Understanding Bias in Data Science

• Understanding Bias

Instructions for addressing bias

• Research bias topics.
• Define the bias.
• Determine the bias's impact on the data science process.
• Provide examples related to class projects

Social Desirability Bias

• A response bias where survey participants provide answers based on societal expectations rather than their own beliefs.
• Impacts the gathering data and communicating results steps in data science.
• Example: In a Unit 4 project about top songs, participants might choose popular songs to avoid judgment.

Survivorship Bias

• Focuses on successful outcomes, ignoring failures.
• Impacts gathering data and communicating results in data science.
• Example: Focusing on wealthy tech entrepreneurs who dropped out of college may create the idea that college is unhelpful to starting a tech career (e.g., Steve Jobs).

Selection Bias

• Occurs when the sample analyzed is not representative of the population.
• An umbrella term encompassing survivorship bias, volunteer bias, etc.
• Impacts the data gathering step.
• Example: Data collected from Duchesne students on water usage may not apply to the general population due to their high-income households.

Volunteer Bias

• Participants choose whether to be part of a survey sample, creating a group of volunteers that differs from others.
• Impacts asking questions, gathering data, and analyzing data.
• Example: A survey sent to the Duchesne community may receive responses mainly from parents, swaying the data.

Funding Bias

• Researchers distort results due to pressure from funders, engaging in questionable research practices.
• Impacts the analyzing, synthesizing, and communicating results steps.
• Example: In a skin tone magazine project, results might be manipulated to say that colorism isn’t an issue.

Recall Bias

• Occurs when participants in a study do not accurately remember past events or leave out details when reporting
• Impacts the gathering, organizing data, and communicating results steps.
• Example: A sick patient may overestimate healthy times and downplay sick times.

Omitted Variable Bias

• Involves excluding a key variable from a model.
• Affects the model, analyze, and synthesis parts of the data science project.
• Example: In a water usage project, excluding data on the number of people living in a house can cause this bias.

Nonresponse Bias

• Occurs when those unwilling/unable to participate in a research study provide different data from those who do take part.
• Impacts gathering data, analyzing, synthesizing data, and end results.
• Example: On a survey, people may not provide an answer about their income.

Reporting Bias

• Involves selectively revealing or suppressing specific information.
• Impacts gathering and organizing data, analyzing and synthesizing data, and communicating results.
• Might arrive at faulty conclusions due to reporting on incorrect data.
• During a skin tone project, there may be inaccurate, skipped pages if there weren't many people present.

Other Bias

• Favoritism toward or prejudice against a particular gender.
• Impacts ask questions, gathering and organizing data, analysis and synthesis, and communicating results steps.
• Medical research typically focuses on only one gender where data may not be applicable to the opposite sex.

Explore bias in data science, including social desirability and survivorship bias. Learn how these biases affect data gathering and communication of results. Examples are provided related to class projects.