Neural Sync Active

Introduction to Statistics

Registry Synced

Introduction to Statistics

3650 words

18 min read

Reading compass

Now · 🎯 Learning Objectives

Introduction to Statistics

🎯 Learning Objectives

After completing this topic, you will be able to:

Define statistics in your own words and explain why it matters in daily life and data science
Distinguish between descriptive statistics and inferential statistics with concrete examples
Explain the difference between a population and a sample, and why sampling is necessary
Identify parameters (population characteristics) vs statistics (sample characteristics)
Recognize the role of statistics in the data science workflow

📋 Prerequisites

No prior statistics knowledge required. This is the starting point.
Basic arithmetic (addition, division, fractions) from high school mathematics.
Coming from BSMA1001-maths-1: The concept of sets and subsets helps when we talk about populations and samples — a sample is literally a subset of the population.

📖 Core Content

1.1 Intuition: What Is Statistics?

Imagine you've baked a giant pot of soup. You want to know if it's salty enough. Do you drink the entire pot? Of course not — you'd be sick! Instead, you stir it well, take one spoonful, taste it, and decide. That one spoonful (a sample) tells you something about the whole pot (the population). Statistics is the science of learning from data. It gives us the tools to:

Collect data in a reliable way (so our "spoonful" fairly represents the pot)
Summarize and describe what we see (is the spoonful salty? how salty?)
Draw conclusions that go beyond the data we have (the whole pot is likely seasoned well enough)
Make decisions under uncertainty (should I add more salt?) Every time you check a weather forecast ("80% chance of rain"), read a news article citing a poll ("62% of voters approve"), or decide whether a new drug works based on a clinical trial, you're relying on statistics. It's not just a math subject — it's the way we make sense of a world full of variation and uncertainty.

🔑 Key Insight: Statistics doesn't give you certainty — it gives you a principled way to measure and communicate uncertainty.

1.2 Formal Definition

Statistics is the science of collecting, organizing, analyzing, interpreting, and presenting data to answer questions and make decisions under uncertainty. This definition breaks into five activities:

Activity	What It Means	Example
Collecting	Gathering data through surveys, experiments, or observations	Asking 500 people their favorite ice cream flavor
Organizing	Structuring data into tables, charts, summaries	Making a bar chart of flavor preferences
Analyzing	Applying mathematical tools to find patterns	Computing the average rating of chocolate flavor
Interpreting	Explaining what the analysis means in context	"Chocolate is preferred by 35% of respondents"
Presenting	Communicating findings clearly	A dashboard showing flavor preferences by age group

1.3 Descriptive vs. Inferential Statistics

This is the single most important distinction in the entire course.

Descriptive Statistics

Descriptive statistics = methods for summarizing and describing data. It's what you do when you have data in front of you and you want to understand what it says. Examples of descriptive statistics:

The average (mean) height of students in a class
A bar chart showing monthly sales
The range of temperatures recorded in a week
The percentage of voters who support a policy Descriptive statistics does not generalize beyond the data you have. It simply describes what's there.

Everyday analogy: You take a selfie — that photo describes how you look at that moment. It doesn't predict how you'll look tomorrow.

Inferential Statistics

Inferential statistics = methods for making predictions or generalizations about a larger group based on data from a smaller group. It's where we use probability to quantify how confident we are in our conclusions. Examples of inferential statistics:

Predicting election results from a poll of 1,000 voters
Estimating the average IQ of all Indians from a sample of 500 people
Testing whether a new drug is more effective than a placebo
Computing a margin of error for a survey

Everyday analogy: You taste one spoonful of soup and conclude the whole pot needs more salt. That's inference — using a small amount of information to make a judgment about a larger amount. The key difference:

Aspect	Descriptive	Inferential
What it does	Summarizes data	Draws conclusions beyond data
Scope	Limited to data at hand	Generalizes to larger group
Uncertainty	None — it's just describing	Central — quantified by probability
Example	"Class average is 72%"	"The population average is likely between 68% and 76%"

1.4 Population and Sample

Whenever we do statistics, we need to be clear about who or what we're studying.

Population

A population is the entire group of individuals, objects, or measurements we are interested in.

It's the "big picture" — everyone or everything we want to draw a conclusion about.
Populations can be large ("all registered voters in India") or small ("all students in this room").
Sometimes the population is hypothetical ("all possible measurements of this chemical reaction").

Sample

A sample is a subset of the population that we actually collect data from.

It's the "spoonful" — the part we can actually measure.
We study samples because populations are usually too large, expensive, or impossible to measure completely.
The sample should represent the population — it should look like a miniature version of the full group.

Parameter vs. Statistic

Term	Definition	Example
Parameter	A numerical summary of the population	The true average height of all Indian women (μ)
Statistic	A numerical summary of the sample	The average height of 500 Indian women we measured (x̄)

Memory trick: Parameter = Population. Statistic = Sample.

Why This Matters

If you confuse a statistic with a parameter, you'll overstate what you know. A sample statistic is an estimate — it has uncertainty. A parameter (if you could measure it) is a fact. Example: A news headline says "42% of Indians support Policy X, based on a poll of 2,000 people."

The 42% is a statistic (computed from the sample)
The true support level among all Indians is the parameter (unknown)
The article should include a margin of error (like ±3%) to show the uncertainty

1.5 Statistics in the Data Science Workflow

Statistics isn't just one step — it's woven throughout the entire data science process: (Diagram) Every stage uses statistical thinking:

Collecting data: How do we avoid bias? How large should the sample be?
Cleaning data: How do we handle missing values? What's an outlier?
Exploring data: What's the typical value? How spread out is it? What patterns exist?
Modeling: How confident are we in our predictions? Which factors are significant?
Interpreting: How do we communicate uncertainty to non-technical audiences?

1.6 Types of Statistical Studies

There are two main types of studies that produce data:

Observational Study

The researcher observes subjects and measures variables of interest without assigning any treatments.

Goal: Identify associations or patterns
Limitation: Cannot establish causation (because of confounding variables)
Example: Surveying 1,000 people about their exercise habits and heart health

Experiment

The researcher actively imposes a treatment on subjects to observe the response.

Goal: Establish cause-and-effect relationships
Key feature: Random assignment of subjects to treatment/control groups
Example: Randomly assigning patients to receive either a new drug or a placebo

Crucial distinction: Association ≠ Causation. Just because two things are related doesn't mean one causes the other. Ice cream sales and drowning rates both increase in summer — buying ice cream doesn't cause drowning! The confounding variable is hot weather (more people swim).

1.7 Worked Examples

Example 1: Descriptive vs. Inferential (Easy)

Scenario: A school principal wants to understand how students performed on the final exam. Data: The scores of all 300 students in the school. Question: The principal computes the average score as 78.4 out of 100. Is this descriptive or inferential statistics? Solution:

Step	Reasoning
1. Identify the population	All 300 students in the school
2. Identify the data	The scores of all 300 students
3. What is being done?	Computing the average — summarizing the data
4. Conclusion	Descriptive — because we're describing the data we have, not generalizing beyond it

If the principal instead used these 300 scores to predict how next year's students will perform, that would be inferential (generalizing beyond the data).

Example 2: Parameter vs. Statistic (Medium)

Scenario: A researcher wants to estimate the average monthly spending on food for households in Mumbai. She surveys 500 households and finds an average of ₹8,500 per month. Questions: a) What is the population? b) What is the sample? c) Is ₹8,500 a parameter or a statistic? Solution:

Part	Answer	Reasoning
a) Population	All households in Mumbai	This is the entire group we want to know about
b) Sample	The 500 households surveyed	This is the subset we actually measured
c) ₹8,500	Statistic	It's computed from sample data. The true population parameter (the real average for all Mumbai households) is unknown

Example 3: Identifying Study Type (Harder)

Scenario: A pharmaceutical company develops a new headache medication. They recruit 200 volunteers and randomly assign 100 to receive the drug and 100 to receive a placebo (sugar pill). After 2 hours, they ask each volunteer to rate their headache on a 1-10 scale. Questions: a) Is this an observational study or an experiment? b) What is the population being studied? c) What is the treatment? d) Can the company conclude causation? Solution:

Part	Answer	Reasoning
a) Type	Experiment	The researchers actively assigned treatments (drug vs. placebo) — they didn't just observe
b) Population	All people who suffer from headaches	The volunteers represent the larger group the drug is meant for
c) Treatment	The new headache medication	The placebo is the "control" — it's what we compare against
d) Causation?	Yes — provided the study was well-designed	Random assignment controls for confounding variables. If the drug group reports significantly less pain, we can attribute the difference to the drug

1.8 Edge Cases & Gotchas

The Entire Population Might Be Unreachable

Sometimes it's literally impossible to measure the whole population. For example:

Hypothetical populations: "All possible outcomes of rolling a die" — infinite!
Destructive testing: If you're testing how much pressure a bridge can hold, testing every bridge would destroy them all
Practical constraints: Even if you could survey all 1.4 billion Indians, it would be too expensive and time-consuming In these cases, a sample isn't a shortcut — it's the only option.

Sample Statistics Have Variability

Here's a crucial fact: if you take different samples from the same population, you'll get different statistics. This is called sampling variability. Example: If 10 different polling agencies each survey 1,000 voters, they'll get 10 different percentages for candidate support — even if they all use perfect methods. No single sample gives the exact population parameter. This variability is why we need inferential statistics — to quantify how much we should trust a single sample's result.

Bias Is Invisible in Your Data

A sample can be biased (systematically different from the population) without you knowing it. If your sample isn't representative, your statistics will be misleading. Classic example: The 1936 Literary Digest poll predicted Alf Landon would win the US presidential election. They surveyed 2.4 million people — a huge sample! But they surveyed their own readers (who were wealthier), car owners, and telephone subscribers. In 1936, this systematically excluded poorer voters, who overwhelmingly voted for Franklin Roosevelt. The poll was spectacularly wrong despite the massive sample size.

Moral: Bigger samples don't fix bias. A small, representative sample is better than a large, biased one.

1.9 Why This Matters

Statistics is the language of data science. Every data science task — from building a recommendation system to training a neural network to A/B testing a website — rests on statistical foundations:

Machine learning is essentially applied inferential statistics (learning patterns from training data to predict on new data)
A/B testing (comparing two versions of a website) is hypothesis testing — a core inferential tool
Data cleaning requires understanding distributions and outliers (descriptive statistics)
Evaluating models uses metrics (accuracy, precision, recall) that are statistical summaries Without statistics, data science is just programming. With statistics, it becomes a disciplined way of learning from data. This topic connects to:
BSMA1002 Week 3: Descriptive measures (mean, median, mode, variance)
BSMA1002 Week 7: Probability — the mathematical foundation for inference
BSMA1004 (Stats 2): Hypothesis testing and confidence intervals
BSMA1001-maths-1: Sets (population/sample as sets)

📐 Key Concepts Summary

Concept	Definition	Example	Key Insight
Statistics	Science of learning from data	Analyzing survey responses	Turns data into decisions
Descriptive Statistics	Summarizing observed data	Mean, median, charts, tables	Describes what's there
Inferential Statistics	Generalizing beyond data	Polling, A/B tests, predictions	Quantifies uncertainty
Population	Entire group of interest	All Indian voters	The "big picture" we care about
Sample	Subset actually measured	1,000 surveyed voters	The "spoonful" we taste
Parameter	Numerical summary of population	True average height (μ)	Unknown constant
Statistic	Numerical summary of sample	Sample average height (x̄)	Known, but varies by sample
Observational Study	Observe without intervention	Survey, cohort study	Finds associations
Experiment	Actively assign treatments	Clinical trial	Can establish causation

⚠️ Common Pitfalls

Pitfall 1: Confusing "Statistically Significant" with "Important"

The mistake: A student hears that a result is "statistically significant" and assumes it's a big, important finding. Why it happens: The word "significant" in everyday language means "important" or "meaningful." The correction: In statistics, "significant" means "unlikely to have occurred by chance" — that's all. A very small effect (e.g., a drug improves recovery time by 0.1%) can be statistically significant if the sample is large enough. Always ask: "Is the effect size meaningful?"

Pitfall 2: Thinking Descriptive Statistics Are "Just Averages"

The mistake: Reducing descriptive statistics to "just computing averages" and missing the richness. Why it happens: The first descriptive statistic we learn is the mean (average), and it seems simple. The correction: Descriptive statistics include measures of spread (variance, IQR), shape (skewness), position (percentiles), and visualization (histograms, box plots). A single average can be misleading — two very different datasets can have the same mean.

Pitfall 3: Overgeneralizing from a Sample

The mistake: Treating a sample statistic as if it were the exact population parameter without acknowledging uncertainty. Why it happens: It's natural to think "42% of respondents support X" means "42% of the population supports X." The correction: Always include a margin of error or confidence interval when generalizing. A sample statistic is an estimate — it has a range of plausible values around it.

📝 Practice Questions

Q1: Descriptive or Inferential?

A teacher records the test scores of all 30 students in her class and finds the median score is 82. Is this descriptive or inferential statistics? Why?

Strategy Hint: Ask yourself: Are we generalizing beyond the data we have, or just summarizing what's in front of us?
<details> Solution
Answer: Descriptive statistics.

Reasoning:

The teacher has scores from all students in the class — that's the complete population she cares about.

She's computing the median, which is a summary of the data she has.

She's not trying to predict how future students will perform or generalize to other classes.

Therefore, this is purely descriptive — describing what the data shows.

</details> > **Q2: Parameter or Statistic?** > > A national health survey measures the blood pressure of 15,000 randomly selected adults. They report that the average systolic blood pressure is 120 mmHg. Is this a parameter or a statistic? > > <details> Solution > > **Answer:** Statistic. > > **Reasoning:** > > 1. The population is "all adults" — this is who we want to know about. > 2. The measurement is based on only 15,000 adults — a **sample** of the full population. > 3. Any numerical summary computed from a sample is a **statistic**. > 4. The true population parameter (the actual average blood pressure of all adults) remains unknown. </details> > **Q3: Population Identification** > > A researcher wants to study the effect of a new fertilizer on tomato plant yield. She applies the fertilizer to 50 tomato plants and measures their fruit production. What is the population? > > <details> Solution > > **Answer:** All tomato plants (or more specifically, all tomato plants that could be treated with this fertilizer). > > **Reasoning:** > > 1. The researcher wants to draw conclusions about the fertilizer's effect on **tomato plants in general**. > 2. The 50 plants she actually treats are just a sample. > 3. The population is the broader group she wants to generalize to. > 4. (Sometimes the population is "all tomato plants of this specific variety under similar growing conditions" — the exact scope depends on the research question.) </details> > **Q4: Observational or Experimental?** > > A researcher analyzes hospital records to see if patients who took a particular medication had fewer complications than those who didn't. No treatment was assigned — the researcher just looked at existing records. Is this an observational study or an experiment? > > <details> Solution > > **Answer:** Observational study. > > **Reasoning:** > > 1. The researcher did **not** assign any treatments — she's looking at existing data. > 2. Patients chose (or their doctors chose) whether to take the medication — that's not random assignment. > 3. Since there's no intervention, this is observational. > 4. **Caveat:** The researcher cannot conclude causation. Patients who took the medication may differ from those who didn't in other ways (healthier lifestyle, better access to healthcare, etc.). </details> > **Q5: Bias Scenario** > > A university wants to know how satisfied students are with campus food. They survey students at the entrance of the only campus dining hall during lunchtime. 85% of respondents say they're satisfied. Is this likely representative of all students? > > <details> Solution > > **Answer:** Probably not representative — the sample is likely biased. > > **Reasoning:** > > 1. **Who's being surveyed?** Only students who eat at the dining hall during lunchtime. > 2. **Who's excluded?** Students who brown-bag lunch, eat off-campus, skip lunch, or have early/late classes. > 3. **Direction of bias:** Students who eat at the dining hall are likely more satisfied (or at least more tolerant of the food) than those who choose not to eat there. > 4. **Result:** The 85% satisfaction rate is likely an **overestimate** of the true satisfaction for all students. > 5. **Better approach:** Survey a random sample of all students, regardless of where they eat. </details> > **Q6: Experiment Design** > > A company wants to test whether a new website layout increases the time users spend on the site. Describe how you would design a statistical experiment to answer this question. > > <details> Solution > > **Answer:** A randomized controlled experiment (A/B test). > > **Outline:** > > 1. **Population:** All users who visit the website. > 2. **Sample:** A subset of users during the test period. > 3. **Random assignment:** When a user visits the site, randomly show them either the **current layout** (control group) or the **new layout** (treatment group). > 4. **Measurement:** Record how long each user stays on the site. > 5. **Comparison:** Compare the average time spent between the two groups. > 6. **Inference:** If the new layout group stays significantly longer, we can attribute it to the layout change (since random assignment controlled for other factors). </details> > **Q7: Sample Size Fallacy** > > A political poll surveys 10,000 people and finds candidate A leads with 52% vs 48% for candidate B. Another poll surveys only 500 people and finds the same 52-48 split. Which result should you trust more — the larger poll or the smaller one? Why? > > <details> Solution > > **Answer:** The larger poll (10,000) is generally more trustworthy, **but only if both polls used proper random sampling.** > > **Reasoning:** > > 1. All else being equal, larger samples give more precise estimates (smaller margin of error). > 2. The 10,000-person poll would have a margin of error of about ±1%, while the 500-person poll would have about ±4.4%. > 3. **However**, sample size isn't everything. If the 10,000-person poll used a biased method (e.g., only calling landlines, excluding younger voters), the smaller poll could be more accurate if it used better methodology. > 4. **Lesson:** Quality of data matters as much as quantity. </details> > **Q8: Real-World Application** > > A fitness app claims its users lose an average of 5 kg in the first month. The claim is based on data from 10,000 users who downloaded the app and used it for at least 30 days. > > a) Is the 5 kg a parameter or a statistic? b) Can we conclude the app causes weight loss? c) What potential biases exist in this data? > > <details> Solution > > **Answer:** > > a) **Statistic** — it's computed from the 10,000 users (a sample of all possible users). > > b) **No, we cannot conclude causation** — this is an observational study (users chose to use the app; there's no control group). > > c) **Potential biases:** > > - **Selection bias**: People who choose to download a fitness app may be more motivated to lose weight regardless of the app. > - **Attrition bias**: Only users who stuck with the app for 30 days are included. Those who quit (and perhaps didn't lose weight) are excluded. > - **Self-reporting bias**: Users might inaccurately report their starting and ending weights. > - **Healthy user bias**: People using fitness apps may also be making other healthy changes (diet, exercise) — the app may not be the sole cause. </details> * * * ## 🔗 Cross-References - **Next topic:** [Data Types and Scales of Measurement](/notes/01-foundation-bsma1002-stats-1-week01-01-data-types-scales) — now that we know what statistics is, we need to understand the kinds of data it works with - **BSMA1001-maths-1**: Sets and subsets — the mathematical foundation for population/sample thinking - **BSMA1002 Week 3**: Descriptive statistics — mean, median, mode - **BSMA1004 (Stats 2)**: Hypothesis testing — applying inferential statistics - **External**: [IITM BS Degree Stats 1 — Week 1 Lecture Videos](https://study.iitm.ac.in/) [Join Discord](https://discord.gg/gE2m4Qrdqv) [Next**Data Types & Scales**](/notes/01-foundation-bsma1002-stats-1-week01-01-data-types-scales)