more / had better jobs / earned 20% more
+Huge Scale up efforts at the national level, across countries  - 40 million children per year
+
+- Reduction in disease = 7.5% increase in schooling
+- Spillover effects - Reduction by other students not treated
+- Health improvements by infants and siblings – cognitive improvements due to reduced disease burden
+- 10 year follow-up – worked more / had better jobs / earned 20% more
+- Huge Scale up efforts at the national level, across countries  - 40 million children per year
+- Deworming outcomes - measured by RCT structure
+
+<details><summary>Speaker notes</summary>
+
+Very influential study
+Now it is being scaled up in kenya nationally by government
+Many countries are doing it
+
+</details>
+
+## Slide 52: Spring Protection - waterborne diarrheal diseases
+
+- Households are disperse in a relatively rural area
+- Draw water from local springs
+- Roll out - randomized from list - project to protect springs and improve water quality
+
+## Slide 53: Spring Protection
+
+<details><summary>Speaker notes</summary>
+
+There is something we can do to protect the water – water quality and availability
+By digging up and protecting the spring head
+
+</details>
+
+## Slide 54: Protected Spring
+
+<details><summary>Speaker notes</summary>
+
+There is something we can do to protect the water – water quality and availability
+By digging up and protecting the spring head
+
+</details>
+
+## Slide 55: How to make this a Randomized Trial
+
+- List of 200 springs
+- Protect 50 springs each year
+- Survey before and after Springs are protected
+- All springs will be protected by the end of the project
+
+## Slide 56: But water quality not improved in Household
+
+- Recontamination in Transport!?
+- Recontamination in Home!?
+- Treat at community source?
+- Treat at household level?
+
+<details><summary>Speaker notes</summary>
+
+There is something we can do to protect the water – water quality and availability
+By digging up and protecting the spring head
+
+</details>
+
+## Slide 57: ‹\#›
+
+- WHO. 2016.  “Results of Round 1 of the WHO International Scheme to Evaluate Household Water Treatment Technologies.” World Health Organization.
+- How many people practice Household Water Treatment (HWT)?
+
+<details><summary>Speaker notes</summary>
+
+Boiling is still the most common around the world
+
+</details>
+
+## Slide 58: Options for Household Water Treatment
+
+- Boiling
+- Ceramic Filters
+- Slow Sand Filters
+- Membrane Filters
+- Solar Disinfection
+- Flocculant Sachets
+
+<details><summary>Speaker notes</summary>
+
+So you can treat the water at the household level
+
+</details>
+
+## Slide 59: Options for Household Water Treatment
+
+- Dilute Chlorine at household level
+  - Water Guard
+- Sell in local stores that carry oil, soap, rice
+- One month supply per bottle, one capful per 20l water
+
+<details><summary>Speaker notes</summary>
+
+So you can treat the water at the household level
+
+</details>
+
+## Slide 60: Social Marketing - sell it at the store, with ads
+
+<details><summary>Speaker notes</summary>
+
+And tablet – Waterguard cross marketing with PUR
+
+</details>
+
+## Slide 61: Social Marketing
+
+## Slide 62: Water Guard - Use and Cost Study
+
+- Set up study
+- Lists of springs List of Households
+- Randomization of 200 springs
+- Randomly select 8 hhs per spring
+- Protect 50 springs
+- Protect 50 springs
+- Baseline Survey
+- Follow-up 1 survey
+- Follow-up 2 survey
+- Follow-up 3 survey
+- Study Design !
+
+## Slide 63: Chlorine Testing
+
+- Are they using Water Guard?
+  - After study has passed
+  - At a random visit
+- Self – Reported
+- Validated measure
+  - Find chlorine in water
+
+## Slide 64: Hach Pocket Colorimeter
+
+- Add sachet of reagent to water sample
+- Does water turn pink
+- How pink does it turn
+
+## Slide 65: Water Guard Promotion study – 7 armed RCT
+
+- Add an intervention arm that was the most possibly convenient and lowest cost to household
+- Incentivized Promoter
+- Free Chlorine
+- Free Delivery
+- Reminders and Reinforcement
+
+## Slide 66: Water Guard Promotion Study
+
+- Usually we think of RCTs as 2 arms
+- But this study has 1 control and 6 possible treatments
+- How long afterwards is there an effect?
+
+## Slide 67: Treatment Arm 7 - Proto Dispenser
+
+- Locate it at the protected Spring.
+- Provide a secure locked supply of chlorine
+- Be rugged under local conditions
+- Provide a way to have 3ml ( capful) delivered into the neck of a jerrycan
+- Provide a way to get it periodically refilled
+- Provide a way for community ownership / guardianship
+
+## Slide 68: Inventing the Dispenser
+
+## Slide 69: Results looked very promising!
+3 Month follow-up had a very high rate of usage
+
+- Results looked very promising!
+- 3 Month follow-up had a very high rate of usage
+
+## Slide 70: Local Promoter
+
+- Local person trained & paid
+- In charge of maintaining, filling,
+- Reporting broken dispensers
+- Encouraging use
+- May be part of a local committee
+- Local promoter has always been part of dispensers from the beginning to long term rollout
+
+## Slide 71: Behavioral Economics
+
+- Economics with Psychology and Cognitive Science
+- How do people make decisions
+- Bounded Rationality
+- People are limited in information, ability to process information
+- Heuristics
+- People use rules of thumb or mental shortcuts to make decisionsStatus Quo Bias
+- Cognitive Burden of changing
+- Prospect Theory
+- Choice under uncertainty , Bounding, Loss Aversion
+- Choice Architecture - Nudge
+- Set up how people make small decisions to change behavior
+
+## Slide 72: Peer Effects in Dispensers
+
+- You do it because others are doing it
+- Visibly see others doing it
+- Discuss it , get knowledge about it, becomes “normal”
+
+## Slide 73: Salience  - noticeable, important, prominent
+
+- Make sure that a problem , and solution are visible
+  - Diarrhea – not so visible
+  - Microbes in drinking water – not visible
+- Human behaviors that reduce risks – not always visible
+- Remember to Use it – reinforcement of a daily behavior
+- See it while doing the water collecting moment
+- Visible / standing out because of the placement
+
+## Slide 74: Dispensers are now huge
+
+- USAID financing
+- Carbon Credits in some cases
+- Part of a very large RCT comparing many types of Water and Sanitation interventions
+- Many different implementations besides springs in Western Kenya
+
+## Slide 75: Kremer donates dispenser to Nobel Museum
+
+## Slide 76: https://www.evidenceaction.org/programs/safe-water-now
+
+## Slide 77: GiveWell - $65mn in January 2022  - evaluate charities
+
+## Slide 78: Effective Altruism - evidence based philanthropy
+
+- https://eaberkeley.com/
+
+## Slide 79: Lab 7 - Household Surveys - What and Why?
+
+- Big Picture – how does a survey flow and look all together
+  - Experimental Design
+  - Survey Design
+- Details – how do different modules work specifically
+- Critically – does this make sense, will the answers be useful?  What have the survey authors done to help to make answers useful?
+- I want you to see a big messy dataset or 2 or 3 that are part of a real research project, that leads to a graph at the end
+  - (that is part of someone winning a Nobel Prize?)
+
+## Slide 80: Start with Springs Sample and Map
+
+- Google Maps of Western Kenya
+- Look at how Springs as Treatment Arms are scattered
+- Think about spatial randomization
+- Tips
+- If maps not displaying reload the browser page
+- Click to Satellite within map window
+
+## Slide 81: (untitled)
+
+- LINK to SURVEY
+
+## Slide 82: Closed/Limited Answers
+
+- Want to make a specific set of answers
+  - Not Open Ended
+  - Not Text
+- Simple codes are in survey - WGP  Baseline
+  - 1=yes 2=no
+  - 88= don’t know
+- Code sheet – with numbers for fixed answers for most common responses – when there are many
+- Want answers to be in a certain range, or of a certain type
+
+## Slide 83: (untitled)
+
+## Slide 84: Treatment Arms
+
+- This variable is added to dataset afterwards
+- Arm 1 - Control
+- Arm 2 - Household Script
+- Arm 3 - Community Script
+- Arm 4 - HH + Community Script
+- Arm 5 - Flat-Fee Promoter + Coupons
+- Arm 6 - Incentivized Promoter + Coupons
+- Arm 7 - Incentivized Promoter + Dispenser at Spring
+
+## Slide 85: Elements of Survey Design
+
+- Start with  map of region, list of houses
+- Number and randomize springs
+- Number and randomize houses around each spring
+- Important for later checks
+  - Who is doing survey
+  - Who is doing data entry
+  - What if there is variability in the reliability of the answers?
+
+## Slide 86: Missing Data and skip patterns
+
+- There is missing data - and what to do about it!!!
+- NAN - pandas has routines for NAN
+- There are skip patterns in surveys
+- e.g. Depending on the Answer to Question 1
+- If answer yes go to question 1b, 1c, 1d
+- If answer no go to question 2
+
+## Slide 87: Household Construction -
+
+<details><summary>Speaker notes</summary>
+
+Traditional thatched house
+
+</details>
+
+## Slide 88: Assets?   ( D Variables in dataset)
+
+<details><summary>Speaker notes</summary>
+
+I like this picture, simple furniture
+Water storage container – jerry can – which we will see a lot of
+
+</details>
+