Introduction to Statistics for Data Analytics

Why statistics matters

In data analytics, statistics helps you:

• Summarize data reliably (not just “eyeballing” charts)
• Quantify uncertainty (confidence intervals instead of single numbers)
• Compare groups fairly (hypothesis tests)
• Understand relationships (correlation vs causation)

Core vocabulary

• Population: the full set you care about
• Sample: observed subset of the population
• Parameter: a population quantity (true mean, true proportion)
• Statistic: a sample-based estimate (sample mean, sample proportion)
• Bias: systematic error (wrong sampling, leakage)
• Variance: how much an estimator varies across samples

A simple mental model

You rarely see the population. You take a sample and estimate.

• Your estimate is not exact.
• Your estimate changes if you resample.

That’s why we use:

• Distributions
• Standard error
• Confidence intervals

Common mistakes to avoid

1. Correlation ≠ causation
2. P-hacking (trying many tests until something is “significant”)
3. Ignoring base rates (rare events)
4. Selection bias (your sample isn’t representative)
5. Over-trusting averages without checking spread/outliers

Minimal Python setup

Imports

import numpy as np
import pandas as pd
import scipy.stats as stats

If SciPy isn’t available in your environment, you can still follow most concepts using NumPy.

How the pieces fit together

Every statistics workflow in this phase follows roughly the same shape: you start with raw data, describe it, then use that description to reason about a population you can’t fully observe.

diagram From data to decision mermaid

  flowchart LR
  A["Population
(unknown, huge)"] --> B["Sample
(what you collect)"]
  B --> C["Descriptive stats
(mean, median, spread)"]
  C --> D["Inference
(CI, hypothesis tests)"]
  D --> E["Decision
(ship it? investigate more?)"]

The general path statistics takes: raw data leads to a summary, then to inference, then to a decision.

From the book

McKinney’s Python for Data Analysis spends most of its pages on wrangling data because, as he puts it, that’s where analysts actually spend their time — modeling and statistics libraries like statsmodels and scikit-learn only pay off once the data feeding them is clean. Treat this statistics phase the same way: the fanciest test is worthless on a messy or biased sample.

🧪 Try It Yourself

Exercise 1 – Population vs. sample mean

pch.exerciseTag pch.exerciseTitle

Exercise 2 – Statistic vs. parameter

pch.exerciseTag pch.exerciseTitle

Exercise 3 – Spotting selection bias

pch.exerciseTag pch.exerciseTitle

Next

Continue to Descriptive Statistics to learn how mean, median, and variance summarize a sample before you do any inference.