Midterm 1 - Python for Behavioral Data Analysis

Vocabulary flashcards covering core causal analytics concepts, causal diagram topologies, confounding criteria, and methods for addressing missing data.

Causal Analytics

The practice of understanding what is causing a particular behavior.

Five Components of Causal Analytics Models

Personal characteristics

Intentions

Actions

Business behavior

Cognition and emotion.

The division of the nervous system responsible for controlling involuntary motion

Autonomic Nervous System

Business Behaviors

Observable company actions and interventions directed at customers, including emails, marketing, pop-ups, revisions to the product, and coupons.

The region of the brain that has been correlated with involvement in subjective feelings.

Insular Cortex

What two ratings are considered to be satisfactory in a CSAT?

Ratings of $4$ or $5$ on a customer satisfaction survey, corresponding to "Satisfying" and "Very satisfying."

Confounder

An unaccounted-for factor that affects both the variable being studied and the outcome, potentially providing an alternative explanation for observed results.

Covariate

A variable that researchers measure and account for because it could influence the outcome, which can also serve as a substitute if it moves alongside the data.

The experimental apparatus that first demonstrated how behavior could be shaped through technology.

Skinner Box Experiment

What are the 3 Connection Topologies of Casual Diagrams?

Chain

Fork

Collider

What does “Slicing” a variable mean?

The process of separating variables by cutting out a specific portion of the variable.

How are variables aggregated?

Combining or summarizing data from multiple measurements using functions such as adding, finding the average, mean, minimum, maximum, or count.

How can previous actions be incorporated into a causal diagram

Binary encoding → As simple as possible (e.g., has children: yes = $1$, no = $0$) before evaluating if quantity is important.

How can customer segments be incorporated into a causal Diagram?

Past actions

Intentions (Reason for trip, reason for cancellatio)

Cognition and emotions

Personal characteristics (Country, ADR (Average Daily Rate), number of children)

Observable variables

Business behaviors

Time trends.

The statistical coefficient commonly utilized to quantify the strength of association between two numeric variables.

Pearson's R

• How can the absolute value of the correlation coefficient be used to simplify causal diagrams?

• 
  

Correlation Thresholding

The method of simplifying causal diagrams by selecting an order-of-magnitude cutoff for the absolute value of correlation coefficients (e.g., retaining values above $0.1$ and dropping values below).

What is the Disjunctive Cause Criterion

A rule stating that adding all variables that directly cause either or both variables of interest (excluding mediators between them) into a regression eliminates confounding.

Backdoor Criterion

A standard stating that a relationship between two variables is confounded if there is at least one unblocked non-causal path between them starting with an arrow pointing toward the cause of interest.

Blocked Path

• A non-causal path in a causal diagram that is neutralized either when a non-collider intermediary variable is included in the regression,

• When an intermediary collider's central variable is omitted from the regression.

Collider

A variable in a causal diagram where two arrows converge head-to-head (\rightarrow q \rarw), naturally blocking the path and preventing confounding unless conditioned on.

M-Pattern

A structural configuration that acts as a collider and must be omitted from regression analysis to prevent introducing confounding.

Missingness Correlation Matrix

An analytical matrix used to evaluate whether patterns of missing data across variables are correlated with one another.

What is the first step in determining whether your data is MAR or MNAR?

•  Does the variable's missingness appear to be affected by the variables values

• Yes = MNAR No = MAR

What are 3 techniques that can be used to handle missing data?

• Predicting min maxing: A procedure for testing if missing data can be dropped by imputing minimum and maximum values into two datasets; if regression coefficients across the datasets do not materially change, the missing data can be discarded.

• 
  

• Run a regression for that variable's most important relationship with each of the 3 data sets you now have

How does a collider effect cofounding in a causal diagram?

Blocks the path, avoiding it from becoming a cofounder

• Why should variables in an M-pattern be removed from your regression analysis?

It is a collider, which will cofound your analysis.

How can you determine if missing data is correlated?

Missingness correlation matrix

What are 3 techniques that can be used to handle missing data?