UGC NET

STATISTICS
Comprehensive Study Notes
From Basics to Standardised Forms | With Formulae & Examples

UGC NET Statistics Notes | Page 1 of 21

UNIT 1: Measures of Central Tendency, Dispersion and
Inequalities
1.1 Introduction
Statistics is the science of collecting, organising, analysing, interpreting, and presenting data.
Two fundamental aspects of describing a dataset are its centre (Central Tendency) and its spread
(Dispersion).

1.2 Measures of Central Tendency

A measure of central tendency is a single value that attempts to describe a dataset by identifying
the central position within that dataset.

A. Arithmetic Mean (AM)

The arithmetic mean is the sum of all observations divided by the number of observations. It is
the most commonly used measure of central tendency.
Simple Mean: AM = (x1 + x2 + ... + xn) / n = Sum(xi) / n

Weighted Mean: Weighted Mean = Sum(wi * xi) / Sum(wi)

Grouped Mean: For grouped data: AM = Sum(fi * mi) / Sum(fi)

Where fi = frequency, mi = midpoint of class interval.

Example: Arithmetic Mean
Data: 10, 20, 30, 40, 50. AM = (10+20+30+40+50)/5 = 150/5 = 30. For weighted mean: if
marks in 3 subjects are 70, 80, 90 with weights 2, 3, 5 respectively: Weighted Mean = (2x70
+ 3x80 + 5x90)/(2+3+5) = (140+240+450)/10 = 830/10 = 83.

B. Median
The median is the middle value of an ordered dataset. It divides the distribution into two equal
halves.
• For odd n: Median = value of ((n+1)/2)th observation
• For even n: Median = average of (n/2)th and ((n/2)+1)th observations
Grouped Median: Median (grouped) = L + [(n/2 - cf) / f] x h

Where L = lower boundary of median class, n = total frequency, cf = cumulative frequency before
median class, f = frequency of median class, h = class width.
Example: Median
Data (sorted): 4, 7, 10, 15, 22. n=5 (odd). Median = 3rd value = 10. For even: 4, 7, 10, 15 ->
Median = (7+10)/2 = 8.5

C. Mode
Mode is the value that occurs most frequently in a dataset.
Grouped Mode: Mode (grouped) = L + [(f1 - f0) / (2f1 - f0 - f2)] x
h

UGC NET Statistics Notes | Page 2 of 21

Where L = lower limit of modal class, f1 = frequency of modal class, f0 = frequency of class
before modal class, f2 = frequency of class after modal class, h = class width.
Example: Mode
Data: 2, 4, 4, 5, 7, 7, 7, 9. Mode = 7 (appears 3 times). A bimodal dataset has two modes;
multimodal has more than two.

D. Geometric Mean (GM)

Geometric Mean: GM = (x1 * x2 * x3 * ... * xn)^(1/n)

Log form of GM: log(GM) = [1/n] * Sum[log(xi)]

GM is used for averaging ratios, rates of growth, and index numbers. It is always <= AM.
Example: Geometric Mean
Data: 2, 8, 32. GM = (2x8x32)^(1/3) = (512)^(1/3) = 8.

E. Harmonic Mean (HM)

Harmonic Mean: HM = n / Sum(1/xi)

HM is used for averaging rates and speeds. Relationship: HM <= GM <= AM.
Example: Harmonic Mean
Speeds: 60 km/h and 40 km/h. HM = 2/(1/60 + 1/40) = 2/(0.0167+0.025) = 2/0.0417 = 48
km/h. This is the correct average speed for equal distances.

1.3 Applications of Measures of Central Tendency

Measure Applications
Arithmetic Mean Average income, average test scores, economic data, any ratio-
level data
Median Income inequality analysis, house prices (skewed data), education
levels
Mode Most popular product, most common disease, market research
Geometric Mean Population growth rates, compound interest, index number
construction
Harmonic Mean Average speed/rate, price-earnings ratios, electrical resistance in
parallel

UGC NET Statistics Notes | Page 3 of 21

1.4 Measures of Dispersion
Measures of dispersion describe the spread or variability of data around the central value. Two
datasets can have the same mean but different dispersions.

A. Range
Range: Range = Maximum Value - Minimum Value

Simple but affected by extreme values. Coefficient of Range = (Max - Min)/(Max + Min).

B. Mean Deviation (MD)

Mean Deviation: MD (about mean) = Sum|xi - x_bar| / n

Relative MD: Coefficient of MD = MD / Mean

Example: Mean Deviation

Data: 2, 4, 6, 8, 10. Mean=6. Deviations: |2-6|=4, |4-6|=2, |6-6|=0, |8-6|=2, |10-6|=4. MD =
(4+2+0+2+4)/5 = 12/5 = 2.4.

C. Variance and Standard Deviation

Variance is the average of squared deviations from the mean. Standard deviation is its square
root.
Population Variance: Population Variance (sigma^2) = Sum(xi - mu)^2 /
N

Sample Variance: Sample Variance (s^2) = Sum(xi - x_bar)^2 / (n-1)

Standard Deviation: Standard Deviation (s) = sqrt[Sum(xi -

x_bar)^2 / (n-1)]

CV: Coefficient of Variation (CV) = (s / x_bar) x 100%

CV is used to compare variability of two datasets with different units or magnitudes. Higher CV
means greater relative variability.
Example: Variance & SD
Data: 2, 4, 6, 8, 10. Mean=6. Squared deviations: 16, 4, 0, 4, 16. Variance =
(16+4+0+4+16)/(5-1) = 40/4 = 10. SD = sqrt(10) = 3.16. CV = (3.16/6) x 100 = 52.7%.

D. Quartile Deviation (Semi-Interquartile Range)

Quartile Deviation: QD = (Q3 - Q1) / 2

Coefficient of QD: Coefficient of QD = (Q3 - Q1) / (Q3 + Q1)

Q1 = 25th percentile, Q3 = 75th percentile. IQR = Q3 - Q1 contains the middle 50% of data.

UGC NET Statistics Notes | Page 4 of 21

1.5 Inequalities
A. Chebyshev's Inequality
Chebyshev's inequality gives a lower bound on the probability that a random variable falls within k
standard deviations of the mean, for any distribution.
Chebyshev's Inequality: P(|X - mu| < k*sigma) >= 1 - 1/k^2

Example: Chebyshev's Inequality

For k=2: At least 1 - 1/4 = 75% of data lies within 2 SDs of mean. For k=3: At least 1 - 1/9 =
88.9% of data lies within 3 SDs. This applies to ANY distribution.
Note: For a normal distribution, 68% lies within 1 SD, 95% within 2 SDs, and 99.7% within 3
SDs - the empirical rule.

B. Lorenz Curve and Gini Coefficient

The Lorenz curve shows the degree of inequality in a frequency distribution (e.g., income,
wealth). The x-axis is cumulative % of population; y-axis is cumulative % of income.
Gini Coefficient: Gini Coefficient = Area between Lorenz curve and
line of equality / Total area under diagonal

Gini = 0 means perfect equality; Gini = 1 means perfect inequality. India's Gini coefficient for
income is approximately 0.35-0.38.

UGC NET Statistics Notes | Page 5 of 21

UNIT 2: Sampling, Sampling Procedure and Hypothesis
Testing
2.1 Basic Concepts of Sampling
Term Definition
Population (Universe) The complete set of all items or individuals of interest in a study
Sample A subset of the population selected for study
Sampling Frame A list of all elements in the population from which the sample is
drawn
Parameter A numerical measure that describes a characteristic of a
POPULATION (e.g., mu, sigma)
Statistic A numerical measure that describes a characteristic of a SAMPLE
(e.g., x-bar, s)
Sampling Error Difference between sample statistic and population parameter due
to chance
Non-Sampling Error Errors from data collection, measurement, or processing; not
reduced by larger samples

2.2 Sampling Methods

A. Probability Sampling
• Simple Random Sampling (SRS): Every unit has an equal chance of selection. Done
by lottery or random number tables. Suitable for homogeneous populations.
- With replacement (SRSWR): Unit can be selected more than once
- Without replacement (SRSWOR): Unit selected only once
• Systematic Sampling: Select every kth element after a random start. k = N/n
(sampling interval). Easy to implement; may have periodicity bias.
Example: Systematic Sampling
Population N=500, Sample n=50. k = 500/50 = 10. Random start = 4. Sample: 4, 14, 24,
34, ... 494.
• Stratified Sampling: Population is divided into homogeneous subgroups (strata)
and samples are drawn from each stratum.
- Proportionate: ni = n x (Ni/N). Better representation.
- Disproportionate: ni differs from proportion; used when variability differs across strata.
• Cluster Sampling: Population is divided into clusters (usually geographic). All units
in selected clusters are included. Cheaper but less precise than SRS.
• Multi-Stage Sampling: Sampling in multiple stages - first select clusters, then sub-
units within clusters. Used in large national surveys.
• Multiphase Sampling: Information collected in phases - basic info from all, detailed
from sub-sample.

B. Non-Probability Sampling
• Convenience Sampling: Select the most easily accessible units

UGC NET Statistics Notes | Page 6 of 21

 • Judgment / Purposive Sampling: Expert judgment used to select units
• Quota Sampling: Pre-specified quotas for different subgroups
• Snowball Sampling: Existing subjects recruit future subjects; used for rare populations

2.3 Sampling Distribution

The sampling distribution is the probability distribution of a statistic (like the sample mean)
computed from all possible samples of size n drawn from a population.
Standard Error of Mean: SE of mean = sigma / sqrt(n)

SE of Proportion: For proportion: SE(p) = sqrt[P(1-P)/n]

Note: The Central Limit Theorem (CLT) states: For large n (>=30), the sampling distribution
of the sample mean is approximately normal, regardless of the shape of the population
distribution.

UGC NET Statistics Notes | Page 7 of 21

2.4 Hypothesis Testing - Framework
Hypothesis testing is a formal procedure for deciding whether a hypothesis about a population
parameter is supported by sample data.

Key Terms
Term Explanation
Null Hypothesis (H0) The hypothesis of no effect or no difference; assumed true until
evidence against it
Alternative Hypothesis (H1 or The hypothesis we seek evidence for; opposite of H0
Ha)
Level of Significance (alpha) Probability of Type I error; typically 0.05 or 0.01
p-value Probability of observing results as extreme as the sample,
assuming H0 is true
Type I Error (alpha) Rejecting H0 when it is true (False Positive)
Type II Error (beta) Failing to reject H0 when it is false (False Negative)
Power of Test (1-beta) Probability of correctly rejecting a false H0
Critical Region Set of values of the test statistic that leads to rejection of H0

Decision Rule: If p-value < alpha, reject H0. If calculated test statistic > critical value, reject H0.

2.5 Chi-Square Test (χ²)

The chi-square test is used for categorical data to test whether observed frequencies differ from
expected frequencies, or whether two categorical variables are independent.

A. Chi-Square Test of Independence (Goodness of Fit / Association)

Chi-Square Statistic: chi^2 = Sum [(O - E)^2 / E]

Where O = Observed frequency, E = Expected frequency.

Expected Frequency: E(ij) = (Row Total_i x Column Total_j) / Grand
Total

df for Chi-Square: Degrees of Freedom (df) = (r-1)(c-1)

Where r = number of rows, c = number of columns.

Conditions: n >= 50, no expected frequency < 5, random sample, nominal/categorical data.
Example: Chi-Square Test
Survey: 100 people on preference for Tea/Coffee by gender. Observed table: Male(Tea=30,
Coffee=20), Female(Tea=25, Coffee=25). Grand total=100. Expected for Male-Tea =
(50x55)/100 = 27.5. Calculate chi^2 = sum[(O-E)^2/E]. df=(2-1)(2-1)=1. Compare with critical
value at alpha=0.05 (3.841). If chi^2 > 3.841, reject H0 of independence.

B. Yates' Correction for Continuity (2x2 tables)

Yates' Correction: chi^2 = Sum [(|O - E| - 0.5)^2 / E]

Applied when df=1 and any expected frequency is between 5 and 10.

UGC NET Statistics Notes | Page 8 of 21

2.6 t-Test (Student's t-test)
The t-test compares means when the population standard deviation is unknown and sample size
is small (n<30). Assumes normal distribution and uses t-distribution.

A. One-Sample t-test
One-Sample t: t = (x_bar - mu0) / (s / sqrt(n))

Tests if the sample mean differs from a known/hypothesized population mean mu0. df = n-1.
Example: One-Sample t-test
H0: mu=70. Sample: n=16, x-bar=74, s=8. t = (74-70)/(8/sqrt(16)) = 4/2 = 2.0. df=15. t-critical
at alpha=0.05, two-tail = 2.131. Since |2.0| < 2.131, fail to reject H0.

B. Independent Two-Sample t-test

Two-Sample t: t = (x1_bar - x2_bar) / [sp * sqrt(1/n1 + 1/n2)]

Pooled SD: Pooled variance: sp^2 = [(n1-1)s1^2 + (n2-1)s2^2] /

(n1+n2-2)

df = n1 + n2 - 2. Tests if two independent group means are equal.

C. Paired t-test (Dependent Samples)

Paired t: t = d_bar / (sd / sqrt(n))

Where d_bar = mean of differences, sd = standard deviation of differences, n = number of pairs.

df = n-1. Used for before-after studies, matched pairs.
Example: Paired t-test
Pre-test scores: 60,65,70,55,80. Post-test: 70,72,75,65,85. Differences: 10,7,5,10,5. d_bar =
7.4. sd = 2.41. t = 7.4/(2.41/sqrt(5)) = 7.4/1.078 = 6.86. df=4. t-critical (alpha=0.05) = 2.776.
Since 6.86 > 2.776, reject H0 - training was effective.

UGC NET Statistics Notes | Page 9 of 21

2.7 Analysis of Variance (ANOVA)
ANOVA tests whether means of three or more groups are equal. It partitions the total variation in
data into components due to different sources.

A. One-Way ANOVA
Tests if there is any statistically significant difference between the means of three or more
independent groups (one factor).
H0: mu1 = mu2 = ... = muk (all group means are equal)
H1: At least one group mean is different
F-statistic (One-Way ANOVA): F = MSB / MSW = (SSB/(k-1)) / (SSW/(N-
k))

SST = SSB + SSW (Total SS = Between SS + Within SS)

Where SSB = Sum of Squares Between groups, SSW = Sum of Squares Within groups, k =
number of groups, N = total observations.
Source SS df MS F
Between Groups SSB k-1 MSB=SSB/(k- MSB/MSW
1)
Within Groups (Error) SSW N-k MSW=SSW/ -
(N-k)
Total SST N-1 - -

Example: One-Way ANOVA

Three fertilizers (A, B, C) tested on crop yield (5 plots each). If F-calculated > F-critical (from
F-table with df1=2, df2=12 at alpha=0.05), we conclude at least one fertilizer gives a different
mean yield. Post-hoc tests (Tukey, LSD) then identify which pairs differ.

B. Two-Way ANOVA
Tests the effect of two factors simultaneously and their interaction.
Two-Way ANOVA F-ratios: F_A = MSA/MSE, F_B = MSB/MSE, F_AB =
MSAB/MSE

SST = SSA + SSB + SSAB + SSE. This tests main effects of Factor A, Factor B, and their
interaction.
Note: ANOVA Assumptions: (1) Observations are independent; (2) Data is normally
distributed within groups; (3) Homogeneity of variances (Levene's or Bartlett's test). Post-hoc
tests needed if H0 is rejected.

UGC NET Statistics Notes | Page 10 of 21

UNIT 3: Time Series Analysis
3.1 Introduction to Time Series
A time series is a set of observations on a variable measured at successive points in time or over
successive periods of time (e.g., monthly rainfall, annual GDP, daily stock prices).

3.2 Components of Time Series

Component Description & Example
Secular Trend (T) Long-term increase or decrease over a long period. E.g., rising
GDP over 20 years, increasing population.
Seasonal Variation (S) Regular fluctuations within a year due to seasons, customs,
climate. E.g., higher ice cream sales in summer.
Cyclical Variation (C) Medium-term fluctuations over several years linked to business
cycles (expansion, peak, contraction, trough). Duration: 2-10 years.
Irregular/Random (I) Unpredictable, erratic short-term fluctuations due to floods, strikes,
pandemics, etc.

3.3 Models of Time Series

Additive Model: Additive Model: Y = T + S + C + I

Multiplicative Model: Multiplicative Model: Y = T x S x C x I

The additive model is used when seasonal variation is constant in absolute terms. The
multiplicative model is used when seasonal variation is proportional to the trend.

3.4 Measurement of Trend

A. Moving Averages Method
A moving average smooths out short-term fluctuations to reveal the underlying trend. For an odd
period (e.g., 3-year), center is clear. For even period (e.g., 4-quarter), a second 2-point moving
average is needed.
Moving Average: 3-Year Moving Average: MA(t) = [Y(t-1) + Y(t) +
Y(t+1)] / 3

Example: Moving Average

Annual data: 2019=50, 2020=55, 2021=60, 2022=70, 2023=65. 3-year MA for 2021 =
(55+60+70)/3 = 61.67.

B. Method of Least Squares

Fits a mathematical equation (linear or non-linear) to the time series data to obtain a trend line.
Most objective method.
Linear Trend Equation: Linear Trend: Y_hat = a + bX

Slope (b): b = [n*Sum(XY) - Sum(X)*Sum(Y)] / [n*Sum(X^2) -

(Sum(X))^2]

UGC NET Statistics Notes | Page 11 of 21

 Intercept (a): a = (Sum(Y) - b*Sum(X)) / n

If X is coded as 0 for middle year (or Sum(X)=0), then: b = Sum(XY)/Sum(X^2) and a = Sum(Y)/n.
Example: Least Squares Trend
Sales (in lakhs): 2019=10, 2020=12, 2021=14, 2022=16, 2023=18. Coded X: -2,-1,0,1,2.
Sum(XY) = (-2x10)+(-1x12)+(0x14)+(1x16)+(2x18) = -20-12+0+16+36=20. Sum(X^2)=10.
b=20/10=2. a=70/5=14. Trend: Y=14+2X. Forecast for 2024 (X=3): Y=14+6=20 lakhs.

C. Measurement of Seasonal Variation

Seasonal Indices measure the amount of seasonal variation in a time series, expressed as a ratio
or percentage.
Seasonal Index: Seasonal Index = (Average for that period / Overall
average) x 100

Methods: Simple Average Method, Ratio-to-Moving Average Method (most common), Ratio-to-
Trend Method, Link Relative Method.

UGC NET Statistics Notes | Page 12 of 21

UNIT 4: Correlation and Regression Analysis
4.1 Correlation Analysis
Correlation measures the strength and direction of the linear relationship between two variables.

A. Karl Pearson's Coefficient of Correlation (r)

Pearson's r: r = Sum[(xi - x_bar)(yi - y_bar)] / sqrt{Sum(xi-
x_bar)^2 * Sum(yi-y_bar)^2}

Computational Formula: r = [n*Sum(XY) - Sum(X)*Sum(Y)] /

sqrt{[n*Sum(X^2)-(Sum(X))^2][n*Sum(Y^2)-(Sum(Y))^2]}

Range: -1 <= r <= +1. r=+1: perfect positive, r=-1: perfect negative, r=0: no linear correlation.
Example: Pearson's r
X: Height(cm): 160, 165, 170, 175, 180. Y: Weight(kg): 55, 60, 65, 70, 75. There is a strong
positive correlation (r close to +1) as both increase together.

B. Spearman's Rank Correlation (rs)

Used when data is ordinal or when assumptions of Pearson's r are not met.
Spearman's rs: rs = 1 - [6*Sum(d^2)] / [n(n^2-1)]

Where d = difference in ranks of corresponding pairs, n = number of pairs.

Example: Spearman's Rank
Two judges rank 5 films: Judge1: 1,2,3,4,5. Judge2: 2,1,4,3,5. d values: -1,1,-1,1,0. d^2:
1,1,1,1,0. Sum(d^2)=4. rs = 1-(6x4)/(5x24) = 1-24/120 = 1-0.2 = 0.80 (strong positive
agreement).

C. Properties and Interpretation of r

• r is dimensionless (unit-free)
• r is symmetric: r(X,Y) = r(Y,X)
• r measures only LINEAR relationship
• |r| close to 1: strong; close to 0: weak; 0.5-0.7 moderate
• r^2 (coefficient of determination): % of variation in Y explained by X
Note: Correlation does not imply causation. Two variables can be correlated due to a third
confounding variable.

4.2 Regression Analysis

Regression analysis determines the mathematical equation relating a dependent variable (Y) to
one or more independent variables (X). It is used for prediction.

A. Simple Linear Regression

Regression Equation: Y_hat = a + bX (Regression of Y on X)

Regression Coefficient (byx): b_yx = r * (sy/sx) [Regression

coefficient of Y on X]

UGC NET Statistics Notes | Page 13 of 21

 Regression Coefficient (bxy): b_xy = r * (sx/sy) [Regression
coefficient of X on Y]

Where sx, sy = standard deviations of X and Y respectively, r = correlation coefficient.

The regression line always passes through the point (x-bar, y-bar).
Relationship between r and b: r = sqrt(b_yx * b_xy)

B. Least Squares Estimates

Slope in deviations: b = Sum(xy) / Sum(x^2) where x = X - X_bar, y =
Y - Y_bar

Intercept: a = Y_bar - b * X_bar

C. Standard Error of Estimate

Standard Error of Estimate: Se = sqrt[Sum(Y - Y_hat)^2 / (n-2)]

Se measures the accuracy of predictions. Smaller Se = better fit.

D. Multiple Regression
Multiple Regression: Y_hat = a + b1*X1 + b2*X2 + ... + bk*Xk

R-squared: R^2 = 1 - [SSE/SST] (Multiple Coefficient of

Determination)

Adjusted R^2 = 1 - [(1-R^2)(n-1)/(n-k-1)] accounts for number of predictors.

Example: Regression
Predict house price (Y) from area (X1) and age (X2). Y_hat = 50000 + 2500*X1 - 1000*X2.
For a 100 sq.m, 5-year old house: Y_hat = 50000 + 250000 - 5000 = Rs 2,95,000.

UGC NET Statistics Notes | Page 14 of 21

UNIT 5: Measurement of Indices, Scale-Free Indicators and
Composite Index
5.1 Index Numbers
An index number is a statistical measure designed to show changes in a variable or a group of
related variables with respect to time, geographic location, or other characteristics. It is always
relative to a base (base = 100).

5.2 Types of Price Indices

A. Simple (Unweighted) Index Numbers
Price Relative: Simple Price Relative = (P1/P0) x 100

Aggregate Price Index: Simple Aggregate Index = [Sum(P1) / Sum(P0)] x

100

B. Weighted Index Numbers

Laspeyre's Price Index: Laspeyre's Index = [Sum(P1*Q0) / Sum(P0*Q0)] x
100

Paasche's Price Index: Paasche's Index = [Sum(P1*Q1) / Sum(P0*Q1)] x

100

Fisher's Ideal Index: Fisher's Ideal Index = sqrt(Laspeyre's x

Paasche's)

Laspeyre's uses base period quantities (Q0), Paasche's uses current period quantities (Q1).
Fisher's is the geometric mean of both and satisfies the Time Reversal and Factor Reversal tests.
Example: Laspeyre's Index
Commodity A: P0=10, P1=12, Q0=50. Commodity B: P0=20, P1=25, Q0=30. Sum(P1*Q0) =
12x50 + 25x30 = 600+750 = 1350. Sum(P0*Q0) = 10x50 + 20x30 = 500+600 = 1100.
Laspeyre's = (1350/1100)x100 = 122.7.

C. Tests for Index Numbers

Test Description
Time Reversal Test P01 x P10 = 1. Index for period 1 to 0 times index 0 to 1 should
equal 1. Fisher's satisfies this.
Factor Reversal Test P01 x Q01 = V01 (Value Index). Fisher's satisfies this.
Circular Test P01 x P12 x P20 = 1. Useful for chain indices.
Unit Test Index should be independent of units of measurement.

5.3 Consumer Price Index (CPI)

CPI (Weighted): CPI = [Sum(P1*W) / Sum(P0*W)] x 100

UGC NET Statistics Notes | Page 15 of 21

CPI measures changes in the price level of a representative basket of consumer goods and
services. Used to calculate real wages, inflation rate, and deflate monetary values.
Inflation Rate: Inflation Rate = [(CPI_current - CPI_previous) /
CPI_previous] x 100%

Real Wage: Real Wage = (Nominal Wage / CPI) x 100

5.4 Making Indicators Scale-Free (Normalisation)

When combining indicators measured on different scales, we must first make them scale-free
(dimensionless). This is called normalisation or standardisation.

A. Min-Max Normalisation
Min-Max Normalisation: X_norm = (X - X_min) / (X_max - X_min)

Scales all values to the range [0, 1]. Used in Human Development Index (HDI) construction.

B. Z-Score Standardisation
Z-Score: Z = (X - mu) / sigma

Transforms data to have mean=0 and SD=1. Z-scores are comparable across different scales.

C. Ratio to Reference
Ratio to Reference: Index_i = (X_i / X_reference) x 100

Expresses each value as a percentage of a reference value (e.g., national average).

5.5 Computation of Composite Index

A composite index combines multiple indicators into a single summary measure.

Steps to Construct a Composite Index:

• Step 1: Select relevant indicators based on conceptual framework
• Step 2: Normalise each indicator to make them scale-free (Min-Max, Z-score, etc.)
• Step 3: Assign weights to each indicator (equal weights, PCA-based weights, expert
weights)
• Step 4: Aggregate: Composite Index = Sum(wi * Normalised Indicator_i)
• Step 5: Validate and perform sensitivity analysis

Composite Index: CI = Sum(wi * Ii) where Sum(wi) = 1

Example: Human Development Index (HDI)

HDI combines 3 normalised dimensions: (1) Health (Life Expectancy Index), (2) Education
(Education Index), (3) Standard of Living (GNI Index). HDI = (LEI + EI + II)^(1/3) - geometric
mean with equal weights. India HDI ~0.633 (2021-22).

UGC NET Statistics Notes | Page 16 of 21

UNIT 6: Principal Component Analysis (PCA)
6.1 Introduction to PCA
PCA is a dimensionality reduction technique that transforms a large set of correlated variables
into a smaller set of uncorrelated variables called Principal Components (PCs), while retaining
most of the original variance.

6.2 Objectives of PCA

• Reduce the number of variables while preserving maximum information
• Remove multicollinearity among variables
• Identify underlying structure or patterns in data
• Create composite scores for further analysis
• Visualise high-dimensional data in 2D or 3D

6.3 Steps in PCA

• Step 1: Standardise the data (Z-score each variable) to remove scale effects
• Step 2: Compute the Covariance Matrix (or Correlation Matrix for standardised data)
• Step 3: Compute Eigenvalues and Eigenvectors of the covariance/correlation matrix
• Step 4: Sort eigenvalues in descending order; corresponding eigenvectors are Principal
Components
• Step 5: Select number of PCs (eigenvalue > 1, or cumulative variance > 70-80%)
• Step 6: Transform original data: PC scores = Z * Eigenvector matrix

6.4 Key Concepts in PCA

Eigenvalues and Eigenvectors
Eigen Equation: C * v = lambda * v

Where C = covariance/correlation matrix, v = eigenvector, lambda = eigenvalue. Each PC

corresponds to one eigenvalue-eigenvector pair.

Variance Explained
Proportion of Variance: Proportion of Variance = lambda_i / Sum(all
lambdas)

Cumulative variance explained by first k PCs = Sum(lambda_1 to lambda_k) / Sum(all lambdas).

Factor Loadings
Loadings are the correlations between original variables and PCs. High loading (positive or
negative) indicates the variable contributes strongly to that PC.

Scree Plot
A plot of eigenvalues (y-axis) vs. PC number (x-axis). The 'elbow' point (where the curve bends
sharply) suggests the number of meaningful PCs to retain.
PC1 Equation: PC1 = a11*Z1 + a12*Z2 + ... + a1p*Zp (Linear
combination)

UGC NET Statistics Notes | Page 17 of 21

Note: PC1 captures the maximum variance. PC2 is orthogonal (uncorrelated) to PC1 and
captures the second-most variance. Each subsequent PC is orthogonal to all previous ones.
Example: PCA Application
5 economic indicators for 20 states: GDP, Per-capita income, Industrial output, Agricultural
output, Employment rate. After PCA, PC1 (which loads high on GDP, income, industrial
output) may represent 'Economic Development' and explain 65% of variance. PC2 may
represent 'Agricultural Profile'. Using just PC1 and PC2 (explaining 85% of variance)
simplifies analysis from 5 to 2 dimensions.

UGC NET Statistics Notes | Page 18 of 21

UNIT 7: Cluster Analysis
7.1 Introduction to Cluster Analysis
Cluster analysis is an unsupervised multivariate technique that groups observations (or variables)
into clusters such that objects within the same cluster are more similar to each other than to
objects in other clusters.

7.2 Types of Clustering Methods

A. Hierarchical Clustering
Builds a tree-like hierarchy of clusters (dendrogram). Can be agglomerative (bottom-up) or
divisive (top-down).
• Agglomerative: Start with each observation as its own cluster, then repeatedly merge the
two most similar clusters.
• Divisive: Start with all observations in one cluster, then repeatedly split.
Linkage Methods for defining distance between clusters:
Linkage Method Description
Single Linkage (Nearest Distance = shortest distance between any two points in the
Neighbour) clusters. Tends to create elongated chains.
Complete Linkage (Farthest Distance = maximum distance between any two points. Creates
Neighbour) compact spherical clusters.
Average Linkage (UPGMA) Distance = average of all pairwise distances. Balance between
single and complete.
Ward's Method Minimises total within-cluster variance. Most widely used; creates
compact equal-sized clusters.
Centroid Method Distance = distance between cluster centroids (means).

B. Non-Hierarchical Clustering (K-Means)

K-Means partitions n observations into k clusters where each observation belongs to the cluster
with the nearest centroid (mean).
• Step 1: Specify k (number of clusters) a priori
• Step 2: Randomly assign k initial centroids
• Step 3: Assign each observation to the nearest centroid (Euclidean distance)
• Step 4: Recalculate centroids as mean of all observations in each cluster
• Step 5: Repeat steps 3-4 until convergence (centroids don't change)

7.3 Distance Measures

Euclidean Distance: Euclidean: d(x,y) = sqrt[Sum(xi - yi)^2]

Manhattan Distance: Manhattan: d(x,y) = Sum|xi - yi|

Mahalanobis Distance: Mahalanobis: d = sqrt[(x-mu)' * C^(-1) * (x-

mu)]

Mahalanobis distance accounts for correlations between variables and is scale-invariant.

UGC NET Statistics Notes | Page 19 of 21

7.4 Determining Optimal Number of Clusters
• Elbow Method: Plot Within-Cluster Sum of Squares (WCSS) vs. k; choose k where the
curve bends (the 'elbow').
• Silhouette Score: Measures how similar each point is to its own cluster vs. other clusters.
Range: -1 to +1; higher is better.
Silhouette Score: Silhouette = (b - a) / max(a, b)

Where a = mean intra-cluster distance, b = mean nearest-cluster distance.

• Dendrogram: For hierarchical clustering, cut the dendrogram at a height that gives the
desired number of clusters.

7.5 Applications of Cluster Analysis

Field Application
Market Segmentation Group customers by purchasing behaviour for targeted
marketing
Development Studies Classify states/districts by socio-economic indicators for policy
making
Biology Classify species by characteristics (taxonomy)
Finance Group stocks by performance patterns
Education Identify student learning profiles for personalised instruction
Public Health Identify disease clusters by region for epidemic surveillance
Example: Cluster Analysis in Social Science
30 districts classified by 5 indicators: literacy rate, per-capita income, infant mortality, access
to healthcare, urbanisation. Ward's hierarchical clustering yields 3 clusters: Cluster 1 (high
development, 8 districts), Cluster 2 (medium development, 12 districts), Cluster 3 (low
development, 10 districts). Policymakers use this to target interventions.

UGC NET Statistics Notes | Page 20 of 21

Quick Reference Summary
Topic Key Formula/Concept UGC NET Focus Area
Mean/Median/Mode AM = Sum(xi)/n; Med = L+(n/2-cf)/f*h Which measure for skewed data,
types of data
SD & CV s = sqrt[Sum(xi-x_bar)^2/(n-1)]; Compare variability, homogeneity
CV=(s/x_bar)*100
Chebyshev P(|X-mu|<k*sigma) >= 1-1/k^2 Distribution-free bounds
Chi-Square chi^2 = Sum[(O-E)^2/E]; df=(r-1)(c-1) Independence, GoF, df calculation
t-test t = (x_bar - mu)/(s/sqrt(n)); paired & 2- Small samples, before-after studies
sample
ANOVA F = MSB/MSW; SST=SSB+SSW 3+ group comparison, F-table use
Time Series Y = T+S+C+I (Additive); Y=TSCx I Components identification, MA
(Multiplicative) method
Correlation r = Sum(xy)/sqrt[Sum(x^2)Sum(y^2)] Range, interpretation, Spearman's
rs
Regression Y_hat = a+bX; b_yx = r*(sy/sx) Prediction, SE of estimate, R^2
Fisher's Index sqrt(Laspeyre's x Paasche's) Time & Factor Reversal test
satisfaction
HDI / Composite Min-Max norm + equal/PCA weights Normalisation steps, aggregation
PCA Eigenvalue > 1; % variance explained Dimensionality reduction, scree plot
Cluster Analysis Euclidean dist; Ward's linkage; K- Linkage methods, elbow method,
Means applications

Best of Luck for UGC NET!

UGC NET Statistics Notes | Page 21 of 21