Readings:

●    Chapter 4: Weapons of Math Destruction (Propaganda Machine: Online Advertising)

●    Optional:  Darrel Huff, Chapter 1: How to lie with statistics, Norton, New York, 1954 - [Internet Search – Huff + “How to lie with statistics” + pdf]

In this assignment, you’ll begin the process of exploring relationships in data. You’ll accomplish this task by computing some basic statistical measures on one of three datasets. This is a goodtime to learn or reboot your Python coding skills.

Step 1 - Select one of the datasets available on CANVAS for completion of this assignment:

●    [mental-health-in-tech-survey-2019.csv] Mental Health in Tech Survey: Survey on Mental Health in the Tech Workplace in 2019 -https://osmhhelp.org/research.html

Dependent Variables:

o treatment: Have you ever sought treatment for a mental health disorder from a mental health professional? (FALSE/TRUE)

o mental_health_disclosure: Would you feel comfortable discussing a mental health issue with your direct supervisor(s)? (Yes/Maybe/No)

o mental_health_support: Overall, how well do you think the tech industry supports employees with mental health issues? (numerical: 1-5)

●    [hospital-discharge-2017.csv] New York Hospital Inpatient Discharge Information in 2017:

https://health.data.ny.gov/dataset/Hospital-Inpatient-Discharges-SPARCS-De-Identified/22g3- z7e7

Dependent Variables:

o Length of Stay: a numeric value representing number of days between admission and discharge

o APR Severity of Illness Code: a numeric value representing severity of illness

●    [deaths-in-custody.csv] Information on deaths that occur in custody or during the process of arrest in California -https://openjustice.doj.ca.gov/data

Dependent Variables:

o manner_of_death: Suicide/Natural/Accidental/Homicide/Cannot be Determined/Other (for all others)

o custody_status: Sentenced/Awaiting Booking/Booked – Awaiting Trial/ Booked – No Charges Filed/Other (for all others)

Step 2 - Explore the data by answering the following questions:

●    Which dataset did you select?

●    How many observations are in the dataset?

●    How many variables in the dataset?

●    Does this dataset seem to belong to a regulated domain in lawas discussed in the lectures? If yes, which one?

●    How many variables in the dataset are associated with a legally recognized protected class? In a table format, list those variables associated with a protected class, identify the protected class and the

associated legal precedence/lawas discussed in the lectures.

Example Output (associated with a different dataset) -

Dataset: Housing Decisions in Metro-Atlanta

Number of Observations: 1,400

Number of Variables: 16

Regulated Domain in Law: Housing (Fair Housing Act)

Number of Protected Class Variables: 2

Step 3 - Determine the relationships between dependent and independent variables

The frequency of a value represents the number of times a value occurs in a data set. Compute the frequency of each value associated with each dependent variable (listed in Step 1) as a function of all of   the protected class variables (independent variables) identified in Step 2. Create table(s) and histogram(s) comparing the frequency values of the dependent variable as a function of the independent variable.

Hint: For variables that are continuous, you might consider creating intervals that represent the data. For categorical/ordinal/nominal values, you might consider converting to numerical values based on a reasonable (albeit subjective) ordering. Don’t forget to do this for EACH dependent variable against all of the protected class variables.

Example Output for One Dependent-Independent Variable Combination:

Step 4 - Show how to manipulate with data

Select one protected class variable (independent variable) and one dependent variable.

•    Create a graph to support the “fairness” hypothesis (The system is fair. There is no difference in the outcomes.)

•    Create a graph to support the bias hypothesis (The system is biased). There is a difference in the outcomes.

•    For each, provide a brief description of your manipulations.

Example Output:

1)   Fair Hypothesis: As seen from this graph, housing decisions are not dependent on the pregnancy status of women. [Manipulations: Used line graph; Increased Scale to +-50; Mapped the ratio of

positive Y decisions (i.e. 50/180 versus 130/180); No label on the Y-Axis].

2)   Bias Hypothesis: As seen from this graph, housing decisions are significantly dependent on the

pregnancy status of women. [This hypothesis was easily supported with the data so didn’t require much in manipulations: Used stacked bar graph; Reduced Scale; Reworded labels].

Step 5:

•    Given your selected protected class variable (independent variable), calculate the average (using mean, median, and mode) of the protected class group (Hint: Variables might need to be converted to numerical values as needed).

•    Run the random sampling method using 50% of the data to create a reduced dataset. Calculate the average (mean, median, and mode) of the protected class group. Indicate if there is a difference (or not) between the original dataset and the reduced dataset for any of the averages.

•    Provide all results.

Example Output Illustrating Possible Average Computations: