Statistics for Data Analysis & Data

Science
Complete Lecture Notes · All Major Topics

Contents
01 Introduction to Statistics 07 Correlation & Covariance

02 Types of Data & Variables 08 Probability

03 Population & Sampling 09 Probability Distributions

04 Descriptive Statistics 10 Normal Distribution

05 Measures of Dispersion 11 Inferential Statistics

06 Graphical Representation 12 Hypothesis Testing

01 Introduction to Statistics
A branch of mathematics involving collecting, analyzing, interpreting, and drawing conclusions from data.

Two Types of Statistics

Descriptive Statistics Inferential Statistics

Summarizes and describes data using measures, Draws conclusions about a population based on a
graphs, and tables. First step in any analysis. sample. Includes estimation and hypothesis testing.

Data Analysis Workflow

Step Action

1 Define the problem / research question

2 Collect data (surveys, experiments, existing datasets)

3 Clean data — remove inconsistencies and handle missing values

4 Exploratory Data Analysis (EDA) — understand distributions and patterns

5 Transform data (normalize, encode categorical variables)

6 Formulate hypothesis (null and alternate)

7 Apply statistical tests

8 Interpret results — accept or reject hypothesis

9 Document analysis and create a report

02 Types of Data & Variables
Data Categories
Type Description Example

Structured Organized in rows and columns Excel sheets, databases

Unstructured No fixed structure Images, audio, emails, web

pages

Cross-sectional Collected at a single point in time Survey data, exam marks

Time Series Collected over a sequence of time intervals Daily stock prices, monthly
sales

Univariate Single variable in dataset Height of students

Multivariate Two or more variables Age, salary, and height

together

Types of Variables
Variable Type Description Example

Nominal Categories with no order Gender (Male/Female),

Colours

Ordinal Categories with order; unequal intervals Education level, Ratings 1-5

Discrete (Numerical) Countable integer values Number of students

Continuous (Numerical) Any value in an interval Height, Weight, Temperature

Interval Meaningful intervals; no true zero Temperature in Celsius, IQ

score

Ratio Meaningful intervals plus true zero Height, Weight, Age

Note: Numerical data uses Interval and Ratio types. Categorical data uses Nominal and Ordinal types.

03 Population & Sampling

Population Sample

The entire group being studied. e.g. All people in A representative subset of the population.
India Recommended: n > 30 for reliable inference.

Sampling Techniques
Technique How it works Key distinction

Random Every person has equal chance — no system Simplest method

Stratified Divide into sub-groups (strata), then randomly pick Picks individuals from
individuals from each every group
 Technique How it works Key distinction

Systematic Select every k-th element (e.g. every 10th person) Follows a fixed interval rule

Cluster Divide into clusters, then randomly select entire clusters Picks whole groups, not
individuals

Note: Stratified vs Cluster: Stratified selects individuals from each group; Cluster selects entire groups at
random.

04 Descriptive Statistics — Central Tendency

Measure Definition Formula / Notes

Mean Sum of all values divided by count. Sensitive to x-bar = Sum(xi) / n

outliers.

Median Middle value when sorted. For even n: average of Less affected by outliers — use
two middle values. when outliers exist

Mode Most frequently occurring value. Best for categorical variables.

Can be none, one, or multiple.

When to Use Which Measure

Situation Recommended Measure

Numerical data, no outliers Mean

Numerical data with outliers or skewed distribution Median

Categorical data Mode

Frequency Concepts
Term Definition

Frequency Number of times a value appears in the dataset

Relative Frequency Frequency divided by total observations (gives proportion or percentage)

Cumulative Frequency Running total of frequencies — always non-decreasing

05 Measures of Dispersion
Describe the spread or variability of data around the central tendency.
Measure Definition Formula

Range Difference between max and min. Simple but Max - Min
sensitive to outliers.

IQR Spread of the middle 50% of data. Less sensitive Q3 - Q1

to outliers.
 Measure Definition Formula

Variance Average squared deviation from the mean. Units o^2 = Sum(xi - u)^2 / n
are squared.

Standard Deviation Square root of variance. Same unit as original o = sqrt(Variance)

data.

Quartiles & Percentiles

Quartile Percentile Formula (gives term position)

Q1 (lower quartile) 25th percentile (n+1) / 4

Q2 (median) 50th percentile (n+1) / 2

Q3 (upper quartile) 75th percentile 3(n+1) / 4

Any percentile P -- (P/100) x (n+1)

06 Graphical Representation & Outliers

Histogram
Displays distribution of continuous data using bins. Y-axis = frequency or density.
Type Shape Mean vs Median

Symmetric (Normal) Bell-shaped, balanced on both sides Mean = Median = Mode

Right Skewed (+) Long tail on the right; outliers on right Mean > Median

Left Skewed (-) Long tail on the left; outliers on left Mean < Median

Mode-based type Description

Unimodal One peak in the distribution

Bimodal Two peaks

Multimodal Many peaks — no single dominant mode

Box Plot (Whisker Plot) — 5-Number Summary

Element Value / Formula

Minimum Lowest non-outlier value

Q1 (lower quartile) 25th percentile

Median (Q2) 50th percentile — line inside the box

Q3 (upper quartile) 75th percentile

Maximum Highest non-outlier value

Upper whisker Q3 + 1.5 x IQR

 Element Value / Formula

Lower whisker Q1 - 1.5 x IQR

Outliers Values beyond either whisker — plotted as individual points

Scatter Plot
Visualizes the relationship between two continuous variables. Identifies positive, negative, or no
correlation, and is also useful for detecting outliers.
Pattern Meaning

Points trend upward (left to right) Positive correlation — both variables increase together

Points trend downward (left to right) Negative correlation — one increases as the other decreases

No visible pattern No correlation — variables are unrelated

Outlier Detection Methods

Method Rule

IQR Method Outlier if value < Q1 - 1.5xIQR or value > Q3 + 1.5xIQR

Z-score Method Outlier if |z| > 3 (beyond 3 standard deviations from mean)

07 Correlation & Covariance

Covariance Correlation Coefficient (r)

Shows direction of relationship (positive/negative). Shows both direction and strength. Range: -1 to +1.
Does not indicate strength.
Formula: r = Cov(x,y) / (sigma-x x sigma-y)
Formula: Cov(x,y) = Sum[(xi - x-bar)(yi - y-bar)] / n

Correlation Value Interpretation

r close to +1.0 Strong positive correlation

r close to -1.0 Strong negative correlation

r close to 0 No or very weak correlation

+/- 0.7 to +/- 1.0 Strong correlation

-0.5 to +0.5 Weak correlation

Note: Correlation does not imply Causation. Two variables can be correlated without one causing the
other.
08 Probability
Probability is the measure of the likelihood of an event occurring. Range: 0 (impossible) to 1 (certain).
P(A) = Favorable outcomes / Total outcomes

Key Concepts
Concept Definition / Formula

Sample Space (S) Set of all possible outcomes of a random experiment

Event A subset of the sample space

Complement (A') All outcomes NOT in A. P(A') = 1 - P(A)

Conditional P(A|B) Probability of A given B has occurred. P(A intersect B) / P(B)

Types of Events
Type Description P(A and B)

Joint / Non-disjoint Have common outcomes; can occur together P(A) x P(B)

Disjoint / Mutually exclusive No common outcomes; cannot occur 0

simultaneously

Independent Occurrence of one does not affect the other P(A) x P(B)

Dependent Occurrence of one affects the other P(A) x P(B|A)

Bayes' Theorem
Updates the probability of an event based on new evidence. Widely used in machine learning, spam
filters, medical diagnosis, and recommendation systems.
P(A|B) = P(B|A) x P(A) / P(B)

09 Probability Distributions
Discrete Random Variable Continuous Random Variable

Finite or countable values. Use PMF (Probability Infinite values over an interval. Use PDF (Probability
Mass Function). Represented as bar charts. Density Function). Represented as smooth curves.

Discrete Distributions
Distribution Description Mean / Variance

Bernoulli Single trial; binary outcome (0 = failure, 1 = success). Mean = p, Var =

p(1-p)

Binomial n independent Bernoulli trials. If n=1, reduces to Bernoulli Mean = np, Var =
distribution. np(1-p)
Continuous Distributions
Distribution Description Key Formula

Uniform All values equally likely over interval [a, b]. f(x) = 1/(b-a), Mean =
(a+b)/2

Normal (Gaussian) Bell-shaped, symmetric. Most common in nature. See Section 10 for full
Defined by mean and standard deviation. detail

10 Normal Distribution, Standardization & Normalization

Standard Normal Distribution (Z-Distribution)
Special case where mean = 0 and standard deviation = 1. Convert any value to a Z-score to compare
across different distributions.
z = (x - mean) / standard deviation
Z-score Meaning

z>0 Value is above the mean

z<0 Value is below the mean

z=0 Value equals the mean

Empirical Rule (68-95-99.7 Rule)

Applies only to normal (bell-shaped) distributions:
Range from mean Data Coverage

+/- 1 standard deviation 68% of all data

+/- 2 standard deviations 95% of all data

+/- 3 standard deviations 99.7% of all data

Note: Values beyond +/- 3 standard deviations (approximately 0.3% of data) are treated as outliers.

Standardization vs Normalization
Standardization (Z-score) Normalization (Min-Max Scaling)

Formula z = (x - mean) / std dev x_new = (x - x_min) / (x_max -

x_min)

Output range Unbounded (no fixed range) Always 0 to 1

Outlier sensitivity Less sensitive (uses mean and std More sensitive (uses min and max)
dev)

Use when Distribution is known to be normal Features are on different scales

Preserves distribution? Yes Not necessarily

11 Inferential Statistics & Estimation
Uses sample data to draw conclusions and make predictions about a larger population.

Population Parameters vs Sample Statistics

Measure Population symbol Sample symbol

Mean mu x-bar

Standard deviation sigma s

Size N (capital) n (lowercase)

Proportion P p-hat

Correlation coefficient rho r

Types of Estimation
Point Estimation Interval Estimation

Single best-guess value for a population parameter. A range of values likely to contain the true
Simple but provides no reliability information. parameter. More accurate and reliable than point
estimates.

Confidence Interval
Expresses how confident we are (e.g. 95%) that the true population parameter lies within the estimated
interval.
CI = Point Estimate +/- Margin of Error
Margin of Error = Critical Value x (sigma / sqrt(n))

Confidence Level Alpha (significance) Critical Value (z)

90% 0.10 1.645

95% 0.05 1.960

99% 0.01 2.576

Note: Use Z-distribution when n > 30 and population std dev is known. Use T-distribution when n <= 30 or
std dev is unknown. Degrees of freedom = n - 1.

12 Hypothesis Testing
A method to evaluate a claim about a population parameter using sample data as evidence.
Null Hypothesis (H0) Alternate Hypothesis (H1)

Default/baseline statement to test. e.g. 'No Opposite of H0. Accepted when H0 is rejected. e.g.
significant difference exists.' Symbol: H0 'A significant difference exists.' Symbol: H1
Decision Rule
Condition Decision

p-value < alpha (significance level) Reject H0 — Accept H1

p-value >= alpha Fail to reject H0

Common alpha values 0.05 (most fields), 0.01 (medical or critical

decisions)

Types of Errors
Error Type Description Also called

Type I Error (alpha) Reject H0 when it is actually TRUE False Positive

Type II Error (beta) Accept H0 when it is actually FALSE False Negative

Tail Tests
Test type Description When to use

One-tailed Alpha is entirely in one tail (left or right) Testing strictly greater than
OR less than a value

Two-tailed Alpha is split as alpha/2 in both tails Testing for any difference in
either direction

Statistical Tests — Quick Reference

Test When to use Sample size

Z-test (one or two sample) Compare sample mean to population mean; population std n > 30
dev known

Independent T-test Compare means of two independent groups; std dev n <= 30
unknown

Paired T-test Compare same group before and after a condition n <= 30

ANOVA / F-test Compare means of 3 or more groups simultaneously Any

Chi-square (independence) Test association between two categorical variables Any

Chi-square (goodness of fit) Test if observed frequencies match an expected Any

distribution

ANOVA Types
Type Description Example

One-Way ANOVA One factor variable; one response variable 3 teaching methods -->
exam scores

Two-Way ANOVA Two factor variables; one response variable Fertilizer + Planting density
--> crop yield
Note: ANOVA null hypothesis: all group means are equal (H0: u1=u2=u3). Alternate: at least one mean is
significantly different.

End of Notes — Statistics for Data Analysis & Data Science