Module 05: Data Analysis for Any Major — Python, pandas, Visualization & AI-Assisted Analytics | College AI Track

"A chart is an argument about what deserves attention—and honest scales are part of that argument." — Paraphrasing stats and visualization educators

Duration: 7–9 hours · Difficulty: Intermediate · Prerequisites: Intro stats comfort; Python 3.10+ in a venv.

Learning Objectives

By the end of this module, you will be able to:

Load, inspect, and clean CSV/Excel-style tables with pandas.
Compute descriptive stats and group summaries appropriate to your domain.
Visualize with matplotlib (optional seaborn) using honest scales and source notes.
Use AI as a calculator assistant—you specify design and verify assumptions.
Compare Excel, Python, and “full AI pipeline” workflows—and know where AI misleads on causality and joins.

Concept 1: Why Python + pandas

Dimension	Excel	Python/pandas	AI assistant
First plot	Fast	Slower setup	Fast if schema clear
Reproducibility	Manual risk	Git + scripts	Weak unless you save code
Complex joins	Error-prone at scale	Explicit	May hallucinate columns
Stats	Toolpack / formulas	scipy/statsmodels	May pick wrong test

Concept: Print df.shape after every merge—row-count surprises catch silent drops or explosions.

Messy data → first check

Symptom	First check	Typical fix
Weird summaries	`describe(include="all")`	Coerce dtypes
Merge exploded	len before/after	Key duplicates
Empty chart	NA fraction	Filter or document imputation

Concept 2: Question → Report

Define question and unit of analysis (one row = ?). 2. Ingest with source + date. 3. Profile missingness, dtypes, dupes. 4. Transform with formulas in comments. 5. Analyze; visualize; report magnitude and limitations.

Associational trap: “Library hours vs GPA” ignores self-selection—frame as association, not cause, unless design supports more.

Pitfalls vs AI

Pitfall	AI might tempt you to…
Confounding	Oversimple regression
p-hacking	Run many tests silently
Ecological fallacy	Confuse group and individual

Concept 3: Visualization with AI

Sketch chart intent in chat; implement in your matplotlib. Checklist: meaningful zero baseline for counts; units on axes; n in caption; source line.

Environment Setup (plain instructions)

bash

python -m venv .venv
,[object Object],
pip install pandas matplotlib openpyxl

Hands-On: Load and Profile

Key Example: This pattern is the spine of most course projects—everything else is filtering, grouping, and plotting.

python

[object Object], pandas ,[object Object], pd

df = pd.read_csv(,[object Object],)
,[object Object],(df.shape)
,[object Object],(df.columns.tolist())
,[object Object],(df.dtypes)
,[object Object],(df.isna().mean().sort_values(ascending=,[object Object],).head())

In words: After read_csv, always check shape, dtypes, and missingness before plotting—otherwise charts lie because "12" is still a string. Group with groupby(...).agg(...), merge with explicit how= and row-count checks, plot with title, xlabel, ylabel, and savefig at 150–300 DPI when final.

Try This! Ask ChatGPT for pandas for your schema—then run head(), assert row counts, and grep column names against the real CSV header.

Fun Fact: Mean of ratios ≠ ratio of means—your caption should say which denominator you used.

Dashboards for coursework

A “dashboard” can be one figure with 2×2 panels: KPI text, distribution, breakdown by category, missingness bars—or a short HTML/PDF export from Jupyter/Quarto.

Presenting findings

Slides: Question → one clean figure → methods in five bullets → limitations → one “so what.” Executive summary: three numbers max with context for a non-technical reader.

Try This Now

Data dictionary before any plot.
Join lab — predict inner vs outer counts on toy CSVs.
Misleading chart online → redraw ethically; two sentences on what changed.

Reproducibility is kindness to graders and future you.
Visualization is argument—axes and n matter.
Statistics without design is storytelling—name bias early.

Key Takeaway

Use Excel for speed, Python for repeatability, AI for boilerplate—you always verify joins and tests.
Label charts with units, source, and sample definition; call out non-causal scatter plots honestly.
Dtypes and missingness before pretty pictures—fix coercion first.
Treat AI-suggested analyses as drafts until assumptions match your study design.
Ship a limitations paragraph—mature projects own what the data cannot prove.

Resources

pandas: https://pandas.pydata.org/docs/ · matplotlib gallery: https://matplotlib.org/stable/gallery/index.html
Wilke, Fundamentals of Data Visualization — principles

Appendix: Sample public data sources

Source	Note
data.gov	Check update frequency
Campus open data	Aggregate; FERPA norms
Kaggle	Cite honestly in coursework

Record access date and license in methods.

Module 05: Data Analysis for Any Major — Python, pandas, Visualization & AI-Assisted Analytics