Table of contents
  1. Overview
  2. Unit 1
    1. Part 1 (20 points)
      1. Question 1.1 (5 points)
      2. Question 1.2 (5 points)
      3. Question 1.3 (5 points)
      4. Question 1.4 (5 points)
    2. Part 2 (40 points)
      1. Question 2.1 (20 points)
      2. Question 2.2
        1. Question 2.2.1 (10 points)
        2. Question 2.2.2 (10 points)
    3. Part 3 (40 points)
    4. Part 4 (No points awarded without this)
  3. Unit 2
    1. Part 1 ($20$ points)
      1. Question 1.1 ($5$ points)
      2. Question 1.2 ($5$ points)
      3. Question 1.3 ($5$ points)
      4. Question 1.4 ($5$ points)
    2. Part 2 ($40$ points)
      1. Question 2.1 ($3\times 2=6$ points)
      2. Question 2.2.1 ($3\times 5=15$ points)
      3. Question 2.2.2 ($19$ points)
    3. Part 3 (40 points)
    4. Part 4 (No points awarded without this)
  4. Unit 3
    1. Part 1 ($40$ points)
      1. Part 1.1 - Simulation Playing (12 points)
      2. Part 1.2 - Thinking about the simulation parameters (12 points)
      3. Part 1.3 - Expanding the simulation (16 points)
    2. Part 2 - Using and Critiquing Existing Simulations ($30$ points)
      1. Part 2.1 (15 points) - Using a System Dynamics Model
      2. Part 2.2 (15 points) - Critiquing an Agent-based Model
    3. Part 3 (30 Points)
      1. Part 3.1 (15 points)
      2. Part 3.2 (15 points)
    4. References

Overview

Every unit, your team will be required to submit a Unit Midpoint Assignment. These will be team assignments with only your teammate(s) in ML & Society.

Unit midpoint submissions will always have four parts:

  1. In Part One, you’ll gain a better understanding of the unit’s case study by answering a series of “Why” questions that move us beyond just “the algorithm is biased.”
  2. In Part Two, you will complete an assignment that does not focus specifically on your semester-long project, but is rather an extension of the material that you learned in class.
  3. In Part Three, you will be asked to connect what you have learned in class to your semester-long project.
  4. Finally, you will be required to provide a Team Participation Statement. This statement, which all team members need to agree on, will provide information on which team members completed which parts of the assignment. This will not be graded, but submissions without the Team Participation Statement will receive an automatic zero (0) on the assignment.

This part is specific to ML and Society

While Part Three will be related to your Impossible Project, the Unit Midpoint submissions are ML-Soc specific. This means that your RAGE teammates are not responsible for helping you with it.

Submitting the Unit Midpoint Submissions

Your submissions are due as a PDF on Autolab. The submission is due by 11:59pm of the Saturday BEFORE the third week of each unit; see the schedule for exact dates. Only one teammate needs to submit.

Each Unit Midpoint Submission is worth $7\%$ of your grade

This is distributed across the three parts, see above.

Make it easy for us!

Please make sure you explicitly mark where you are answering each question!

Unit 1

Start early, ask questions, check Piazza!

This is a new assignment and is likely to require some clarifications. Please make sure that you start early to give time for this, to ask questions when you have them (on Piazza!), and to check Piazza for answers to other people’s questions

Part 1 (20 points)

Unit 1 is focused on racial inequality in healthcare, as exposed by a biased machine learning algorithm.

For each of the questions below, provide an answer by 1) giving a quote from the required reference that supports your answer, and 2) giving a 1-2 sentence rationale as to why this quotation supports the answer given. If you need additional content to support your argument, you may also use another reference of your choice.

Question 1.1 (5 points)

Q: Why were there racial disparities in the algorithm’s recommendation? Note: your answer should use the term proxy variable.

Question 1.2 (5 points)

Q: Why was the underlying proxy variable racially biased?

Question 1.3 (5 points)

Q: Why did these people make racially biased decisions (note, this question should be a hint as to what we expect for 1.2…)?

Required Reference (video): The IP student video

Question 1.4 (5 points)

Q: Why are racial bias and discrimination built into our society at a structural level (note, this question should be a hint as to what we expect for 1.3…)?

Part 2 (40 points)

The second part of your assignment asks you to take some of the tools you’ve learned about in class—causal graphical models, literature searching, causal inference, and experimentation—to help 1) characterize the problems created by an algorithmic intervention in healthcare, and 2) to develop solutions, to fixing the algorithm and to fixing the underlying inequalities in the U.S. healthcare system.

Question 2.1 (20 points)

Draw a causal diagram that has at least 10 nodes and 20 edges that explains why the algorithm was racially biased

Make it easy on you and us!

There are no requirements for what tool you use to draw this. Hand-drawn, Powerpoint, Miro - they are all fine. Just make sure we can read what you wrote!

For full credit, you need to do the following:

  • You include the nodes “patient race,” “economic inequality,” “segregation,” and “insurance.”
  • For at least 5 edges, you provide an explanation of why you believe that causal link exists, with reference to an academic article. Note: the article can simply show a correlation between variables if you have a reasonable causal explanation underpinning it (i.e. correlation + theory).
  • For at least two sets of edges, you explain why you believe that complex causal pathway exists, with reference to an academic article. This can be either a chain (e.g. A -> B -> C), or a combined/moderated causal path (e.g. both A and B cause C). Note: the article can simply show a correlation between variables if you have a reasonable causal explanation underpinning it (i.e. correlation + theory).

Referencing academic articles

Any form of proper referencing (e.g. APA format) is acceptable. Please do not, however, just give us a link.

Question 2.2

Use your diagram to propose an intervention that might “fix” the following things. Details are below on what your “fix” should include.

Propose a “fix” for…

Question 2.2.1 (10 points)

…the algorithm - that is, how would you fix this problem from a technical/data perspective?

Question 2.2.2 (10 points)

…totally eliminating racial inequalities in healthcare

For full credit on Questions 2.2.1 and 2.2.2, you should provide:

  • A set of nodes or edges that you add or remove from the causal graph you have created
  • A (set of) real-world intervention(s) (even if far-fetched!) that would be required to add/remove those nodes/edges, explaining how the intervention directly aligns with those nodes/edges
  • Two different possible evaluation frameworks for how you would determine whether or not the fix actually worked:
    • One of the frameworks should make use of an experimental approach. You should explain then why this experiment is, or is not, ethical, i.e. whether or not the benefits outweigh the costs.
    • The second should make use of observational data. You should justify why you believe this approach will provide useful data and what limitations will arise because you use observational data.

Part 3 (40 points)

For Part 3, you will complete Parts 1-6 of the Impossible Project group submission without input from your RAGE teammate. You may use elements of your answer for Part 2 for Part 3, but can’t just re-use all of it. See below!.

Also, note for Part 4 that you need only address the “Tech” part, not the “Historical Roots” part, and for Part 6 you need only focus on “our tools” (causal inference, or anything covered in our lectures this unit), and not on RAGE tools!

For full credit on you need to

  • Use elements of the tool we learned about in this unit (tools of causal inference and causal modeling) in at least some of your answers, and explicitly point out how you did so.
  • Address a problem and solution that are different than what you talked about for Part 2. That is, your problem should not be relevant to the algorithm used in Parts 1/2 of this assignment, and your solution should not be one of the interventions you came up with for Questions 2.2.1 or 2.2.2.
  • Add an additional, seventh part to your response, that discusses the limitations of the tools you used, and what you think you need to move beyond those limitations.

What is the problem?

We want you to address a problem that is narrow- in the sense that it addresses some component of racial inequality in healthcare- and broad, in that you can argue that addressing this problem is a critical step in dismantling white supremacy. This is hard to do! We’ll help you, and will be kind on this first submission as long as you put in the effort.

Part 4 (No points awarded without this)

Please provide a Team Participation Statement. To receive any credit for the (entire) assignment, the team participation statement should have the following:

  • Information on the specific parts of the assignment that each team member contributed to. This should cover all questions. Note that we do not need significant details, a few sentences should be enough
  • A statement by each team member that expresses their explicit agreement for the above. I.e. something like “I agree that this statement reflects the distribution of work in our group. -Your Name”.

Unit 2

Start early, ask questions, check Piazza!

This is a new assignment and is likely to require some clarifications. Please make sure that you start early to give time for this, to ask questions when you have them (on Piazza!), and to check Piazza for answers to other people’s questions

Part 1 ($20$ points)

Unit 2 is focused on racial inequality in policing/criminal justice system, as exposed by a biased machine learning algorithm, specifically PredPol.

For each of the questions below, provide an answer by 1) giving a quote from the required reference that supports your answer, and 2) giving a 1-2 sentence rationale as to why this quotation supports the answer given. If you need additional content to support your argument, you may also use another reference of your choice.

Question 1.1 ($5$ points)

Q: Why were there racial disparities in the algorithm’s recommendation? Note: your answer should use the term proxy variable.

Question 1.2 ($5$ points)

Q: Why was the underlying proxy variable racially biased?

Question 1.3 ($5$ points)

Q: Why did these people make racially biased decisions (note, this question should be a hint as to what we expect for 1.2…)?

Question 1.4 ($5$ points)

Q: Why are racial bias and discrimination built into our society at a structural level (note, this question should be a hint as to what we expect for 1.3…)?

Part 2 ($40$ points)

The main paper for part 2

To get started you have to first read the paper by Manish Raghavan titled What Should We Do when Our Ideas of Fairness Conflict? , which appeared in the January 2024 edition of Communications of the ACM

Recall that in class we saw an impossibility result that basically says we cannot find one fairness definition to “rule them all.” I.e., we cannot have a fairness definition that has three natural properties (unless there is no bias in the world). Raghavan in the above paper surveys how researchers have proposed ways to deal with this impossibility result.

In this part y’all will think through about deciding on fairness definition(s) for a proposed algorithm for predictive policing. Specifically:

The setup

Your group is part of a larger review board that will be considering new proposals for predictive policing algorithms that your city has decided to adopt.

Your job in part 2 of this submission would be to put forth certain fairness definitions that any proposed algorithm should have the ability to satisfy.

Some clarifications

While, as you’ll see in class, you could make a (very) strong case for not having any predictive policing being used in your community, that is not a choice for this part.

You are expected to read at least one paper cited in pages 92-97 in the Raghavan paper and use it to answer the question(s) below.

Now on to the questions.

Question 2.1 ($3\times 2=6$ points)

Identify three stakeholders from whose point of view your group will propose fairness definitions. Justify the stake of these three stakeholder groups in your city adopting a new predictive policing algorithm.

Choose your stakeholder groups carefully

The questions that follow have a bearing on what stakeholder groups you can choose. We recommend that you answer Question 2.1 in concert with Question 2.2

Question 2.2.1 ($3\times 5=15$ points)

For each of the three stakeholder groups pick a definition of fairness. Your fairness definitions must come from academic papers referred to in the paper by Raghavan that propose fairness definition that work around the fairness impossibility results. Your answer should use at least two different fairness definitions among the three choices.

Clarifications

Your fairness definitions must be precise. Giving mathematical definition is one way to do this but you can describe the fairness definition in English as well.

Make sure to give proper citation for the references.

Question 2.2.2 ($19$ points)

For each of the choices of fairness definitions for the three stakeholders, answer the following questions:

  • Why y’all choose the specific fairness definition? Why does the specific fairness definition chosen by your group “make sense” for the corresponding stakeholder group?
  • Mention another fairness definition that y’all considered for this stakeholder (this definition does not need to be precisely defined: just an overview is enough but it does have to come from an academic paper referred to in the paper by Raghavan). Why did y’all ultimately decide not to pick this definition?

Part 3 (40 points)

For Part 3, you will complete Parts 1-6 of the Impossible Project group submission but ignoring the History parts of the questions. You may use elements of your answer for Part 2 for Part 3, but can’t just re-use all of it. See below!.

Also, note for Part 4 that you need only address the “Tech” part, not the “Historical Roots” part, and for Part 6 you need only focus on “our tools” (impossibility results in computing, or anything covered in our lectures this unit), and not on RAGE tools!

For full credit on part 3 you need to

  • Use elements of the tool we learned about in this unit (impossibility results in computing and mathematical modeling) in at least some of your answers, and explicitly point out how you did so.
  • Address the problem picked by your group in the Unit 1 groups submission, OR change your problem, but explicitly state why your group decided to change the problem from your Unit 1 group submission.
  • Add an additional, seventh part to your response, that discusses the limitations of the tools you used, and what you think you need to move beyond those limitations.

What is the problem?

We want to see your problem statement at three levels of detail:

  • Narrow problem: This is the problem that is most narrow in scope and something for which you propose the first step of your solution in your plan
  • Medium problem: This is the next level up from your narrow problem but is not the same as ending white supremacy.
  • High level problem: This has to be ending white supremacy.

What should be in the plan?

We want y’all to start off with your initial plan to tackle the narrow problem. Then show us your plan on how after solving the narrow problem to solve your medium problem. Finally, show us your plan on how after solving medium problem to end white supremacy.

Your plans will get less specific as y’all move from narrow problem to medium problem to the high level problem and that is perfectly fine!

What about causal diagrams?

We would encourage y’all to use causal diagrams to talk about the Current world and the Future world (and possibly even part of your Plan).

Remember that for unit 2, your causal diagrams can have cycles in them.

Do we have to prove anything?

While Atri will love that, proofs are definitely not what we are expecting. Rather what we expect is for y’all to consider the various notions of impossibilities (and their “work arounds”) and see how you can use them in your chosen problem.

Part 4 (No points awarded without this)

Please provide a Team Participation Statement. To receive any credit for the (entire) assignment, the team participation statement should have the following:

  • Information on the specific parts of the assignment that each team member contributed to. This should cover all questions. Note that we do not need significant details, a few sentences should be enough
  • A statement by each team member that expresses their explicit agreement for the above. I.e. something like “I agree that this statement reflects the distribution of work in our group. -Your Name”.

Unit 3

Format has changed!!

We decided to change up the format for Unit Midpoint 3. We think this setup is more appropriate for this unit’s domain and tool. If you disagree, blame Kenny (but you still have to do the assignment).

Part 1 ($40$ points)

More specifically, the domain for Unit 3 is misinformation, and the tool we’re working through is simulation. In Part 1 of this unit midpoint, you’re first going to play with a simulation of [mis]information and answer a few questions about it. This should be quick and fun. You’re then going to think more about that model: about its parameters and how you can adapt it to a new setting. You’ll then work to parameterize your adaptation with content from research papers.

Part 1.1 - Simulation Playing (12 points)

Open up the (absolutely fabulous) simulation of (complex) contagions by Nicky Case. As we discussed in class, this is the section of this submission where we’ll be dealing directly with the concept of misinformation.

As you work through it, answer the following questions to guide your intuitions.

Note

Y’all don’t need more than 1-2 sentences for any of these answers, don’t overdo it!

  • Question 1.1.1 (2 points) What is the minimum number of edges you have to draw (i.e. to add to the edges that are already there) to “Fool everyone into thinking the majority of their friends are drinkers”?
  • Question 1.1.2 (2 points) What is a contagion in the context of the simulation?
  • Question 1.1.3 (2 points) What is the difference between a simple contagion and a complex contagion?
  • Question 1.1.4 (2 points) What is the minimum number of edges you need to infect everyone with complex wisdom in the first puzzle in Part 4 of the simulation? How did you figure that out?
  • Question 1.1.5 (2 points) What is the minimum number of edges you need to add to the edges that are already there to infect everyone with complex wisdom in the second puzzle in Part 4 of the simulation? How did you figure that out? Provide a screenshot if you are able to get less than 10.
  • Question 1.1.6 (2 points) Finally, answer the same question for the final puzzle (the puzzle in Part 5 of the simulation) - what is the minimum number of edges you need to add to the edges that are already there to infect everyone, and how did you figure it out? Provide a screenshot of your solution.

Part 1.2 - Thinking about the simulation parameters (12 points)

OK, now we’re going to actually think through this simulation model a bit more:

  • Question 1.2.1 (4 points) In class we talked about model parameters. Identify four parameters of the final model in the simulation model from Part 1.1. For each, provide the range of the parameter.

Note

You’ll have to think beyond what the game actually lets you change to come up with four.

  • Question 1.2.2 (4 points) Simulations are only useful if we can come up with an interesting outcome to measure. Specify two different outcome measures that we might be interested in with respect to a simulation of the spread of [mis]information, and justify why they’re good outcomes.
  • Question 1.2.3 (4 points) Now, link each of your parameters to each of your outcomes. That is, for each parameter, tell us how the outcome changes as the parameter changes, holding all other parameters constant. You’re welcome to use a table if that’s easier.

Part 1.3 - Expanding the simulation (16 points)

Now we’re going to develop an extension of the simulation model above (which we will call base model) that can help us to think about misinformation spread on social media. In both Question 1.3.1 and 1.3.2, you will be asked to make use of at least one academic paper to inform your modeling decisions; the relevant papers you may choose from are at the end of this page here. For both Questions 1.3.1 and 1.3.2, you should include in your answer:

  1. Any changes to the base model
  2. A description of any new parameters you need to add to explore different potential assumptions
  3. How default parameterizations and new model assumptions are informed by findings in at least one of the research papers listed here.

  • Question 1.3.1 (8 points) - Amend the simulation to be more representative of a social media site like Twitter. Specifically, develop a way to extend the base model to account for the role of recommendation algorithm.

  • Question 1.3.2 (8 points) - Amend your model in 1.3.2 to account for the role of AI-generated content. You may be as liberal as you want in imagining how AI-generated content changes the information space, but again, your assumptions must be informed by the literature.

Part 2 - Using and Critiquing Existing Simulations ($30$ points)

Part 2.1 (15 points) - Using a System Dynamics Model

PRISM is a system dynamics model used to simulate possible interventions to improve long-term health in America. We’re going to use it to conduct a brief analysis of hypothetical interventions. Specifically, your task in Part 2.1 is to construct a set of three possible interventions and to compare and contrast the success of those interventions under the assumptions of the PRISM model.

Note

An “intervention” in the context of this PRISM tool is a “Scenario”. So you’ll be creating and comparing three different scenarios.

Note

You can set the “Demographic Profile” to be whatever you want (e.g. the national average)

You need to:

  • (2 points / intervention) Specify, for each intervention, what the “levers” are that you pull (i.e. which model parameters you change), and how those are informed by the causal map that describes the assumptions of the model.
  • (6 points) Compare and contrast the outcomes for each intervention, considering at least three different outcome metrics. You should provide a chart for each of the outcomes and explain (to the best of your ability) which intervention was the most effective for that metric and why you think that was the case.
  • (3 points) Select which intervention is the “best,” and why you believe that to be the case (note that subjectivity is fine for this question)

Part 2.2 (15 points) - Critiquing an Agent-based Model

Now, we’re going to take a new simulation model and critique it. Specifically, open up this paper and, after reading it, answer the following questions:

  • Question 2.2.1 (7 points)- In 15 lines or less of pseudocode, outline the simulation “loop” (going back to our class notes as necessary).
  • Question 2.2.2 (2 points)- What was the primary research question the authors were interested in answering with their simulation model?
  • Question 2.2.3 (2 points)- What was the virtual experiment they designed to answer this question?
  • Question 2.2.4 (2 points)- What were the outcome metrics they used to assess the results of their experiment?
  • Question 2.2.5 (2 points)- What did they conclude from their findings?

Part 3 (30 Points)

We’re going to go back to Nicky Case for Part 3 and their Loopy tool. Using Loopy, your group is going to implement a system dynamics model that represents your project:

Part 3.1 (15 points)

Using Loopy, create a model of the problem, and provide 1) a link to your model and 2) a 1-3 paragraph explanation of your model, expanding where you think Atri and/or Kenny needs more context to understand your diagram. Your model (and supporting explanation) should lay out:

  1. The nodes (along with their representative level/value) and positive and negative feedback loops associated with the current world, representing your narrow problem and how it connects to white supremacy.
  2. The feedback processes that sustain the current world. For this, you will need a dynamic representation (i.e. you should tell us what to click on in the linked-to model to see white supremacy in action).

Part 3.2 (15 points)

Now, you’ll show us how your solution will get us from this current world to the future world:

  • Question 3.2.1 (5 points)- Describe what the future world looks like in the context of an end state for your future model. That is, what does a world without white supremacy look like in terms of the levels for (a subset of) the nodes in your model??
  • Question 3.2.2 (5 points)- Explain, from the perspective of your model, what would need to be done to get to this end state. That is, tell us what we need to click on/do in your model to see the end state from Question 3.2.1.
  • Question 3.2.3 (5 points)- Explain how your solution actually acts to cause those changes to the model. That is, there should be an easy mapping that Atri/Kenny can understand from “click on this” to your actual solution. Note that your solution might have to (and probably should have to) change from your Unit 2 group submission in order to make this happen.

References

  • Ali, M., Sapiezynski, P., Bogen, M., Korolova, A., Mislove, A., & Rieke, A. (2019). Discrimination through Optimization: How Facebook’s Ad Delivery Can Lead to Biased Outcomes. Proceedings of the ACM on Human-Computer Interaction, 3(CSCW), 1–30. Link
  • Boeker, M., & Urman, A. (2022). An Empirical Investigation of Personalization Factors on TikTok. Proceedings of the ACM Web Conference 2022, 2298–2309. Link
  • Brashier, N. M. (2024). Fighting Misinformation Among the Most Vulnerable Users. Current Opinion in Psychology, 101813. Link
  • Ecker, U. K. H., Lewandowsky, S., Cook, J., Schmid, P., Fazio, L. K., Brashier, N., Kendeou, P., Vraga, E. K., & Amazeen, M. A. (2022). The psychological drivers of misinformation belief and its resistance to correction. Nature Reviews Psychology, 1(1), 13–29. Link
  • Geers, M., Swire-Thompson, B., Lorenz-Spreen, P., Herzog, S. M., Kozyreva, A., & Hertwig, R. (2024). The Online Misinformation Engagement Framework. Current Opinion in Psychology, 55, 101739. Link
  • González-Bailón, S., Lazer, D., Barberá, P., Zhang, M., Allcott, H., Brown, T., Crespo-Tenorio, A., Freelon, D., Gentzkow, M., Guess, A. M., Iyengar, S., Kim, Y. M., Malhotra, N., Moehler, D., Nyhan, B., Pan, J., Rivera, C. V., Settle, J., Thorson, E., … Tucker, J. A. (2023). Asymmetric ideological segregation in exposure to political news on Facebook. Science, 381(6656), 392–398. Link
  • Guess, A. M., Malhotra, N., Pan, J., Barberá, P., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., González-Bailón, S., Kennedy, E., Kim, Y. M., Lazer, D., Moehler, D., Nyhan, B., Rivera, C. V., Settle, J., Thomas, D. R., … Tucker, J. A. (2023a). How do social media feed algorithms affect attitudes and behavior in an election campaign? Science, 381(6656), 398–404. Link
  • Guess, A. M., Malhotra, N., Pan, J., Barberá, P., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., González-Bailón, S., Kennedy, E., Kim, Y. M., Lazer, D., Moehler, D., Nyhan, B., Rivera, C. V., Settle, J., Thomas, D. R., … Tucker, J. A. (2023b). Reshares on social media amplify political news but do not detectably affect beliefs or opinions. Science, 381(6656), 404–408. Link
  • Kendeou, P., & Johnson, V. (2024). The nature of misinformation in education. Current Opinion in Psychology, 55, 101734. Link
  • Martel, C., & Rand, D. G. (2023). Misinformation warning labels are widely effective: A review of warning effects and their moderating features. Current Opinion in Psychology, 54, 101710. Link
  • Matias, J. N. (2023). Influencing recommendation algorithms to reduce the spread of unreliable news by encouraging humans to fact-check articles, in a field experiment. Scientific Reports, 13(1), 11715. Link
  • Newman, E. J., & Schwarz, N. (2024). Misinformed by images: How images influence perceptions of truth and what can be done about it. Current Opinion in Psychology, 56, 101778. Link
  • Nyhan, B., Settle, J., Thorson, E., Wojcieszak, M., Barberá, P., Chen, A. Y., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., González-Bailón, S., Guess, A. M., Kennedy, E., Kim, Y. M., Lazer, D., Malhotra, N., Moehler, D., … Tucker, J. A. (2023). Like-minded sources on Facebook are prevalent but not polarizing. Nature, 620(7972), 137–144. Link
  • Robertson, R. E., Green, J., Ruck, D. J., Ognyanova, K., Wilson, C., & Lazer, D. (2023). Users choose to engage with more partisan news than they are exposed to on Google Search. Nature, 618(7964), 342–348. Link
  • Swire-Thompson, B., & Lazer, D. (2022). Reducing Health Misinformation in Science: A Call to Arms. The ANNALS of the American Academy of Political and Social Science, 700(1), 124–135. Link
  • Joseph, K., Shugars, S., Gallagher, R., Green, J., Mathé, A. Q., An, Z., & Lazer, D. (2021). (Mis) alignment between stance expressed in social media data and public opinion surveys. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 312–324. Link
  • Udry, J., & Barber, S. J. (2024). The illusory truth effect: A review of how repetition increases belief in misinformation. Current Opinion in Psychology, 56, 101736. Link