# Introduction

Welcome to Fall 2022 Field Methods. This class will teach you about designing social science RCTs. This involves the following steps:

- Posing a normatively-defined
**problem**out in the real world. That is, tell us about “something that is wrong in the world” that deserves attention and resources to try to rectify. - Using evidence and theory to identify a
**mechanism**that is an important driver of this problem. - Designing an
**intervention**that can disrupt the mechanism. - Designing an experimental
**test**of the intervention.

The class will be hands-on and driven by research interests of people in the class. We will have two introductory sessions, and then you will go through 3 cycles of an RCT design cycle:

- 1st week of cycle (intervention concept): Use theory and data to argue for the importance of a particular mechanism, use it to define an intervention strategy, and present an intervention concept.
- 2nd week of cycle (outcomes measurement strategy): Define primary and secondary outcomes of interest and propose a strategy for measuring these outcomes. Defend the reliability and validity of these measures.
- 3rd week of cycle (experimental test): Define the causal effects of interest, a randomization strategy for identifying them, and an estimation and testing strategy for assessing. Defend your experimental design on the basis of unbiasedness and statistical power.

After the first two weeks, all of your work will be done in groups of 3 or 4 people. (I will assign the groups, and we may change the group assignments over the course of the semester.) Each week will involve making progress in successive steps of RCT design ass outlined above. During class time, the groups will present material to each other for discussion.

# Logistics

- We will meet in person in Room 217 Fridays 11-noon. Because much of the class involves group discussion I am not planning a Zoom hybrid option.
- No class on the following dates:
- Sept 16 (APSA)
- Oct 28 (EGAP meeting at Cornell)
- Nov 18 (EGAP meeting in Abidjan)

- If you need to miss a class, coordinate with your group so that your contributions are incorporated in that week’s group presentation.
- I will add readings in the weekly plan entries below.
- If you want to meet to discuss your class work or to discuss accommodations, please sign up for office hours using Calendly: https://calendly.com/cdsamii
- Pre-requisites are Quant I and II or equivalent. Please discuss if you are unsure.

# Weekly Plan

## Sept 2-9

On Sept 9, we will be meeting in class with the goal of selecting the 3 “problems” that will serve as the basis of the group projects over the course of the semester. To do this, each of you will do a 5-minute “lightening talk” to try to persuade us that the problem you have selected is one of the problems we should select as a class. I want you to prepare **at most three slides** (really, no more than three) in which you do the following:

- State the problem (along the lines of what we did in class today) and defend your stance that it is indeed a problem, on normative grounds.
- Show us some compelling statistics that demonstrate that the problem is serious. This can be shown through indication of how extensive the problem is, in terms of really strong correlations with other problems, etc.

Your goal is to take 5 minutes to convince your skeptical peers that this problem matters.

**By Thursday evening the day before class, email your three slides to me in PDF form, so that I can compile them into one slide deck.**

Then, we will deliberate and ultimately vote to select three problems that will serve as the basis of the group projects for the semester.

Given what we heard today, there does appear to be substantial overlap in research interests, so it is possible that we will combine some people’s problems together into one.

Now, you are about a month away from when you will start having to design your first RCT. You should start doing some reading in preparation for that. Here are readings you should start working through, and continue to work through over the coming weeks:

- Gerber, Alan S., and Donald P. Green. Field experiments: Design, analysis, and interpretation. WW Norton, 2012. Ch 1-4. Available from Library or order a hard copy. Materials for replicating their examples is here: https://isps.yale.edu/FEDAI
- Duflo, Esther, Rachel Glennerster, and Michael Kremer. “Using randomization in development economics research: A toolkit.” Handbook of development economics 4 (2007): 3895-3962. link
- Athey, Susan, and Guido W. Imbens. “The econometrics of randomized experiments.” Handbook of economic field experiments. Vol. 1. North-Holland, 2017. 73-140. Skim this alongside Gerber and Green Chapter 3. link
- Learn about the estimatr R package, which will be our workhorse package when we get to analyzing experimental data: link
- Evidence in Governance and Politics (EGAP) has an online resource book that has various resources for field experimental design. Start perusing the material there: link
- Coppock, A. 2020. “10 Things to Know About Statistical Power.” EGAP Resource. link

I obviously don’t expect you to read all of this by Sept 9. Rather, make a plan for yourself to make your way through this material over the coming month.

First week slides: link

## Sept 9-16

Note that we do not have class on September 16 (APSA). Nonetheless, in addition to continuing with the readings listed above, I want you to work through the code in this design effects exercise: link

I know it’s sort of silly to just copy and paste code, but I want you to understand what the code is doing exactly. It shows steps that you could use to test out experimental designs that include clustering and stratification. Send to me a PDF, ideally produced using RMarkdown, in which you replicate the exercise and comment on what you are doing in each code chunk.

In class on Sept 9 we selected the “problems” that will motivate our RCT design group projects for this semester:

- Participation in extremist movements
- Failure of integration for survival migrants
- Non-material performance motivation for local bureaucrats and public servants

We will proceed through these in the order listed above. So, we will spend the next three weeks on “Participation in extremist movements,” working through the three-part cycle listed above, and then we will move onto “Failure of integration for survival migrants” after that, and so.

First thing is the create the groups. We will have three groups of 4 working together. I will be sending an email with a proposed grouping, and then we can discuss.

## Sept 16-23

First, please be sure that you have completed the design exercise. If you haven’t done so already, please send me a PDF of your RMarkdown output so that I can check that you are implementing and understanding things correctly. Please send those to me by Wednesday. We will take the first 15 minutes or so to discuss.

Then, we will turn to our first group project, focusing on the problem of “Participation in extremist movements.” Here are the group assignments that I am proposing:

Group 1 Avi Daniel Jimmy Dylan

Group 2 Mengfan Alper Henry Julieta

Group 3 Sangyong Lanie Sylvan Jona

I based the groups on overlap in the interests and contexts that I saw in each of the problems that you presented last week.

As discussed last week, each group can interpret and apply the theme of “Participation in extremist movements” in different ways. So, group 1 could stick with Jimmy’s original emphasis on Islamic extremism in Mozambique (that’s what I would recommend for them), but other groups could consider manifestations in another region/country and on the basis of other ideologies.

For Friday, each group should prepare to do a 15-20 minute presentation. What I would like each group to present is their version of the problem (2-4 slides). Then, I would like 2-4 slides that review possible causal processes or mechanisms that lead to participation in extremist movements within the context they are studying. For this part, you should consult the current literature and give appropriate references in the slides. Any intervention to prevent participation in extremist movements should seek to intervene on a mechanism that existing theory and evidence suggests is an important explanation for why people join extremist movements in the context that you are studying. Your assignment is to consult existing literature and evidence to see what some of these mechanisms might be. It’s possible that the mechanisms that are important for participation in Islamic extremism in Mozambique differ from, say, participation in rightwing extremist groups in the US, for example. It will be interesting to see the relationships between what different groups find.

Given the short amount of time, I don’t think you will be at the point where you can design your intervention yet. That will have to wait until the following week. (There is enough flexibility in the schedule to accommodate taking an extra week to do so.)

## Sept 24-30

For this coming Friday, each group will be doing presentations in which they define their intervention. This will involve the following:

- 1 slide reminding us of your framing of the problem in your context,
- 1 slide telling us the precise population and outcomes (behaviors, attitudes, or beliefs) that are implicated in your framing of the problem and that will be the focus of your approach to intervention,
- 1 slide telling us the mechanism that you have selected, explaining why you have selected it (such as evidence for its demonstrated relevance, how compelling it is in terms of theory, whether it lends itself to intervention both in terms of feasibility and ethics, etc.) and how the mechanism explains the pattern of outcomes in the population that are your focus,
- 1 slide describing the intervention, including what kinds of tasks or incentives are being introduced and who exactly will be doing the intervening (e.g., will there be a partnership with an organization or agency?),
- 1 slide explaining how you propose that the intervention could work to change outcomes among your population of interest, based on your theoretical framing (mechanism) and existing evidence or examples of interventions already in use, and
- 1 slide explaining any potential political obstacles or ethical issues that you will be navigating.

At least one person in your group should have the task of searching for examples of existing interventions, to help to motivate and distinguish what your group is proposing to do. Others can focus on developing the content of your group’s intervention.

We will be discussing each group’s proposed interventions in terms of how well motivated they are on the basis of their theoretical framing and then in terms of feasibility, politics, and ethics.

Looking ahead to the following two weeks, we will next be turning to outcome measurement strategies and then experimental design strategies in the two weeks after this Friday.

## Oct 1-7

For this Friday, you will be doing presentations on your outcome measurement strategies. Given that topic, an issue that your outcome measurement strategy should address is the possibility that subjects will hide their true attitudes or behaviors, given the sensitive nature of what we are studying. Here are some readings to help addressing those issues (all of the readings can be found online via open source or through NYU libraries):

- De Quidt, Jonathan, Lise Vesterlund, and Alistair J. Wilson. “Experimenter demand effects.” In Handbook of research methods and applications in experimental economics. Edward Elgar Publishing, 2019.
- Blattman, Christopher, Julian Jamison, Tricia Koroknay-Palicz, Katherine Rodrigues, and Margaret Sheridan. “Measuring the measurement error: A method to qualitatively validate survey data.” Journal of Development Economics 120 (2016): 99-112.
- Tourangeau, Roger, and Ting Yan. “Sensitive questions in surveys.” Psychological Bulletin 133, no. 5 (2007): 859.

Note that the types of “randomized response” methods (endorsement experiments, list experiments, etc.) that Tourangeau and Yan discuss tend to reduce statistical power. We will be addressing statistical power in the following week.

For “climate” and “bystander” approaches to reducing deviant behaviors, you will want to evaluate effects on both individual attitudes as well as perceived norms.

- Bicchieri, Cristina. “Measuring Norms: Consensus and Conformity.” In Norms in the wild: How to diagnose, measure, and change social norms. Oxford University Press, 2016, Ch. 2.
- Mackie, Gerry, Francesca Moneti, Holly Shakya, and Elaine Denny. “What are social norms? How are they measured.” San Diego, CA (2015).

Your primary outcome of interest may be somewhat abstract, and may therefore require multiple measures that you try to combine together. For dealing with this kind of situation, I would like to remind you of the following papers that we studied last Spring in Quant II:

- Anderson, Michael L. “Multiple inference and gender differences in the effects of early intervention: A reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects.” Journal of the American statistical Association 103, no. 484 (2008): 1481-1495.
- Casey, Katherine, Rachel Glennerster, and Edward Miguel. “Reshaping institutions: Evidence on aid impacts using a preanalysis plan.” The Quarterly Journal of Economics 127, no. 4 (2012): 1755-1812.

To the extent possible, it is nice to try to make use of outcome measures that have been tried and validated in other studies. This sets you up to make comparisons between findings in your context and findings from other contexts. Sometimes there are no good measures that exist, and so you need to be inventive. You should explain such choices in your presentation.

For your presentations, here is what you want to cover:

- 1 slide restating the population, intervention, and outcome (where the outcome is defined in conceptual terms) that are the focus of your study – (“we are studying the effect of a [intervention] on [outcome] among [population]”),
- 1 slide that defines your primary outcome of interest, and discusses challenges in measuring this outcome (e.g., sensitivity, not directly observable, lack of a single gold standard measure, etc.),
- 1-2 slides explaining your strategy for measuring your primary outcome, explaining how your strategy addresses the challenges,
- 1 slide that defines, explains the rationale for, and points out any challenges in measuring at least 1 secondary outcome of interest, where this can be a mediator outcome that helps you evaluate the importance a causal mechanism or it could be a downstream outcome that helps you evaluate the broader significance of whatever effects you find, and
- 1-2 slides explaining your measuring strategy for this secondary outcome of interest.

## Oct 8-14

For this coming Friday, your task is to come up with your randomization design and analysis plan. This involves specifying the following:

- Units of observation,
- Outcome measures on these units,
- Unit-level covariates that you will measure for either improving statistical precision or testing for interaction effects/effect heterogeneity
- Units of randomization, which could be the same as the units of observation (unit-randomization) or clusters of units of observation (cluster-randomization),
- If cluster-randomization, any cluster-level covariates you will measure for either improving statistical precision or testing for interaction effects/effect heterogeneity, and
- Specifications of regression equations that you will use to estimate your treatment effects.

As discussed in class, you can think of this as specifying the elements that will make up your dataset.

When specifying the units of randomization, you want to account for possible spillover effects. If spillovers are just a nuisance, then the design should establish units of randomization that are large enough to contain spillovers and far enough apart to insulate against spillovers from other areas/units. If spillovers are of substantive interest, then this handbook chapter may be of interest:

- Aronow, Peter; Dean Eckles, Cyrus Samii & Stephanie Zonszein. (2020). “Spillover Effects in Experimental Data.” In James Druckman & Donald Green, Eds. Advances in Experimental Political Science. Cambridge: Cambridge University Press. arxiv

Then, once these design elements are specified, I want you to conduct a power analysis. The goal of the power analysis is determine the sample size necessary to achieve acceptable power (e.g., 80% power at 95% confidence) for the “minimally meaningful effect size.” This is the smallest effect size that you and other people interested in your research would consider “meaningful.” This is a judgment call based on substantive considerations. If the outcome measure is a continuous measure, then you could state it in terms of standard deviations. Then you would need to reason as to whether, for example, a 0.25sd effect would still be considered meaningful? How about a 0.10sd effect? And so on. If it is a binary outcome, you could state it in terms of percentage point differences given what we understand to be the baseline level. 10 percentage points? 5 percentage points? Etc.

The Duflo et al. piece that I asked you to read contains a good, compact discussion of power analysis, emphasizing the concept of the “minimal detectable effect” :

- Duflo, Esther, Rachel Glennerster, and Michael Kremer. “Using randomization in development economics research: A toolkit.” Handbook of development economics 4 (2007): 3895-3962.

A well-powered study is one for which the “minimal detectable effect” equals the “minimally meaningful effect size.”

Your power analysis should be done using simulation in R. Minimal detectable effects can, in principle, be calculated using analytical formulas. However, I like to use simulation because it forces you to write out the code that you would use to implement your randomization and analysis plan, which removes any ambiguity about what you are doing and also allows you to check to be sure that everything will run correctly.

Coding a simulation involves a lot of judgment calls about how to specify the potential outcome distributions, where to get information about the distributions of covariates, etc. Do your best, using the design replication exercise from earlier this semester, and then in class we will discuss some nuances. A nice references is the following:

- Coppock, A. 2020. “10 Things to Know About Statistical Power.” EGAP Resource. link.

The declareDesign package referenced in Coppock’s tutorial contains nice tools, although you will need to invest in learning the syntax.

A few weeks ago we discussed the question of whether it is better, for purposes of statistical power, to use covariates to stratify the randomization or control for the covariates after the fact. Generally stratifying is superior although the difference between the two diminishes in sample size. Here is the reference for that

- Miratrix, Luke W., Jasjeet S. Sekhon, and Bin Yu. “Adjusting treatment effect estimates by post-stratification in randomized experiments.” Journal of the Royal Statistical Society: Series B (Statistical Methodology) 75, no. 2 (2013): 369-396.

Your group’s presentation should include the following:

- 1 slide specifying your units of observation, unit-level outcomes, and unit-level covariates and explaining how you will obtain the data on them,
- 1 slide explaining the units of randomization, and then if they are different than the units of observation, how exactly they will be defined (including any relevant discussion of spillovers) and how you will obtain data on them for the purposes of randomization,
- 1-2 slides explaining your regression specifications, including discussion of any stratum fixed effects, weighting to account for unequal assignment probabilities, clustering of standard errors, any other covariate controls that you will use, and then any special specifications (e.g. interaction specifications, mediation models, spillover models, distribution regressions) that you will use to estimate interaction effects, mediation effects, spillover effects, distributional effects, etc. (if relevant).
- 1 slide explaining how you set up your simulation code to run a power analysis, explaining how you generated your potential outcomes, what kinds of heterogeneity you coded in terms of level outcomes, effects, etc.,
- 1 slide explaining your “minimally meaningful effect size” for your primary outcome,
- 1 slide that shows a power curve, where the X axis would be a measure of sample size, and the Y axis would be either the minimally detectable effect or the statistical power.

## Oct 15-21

First, for the group members “transfers,” I implemented the random assignment among those who submitted requests and we ended up with two reassignments. Thus, the groups for the second project will be

- Group 1 Avi Daniel Jimmy Dylan
- Group 2 Mengfan Henry Jona Julieta
- Group 3 Alper Lanie Sangyong Sylvan

Second, with respect to our discussion of using baseline outcome measures as regression controls (known as the “ANCOVA” estimator) rather than using difference-in-differences, here is a paper that explains things very clearly (while also making the argument for having multiple rounds of endline measurement if possible):

- McKenzie, David. “Beyond baseline and follow-up: The case for more T in experiments.” Journal of Development Economics 99, no. 2 (2012): 210-221.

And then for an up to date and technical statement on the question of “how fine-grained should I go with my stratification,” you can check out this paper as well as the work it cites:

- Fogarty, Colin B. “On mitigating the analytical limitations of finely stratified experiments.” Journal of the Royal Statistical Society: Series B (Statistical Methodology) 80, no. 5 (2018): 1035-1056.

Third, looking at the class sessions remaining, we have 10/21, 11/4, and 11/11 for the “migrant integration” topic, and then 12/2, 12/9, and then we can try to have one session during the exam period to work through the “non-material performance motivation for bureaucrats” topic.

Now, we are starting the “failure of integration for survival migrants” project this Friday. Such failure can emerge as a result of host country citizens or host community members being unwilling to accept migrants. Or it could occur due to the migrants themselves lacking certain endowments that are necessary to integrate. This is quite a general problem that emerges across societies. Thus, there should be ways to construe the problem so that it is well motivated within the various geographic domains of expertise among members of the class.

So, in a manner that is similar to what we did last month, for Friday, each group should prepare to do a ca. 20-30 minute presentation on their version of the problem and ALSO their intervention concept (i.e. we are combining into one session what we had done in two sessions previously):

- 1 slide stating the problem as it pertains to their geographic area of focus, trying to be precise about the population and outcomes (behaviors, attitudes, or beliefs).
- 2 slides describing possible causal mechanisms that sustain the problem of failed migrant integration. Try to present these in the form of directed acyclic graphs (DAGs). Then tell us which mechanism or component of a mechanism will be the focus of your intervention and why (evidence, amenability to intervention, etc.).
- 1 slide describing the intervention, including what kinds of tasks or incentives are being introduced and who exactly will be doing the intervening (e.g., will there be a partnership with an organization or agency?).
- 1 slide explaining how you propose that the intervention could work to change outcomes among your population of interest, based on your theoretical framing (mechanism) and existing evidence or examples of interventions already in use.
- 1 slide explaining any potential political obstacles or ethical issues that you will be navigating.

Looking forward to learning more about each group’s intervention concepts.

## Oct 22-Nov 4

For this coming Friday, your presentations are focusing on outcome measurement strategies. The presentations should be structured similarly to the say you did this for the last cycle: 1 slide restating the population, intervention, and outcome (where the outcome is defined in conceptual terms) that are the focus of your study – (“we are studying the effect of a [intervention] on [outcome] among [population]”),

- 1 slide that defines your primary outcome of interest, and discusses challenges in measuring this outcome (e.g., sensitivity, not directly observable, lack of a single gold standard measure, etc.),
- 1-2 slides explaining your strategy for measuring your primary outcome, explaining how your strategy addresses the challenges,
- 1 slide that defines, explains the rationale for, and points out any challenges in measuring at least 1 secondary outcome of interest, where this can be a mediator outcome that helps you evaluate the importance a causal mechanism or it could be a downstream outcome that helps you evaluate the broader significance of whatever effects you find, and
- 1-2 slides explaining your measuring strategy for this secondary outcome of interest.

## Nov 5-11

For this coming Friday, you will be presenting your randomization and analysis plans as well as your power analyses. As before, your presentations should be structured as follows:

- 1 slide specifying your units of observation, unit-level outcomes, and unit-level covariates and explaining how you will obtain the data on them,
- 1 slide explaining the units of randomization, and then if they are different than the units of observation, how exactly they will be defined (including any relevant discussion of spillovers) and how you will obtain data on them for the purposes of randomization,
- 1-2 slides explaining your regression specifications, including discussion of any stratum fixed effects, weighting to account for unequal assignment probabilities, clustering of standard errors, any other covariate controls that you will use, and then any special specifications (e.g. interaction specifications, mediation models, spillover models, distribution regressions) that you will use to estimate interaction effects, mediation effects, spillover effects, distributional effects, etc. (if relevant),
- 1 slide explaining how you set up your simulation code to run a power analysis, explaining how you generated your potential outcomes, what kinds of heterogeneity you coded in terms of level outcomes, effects, etc.,
- 1 slide explaining your “minimally meaningful effect size” for your primary outcome,
- 1 slide that shows a power curve, where the X axis would be a measure of sample size, and the Y axis would be either the minimally detectable effect or the statistical power.

When describing your simulation code, key parameters to report include

- Effect sizes expressed in terms of control group standard deviations,
- R-squared for a regression of the primary outcome potential outcomes on the covariate(s) that you are using for either blocking/stratification or ex post regression control, and
- For cluster randomized experiments, the intra-class correlation (ICC) of the primary outcome potential outcomes.

To interpret the R squared (as in, is this high? or low?) it’s a good idea to look at work in the domain that you are studying. So, if you are using a set up that has an R squared of 0.25 for both treated and control potential outcomes, then you need to assess whether it is common for the topic, level of analysis, and outcome that you are studying whether regressions with R squareds that high. Or would that be unusually high? It’s one of those things for which you gain some intuition as you see and conduct more quantitative research. In many of the areas in which I work, R squareds are pretty low, maybe 0.25 would be an upper bound. But it does depend on the topic, and so it’s a matter of referencing work on your topic.

In R, you can use the “deff()” function in the Hmisc package to estimate the ICC. In Stata you can use the “loneway” command.

Then, I noted the following for each of the groups, as things to try to resolve in their designs:

- For the “anti-Russia attitudes” team, the question was how one could evaluate the effect of participating in the volunteer activity per se as compared to merely receiving information about it. A reasonable approach would be a mediation analysis, which should be incorporated into the analysis plan and power analysis.
- For the “North Korean refugee integration” group, the question was how to account for noisy estimation of a latent factor using PCA in one’s power analysis. A reasonable approach would be to consider effect sizes on the latent index, simulation outcome variables that are noisy measures of this latent index, apply PCA to estimate the latent index, and then see how power looks.
- For the “Venezuelan migrant workers” group, the questions were how to address the infrequent/periodic nature of hiring and potential power issues from audit experiment outcomes

## Nov 28-Dec 2

Looking ahead to the rest of the semester, we have two more class meetings and then the exam period. I think we should plan to wrap things up during these last two sessions. So, this Friday I will have your groups present your problem definitions, diagnoses, and intervention concepts. Then next week you will present full research design and analysis plans, including outcomes, randomization plan, measurement strategy, and analysis plan.

After that, if you have project-related methods ideas or issues that you would like to discuss with me, the exam period is available for you to schedule a meeting with me (using the Calendly). Of course, you can also feel free to meet with me when we return for the Spring semester.

The goal of this class was to get you comfortable with the idea of starting with a problem and diagnosis, translating this into an intervention concept, and then thinking about a test for that intervention concept. Over the past two project cycles, I am seeing that most people are coming to grasp this basic methodological process. This is great – I hope you feel more confident about your ability to pursue this style of research as well.

For these last two weeks, we will turn to our last substantive problem area: “non-material performance motivation for local bureaucrats and public servants.” For a primer on recent research in this area, here are some useful review essays, some of which focus on developing country contexts but have more general relevance:

- Finan, Frederico, Benjamin A. Olken, and Rohini Pande. “The personnel economics of the developing state.” Handbook of economic field experiments 2 (2017): 467-514.
- Pepinsky, Thomas B., Jan H. Pierskalla, and Audrey Sacks. “Bureaucracy and service delivery.” Annual Review of Political Science 20 (2017): 249-268.
- Bertelli, Anthony M., et al. “An agenda for the study of public administration in developing countries.” Governance 33.4 (2020): 735-748.
- Dahlström, Carl, and Victor Lapuente. “Comparative Bureaucratic Politics.” Annual Review of Political Science 25 (2022): 43-63.
- Brierley, Sarah, et al. “Bureaucratic Politics: Blind Spots and Opportunities in Political Science.” https://www.guillermotoral.com/bureaucratic_politics_review.pdf

Your group should situate the problem in a specific context that is well understood by at least one member of your group. You should also define a specific policy area (e.g., “enforcement of traffic violations in context XX”, “responding to licence requests in context YY,” etc.). The more specific and concrete, the better. Surely you all have specific examples of “poor public servant performance motivation” that you think are worth trying to fix!

So, for this Friday, your group’s task is to present your problem definition, diagnosis, and intervention concept. Like in the previous two round, this should consist of the following:

- 1 slide stating the problem as it pertains to your geographic area and policy domain of focus, trying to be precise about the population and outcomes (behaviors, attitudes, or beliefs).
- 2 slides describing possible causal mechanisms that sustain the problem of failed migrant integration. Present these in the form of directed acyclic graphs (DAGs). Then tell us which mechanism or component of a mechanism will be the focus of your intervention and why (evidence, amenability to intervention, etc.).
- 1 slide describing the intervention, including what kinds of tasks or incentives are being introduced and who exactly will be doing the intervening (e.g., will there be a partnership with an organization or agency?).
- 1 slide explaining how you propose that the intervention could work to change outcomes among your population of interest, based on your theoretical framing (mechanism) and existing evidence or examples of interventions already in use.
- 1 slide explaining any potential political obstacles or ethical issues that you will be navigating.