Notes

Heading 1

Heading 2

Heading 3

Heading 4

Heading 5
Heading 6

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.

Block quote

Ordered list

  1. Item 1
  2. Item 2
  3. Item 3

Unordered list

  • Item A
  • Item B
  • Item C

Text link

Bold text

Emphasis

Superscript

Subscript

Confirmation Bias class version
Summary

3-hour interactive presentation. 8 scaffolding lessons teach how confirmation bias affects every step of the research process. Participants learn strategies for addressing bias in hypothesis formation, experimental design, and data analysis. Teach one week of class, or run an intensive workshop.

‍

START Slide deck
Confirmation Bias class version

References:

Confirmation Bias

Bacon, F. (1620). Novum Organum, sive Indicia Vera de Interpretatione Naturae (“New organon, or true directions concerning the interpretation of nature”).

Bebarta, V., Luyten, D., & Heard, K. (2003). Emergency medicine animal research: Does use of randomization and blinding affect the results? Academic Emergency Medicine, 10(6), 684–687. https://doi.org/10.1111/j.1553-2712.2003.tb00056.x

Born, R. T. (2024). Stop fooling yourself! (Diagnosing and treating confirmation bias). eNeuro, 11(10), ENEURO.0415-24.2024. https://doi.org/10.1523/ENEURO.0415-24.2024

Canli, T., Zhao, Z., Desmond, J. E., Kang, E., Gross, J., & Gabrieli, J. D. E. (2001). An fMRI study of personality influences on brain reactivity to emotional stimuli. Behavioral Neuroscience, 115(1), 33–42. https://doi.org/10.1037/0735-7044.115.1.33

Casad, B. J., & Luebering, J. E. (n.d.). Confirmation bias. Encyclopedia Britannica. https://www.britannica.com/science/confirmation-bias

Catalog of Bias. (n.d.). https://catalogofbias.org/biases/

Cwiek, A., Rajtmajer, S. M., Wyble, B., Honavar, V., Grossner, E., & Hillary, F. G. (2022). Feeding the machine: Challenges to reproducible predictive modeling in resting-state connectomics. Network Neuroscience, 1–20. https://doi.org/10.1162/netn_a_00212

Darwin, F. (Ed.). (1892). Charles Darwin: His life told in an autobiographical chapter, and in a selected series of his published letters (abridged edition). https://darwin-online.org.uk/EditorialIntroductions/Freeman_LifeandLettersandAutobiography.html

Ferguson, J., Littman, R., Christensen, G., Paluck, E. L., Swanson, N., Wang, Z., Miguel, E., Birke, D., & Pezzuto, J.-H. (2023). Survey of open science practices and attitudes in the social sciences. Nature Communications, 14(1), 5401. https://doi.org/10.1038/s41467-023-41111-1

Fine, C. (2010). From scanner to sound bite: Issues in interpreting and reporting sex differences in the brain. Current Directions in Psychological Science, 19(5), 280–283. https://www.jstor.org/stable/41038586

Gazzaniga, M. S. (2005). Forty-five years of split-brain research and still going strong. Nature Reviews Neuroscience, 6(8), 653–659. https://doi.org/10.1038/nrn1723

Gelman, A., & Loken, E. (2016). The statistical crisis in science. In M. Pitici (Ed.), The Best Writing on Mathematics 2015 (pp. 305–318). Princeton University Press. https://doi.org/10.1515/9781400873371-028

Gregory, R. L. (2015). Eye and brain: The psychology of seeing (5th ed.). Princeton University Press. https://doi.org/10.1515/9781400866861

Kaanders, P., Sepulveda, P., Folke, T., Ortoleva, P., & De Martino, B. (2022). Humans actively sample evidence to support prior beliefs. eLife, 11, e71768. https://doi.org/10.7554/eLife.71768

Kahneman, D., & Tversky, A. (1996). On the reality of cognitive illusions. Psychological Review, 103(3), 582–591. https://doi.org/10.1037/0033-295X.103.3.582

Kapoor, S., & Narayanan, A. (2023). Leakage and the reproducibility crisis in machine-learning-based science. Patterns, 4(9), 100804. https://doi.org/10.1016/j.patter.2023.100804

Kappes, A., Harvey, A. H., Lohrenz, T., Montague, P. R., & Sharot, T. (2020). Confirmation bias in the utilization of others’ opinion strength. Nature Neuroscience, 23(1), 130–137. https://doi.org/10.1038/s41593-019-0549-2

Little, M. A., Varoquaux, G., Saeb, S., Lonini, L., Jayaraman, A., Mohr, D. C., & Kording, K. P. (2017). Using and understanding cross-validation strategies. GigaScience, 6(5). https://doi.org/10.1093/gigascience/gix020

MacLeod, A. K., Coates, E., & Hetherton, J. (2008). Increasing well-being through teaching goal-setting and planning skills: Results of a brief intervention. Journal of Happiness Studies, 9(2), 185–196. https://doi.org/10.1007/s10902-007-9057-2

Matthay, E. C., & Glymour, M. M. (2020). A graphical catalog of threats to validity: Linking social science with epidemiology. Epidemiology, 31(3), 376–384. https://doi.org/10.1097/EDE.0000000000001161

Monaghan, T. F., Agudelo, C. W., Rahman, S. N., Wein, A. J., Lazar, J. M., Everaert, K., & Dmochowski, R. R. (2021). Blinding in clinical trials: Seeing the big picture. Medicina, 57(7), 647. https://doi.org/10.3390/medicina57070647

Muthukumaraswamy, S. D., Forsyth, A., & Lumley, T. (2021). Blinding and expectancy confounds in psychedelic randomized controlled trials. Expert Review of Clinical Pharmacology, 14(9), 1133–1152. https://doi.org/10.1080/17512433.2021.1933434

Platt, J. R. (1964). Strong inference: Certain systematic methods of scientific thinking may produce much more rapid progress than others. Science, 146(3642), 347–353. https://doi.org/10.1126/science.146.3642.347

Schulz, K. F., Altman, D. G., Moher, D., & for the CONSORT Group. (2010). CONSORT 2010 statement: Updated guidelines for reporting parallel group randomised trials. BMJ, 340(mar23 1), c332. https://doi.org/10.1136/bmj.c332

Toga, A. W., & Thompson, P. M. (2003). Mapping brain asymmetry. Nature Reviews Neuroscience, 4(1), 37–48. https://doi.org/10.1038/nrn1009

Tuyttens, F. A. M., De Graaf, S., Heerkens, J. L. T., Jacobs, L., Nalon, E., Ott, S., Stadig, L., Van Laer, E., & Ampe, B. (2014). Observer bias in animal behaviour research: Can we believe what we score, if we score what we believe? Animal Behaviour, 90, 273–280. https://doi.org/10.1016/j.anbehav.2014.02.007

Verploegh, I. S. C., Lazar, N. A., Bartels, R. H. M. A., & Volovici, V. (2022). Evaluation of the use of p values in neurosurgical literature: From statistical significance to clinical irrelevance. World Neurosurgery, 161, 280–283.e3. https://doi.org/10.1016/j.wneu.2022.02.018

Vesterinen, H. M., Sena, E. S., ffrench-Constant, C., Williams, A., Chandran, S., & Macleod, M. R. (2010). Improving the translational hit of experimental treatments in multiple sclerosis. Multiple Sclerosis Journal, 16(9), 1044–1055. https://doi.org/10.1177/1352458510379612

Vul, E., Harris, C., Winkielman, P., & Pashler, H. (2009). Puzzlingly high correlations in fMRI studies of emotion, personality, and social cognition. Perspectives on Psychological Science, 4(3), 274–290. https://doi.org/10.1111/j.1745-6924.2009.01125.x

Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly Journal of Experimental Psychology, 12(3), 129–140. https://doi.org/10.1080/17470216008416717

Weiss, Y., Simoncelli, E. P., & Adelson, E. H. (2002). Motion illusions as optimal percepts. Nature Neuroscience, 5(6), 598–604. https://doi.org/10.1038/nn0602-858

Ye, J., Li, Y., Lazar, N. A., Schaeffer, D. J., & McDowell, J. E. (2016). Finding common task‐related regions in fMRI data from multiple subjects by periodogram clustering and clustering ensemble. Statistics in Medicine, 35(15), 2635–2651. https://doi.org/10.1002/sim.6906

Zitron-Emanuel, N., & Ganel, T. (2018). Food deprivation reduces the susceptibility to size-contrast illusions. Appetite, 128, 138–144. https://doi.org/10.1016/j.appet.2018.06.006

‍

Confirmation Bias class version

Instructor guide:

Confirmation Bias

This instructor guide is designed as a flexible framework to help you deliver the Confirmation Bias unit effectively. 

  • Each lesson is summarized with key takeaways and includes cues for additional, step-by-step directions for activities. 
  • Use the video and supplementary slide-by-slide annotations with speaker notes provided to quickly familiarize yourself with the material, streamline your lesson preparation, and enhance classroom discussions. 
  • Feel free to adapt and customize the content to fit your teaching style and your students’ needs. Get access to the slides here, then navigate to File -> Make a Copy to get started.

Overview and Introduction

‍

Summary:

This unit on confirmation bias explores the subtle yet powerful ways that our predispositions can influence every stage of scientific research. By understanding how confirmation bias can distort data collection, analysis, and interpretation, students will learn essential strategies to design more rigorous, transparent, and reproducible studies. This unit is ideal for early-career researchers and advanced students who wish to strengthen their critical thinking skills and safeguard against biased reasoning.

‍

Why use this unit:

  • This unit on confirmation bias equips students with a fundamental understanding of cognitive biases, especially confirmation bias, which is often the root of many research errors.
  • Each lesson blends theoretical insights with practical activities, ensuring that learners not only recognize bias but can also implement strategies to minimize its impact in their work.
  • Real-world examples and interactive activities encourage students to reflect on their own decision-making processes and develop more robust research practices.

Lesson 1: Our biased brains

‍

Lesson summary:


In this lesson, instructors introduce the concept of confirmation bias, highlighting how our brains tend to favor information that reinforces existing beliefs. Students will learn why these biases exist from both psychological and neuroscientific perspectives, and how they can inadvertently affect decision-making in research.

‍


Key takeaway:


Students should come away with an awareness of their own cognitive biases and an understanding of how even expert researchers can be influenced by them.

‍


Activity overview:
(~3-5 minutes)

Participants engage in a hypothesis testing exercise (i.e., a modified version of the Wason 2-4-6 task) to experience firsthand how initial choices can bias subsequent evidence sampling.
 

‍

Link to activity:

https://hms-wason-246-v2-grid.vercel.app/?sessionID=3mrrc48ncda 

‍

Step-by-Step activity instructions:


  1. [Instructor] Prompt students to start typing numbers into the 3 input fields. 
    1. Remind them that they will either see “TRUE” or “FALSE” when then hit the “TEST” button.
      1. This indicates if their number sequences match the “secret rule” for this session.
    2. Note: Each individual user will have a random “secret rule,” so if you’d like pairs or teams to work together on the same one, have a representative be the only one who pulls up the activity/have them share a single device for that group.
  2. Type numbers into the 3 input boxes then hit the “TEST” button to see if your guess is “true” or “false” in regards to the “secret rule.”
  3. Repeat until you’re ready to draft a hypothesis.
  4. Hit the “SUBMIT YOUR HYPOTHESIS” button to reveal an input box.
  5. Type your hypothesis for what the “secret rule” is into the input box.
  6. Click “SUBMIT” to discover what the “secret rule” is and if you guessed correctly.
  7. Click “CONTINUE” to review your guesses in relation to the guesses of others.
  8. [Instructor] Spend time with students reviewing their guesses. Make sure that by the end of discussion, there is a conversation about falsifying your hypothesis. 
    1. Use the discussion questions in the unit to help facilitate discussion.

‍

Activity takeaway:

Students will need to figure out that they must falsify their initial hypotheses.

‍

‍

Lesson 2: “Favored” vs. alternative hypotheses

‍

Lesson summary:

This lesson focuses on the pitfalls of designing experiments that only test a single, “favored” hypothesis. Instructors will discuss the importance of developing multiple, mutually exclusive hypotheses to challenge assumptions and reveal hidden biases.

‍


Key takeaway:


Students should learn to critically evaluate their own research designs by ensuring they incorporate alternative explanations to avoid tunnel vision.

‍


Activity overview:
(~10 minutes)

Students will work solo or in small groups to generate a mutually exclusive hypothesis to their initial favored idea, using prompting questions to help them think through different possibilities.
This activity can be repeated multiple times to receive different prompting questions. A list of prompting questions to generate mutually exclusive hypotheses is available here. 

‍

Link to activity:

https://hms-cbi-hyp-bot-v1.vercel.app/ 

‍

Step-by-Step activity instructions:


  1. [Instructor] you will have the option of sharing the activity with your students via QR Code or a direct link.
  2. Decide if you want to write your own hypothesis (and draft it in the input box on the left) or select the starting hypothesis provided then click the “CONTINUE” button.
  3. [Instructor] Remind students that they are drafting an alternative hypothesis to the one they either wrote or were given on the first page of this activity.
  4. Next you should see a chatbox on the left and on the right, your starting hypothesis with an input box below it. Draft an alternative hypothesis in that input box using the first question in the chatbox to help you. Click “Submit” when ready.
  5. You should get a new question in the chatbox. Use it to draft a second alternative hypothesis and hit “submit” again.
  6. [Instructor] Prompt students to continue for a total of 3 rounds.
  7. Repeat this process for a third time.
  8. You will be redirected to a page to see all your hypotheses as well as those of your peers if you worked on this activity among a group.
  9. You can export the results to PDF if you would like.
  10. [Instructor] Lead discussion while on the results screen. Review the 3-round “journey” for select students who volunteer to share or who you choose to call on.
    1. Have students share how they approached their iterations.
    2. Use the discussion questions in the unit to help facilitate discussion.

‍

Activity takeaway:

Students will practice thinking about mutually exclusive hypotheses, and will see how others responded to different prompting questions. 

Lesson 3: Researcher degrees of freedom

‍

Lesson summary:

This lesson explains how subjective decisions, ranging from data cleaning to statistical analysis, can unintentionally favor an expected outcome. Through concrete examples, students learn how “researcher degrees of freedom” can lead to overconfident or skewed results.

‍


Key takeaway:


Students should understand the prevalence of decision points in scientific research and how each decision is a potential point of entry for bias.

‍


Activity overview: (~10 minutes)


In this session, learners are given a case study illustrating various choices that can lead to bias. They will then select strategies that are at the greatest risk for producing biased results.
 If you have a small group, consider having a single person show their screen to all participants so that the group can debate these “would you rather” prompts as they go. If you have a larger group, consider dividing them into small groups to attempt the activity. 

‍

Heads up!

This activity uses some technical language that might feel unfamiliar to those without some knowledge of fMRI research. The point of using this specific example isn’t to intimidate you or your students, it’s to drive home the point that we can still find rigor issues even if we aren’t deeply familiar with a subject area. Here’s how!

  • Analysis Parameter: Smoothing strongly favors h1 because it loosens the parameters of the analysis to find a promising result. 
  • Analyzing the most correlated voxel strongly favors h1 because it double-dips, using the same data to determine the scope of the analysis and perform the analysis.
  • Personality screen strongly favors h1 because it selects a biased sample that seems aligned with that hypothesis.
  • Simulated subjects might favor h1, but it’s more so bad statistical practice to use simulations to inflate the power of an experiment. 
  • A simple control seems like a good idea, but adding in another kind of picture (landscapes), which is an unnecessary source of noise. 
  • Multiple comparison correction actually favors the null hypothesis, rather than either h1 or h2. This correction, called the Bonferroni correction, is a very strong correction that would make it impossible to find significant results when thousands of data points are examined. 

‍

Link to activity:

https://hms-cbi-fav-game-v0.vercel.app/ 

‍

Step-by-Step activity instructions:


  1. [Instructor] you will have the option of sharing the activity with your students via QR Code or a direct link.
  2. Read the H1 for the fMRI case study: “H1 is that emotional images have more peaked fMRI activity.”
  3. Then read the 2 choices provided and click on the option you believe would be the MOST biased action to take in an experiment about this H1.
  4. Then type out your reasoning for why this action would bias the experiment in favor of H1 in the input box and click the “ROUND 2” button.
  5. Repeat then click the “ROUND 3” button.
  6. Repeat then click the “SUBMIT” button.
  7. Compare your results with those of your peers on the final page. You can toggle through everyone’s answers and discuss your thought process.
  8. [Instructor] Lead discussion while on the results screen. Be sure to toggle between the different responses provided in each round.
    1. Hover over the visualization to see a pop-up with text of answers from the game.
    2. Use the “PREVIOUS” and “NEXT” buttons to reveal student responses.
    3. Use the discussion questions in the unit to help facilitate discussion.

‍

Activity takeaway:

Students should begin to build an intuition around the kinds of research practices that can, while well-intentioned, introduce bias into their experiment. 

‍

Lesson 4: Mitigating bias through masking

‍

Lesson summary:

This lesson introduces the concept of masking (or blinding) in experimental design. Instructors will explain how masking can prevent both participants and researchers from inadvertently influencing results with examples from clinical trials.

‍


Key takeaway:


Students should understand how masking techniques improve the objectivity of experimental results and learn the practical steps to implement effective masking.

‍


Activity overview:
(~1-2 minutes)

Students will be able to explore the impact of bias on inflated effect sizes in scientific research.


‍

Heads up!

This activity uses some sophisticated technical terms. Here are some quick definitions for each of these terms:

  • Sample Size: The number of observations or data points included in a study or experiment. Larger sample sizes generally increase the reliability and statistical power of a study.
  • Bias Amount: A measure of systematic error introduced into a study, which can distort the results away from the true effect. Bias can arise from data collection, selection processes, or analysis methods.
  • True Effect Size (d): A quantitative measure of the magnitude of the difference or relationship being studied. A larger effect size means a stronger relationship or difference between groups.
  • Probability of Statistical Significance: The likelihood that an observed result is statistically significant, meaning it is unlikely to have occurred by random chance.
  • p-value (p < 0.05, p < 0.01, p < 0.001):
  • p < 0.05: The probability of observing the result by random chance is less than 5%, often used as a standard threshold for significance.
  • p < 0.01: The probability is less than 1%, a more stringent criterion.
  • p < 0.001: The probability is less than 0.1%, an even more rigorous threshold.
  • Statistical Significance: A determination of whether an observed effect is likely to be genuine rather than due to random variation.

‍

Link to activity:

https://hms-cbi-pub-bia-v0.vercel.app/ 

‍

Step-by-Step activity instructions:


  1. [Instructor] It might make the most sense to treat this activity as an extended discussion. 
    1. Students can still work solo, but the exploration works best with prompting about how different configurations influence effect size while segueing into experiences students have had in their labs or when reading published papers.
  2. You are provided with two sliding bars. One represents “Sample size” and the other represents “Bias” in a given experiment.
  3. Click on each dot to move them along their respective sliders.
  4. Try different pairings to see how these factors distort the relationship between true effect size and the probability of statistical significance.
  5. The relationships you explore will be visualized in the graph below the 2 sliders.
  6. [Instructor] If you haven’t been a part of discussion thus far, make sure you lead a brief discussion, and connect this exploration with the next segment of the unit showcasing the impact of not masking and not randomizing on effect sizes.

‍

Activity takeaway:

Students should understand that there is a causal relationship between bias and the risk that they will fabricate significant results. 

‍

Lesson 5: How good is your mask?

‍

Lesson summary:

In this lesson, the focus shifts to the evaluation of masking effectiveness. Instructors will discuss factors that may inadvertently reveal experimental conditions and how to assess whether masking has been compromised.

‍


Key takeaway:


Learners should grasp the challenges in maintaining effective masking and be introduced to methods for evaluating its success in reducing bias.

‍


Activity overview:
(~12-15 minutes)

Participants will review a set of experimental scenarios and identify clues that might have led to accidental unmasking. They will then propose solutions to improve masking procedures.
Make sure you give all participants ample time to read the experiment, and encourage them to use the hint function to proceed if they get stuck. 

‍

Link to activity:

https://hms-aem-rig-fil-v3.vercel.app/pages/input 

‍

Step-by-Step activity instructions:


  1. [Instructor] Students will have the option to complete this activity as an individual or group. 
    1. If you want to run it for a class to only see their peers’ results, click the “AS A GROUP” button to receive a session link that each member of the class will need to use to have their results coordinate on the results page.
    2. If students select the “AS AN INDIVIDUAL” button, the results page at the end will show results from “global users” outside of your class.
  2. Review the details of a “BEHAVIORAL STUDY ON SLEEP PATTERNS” and, solo or with your group, think through all the ways this study might be accidentally unmasked.
  3. Type your observations into the input box.
  4. [Instructor] Prompt students to use the hint feature and chat about options with each other to continue thinking about risk points to unmasking.
  5. Click “HINT” whenever you feel stuck to help you think of different ways the mask might fail.
  6. When you are done, click the “SUBMIT” button.
  7. Next, along the left side of the screen, you will see the responses of other users along with your own. 
  8. [Instructor] Use this results page as a brief opportunity for discussion where students share about their choices and notes. How and why did they make those decisions?
  9. On the right side of the screen, you can supply advice to the study team in the input box then click “SUBMIT ADVICE” or you can skip giving advice and click “GO TO CHECKLIST” directly.
  10. [Instructor] Remind students that offering advice is optional, but be sure to give students more time if they choose to do this.
  11. If you do provide advice, you will be taken to a screen to review advice you and other users have given. You can then click the “GO TO CHECKLIST” button.
  12. [Instructor] For the checklist, prompt students to take a moment to consider what they will really need to remember/self-check in the future for their own experiments.
  13. On the checklist page, feel free to add custom checklist items to help you evaluate your own research using the “ADD QUESTION” button.
  14. When you are ready, click the “DOWNLOAD” button to save this list as a PDF.
  15. [Instructor] This activity is fairly discussion heavy, but feel free to wrap up the activity with any final thoughts or observations before segueing back into the lesson that picks up on the question of formally assessing if an experiment was properly masked.

‍

Activity takeaway:

Students should build an intuition for places where masking issues can arise. This activity has done its job when students have practiced both critiquing issues in the provided experiment and suggesting creative solutions to overcome those issues. 

‍

‍

Lesson 6: Analytical practices to mitigate bias

‍

Lesson summary:

This lesson differentiates between exploratory and confirmatory data analysis. Instructors will highlight how blending these approaches without clear distinction can lead to confirmation bias and discuss strategies to preserve the integrity of research findings.

‍


Key takeaway:


Students should learn how to structure their analyses to move from exploratory observations to robust, hypothesis-driven testing while avoiding data “double dipping.”

‍


Activity overview:
(~5 min)

Learners will work with a sample dataset and one of two preset hypotheses. They will evaluate the dataset to see if they can know anything “for certain” regarding their given hypothesis. They will ultimately discuss the implications of over-interpreting exploratory results. 

‍

Heads up!

All participants will be randomly assigned to one of two groups. Half will be asked to demonstrate that daily activities do influence student outcomes, while the other will demonstrate that daily activities do not influence student outcomes. 

‍

Link to activity:

https://hms-cbi-gar-for-v0.vercel.app/ 

‍

Step-by-Step activity instructions:


  1. [Instructor] Students will be given 1 or 2 possible hypotheses to inform their thinking. If possible, let students stay in the dark about the existence of a counter hypothesis their peers might have. Part of the “twist” of the experience is to see how far students will go from exploratory thinking to confirmatory thinking.
  2. You’ll see a hypothesis at the top of the screen in relation to the broader research question: “Do daily activities affect student outcomes?”
  3. Looking at the visualization of the data provided, use the radio buttons in the 2 categories (Student outcomes and daily activities) to see what types of relationships you notice between a given activity and an outcome. 
  4. [Instructor] While students are working, gently model/prompt the students to explore different choices and notice how the selecting various combinations reveals different relationships. 
  5. When you’ve noticed a relationship that connects to your given hypothesis, leave the radio dials on those 2 items, and then click the "SUBMIT” button.
  6. Next, you will be able to compare your input with your peers.
  7. [Instructor] When you lead discussion, point out that different students had conflicting hypotheses. Discuss what connections they saw. 
    1. If they selected choices that deliberately “demonstrated” that their hypothesis was “true,” discuss how they decided to bypass selections that revealed data that countered their hypothesis.
    2. Use the discussion questions in the unit to help facilitate discussion. 

‍

Activity takeaway:

Students should grasp how easy it is to find a pattern in data when one goes looking for it. This practice is very important for exploratory analysis, but becomes deeply problematic when it is used in confirmatory research.

‍

Lesson 7: Data masking in machine learning models

‍

Lesson summary:

This lesson draws parallels between experimental masking and data leakage in machine learning. Instructors will explain the concept of data leakage and its impact on model evaluation, emphasizing the importance of proper data partitioning.

‍


Key takeaway:


Students should have a very general understanding of how to design machine learning experiments to prevent data leakage and accurately assess model performance.

‍


Activity overview: (~3 min)


Students will interact with a very basic simulation of machine learning experiment for detecting a condition like Parkinson’s Disease, and identify potential sources of leakage.

‍

Link to activity:

https://hms-cbi-dat-hld-v1.vercel.app/ 

‍

Step-by-Step activity instructions:


  1. [Instructor] Students can work together or solo. The main prompt to give them is to be thoughtful about the distribution of data they assign to the “Training” vs “Testing” sets of data.
    1. Since this is an overly simplified simulation, remind students that selecting a toggle represents assigning each of the 4 subsets of data to the “Training” vs “Testing” sets.
    2. If students ensure their subsets are assigned exclusively to either the “Training” or “Testing” sets, they’ve figured out the right course of action.
    3. If students “double dip” the data, they have fallen for the trick and have “leaked” their data to the model.
  2. Take note of the “Training” vs “Testing” boxes that represent the subset of data you will be parsing out in this simulation.
  3. Use the toggle buttons to select which pieces of data you want to portion into these “Training” vs “Testing” boxes (data subsets).
  4. When you are ready, click the “BUILD MODEL” button to receive a percentage estimate of the accuracy of your model based on the testing data.
  5. Repeat this process once more then click the “PREDICT PERFORMANCE” button.
  6. You should see a bar that provides you with a refined prediction of performance based on your 2 rounds of testing.
  7. Click the “EVALUATE THE MODEL ON HOLDOUT DATA” button to see your model’s performance on other data.
  8. You will see the comparison between your prediction and the actual performance of your model.
  9. Click “SEE HOW OTHERS DID” to get a pop-up of user error distribution.
  10. [Instructor] At this stage, lead discussion and have students share about their experience. 
    1. Use the discussion questions in the unit to help facilitate discussion.

‍

Activity takeaway:

Students should successfully avoid data leakage with their model. Students should learn that reserving a share of their data for testing is a necessary component of building a good model. 

‍

Lesson 8: Bonus biases that disrupt research

‍

Lesson summary:

This final lesson broadens the discussion to other biases that can influence research, including conformity bias and cognitive dissonance. Instructors will tie these concepts back to confirmation bias, illustrating how multiple biases can compound and affect scientific outcomes.

‍


Key takeaway:


Students should come away with a holistic view of how various biases interconnect and strategies to mitigate their combined effects on research integrity.

‍


Activity overview:
(~15 min)

Participants will map out a “bias ecosystem” by connecting different biases and discussing real-life examples where these interactions have led to research pitfalls.


‍

Heads up!

This activity introduces a lot of terms for a variety of biases. If students have questions about any particular bias, you can learn more via searching the catalogue of bias. 

‍

Link to activity:

https://hms-cbi-dat-hld-v1.vercel.app/ 

‍

Step-by-Step activity instructions:


  1. [Instructor] note that there is an option on the landing page to have students work as an individual and as a group. Decide how you would like to have students proceed. 
    1. Individual: 6 preset biases; results view shows submissions from “global” users outside of your class.
    2. Group: you can choose to start with 6 preset biases or a custom start; results view shows the submissions from your class only.
  2. You should see 6 boxes in the playable field with text describing 6 different types of bias.
  3. You can click on any box (bias) to drag it to a different position within the field.
  4. You can click the center of any box to pull out a line that you can connect to another box (bias).
  5. Once you connect 2 biases, click “Click to edit” to type in a description of how these biases are connected.
  6. You can connect more than one bias to another.
  7. Once you’ve completed your “bias map” you can click “EXPORT AS PNG” or “EXPORT AS PDF” to save.
  8. Click “SUBMIT” to advance to the next page and compare your bias map with other users.
  9. [Instructor] At this stage, lead discussion and have students share about their experience and the connections they’ve made. 
    1. Segue to the discussion questions in the unit to connect students to the next portion of the lesson.

‍

Activity takeaway:

Students should build a sense for how different kinds of rigor issues and biases can replicate, cause, inform, or otherwise relate to one another. This activity has done its job when the group has shared 2-3 interesting connections between different kinds of bias!

‍

Observations & final notes

‍

Observations & final notes

Each unit is estimated to comprise approximately 3 hours of instructional time (approx. 15 minutes per lesson), but variances in discussion length, student needs, experiences with the interactive activities, or instructor customization may yield different unit and lesson durations.

Concepts likely to challenge students:

  • Students may have emotional reactions in places where the instructional content differs from their prior experience. This is okay! You’re helping them to learn how to do more rigorous work. 
    • We advise letting disgruntled students express their points of disagreement, then gently encouraging them to consider why the materials might disagree with what they’ve been taught previously. 
    • Remember: There’s nothing wrong with asking a student to hold on to their grievance to let you conclude the unit. 
    • Thank we got it wrong? We want to improve! Email us at c4r@seas.upenn.edu.
  • Differentiating between exploratory and confirmatory analysis and the pitfalls of “double dipping” in data.
  • Recognizing subtle instances of unmasking in experiments, especially when biases operate at multiple levels.
  • Designing research protocols that adequately control for the myriad ways confirmation bias can infiltrate decision-making.
  • Navigating technical details of machine learning modeling, particularly in parsing out different types of cross-validation.

Remember!

This guide is intended as a flexible framework for presenting the Confirmation Bias unit. Instructors are encouraged to adapt the content to best suit their teaching style and the needs of their students. Feel free to expand on any section, incorporate additional examples, or integrate further interactive elements. Remember, this is your presentation: use it as a starting point and customize it to best serve your teaching.

‍