Lesson 2 — Data Literacy & Exploratory Data Analysis (EDA)

Why this matters (motivation)¶

Most mistakes in analytics happen before modeling:

• confusing data types (e.g., numbers stored as strings),

• ignoring missingness,

• trusting outliers without checking,

• using the wrong chart and telling the wrong story.

EDA is your “first conversation” with the data.

The EDA mindset: what questions are we asking?¶

A simple EDA checklist (what we do every time)¶

1. Preview: rows, columns, data types

2. Validate: missing values, duplicates, impossible values

3. Summarize: distributions (center + spread)

4. Compare: by group (category, time, segment)

5. Visualize: choose the right chart for the question

6. Write: 3–5 plain-language insights + 1 caveat

Data literacy essentials (types and pitfalls)¶

Data types that matter¶

• Numeric: revenue, price, age

• Categorical: region, product category, plan type

• Datetime: purchase date, signup date

• Identifiers: customer_id (not a “number” you average)

Missing values: what they might mean¶

Missingness can be:

• random (e.g., occasional logging failures), or

• systematic (e.g., income missing more for high-income respondents).

Descriptive statistics that students actually use¶

Core summary numbers¶

• Mean: average level

• Median: typical value (robust to outliers)

• Standard deviation (SD): overall spread

• Interquartile range (IQR): spread of the middle 50%

What to report (rule of thumb)¶

• If the histogram looks symmetric: mean ± SD

• If it looks skewed: median + IQR

Visualization as a decision tool (not decoration)¶

Quick chart chooser (practical)¶

• Compare categories → bar chart (sorted)

• Trend over time → line chart

• Relationship between two numeric variables → scatter plot

• Distribution → histogram / boxplot

• Two dimensions (category × category) → heatmap

Mini case: “Sales and marketing” dataset (EDA story)¶

Question: “Which product categories grew, and is growth associated with marketing spend?”

EDA steps:

1. Check data types (date as datetime, spend as numeric)

2. Summarize missing values (overall + by category)

3. Plot distribution of sales (is it skewed?)

4. Plot sales over time (overall + by category)

5. Scatter: sales vs marketing (colored by category)

6. Write 3 insights + 1 caveat

Mini-lab (Google Colab)¶

Churn Data: https://drive.google.com/file/d/1CcOCXyGM9GLFmzFbKfmIhioM-yLDkDqU/view?usp=sharing

In-class tasks (checkpoints)

1. Print data types and basic shape

2. Create a missingness summary table

3. Produce: (i) histogram/boxplot, (ii) line chart, (iii) bar chart by category

4. Submit 3 insights + 1 caveat in markdown inside the notebook

Submission

• LINK

AI check (responsible use for EDA)¶

Good prompt examples

• “Suggest 5 EDA checks for a dataset of customer transactions with dates, categories, and revenue.”

• “Given these columns, what plots best answer ‘which category is growing’?”

Bad prompt example

• “Write my EDA conclusions for me” (without your own verification and interpretation)

Review questions (quiz / reflection)¶

1. When would you report median and IQR instead of mean and SD?

2. Give one example of systematic missingness in a business dataset.

3. Which chart would you choose to show: (i) trend, (ii) distribution, (iii) group comparison—and why?

Kindly submit Reflection prompts for Lesson 1 and Lesson 2 here (before next class). https://docs.google.com/forms/d/e/1FAIpQLSfsq2ln4ru9G1Mtt8Huezzl52RbNw0SFnXyeMj9g2wuudt8UQ/viewform?usp=publish-editor