Does perceived labor-market competition increase out-group discrimination between refugees and hosts? Evidence from Uganda and Ethiopia.

Our study consists of a randomized questionnaire module that will be implemented as part of a large labor-market-related survey.
We are targeting refugee populations and host community members living in close proximity in rural and urban settings in Uganda and Ethiopia.

Our survey sample will cover approximately
2400 refugees and 2400 hosts across rural and urban areas of Uganda
2400 refugees and 2400 hosts across rural and urban areas of Ethiopia

As part of this survey, respondents will be randomly exposed to different narratives about a fictitious member of the out-group or in-group with identical or different labor characteristics.
Some respondents of the local host group will thus listen to a narrative about a fictitious individual of the host population and others to a narrative about a refugee. The same applies to respondents of the group of refugees. This fictitious character’s labor market characteristics, such as their occupation, are included in the vignette read to the respondent. Our experiment randomly varies this information across vignettes so that it either matches or is different from the respondent’s own job characteristics.
The study aims to analyze the interplay between perceived labor market competition and out-group discrimination. Using the results of our experiment, we will evaluate prevailing attitudes regarding out-group discrimination among hosts and refugees in Ethiopia and Uganda. Moreover, we will ask whether perceived labor market competition is an important driver of discriminatory attitudes towards the out-group.

Because the experiment is being executed in Uganda and Ethiopia, we can additionally compare the results from the two countries to descriptively analyze whether differential migration policies correlate with perceived labor market competition and discriminatory responses. In an attempt to capture additional effects and take advantage of using two identical designs in two countries sharing similar characteristics, we aim at doing a cross-country analysis, pooling the two sets of data together and measuring the same outcomes. Particular attention will be put on the use of sampling weights.
The intervention is planned to start on 29 January 2022 in Uganda and 20 February 2022 in Ethiopia and end on 28 February 2022 and 15 March 2022 in Uganda and Ethiopia, respectively.

Our experiment evaluates the link between perceived labor market competition and an increase in out-group discrimination.

It relates to two different strands of the literature.
First, we can relate the experiment to a growing body of papers that look at labor market competition and labor market discrimination between refugees and hosts. Sahin Mencutec and Nashwan (2021) and Fakih and Ibrahim (2016) describe multi-faceted perspectives of labor market integration of Syrian refugees in Jordan. Clemens et al. (2018) refer to key contextual factors influencing refugees’ labor market integration and perceived competition from the side of the local population. From a methodological point, Loiacono and Vargas (2019) come closest to our approach. The authors ran a lab-in-the-field experiment in the Ugandan refugee camp Nakivale, focusing on discrimination in hiring decisions of Ugandan employers against refugees.
Our experiment also relates to a large literature on out- versus in-group behavior in the refugee-host context. A non-exhaustive list includes Cattaneo and Grieco (2020), Crisp and Turner (2009), Cameron, et al. (2006), Barron et al. (2021), Chatruc and Rozo (2021), whose research attempts to explore interactions with members of “opposite” groups. They analyze how interaction, perspective-taking, or narratives affect prejudice, pro-social behavior, and most importantly discrimination towards the out-group member.

We are linking those two strands of the literature by not only quantifying perceived labor market discrimination and out-group discrimination between local hosts and refugees - but also analyzing whether high perceived labor market competition is associated to an increase in out-group discrimination.
To do so, we are proposing an innovative experimental approach, a treatment arm matrix.
Individuals will listen to a narrative about a fictitious individual who is either belonging to the in-group or out-group (refugees/ hosts) and who is either sharing or not the same occupation as the respondent’s.

Same occupation Different occupation
In-group T1 T2

Out-group T3 T4

This 2x2 experimental matrix enables us to disentangle the mechanisms behind prejudicial attitudes against the out-group members.
Considering both the horizontal and vertical variation in our treatment arms together will enable us to disentangle whether or not discrimination to the out-group members is, in reality, a fear of loss of livelihood rather than merely a fear “of differences” against out-group members.

Given the matrix, we can test three different hypotheses.

Hypothesis 1: Individuals discriminate against members of the out-group, compared to members of the in-group.

Comparing pooled measures of discrimination (see section “Primary Outcomes”) for treatment groups T1 and T2 to treatment groups T3 and T4 allows us to quantify discriminatory views of the host population towards refugees and vice versa - independently from their labor market status.
In our heterogeneous analysis, we will analyze whether out-group discrimination prevails among the group of refugees or locals and to which extent it is linked with further individual and local background characteristics.

Hypothesis 2: Preconceived notions of labor market competition may have adverse effects on the perception of the other

By comparing pooled measures of discrimination for treatment groups T1 and T3 to treatment groups T2 and T4, we are able to analyze whether higher perceived labor market competition (proxied by a shared occupation) is linked to higher levels of individual discrimination and labor market discrimination (see section “Primary Outcomes”).
In our heterogeneous analysis, we will further explore whether this effect is particularly strong among individuals who exert locally infrequent occupations.
E.g.: The effect of exposing a farmer to a fictitious story about another farmer (high job density in an agricultural setting) might be different than exposing an architect to a fictitious story about another architect (low job density in our settings)

Hypothesis 3: Discrimination against members of the out-group is more pronounced when preconceived notions of labor market competition are strong.

In this hypothesis, we are analyzing the interplay between labor market competition and out-group discrimination.
We hypothesize that discrimination against the out-group is not happening in a universal way, but depends on the perceived threat of labor market competition arising from this group.
We thereby test whether there is targeted discrimination where individuals discriminate against the out-group because of their perception of labor market competition. From a conceptual point, this would imply that the difference in attitudes towards the in-group with the same occupation (T1) and the out-group with the same occupation (T3) is larger than the difference in attitudes towards the in-group with different occupation (T2) and the out-group with different occupation (T4): (T1-T3) is significantly different from (T2-T4).

Our premise is that hosts and refugees may feel more threatened if they face direct competition in the labor market, while they do have more welcoming attitudes when they are relatively far in terms of labor characteristics to the out-group member. Discrimination is then triggered mostly by labor market competition and not out-group membership per se.

REFERENCES:
1. Sahin Mencutek, Z., & Nashwan, A. J. (2021). Perceptions about the labor market integration of refugees: evidences from Syrian refugees in Jordan. Journal of International Migration and Integration, 22(2), 615-633.
2. Fakih, A., & Ibrahim, M. (2016). The impact of Syrian refugees on the labor market in neighboring countries: empirical evidence from Jordan. Defence and Peace Economics, 27(1), 64-86.
3. Clemens, M., Huang, C., & Graham, J. (2018). The economic and fiscal effects of granting refugees formal labor market access. Center for Global Development Working Paper, 496.
4. Loiacono, F., & Vargas, M. S. (2019). Improving Access To Labour Markets for Refugees: Evidence from Uganda. Working paper.
5. Cattaneo, C., & Grieco, D. (2021). Turning opposition into support to immigration: The role of narratives. Journal of Economic Behavior & Organization, 190, 785-801.
6. Crisp, R. J., & Turner, R. N. (2009). Can imagined interactions produce positive perceptions?: Reducing prejudice through simulated social contact. American psychologist, 64(4), 231.
7. Cameron, L., Rutland, A., Brown, R., & Douch, R. (2006).Changing children's intergroup attitudes toward refugees: Testing different models of extended contact. Child development, 77(5), 1208-1219.
8. Barron, K., Harmgart, H., Huck, S., Schneider, S. O., & Sutter, M. (2020). Discrimination, narratives and family history: An experiment with Jordanian host and Syrian refugee children. MPI Collective Goods Discussion Paper, (2020/13).
9. Chatruc, M. R., & Rozo, S. V. (2021). How Does It Feel to Be Part of the Minority? Impacts of Perspective-Taking on Prosocial Behaviors (No. 14303). Institute of Labor Economics (IZA).

2022-01-29

2022-07-31

Our primary measures of interest are;
(1) Measures of discrimination towards the fictitious individual
(2) Measures of perceived labor market competition and labor market discrimination

Questions belonging to measure (1) capture discriminatory attitudes. They cover questions on whether the respondent would accept an out-group member in their community, family, or as a work colleague. We will use these questions to build an index of how welcoming the respondent is towards members of the out-group or in-group.

Moreover, the survey includes a set of questions aimed at gauging labor-market-related discrimination and perceptions of competition (measure (2)). For example, we ask whether respondents feel in competition for the same jobs with the fictitious character. We also include normative questions eliciting the desired degree of labor market integration for the fictitious individual.

Those three sets of questions will allow us to analyze
(1) The degree of perceived labor market competition when individuals are sharing or not the same labor market characteristics
(2) The amount of discrimination towards the out-group compared with the in-group
(3) The interplay between the two: Does discrimination increase in situations where perceived labor market competition is high?

Our experimental intervention is embedded in a survey questionnaire focused on labor market outcomes and interaction between refugees and host communities in Uganda and Ethiopia.

As part of this experiment, respondents are randomly assigned to four equally sized treatment groups. Each respondent will listen to a single narrative about a fictitious individual who is either from the in-group or the out-group and either sharing or not the respondent’s labor market characteristics.
T1: In-group; same labor market characteristics (25%)
T2: In-group; different labor market characteristics (25%)
T3: Out-group; same labor market characteristics (25%)
T4: Out-group; different labor market characteristics (25%)

Due to the matching process along with labor market characteristics, only those respondents who indicate being currently employed (including self-employed) or actively searching for work will administer the experimental module.

After the narrative is read out loud to the respondents, they are asked to respond to questions about the fictitious individual, notably concerning perceived labor market competition and general attitudes (prejudice, discrimination).
The purpose of our study is to better understand discriminatory attitudes towards the out- and in-group, and to gauge the extent to which negative/positive perceptions of refugees among host communities as well as obstacles/drivers to integration among refugees are driven by the perception that both groups are competing or not in the labor market.

The experiment proceeds as follows:

(1) Respondents are randomly assigned to any of the four equally sized treatment groups (T1, T2, T3, and T4) as described above. Randomization is executed by a computer for the entire sample, which will be stratified based on group membership (refugee/ host) and location (Uganda: Isingiro District and Kampala; Ethiopia: Addis Ababa and Somali region).

(2) Within the detailed survey questionnaire, respondents are asked to report whether they are currently working or searching for work and which industry and occupation they are currently working in (or intending to turn to).

The experimental module can thus link back to the group membership of the respondent (refugee/ host), as known ex-ante from the listing, and the relevant occupation, determined by the answers in the respective survey module.

(3) In the experimental section, respondents will be randomly exposed to a standard narrative about a fictitious individual who belongs to group G and is exerting occupation O.

The random narratives stress the background of the fictitious character as a local or refugee as well as its occupational background. Narratives are only differing with respect to single words that are inserted for group G and occupation O.
G = {refugee; local host}
O = {respondent’s occupation; occupation differing from the respondent}

Single words for group G and occupation O will be auto-filled by a computer-based on survey response right before the experimental section.

T1: Respondent listens to a narrative about an in-group member who is working in the same occupation as himself.
T2: Respondent listens to a narrative about an in-group member who is working in a different occupation than himself.
T3: Respondent listens to a narrative about an out-group member who is working in the same occupation as himself.
T4: Respondent listens to a narrative about an out-group member who is working in a different occupation than himself.

The narratives (example for Uganda - symmetrically for Ethiopia) take the following form:

“[AIDA/ROBERT] is a [GROUP G: Ugandan/ refugee living in Uganda]. [She/He] (has lived in Uganda [her/his] entire life and) moved to [Isingiro district/ Kampala] five years ago. [She/He] has been working as a [OCCUPATION O: Same occupation as respondent/ different occupation] for a long time so [she/he] has a lot of experience in [her/ his] occupation. [She/He] also speaks many Ugandan local languages and English very well. [She/He] enjoys working in this profession and would recommend [her/his] friends to work in the same sector. But while being a [OCCUPATION O: Same occupation as respondent/ different occupation] fulfills [her/him], [she/he] is sometimes very tired after work. Due to difficult circumstances, [she/he] has to change jobs while keeping her/his current profession. So far, she/he has struggled finding a job.”

The narrative stresses competition on the local labor market by referring to the fictitious character as a local neighbor, living in the same district/ capital. Stories are comparable across all groups (refugees; hosts; different occupations): all four treatment groups’ narratives have the same length and are containing identical information (e.g. duration of stay in this district - thereby similar knowledge and networks).
The narrative is designed to be politically benign and to avoid raising positive or negative emotions. To make respondents relate with the narrative, the gender of the fictive character in the narrative is set to match that of the respondent.

While inserting a string for the group level G is straightforward, we are implementing one additional step when it comes to occupation group O in order to make narratives exactly comparable in their length (one to three words).

For individuals who have been randomly assigned to treatment groups T1 or T3 (same occupation), the enumerator will be asked to shorten the occupation string indicated by the respondent.
E.g.: If the respondent states he is an “undergraduate teacher for Maths and sports”, the enumerator is trained to shorten the string to “teacher”.
The narrative will then be based on the shortened version of the string.

For individuals who have been randomly assigned to treatment groups T2 or T4 (different occupation), we have prepared a list of different occupations that the computer will randomly draw from.
This list of occupations is split into “high skilled (above secondary degree)” and “low skilled (below secondary degree)” occupations so that respondents will be matched with a fictitious character of the same skill level. This step intends to avoid confounding effects based on hierarchical judgements.
For programming reasons, the enumerators have to confirm that the randomly drawn occupation is indeed different from the string variable that the respondent indicated as his own. If the enumerator denies, a second random draw happens and the procedure is repeated.

(4) After the fieldworker reads the narrative to the respondent, the respondent answers questions that
Make them relate to the story (“Should the character search for work? What would you recommend him/ her to do?”)
Elicit their willingness to interact with the fictitious individual.
Gauge their normative perception concerning labor market integration of the fictitious individual

Computer-based

Individual. We stratify by refugee status (refugee or non-refugee), and location (Uganda: Kampala, Isingiro District; Ethiopia: Addis Ababa, Somali Region).

No

210 Clusters in Uganda (Enumeration areas from the Census of 2014 in Uganda)
210 Clusters in Ethiopia (Enumeration areas from a mapping planned for the upcoming Census)

Uganda: Isingiro District: 1200 Ugandan households and 1200 Refugee households Kampala District: 1200 Ugandan households and 1200 Refugee households Ethiopia: Addis Ababa region: 1200 Ethiopian households and 1200 Refugee households Somali region: 1200 Ethiopian households and 1200 Refugee households Note that we are expecting a certain attrition and non-response rate given the refugee-host context in both countries. We expect attrition to be at random, proportionally distributed across treatment groups. Another correction we will make to our sample for relevance is that we are administering the experiment exclusively to individuals who are currently working (employed, self-employed) or unemployed individuals actively looking for a job. That way, we are excluding out of the labor force individuals, who would struggle to relate to the narrative. Consequently, our sample size is likely to reduce, especially among the refugee population who is more likely to be out of the labor force. However, since randomization is stratified based on groups (refugees/hosts) and location (rural/urban), we expect this to be proportional across our four treatment groups. Besides testing our four hypotheses on the entire sample (refugees and hosts in urban and rural areas of Ethiopia and Uganda), we may do some heterogeneous analyses. Therefore we will divide the sample according to these characteristics to disentangle further between groups’ differences in attitudes. The experimental module is part of a large survey executed in the areas of interest. It consists of a household survey where all the members are asked a set of questions related to their socio-economic conditions. This survey lasts about 20 minutes and aims at drawing an extensive picture of the types of households we are interviewing. The household survey is followed by an RSI (Random Sample Individual) survey where one working-age individual is selected at random among all household members. The survey lasts one hour and extensively goes over the labor market history of the respondent. It also includes our experiment and sections related to integration, perception, and networks. Thus, our experiment is administered to one individual per household. Consequently, we do not have to cluster our standard errors at the household level and we are avoiding spillover effects.

The whole sample is split into 4 treatment arms. When disregarding attrition, non-response, and out-of-the-labor-force individuals by whom the experiment will not be administered, we are left with 2,400 respondents per treatment arm.

To detect a minimum effect of 2% of our outcome variable (translated into a 10% increase from our baseline value), we need a sample size of around 2,200 individuals per variation, with a 90% confidence level. We are confident that our sample size is sufficient to get meaningful and interpretable results (with enough power). Such an MDE is realistic given what we find broadly in the literature on these kinds of outcomes of attitudes and beliefs (see for instance Cattaneo and Grieco (2020)). Our randomization implies 2,400 individuals per treatment group (i.e. 1,200 individuals per treatment group and country), and some of our analysis will be based on comparing outcomes of two pooled treatment groups. Thus further power calculations need to be performed to account for these different sample divisions. REFERENCE: Cattaneo, C., & Grieco, D. (2021). Turning opposition into support to immigration: The role of narratives. Journal of Economic Behavior & Organization, 190, 785-801.