Historical Legacies, Slavery, and Trust

This project investigates the enduring effects of slavery experienced by past generations on interpersonal trust in a contemporary African country. Focusing on the historical territory of a former pre-colonial kingdom, we adopt an epidemiological approach to identify three groups:
i) descendants of families currently living in communities that, during the era of slavery, were repeatedly raided by a nearby kingdom, resulting in the capture and enslavement of some of their relatives;
ii) descendants of former slaves who grew up in communities within the kingdom where their ancestors had been taken as captives; and
iii) descendants of former slave-owning families belonging to that kingdom.

To examine how these historical legacies shape present-day trust dynamics, we conduct an incentivized trust game experiment in which participants interact with anonymous partners whose family backgrounds are known. Specifically, we test whether levels of trust and trustworthiness vary depending on whether the matched partner descends from one of the three historical conditions defined above. This design allows us to capture not only overall levels of trust but also whether mistrust is selectively directed toward specific historical outgroups.

2025-05-07

2026-12-31

Our primary focus is on trust, measured as the coins sent by the first mover in a Trust Game to a second mover.
The two individuals are matched anonymously but are made aware of each others' family and location background (resident in the territory of the historical kingdom & slave origins vs. resident in raided area & slave family vs. slave-owner origins).

Definitions:
Participants are classified in three groups.
SK = descendants of families currently living in communities that, during slavery times, repeatedly suffered slave raids by a nearby kingdom, during which some of their relatives were kidnapped, enslaved, and forcibly taken into the kingdom.
SA = descendants of former slaves grown up in communities within the kingdmo where ancestors were deported as captives.
MS = descendants of former slave-owners families part of the kingdom.

Average total trust Avg trust_i = (trust_i=>SK + trust_i=>SA + trust_i=>MS)/3
Trust ingroup = coins sent by First Mover belonging to group x to Second Mover belonging to the same group.
Trust outgroup = total coins sent by First Mover belonging to group x to Second Mover belonging to the two other group.
Parochial trust = (coins sent by First mover to Second mover from same group) - (coins sent by same First mover to Second mover from a different group(s))
[We will perform a preliminary check testing whether ingroup trust is larger than outgroup trust for the three groups groups (i.e. check whether participants in our sample display parochial trust with respect to the groups as defined by their origins). We expect results consistent with the literature that shows a general tendency to display some level of parochial trust in societies across the globe.]


RQ 1: Are there differences among groups in the levels of trust?

Hp1a: Levels of average total trust today differ among the three groups. We do not have a directional hypothesis.
We test it using a Kruskal-Wallis test and a series of pairwise Wilcoxon rank-sum tests. We also use the following regression model:
Avg trust_i = alpha_0 + \beta1 SK + \beta2 SA + \beta3 X_i + \epsilon_i
SK and SA are dummies indicating groups, and X_i is a set of socio-demographic controls.

Hp1b: Levels of outgroup trust in SK are lower than in SA and MS.
We anticipate in societies that have been historically raided by outsiders, collective and family memory shaped norms of distrust toward outgroups. We have a directional hypothesis.
trust-outgroup_SK < (trust-outgroup_MS + trust_outgroup_SA)/2


RQ 2: What are the trust dynamics within and between specific groups today?

Hp2a: Descendants of enslaved individuals who live in raided communities trust more descendants of enslaved participants who today live in the kingdom communities than descendants of former slave-owners. We anticipate that distrust toward MS is driven by SK's family and collective historical memories. We have a directional hypothesis.
trust_SK=>SA > trust_SK=>MS
We will test the hypothesis using Wilcoxon rank-sum test.
trust-outgroup_SK = alpha_0 + \beta1 SA + \beta2 X_i + \epsilon_i

Hp2b: Descendants of former slave-owners trust more descendants of enslaved individuals today resident within the historical kingdom borders than descendants of enslaved individuals who live today in raided communities. We anticipate that the result is driven by MS participants perceiving as closer, and so more trustworthy, descendants of slave owners who are remained in the kingdom's territory and integrated in the local community even after the abolition of slavery and the kingdom's dissolution. We have a directional hypothesis.
trust_MS=>SA > trust_MS=>SK
We will test the hypothesis using Wilcoxon rank-sum test. We also use the following regression model:
trust-outgroup_MS = alpha_0 + \beta1 SA + \beta2 X_i + \epsilon_i

Hp2c: Descendants of enslaved individuals today resident within the historical kingdom borders display the same level of outgroup trust when paired with descendants of enslaved individuals who live today in raided communities or with descendants of former slave-owners. We do not have a directional hypothesis.
trust_SA=>MS = trust_SA=>SK
We will test the hypothesis using Wilcoxon rank-sum test. We also use the following regression model:
trust-outgroup_SA = alpha_0 + \beta1 MS + \beta2 X_i + \epsilon_i

See above

First, we will test whetherd descendants of enslaved individuals today resident within the historical kingdom borders display the same level of parochial trust of descendants of former slave-owners. We do not have a directional hypothesis: on the one hand, slavery was abolished more than a century ago and the two groups live in the same community seemingly peacefully and well integrated. On the other hand, family and collective historical memories can possibly still generate parochial dynamics across these groups.
parochial trust SA = parochial trust MS

We will test the hypothesis using Wilcoxon rank-sum test. We also use the following regression model:
parochial-trust_i = alpha_0 + \beta1 MS + \beta2 X_i + \epsilon_i
(we do not use data from SK in the analyses above)


Second, we will repeat the main analysis but for using:

1) Trustworthiness as a Secondary Outcome (incentivized measure)
We also examine trustworthiness as a secondary outcome, measured by the share of the endowment returned by the second mover in a Trust Game where the first mover’s family background is disclosed.
While our main analysis focuses on trust, we include trustworthiness for additional insights. This choice is motivated by the challenges of measuring trustworthiness in a field setting and our limited sample size.
Importantly, we opted to inform second movers of the actual decisions made by first movers, rather than using the strategy method with hypothetical scenarios. We made this design choice to enhance the salience of the second movers' decisions. However, this approach comes at a cost: we lose observations when first movers send 0 coins, since no decision is then required from the second mover. While we cannot predict the proportion of first movers who will send 0 coins, this may result in less precise estimates of trustworthiness compared to trust.

2) Other unincentivized trust measures collected:

2.a. Generalized Trust
Participants respond to survey question Q43 from Wave 8 of the World Values Survey (WVS): "In general, how much do you trust people?" This is recorded as a binary outcome.
2.b. Parochial Trust
To measure parochial trust, we include survey questions Q44–Q49 from the WVS Wave 8. These responses are measured on a 1–4 scale.
2.c. Institutional Trust
Participants are also presented with a selected set of questions from WVS Wave 8 (Q50–Q58 and Q69–Q77) to assess trust in specific institutions (e.g., government, police, courts) and organizations (e.g., church, newspapers). These outcomes are also measured on a 1–4 scale.


Mechanism – Sentiment Analysis
To explore mechanisms, we conduct a sentiment analysis of open-ended comments made by participants in response to a text describing the conditions of slaves, which was read aloud to them.

We conduct a trust game experiment with three roughly balanced groups of participants: i) descendants of families currently living in communities that, during slavery times, repeatedly suffered slave raids by a nearby kingdom, during which some of their relatives were kidnapped, enslaved, and forcibly taken into the kingdom; ii) descendants of former slaves grown up in communities within the kingdom where ancestors were deported as captives; iii) descendants of former slave-owners families part of the kingdom.
Each participant plays three trust games that are identical in structure but vary in the group identity of their matched partner.

The trust game proceeds as follows: each participant receives an initial endowment of 5 experimental coins. The first-mover (the trustor) can choose to send either 0 or 5 coins to the second-mover (the trustee). Any amount sent is quadrupled by the experimenters. The second-mover then decides how many of these coins to return to the first-mover, choosing an integer amount between 0 and the full received amount (no fractional coins are allowed). Final payoffs are calculated as follows: the first-mover earns the coins kept plus any returned amount, while the second-mover earns the amount received minus what they return.

Unlike the strategy method, second-movers are only shown the actual decision made by their matched first-mover and respond accordingly. If the first-mover sends 0, the game ends immediately and the second-mover makes no decision. Each participant is randomly assigned to play as either a first- or second-mover at the start of the experiment, and this role remains fixed across the three games.

Crucially, in one game, participants are paired with a randomly selected individual from their own group; in the other two, with someone from the two remaining different groups. The order of these two pairings is randomized. Participants are informed that each game will be paid separately and independently, and that payments will be distributed during the following weeks.

Not available

In office by a computer

Individual

No

Approximately 1800 observations from 600 individuals (this is an estimation: the stopping rule for the data collection will be the termination of available research funds)

Approximately 1800 observations from 600 individuals. Three groups (as specified above), balanced samples.

Approximately 200 participants from each group