Income, cognition, and decision-making among urban traders

The intervention is a disguised income transfer in which a member of the study team visits the traders in the treatment group and makes a purchase from the trader's market stall approximately equal to 400 Ethiopian birr. The purchaser does not reveal themself to be a member of the study team, and the enumerator who interviews the trader does not make mention of the purchase.

2020-02-24

2020-04-04

There are three primary outcomes:
1. Quality of decision-making measures: Proxied by a measure of consistency with GARP using revealed preferences and selection of a non-dominated option
2. Take-up of a free mobile credit program
3. Index of preferences for healthy foods

There are three primary outcomes:
1a. Quality of decision-making measures: I use the same menu choice tool developed by Harbaugh, Krause and Berry (2001), in which respondents are presented with 11 menus which each show a selection of consumption bundles (comprised of oranges and bags of chips). Each menu offers a selection of bundles of goods that come from the same budget line. The menus differ by relative prices and total budget. The subject is asked to select their preferred bundle of goods from each menu.Participants are incentivized to reveal their true preferences via random selection of a menu at the end and providing the participant with their selected bundle. I use this data to construct the swaps index, which measure relative welfare loss stemming from choices that are inconsistent with the trader’s most likely preference relation, developed by Apesteguia & Ballester (2015).
1b. In addition to the eleven menus described above, I add a twelfth menu to the task that has bundles that are not from the same budget line and that includes two dominated options. I use this to construct a binary indicator reflecting selection of a dominated bundle.
2. Take-up of free mobile credit program: Immediately following their interview, traders are informed about an opportunity to receive 50 birr of mobile credit. To receive the credit, they are required to 'register' for the credit by sending a text message the following evening. The outcome is whether or not the traders registered in their assigned registration time window.
3. Preferences for healthy foods: Using the decision-making task described in Outcome 1, I generate a health preferences score that reflects the number of menus in which a trader chooses a bundle that contains at least as many oranges as bags of chips.

1. Cognitive outcomes: Attention score as measured by d2 test, and working memory score as measured by backward digit span test
2. Healthy behaviors (expected income change only): Index of hand-washing frequency, water purification, and teeth-brushing
3. Alternate mechanisms: Food security; sleep; labor hours; stress; locus of control; aspirations

1a. Measure of attention: I use the error-corrected processing speed from the d2 test. This is the number of characters processed minus the number of characters incorrectly processed (incorrect cross-outs or incorrect non-cross-outs)
1b. Working memory score: I use the Weschler total correct score to measure working memory which is the number of lists of digits correctly recited backward from the backward digit span test.
2. Healthy behaviors index: Normalized and weighted index following Anderson (2008) of the following items: (i) Appropriate hand-washing frequency in the last 24 hours (0: never, 1: sometimes, 2: always); (ii) Brushed teeth in last 24 hours (Y/N); (iii) Made use of a water purification method in last 24 hours (Y/N)
3. Quantity of sleep (hours) and quality of sleep (number of sleep disruptions) previous night; Normalized and weighted index of food consumption (whether breakfast was eaten this morning, number of meals eaten in the last 24 hours, and numbers of meals with animal-source protein in the last 24 hours); Alcohol consumption in the 12 hours; Number of hours worked yesterday and time started today; Stress (Likert scale); Depression (PHQ-2 questions, over last 24 hours instead of past two weeks); Pain (Likert scale); Locus of control scale (1-9); Aspirations scale (1-9)

I implement a lab-in-the-field study in urban markets in Addis Ababa in which I recruit low-income urban traders for a behavioral study. I first randomize traders to be interviewed on one of four non-market days (Monday, Tuesday, Thursday, or Friday). I then cross-randomize traders to be assigned with equal probability to a control group or a treatment group. The treatment group receives the intervention described above the morning of their assigned interview day.

Computer-based randomization

Individual (trader)

No

400 traders across 20 markets (sampling is clustered; treatment is not)

400 traders

Control: 200 traders; Treatment: 200 traders

To compare treatment (an unexpected increase in income) to control (200 to 200), the minimum detectable effect size is 0.28 SD.