1 of 36

Biased Beliefs and Product Choice: Theory and Evidence from Zambia

Jie Bai, Harvard Kennedy School

David Sungho Park, KDI School of Public Policy and Management

Ajay Shenoy, U.C. Santa Cruz

This document is an output from the research initiative ‘Private Enterprise Development in Low-Income Countries’ (PEDL), a programme funded jointly by the by the Centre for Economic Policy Research (CEPR) and the Foreign, Commonwealth and Development Office (FCDO), contract reference PEDL_LOA_9507_Shenoy. The views expressed are not necessarily those of CEPR or FCDO.

This project was supported in part by a grant from the Institute for Social Transformation at the University of California, Santa Cruz.

We also gratefully acknowledge financial support from:

2 of 36

Why are some firms more productive than others?

  • Technology
    • Recipe for outputs
    • Information about how to produce
  • Management
    • Motivating workers
    • Monitoring performance
    • Financial planning
    • Product choice, sourcing, and pricing

Does imperfect information / misperception about demand, suppliers, and costs lower productivity?

Do misperceptions persist through a “learning-by-doing” channel?

More productive firms 🡺 Better livelihoods for entrepreneurs

Misperceptions 🡺 high costs, unsold inventory 🡺 lost opportunities for highly marginalized people

3 of 36

This Project

Context: Single-establishment shops/stands in Lusaka, Zambia

4 of 36

This Project

Context: Single-establishment shops/stands in Lusaka, Zambia

  1. Identify similar shops, and products stocked by some but not others
  2. Survey evidence: how do beliefs about demand and costs for those who do stock compare to experience of those who do not?
  3. Randomized trial: risk-free chance to stock the product
  4. Structural model of sourcing and selling with noisy signals and learning by doing/stocking
  5. Next Project: Design and introduce an app that pools data from different shops to automatically make recommendations

5 of 36

Outline

  1. Motivation
  2. Design: Identifying and Learning about “Niche” products
  3. Survey Results
  4. RCT Design

6 of 36

Goal: Choose Very Specific Products Stocked by Some, but Not All

  • “Niche” products
    • Stores similar in main products
    • But niche product is only stocked by ~30%

  • Can compare onions to onions
    • Harmonize goods and units
    • Prices are comparable
    • Supplier to one shop can supply another

7 of 36

Existing Data: From Lusaka Productivity Study

  • Sept 2022: Screener (3000 shops)
    • Locations
    • Images of Inventory

  • Jan—Mar 2023: In-Person Survey (1000 shops)
    • Sourcing (regular suppliers and markets)
    • Main products
    • Niche products
    • Lots more….

8 of 36

Identifying Product Choice

50202309 - Sport or energy drink

50202310 - Spring or mineral water

50202305 - Fresh juice

50202306 - Soft drinks

50131701 - Fresh milk or butter products

2 Zambia-based enumerators working separately & reconciled by third

9 of 36

Identifying Product Choice

47131811 - Laundry products

53131602 - Hair care supplies

53131607 - Hand or body lotion or oil

50161509 - Natural sugars or sweetening products

50151513 - Edible vegetable or plant oils

50171551 - Cooking or table salt

53131641 - Petroleum jelly

50221300 - Flour and milled products

51171608 - Glycerine

50192902 - Shelf stable plain pasta or noodles

53102305 - Infant diapers

50193103 - Gravy mix

50181709 - Baking supplies

50201714 - Non dairy creamers

51142400 - Drugs used for vascular and migraine headaches

50201709 - Instant coffee

91101601 - Facial or body treatments

26111700 - Batteries and cells and accessories

53131619 - Cosmetics

50201710 - Leaf tea

12131706 - Matches

24111503 - Plastic bags

10 of 36

Vectorization, Step 1: Full-Dimensional Dummies

stock1

0

stock2

0

stock12

1

stock13

0

stock15

0

stock30

1

stock31

0

stock32

0

stock33

0

stock34

0

stock35

0

stock36

1

47131811 - Laundry products

53131602 - Hair care supplies

53131607 - Hand or body lotion or oil

50161509 - Natural sugars or sweetening products

50151513 - Edible vegetable or plant oils

50171551 - Cooking or table salt

53131641 - Petroleum jelly

50221300 - Flour and milled products

51171608 - Glycerine

50192902 - Shelf stable plain pasta or noodles

53102305 - Infant diapers

50193103 - Gravy mix

50181709 - Baking supplies

50201714 - Non dairy creamers

51142400 - Drugs used for vascular and migraine headaches

11 of 36

Vectorization, Step 2: Dimensional Reduction through Principal Component Analysis

Credit: Casey Cheng, Towards Data Science

12 of 36

Vectorization, Step 2: Dimensional Reduction through Principal Component Analysis

com1

3.426393

com2

-1.07705

com3

-0.66823

com4

-0.09338

com5

-0.90672

stock1

0

stock2

0

stock12

1

stock13

0

stock15

0

stock30

1

stock31

0

stock32

0

stock33

0

stock34

0

stock35

0

stock36

1

13 of 36

Clustering Shops by Product Choice

Credit: java T point

14 of 36

Chosen Clusters and Products

15 of 36

Survey Modules: Getting at Why They Stock/Don’t Stock

Stock?

What varieties?

yes

For each…

Qty Sold

Price

Cost

Suppliers

Markets

Customers ask?

Other shops stock?

Know where to buy?

Beliefs:

What would you charge?

What would it cost?

Belief distribution: sales

We know where they already source

Better place to source?

Accurate?

no

Know average (time) cost of travel

16 of 36

Outline

  1. Motivation
  2. Design: Identifying and Learning about “Niche” products
  3. Survey Results
  4. RCT Design

17 of 36

Stock vs. Not Stock

At Baseline

At screening

 

NO STOCK

YES STOCK

No response

Total

NO STOCK

283

169

13

465

YES STOCK

57

177

6

240

Total

340

346

19

705

Can’t be explained away by physical location or size of inventory

Significant predictor: How many new products have you introduced in the past year?

18 of 36

Among those who DON’T STOCK:�Know of the Product and Where to Source

  • 63% say a customer asked about the good
    • 19% daily
    • 47% weekly
    • 25% monthly
  • 74% know of similar shops that stock the product
  • 302 out of 340 know some markets where they can source the product.
    • 241 are already visiting that market for their main goods

19 of 36

But They Expect Lower Markups than are Realized by Shops that DO STOCK

If you did start stocking this product, what do you expect would be the order price for one ${niche_prod_unit}?

And what would you charge customers for one ${niche_prod_unit}?

What is the sales price per unit?

When you buy from this source: What is the order price per unit?

Can’t be explained by location!

Beliefs about target product can’t be explained by markup on main products

20 of 36

Outline

  1. Motivation
  2. Design: Identifying and Learning about “Niche” products
  3. Survey Results
  4. RCT Design

21 of 36

RCT: Objectives

Treatment: 100% subsidy to stock the product

  1. Do shops who take up the product earn higher returns than on their existing portfolio of products (adjusting for travel costs) ?
    • If yes: They could earn higher returns by stocking the product
  2. Do they continue stocking [after incentives run out]?
    • If yes: They update beliefs
  3. Are shops given more time to experiment more likely to keep stocking?
    • If yes: learning takes time, and we can estimate the rate of learning

22 of 36

Preliminary Design: Intervention

Control

110

1-week subsidy

110

2-week subsidy

110

Recommend & Offer Reimbursement

Survey:

  • Ask about 3 goods (including target good)
  • Do NOT specify which is the target good

1 Week later: return to reimburse purchases

1 Week later: return to reimburse purchases

23 of 36

Preliminary Design: Data Collection

Month 1

Month 2

Month 3

Pre-Intervention Survey

Intervention

Each Month

Week 1:

Mystery shopper

Week 2:

Mystery shopper

Phone survey

Week 3:

Mystery shopper

Week 4:

Mystery shopper

Phone survey

Monthly survey: overall sales, customers, assets, prices of main products

24 of 36

Summary

  • Research questions:
    • Does imperfect information / misperception about demand, suppliers, and costs lower productivity?
    • Do misperceptions persist through a “learning-by-doing” channel?
  • Descriptive evidence
    • Shops that stock similar products: still differ in whether they stock some products
    • Can’t be explained by location or transit costs
    • Shops that don’t stock have pessimistic beliefs
  • Next stage: RCT to allow subsidized experimentation
    • Do shops that stock earn similar returns?
    • Can shops update their beliefs about returns, and make new choices?

25 of 36

Bonus

26 of 36

Principal Component 1: Store vs. Veg. Stand

47131811 - Laundry products (laundry deterge

50202304 - Shelf stable juice

50181905 - Sweet biscuits or cookies

50202306 - Soft drinks

53131607 - Hand or body lotion or oil

50161509 - Natural sugars or sweetening prod

53131641 - Petroleum jelly

53131608 - Soaps

50151513 - Edible vegetable or plant oils (e

14111704 - Toilet tissue / toilet paper

50171830 - Dipping sauces or condiments or s

53131602 - Hair care supplies

50202310 - Spring or mineral water

47131810 - Dishwashing products (dish soap,

53131615 - Feminine hygiene products

53131502 - Toothpaste

50192402 - Nut or mixed spreads (e.g. peanut

50192902 - Shelf stable plain pasta or noodl

51171608 - Glycerine

50202300 - Non alcoholic beverages

 

50406500 - Tomatoes

50405300 - Onions

50407000 - Nominant vegetables

50405600 - Peppers

50421800 - Dried beans

50403000 - Chinese cabbages

50403400 - Cucumbers

50402300 - Cabbages

50403500 - Eggplants

50407030 - Ginger root

50405700 - Potatoes

50425300 - Dried peanuts

50401824 - Green beans

50402500 - Carrots

50304100 - Lemons

50405200 - Okras

50305000 - Oranges

50301500 - Apples

50424100 - Dried herbs

99999909 - Edible soil / stones

27 of 36

Principal Component 2: Food vs. Not Food

53131604 - Hair combs or brushes

53131629 - Makeup kits

54101604 - Earrings

53131638 - Nail polish

53102513 - Headbands

99999908 - Clothespin / clothes peg

54111704 - Watch straps or bands or bracelet

53121804 - Makeup or manicure cases

53131630 - Lip balm

53102402 - Socks

52152000 - Domestic dishes and servingware a

52121700 - Towels

52151700 - Domestic flatware and cutlery

60122900 - Beads or beading accessories

52151807 - Domestic stock pots

48101905 - Food service cups or mugs

53131507 - Toothpicks

24112602 - Jars

54101602 - Necklaces

52152200 - Dishwashing and dish storage acce

 

50181905 - Sweet biscuits or cookies

50202304 - Shelf stable juice

50151513 - Edible vegetable or plant oils (e

50131702 - Shelf stable milk or butter produ

50161509 - Natural sugars or sweetening prod

50192109 - Crisps or chips or pretzels or mi

47131811 - Laundry products (laundry deterge

53102305 - Infant diapers

50151514 - Edible vegetable or plant fats

50202306 - Soft drinks

50131701 - Fresh milk or butter products

53131502 - Toothpaste

50171830 - Dipping sauces or condiments or s

50181901 - Fresh bread

14111704 - Toilet tissue / toilet paper

50192402 - Nut or mixed spreads (e.g. peanut

50161814 - Sugar or sugar substitute candy

50192902 - Shelf stable plain pasta or noodl

50202310 - Spring or mineral water

50131600 - Eggs and egg substitutes

28 of 36

Principal Component 3: Cosmetics vs. Food

53131638 - Nail polish

53131630 - Lip balm

53121804 - Makeup or manicure cases

53131629 - Makeup kits

53131604 - Hair combs or brushes

54101604 - Earrings

54101602 - Necklaces

53131602 - Hair care supplies

53131620 - Perfumes or colognes or fragrance

53102513 - Headbands

99999901 - Hair extensions / wigs

31201610 - Glues

53131625 - Hair or beard nets

53121606 - Lipstick cases

54111704 - Watch straps or bands or bracelet

60122909 - Plastic beads

53131607 - Hand or body lotion or oil

44121618 - Scissors

60122907 - Assorted or decorative beads

60141000 - Toys

 

50406500 - Tomatoes

50221300 - Flour and milled products

50405300 - Onions

50121538 - Shelf stable fish (e.g. dried fis

99999907 - Mealed vegetable (e.g. soya mince

50151513 - Edible vegetable or plant oils (e

50407000 - Nominant vegetables

50192902 - Shelf stable plain pasta or noodl

50405600 - Peppers

50421800 - Dried beans

50192402 - Nut or mixed spreads (e.g. peanut

50221201 - Ready to eat or hot cereals

50192401 - Jams or jellies or fruit preserve

50202304 - Shelf stable juice

50171832 - Salad dressing or dips

50171551 - Cooking or table salt

50171830 - Dipping sauces or condiments or s

47131810 - Dishwashing products (dish soap,

50202309 - Sport or energy drink

50201709 - Instant coffee

29 of 36

Principal Component 4: Pharmacy vs. Food

51102000 - Antitubercular drugs

51161800 - Cough and cold and antiallergy pr

51102300 - Antiviral drugs

51142400 - Drugs used for vascular and migra

51171900 - Antiulcer and related gastrointes

51101900 - Antimalarial drugs

51241200 - Dermatologic agents (e.g. medicin

51102700 - Antiseptics

53131622 - Condoms

11121802 - Cotton

51191905 - Vitamin supplements

53131607 - Hand or body lotion or oil

53131608 - Soaps

53131620 - Perfumes or colognes or fragrance

53102504 - Gloves or mittens

53131616 - Para pharmaceutical creams or lot

53131606 - Deodorants

53131615 - Feminine hygiene products

53131613 - Skin care products

51101800 - Antifungal drugs

 

50171551 - Cooking or table salt

50161509 - Natural sugars or sweetening prod

50121538 - Shelf stable fish (e.g. dried fis

50221300 - Flour and milled products

50181905 - Sweet biscuits or cookies

50405300 - Onions

50192109 - Crisps or chips or pretzels or mi

50131600 - Eggs and egg substitutes

99999907 - Mealed vegetable (e.g. soya mince

50406500 - Tomatoes

50151513 - Edible vegetable or plant oils (e

50192902 - Shelf stable plain pasta or noodl

50202304 - Shelf stable juice

50421800 - Dried beans

50405600 - Peppers

50161814 - Sugar or sugar substitute candy

50192402 - Nut or mixed spreads (e.g. peanut

50181709 - Baking supplies

50131702 - Shelf stable milk or butter produ

50171830 - Dipping sauces or condiments or s

30 of 36

Principal Component 5: Food vs. Convenience

50405300 - Onions

50405600 - Peppers

50406500 - Tomatoes

50421800 - Dried beans

50403400 - Cucumbers

50407000 - Nominant vegetables

50425300 - Dried peanuts

50121538 - Shelf stable fish (e.g. dried fis

50407030 - Ginger root

50304100 - Lemons

50305000 - Oranges

50221300 - Flour and milled products

50402500 - Carrots

50301500 - Apples

50401824 - Green beans

99999907 - Mealed vegetable (e.g. soya mince

50403500 - Eggplants

50171551 - Cooking or table salt

50402300 - Cabbages

50301700 - Bananas

 

53131602 - Hair care supplies

99999901 - Hair extensions / wigs

53131630 - Lip balm

53121804 - Makeup or manicure cases

53131638 - Nail polish

53121606 - Lipstick cases

54101604 - Earrings

53131629 - Makeup kits

53131620 - Perfumes or colognes or fragrance

54101602 - Necklaces

91101601 - Facial or body treatments

53131607 - Hand or body lotion or oil

53131604 - Hair combs or brushes

53131613 - Skin care products

53131608 - Soaps

53131606 - Deodorants

54111704 - Watch straps or bands or bracelet

51171608 - Glycerine

53102303 - Underpants

53131616 - Para pharmaceutical creams or lot

31 of 36

Challenge: “Everything Else” Cluster

Everything else….

32 of 36

Solution: Multi-Stage Clustering

Set Aside

33 of 36

Solution: Multi-Stage Clustering

  •  

Stage 1

Stage 2

Stage 3

Candidate clustering 1

Candidate clustering 2

Candidate clustering 3

Candidate clustering 4

34 of 36

Winnowing Candidate Clusters

Start with 512 candidate clusterings

Keep only those that survive these steps:

  1. Keep only clusters where top product is stocked by at least 50%
  2. Count # of shops remaining: drop candidate if N < 1000
  3. Drop candidate if number of clusters > 10
  4. Drop if any cluster has < 50 stores

35 of 36

Most Relevant Literature

  • Huntington, Samaniego de la Parra, Shenoy (ongoing)
    • Model of retail, Productivity estimator, mentorship intervention
  • Information and imperfect management
    • Kremer, Robinson, Rostapshova (2013;2017): Stockouts and product choice
    • Conley and Udry (2010); Hanna et al. (2014): Learning about production
    • Tanaka et al. (2020): Information/forecasting about macro aggregates
  • Variation in local knowledge and management
    • Brooks et al (2018): Mentorship (about product sourcing)
    • Dalton et al. (2021): Curating local knowledge

36 of 36

Varieties

Exercise books, like those used by school children

Onions

cream for treating common skin ailments like acne or itching