Easy2Siksha.com
GNDU Question Paper-2021
B.A 5
th
Semester
QUANTITATIVE TECHNIQUES
(Quantitative Technqiues-V)
Time Allowed: Three Hours Max. Marks: 100
Note: Attempt Five questions in all, selecting at least One question from each section. The
Fifth question may be attempted from any section. All questions carry equal marks.
SECTION-A
1. What are the main features of Poisson distribution? Explain by giving suitable
examples.
2. Define Hypothesis in the statistical parlance. What is the difference between NULL and
ALTERNATE Hypothesis ? Write a note on hypothesis testing.
SECTION-B
3. Derive the basic properties of t-distribution.
4. Highlight the characteristic features of Chi-square-distribution. Highlight its use by
giving a suitable example.
SECTION-C
5. What is the difference between Paired t-test and Non-paired t-test? Enlist four
situations (two each) where these can be applied.
Easy2Siksha.com
6. A certain drug was administered to 200 people out of a total 500 included in the sample
to test its efficacy against Dengue. The results are as follows:
Incidence of Dengue
Incidence of Dengue
Total
Yes
No
Drug
50
150
200
No Drug
250
50
300
Total
300
200
500
Can you say that drug is effective in preventing Dengue ?
SECTION-D
7. A manufacturer appoints 3 workers A, B and C and observes their production in terms
of number of units produced with the use of three different machines X. Y. Z. Perform a
Two-way ANOVA on the data given below and interpret your result on average production
status:
Workers
Machines
X
Y
Z
A
16
64
40
B
56
72
56
C
12
56
28
8. Is the Analysis of Variance (ANOVA) technique an extension of the tests used for testing
the difference between two means? Support your agreement/disagreement with the
details. Enlist the assumptions of ANOVA technique for CRD and RBD of experiment.
Easy2Siksha.com
GNDU Answer Paper-2021
B.A 5
th
Semester
QUANTITATIVE TECHNIQUES
(Quantitative Technqiues-V)
Time Allowed: Three Hours Max. Marks: 100
Note: Attempt Five questions in all, selecting at least One question from each section. The
Fifth question may be attempted from any section. All questions carry equal marks.
SECTION-A
1. What are the main features of Poisson distribution? Explain by giving suitable
examples.
Ans: The Curious Case of Mr. Poisson and the “Rare Event Story”
Once upon a time, in a small town full of curious minds, there lived a young mathematician
named Siméon Denis Poisson. He was fascinated by the little surprises that life threw at
people — unexpected phone calls, sudden arrivals of customers at a shop, the number of
raindrops hitting a small patch of ground, or even how many times someone sneezes in an
hour!
Poisson wondered:
“Is there a way to predict how often these rare or random things happen?”
This question gave birth to one of the most useful ideas in statistics — The Poisson
Distribution.
The Beginning: Understanding the Idea
Imagine you are standing in front of your favorite bakery. You love their chocolate cake so
much that you start noticing how many customers arrive every ten minutes. Sometimes 2
come, sometimes 3, sometimes none at all.
Easy2Siksha.com
You think:
“Hmm... customers come randomly, but maybe there’s a pattern hidden in this
randomness!”
This is exactly what the Poisson distribution helps us find — the probability of a certain
number of events happening within a fixed period of time, when these events occur
independently and rarely.
In simpler terms, Poisson distribution tells us things like:
• How many customers might visit the bakery in an hour?
• How many calls might a call center receive per minute?
• How many printing errors might appear on one page of a book?
• How many accidents might happen at a traffic signal in a day?
The Magic Formula
Poisson found a simple formula to calculate such probabilities:
Here’s what this means in everyday language:
• → Probability of exactly x events happening (for example, 3 phone calls in
5 minutes).
• (lambda) → The average number of events in that fixed time or space (for example,
if on average 4 calls come every 5 minutes, then λ = 4).
• → The magical mathematical constant (approximately 2.718).
• → The factorial of x (the product of all numbers from 1 to x).
So Poisson’s formula tells us:
If you know the average number of times something usually happens, you can predict the
probability of any number of occurrences within that same period.
Let’s Bring It to Life — With an Example
Example 1: The Bakery Story
Suppose on average, 3 customers visit a bakery every 10 minutes (so λ = 3).
Now, you want to know — what’s the probability that exactly 2 customers will arrive in the
next 10 minutes?
Easy2Siksha.com
Using the formula:
Step-by-step:
•
•
•
So,
That means there is about a 22.4% chance that exactly two customers will arrive in the next
10 minutes.
Isn’t it amazing that we can calculate the probability of such random events with a simple
formula?
The Main Features (Explained Like a Story)
Now that we understand what Poisson distribution is, let’s explore its main features — but
let’s do it with relatable examples so it feels real.
It Describes Rare Events
Imagine a post office in a small town. Most of the time, it’s quiet, but once in a while, a new
letter or parcel arrives unexpectedly.
Such rare events — like accidents, defects, misprints, or arrivals — are what Poisson
distribution explains beautifully.
So, the first feature is:
Poisson distribution deals with rare and independent events that occur at random over a
fixed time or area.
Easy2Siksha.com
Only One Parameter (λ)
Poisson distribution is a single-parameter distribution.
That parameter is λ (lambda), which represents the mean (average) number of occurrences
of an event in the given interval.
For example, if a hospital receives 5 emergency calls every night on average, then λ = 5.
Everything about the distribution — its shape, spread, and probability — depends on this
single value.
So, the second feature is:
Poisson distribution is controlled by only one parameter λ (mean = variance = λ).
The Mean and Variance are Equal
This is one of the most interesting and unique properties.
In most statistical distributions (like the Normal distribution), the mean and variance are
different.
But in Poisson’s world, they are exactly the same!
So, if the average number of traffic accidents per day (λ) is 4, then both the mean and the
variance are 4.
It’s like saying the “average surprise” and the “spread of surprises” are the same.
Feature 3:
Mean (μ) = Variance (σ²) = λ
Events are Independent
Imagine a telephone exchange where calls come in randomly.
One person calling does not make it more or less likely that another person will call — each
event is independent.
This independence is crucial. If one event affects another, Poisson distribution no longer
applies.
Feature 4:
Each event happens independently of the others.
Easy2Siksha.com
No Two Events Occur at the Exact Same Instant
In real life, two calls can’t come at the exact same millisecond.
Poisson assumes that events occur one at a time — not simultaneously.
Feature 5:
Only one event can occur in a tiny instant; two simultaneous events are practically
impossible.
Applicable for Discrete Events
Poisson distribution is discrete, which means it only counts whole numbers (0, 1, 2, 3…).
You can’t have 2.5 customers arriving or 1.3 phone calls — that makes no sense!
Feature 6:
Poisson distribution applies to countable, whole-number events.
It’s a Limiting Case of the Binomial Distribution
You might remember the Binomial Distribution, which deals with the number of successes
in a fixed number of trials.
Now, when the number of trials (n) becomes very large, and the probability of success (p) is
very small, but their product (np) remains constant — this constant becomes λ.
In that case, the Poisson distribution is born!
So, mathematically:
If and then Binomial Poisson.
Feature 7:
Poisson distribution is a special (limiting) form of the Binomial distribution.
Shape of the Distribution
The shape of a Poisson distribution depends entirely on λ.
• If λ is small (say 1 or 2), the curve is skewed to the right — meaning there are more
chances of getting smaller numbers (like 0, 1, or 2).
Easy2Siksha.com
• As λ increases (say 10 or more), the curve starts looking more symmetrical — almost
like a Normal Distribution.
So, when rare events become less rare, Poisson’s world starts behaving like the real,
balanced world.
Feature 8:
The shape is positively skewed for small λ and nearly symmetric for large λ.
Another Example: The Call Center Story
Let’s make this even more fun.
You are the manager of a small call center. On average, your team receives 6 calls every 10
minutes. You want to know the probability of getting exactly 4 calls in the next 10 minutes.
Using the formula:
Calculating step by step:
•
•
•
So, there’s a 13.39% chance of receiving exactly 4 calls in the next 10 minutes.
Through this, the call center can plan — for example, how many employees should be on
shift at a given time.
Real-Life Applications (Where Poisson Rules the World)
Poisson distribution is everywhere once you start looking for it:
1. Traffic Management – Number of vehicles passing a signal per minute.
2. Telecommunication – Number of incoming calls to a help desk.
3. Natural Events – Number of meteors visible in an hour.
Easy2Siksha.com
4. Finance – Number of defaults on loans in a day.
5. Healthcare – Number of patients arriving at an emergency room.
6. Publishing – Number of misprints or typos in a printed page.
7. Sports – Number of goals scored in a football match.
It’s like a magical lens that allows us to see patterns in the randomness of life.
A Gentle Summary (Like Wrapping Up a Story)
Let’s imagine life as a big canvas of random events — customers walking in, phone calls
ringing, stars falling, or buses arriving.
Poisson Distribution is the mathematician’s paintbrush that helps us capture the probability
of these little surprises.
In short:
Feature
Description
1
Deals with rare, independent events
2
Only one parameter (λ)
3
Mean = Variance = λ
4
Events occur independently
5
No two events occur simultaneously
6
Discrete distribution
7
Limiting form of Binomial Distribution
8
Shape depends on λ (skewed for small λ, symmetric for large λ)
Final Thought
Poisson’s discovery reminds us of something beautiful — even in randomness, there is
rhythm.
Whether it’s raindrops, calls, or footsteps, the universe dances to hidden mathematical
beats.
Through the Poisson distribution, we learn not just to count events, but to understand
uncertainty.
And that’s the true magic of statistics — turning chaos into clarity.
Easy2Siksha.com
2. Define Hypothesis in the statistical parlance. What is the difference between NULL and
ALTERNATE Hypothesis ? Write a note on hypothesis testing.
Ans: Hypothesis, Null vs. Alternate Hypothesis & Hypothesis Testing
A Story of Questions, Evidence, and Decisions
A Fresh Beginning
On a rainy evening, a group of young researchers sat in a café, debating over a simple
question: “Does drinking green tea really improve memory?”
One student said, “Of course it does! My grandmother swears by it.” Another replied,
“That’s just anecdotal. We need proof.”
At that moment, their professor, who happened to walk in, smiled and said: “What you’re
discussing is the very heart of statistics. You’re forming a hypothesis—a statement that can
be tested with data. And to test it, you’ll need to understand the difference between a null
hypothesis and an alternate hypothesis, and the process of hypothesis testing.”
And so begins our journey into one of the most fascinating ideas in statistics: how to move
from beliefs to evidence-based conclusions.
Chapter 1: What Is a Hypothesis in Statistics?
In everyday life, a hypothesis is simply an assumption or guess. But in statistical parlance, a
hypothesis is a specific, testable statement about a population parameter (like a mean,
proportion, or variance).
In simple words: A hypothesis is a claim about reality that we want to test using data.
Examples:
• “The average height of students in this college is 170 cm.”
• “The proportion of voters supporting Candidate A is 60%.”
• “Green tea improves memory scores.”
Each of these is a hypothesis—a statement that can be checked with data.
Chapter 2: Types of Hypotheses
In statistics, we usually deal with two competing hypotheses:
1. Null Hypothesis (H₀)
• Symbol:
Easy2Siksha.com
• Definition: The null hypothesis is a statement of no effect, no difference, or no
relationship.
• It assumes that any observed difference is due to chance.
• It is the hypothesis we try to disprove or reject.
Example:
• “Green tea has no effect on memory.”
• “The average height of students = 170 cm.”
2. Alternate Hypothesis (H₁ or Hₐ)
• Symbol:
or
• Definition: The alternate hypothesis is a statement that contradicts the null
hypothesis.
• It represents the researcher’s claim or expectation.
• If data provides enough evidence, we accept the alternate hypothesis.
Example:
• “Green tea improves memory.”
• “The average height of students ≠ 170 cm.”
Chapter 3: Key Differences Between Null and Alternate Hypothesis
Aspect
Null Hypothesis (H₀)
Alternate Hypothesis (H₁)
Meaning
Assumes no effect or no difference
Assumes there is an effect or difference
Symbol
or
Nature
Conservative, status quo
Researcher’s claim
Example
“The mean = 50”
“The mean ≠ 50”
Goal
To be tested and possibly rejected
To be accepted if H₀ is rejected
In short:
• H₀ = “Nothing new is happening.”
• H₁ = “Something new is happening.”
Chapter 4: What Is Hypothesis Testing?
Hypothesis testing is the statistical process of deciding whether to accept or reject the null
hypothesis, based on sample data.
It’s like a courtroom trial:
• H₀ (null hypothesis) = The accused is innocent.
• Evidence (data) = Collected through experiments or surveys.
Easy2Siksha.com
• Decision = If evidence is strong, reject H₀ (declare guilty). If not, fail to reject H₀
(remain innocent).
Notice: We never “prove” H₀ true. We only decide whether there is enough evidence to
reject it.
Chapter 5: Steps in Hypothesis Testing
Let’s walk through the process step by step, like a story unfolding:
Step 1: State the Hypotheses
• Null Hypothesis (H₀): No effect or difference.
• Alternate Hypothesis (H₁): There is an effect or difference.
Example: H₀: The average exam score = 50 H₁: The average exam score ≠ 50
Step 2: Choose the Significance Level (α)
• This is the probability of rejecting H₀ when it is actually true (Type I error).
• Common choices: 0.05 (5%) or 0.01 (1%).
Example: α = 0.05 means we are willing to take a 5% risk of wrongly rejecting H₀.
Step 3: Select the Test Statistic
• Depending on the data, we choose a test:
o t-test (for means)
o z-test (for large samples)
o chi-square test (for categorical data)
o ANOVA (for comparing multiple means)
Step 4: Compute the Test Statistic
• Use formulas to calculate the test statistic (like t or z value).
• This measures how far the sample result is from the null hypothesis.
Step 5: Find the Critical Value or P-Value
• Critical Value Method: Compare test statistic with a threshold value.
• P-Value Method: Calculate the probability of observing the data if H₀ is true.
Step 6: Make the Decision
• If test statistic > critical value → Reject H₀.
• If p-value < α → Reject H₀.
• Otherwise → Fail to reject H₀.
Easy2Siksha.com
Step 7: Draw the Conclusion
• Translate the statistical decision into plain language.
Example: If p-value = 0.01 and α = 0.05 → Reject H₀. Conclusion: “There is significant
evidence that the average exam score is not 50.”
Chapter 6: Errors in Hypothesis Testing
Like any decision-making process, hypothesis testing can go wrong:
1. Type I Error (False Positive)
• Rejecting H₀ when it is actually true.
• Example: Concluding green tea improves memory when it doesn’t.
2. Type II Error (False Negative)
• Failing to reject H₀ when it is false.
• Example: Concluding green tea has no effect when it actually does.
Balancing these errors is the art of good statistical design.
Chapter 7: A Humanized Example
Let’s revisit Aarav, our student researcher. He wants to test if a new teaching method
improves math scores.
• H₀: The new method has no effect (mean score = 70).
• H₁: The new method improves scores (mean score > 70).
He collects data from 30 students, calculates the test statistic, and finds a p-value of 0.02.
Since p < 0.05, Aarav rejects H₀. Conclusion: The new method significantly improves
scores.
This is how statistics transforms curiosity into evidence.
Chapter 8: Why Examiners Love This Question
This question checks:
• Conceptual clarity (definition of hypothesis)
• Understanding of null vs. alternate hypothesis
• Knowledge of hypothesis testing steps
• Ability to explain with examples
Easy2Siksha.com
When written in a story-like, humanized way, it becomes enjoyable to read and easy to
grade.
Chapter 9: Real-Life Applications
Hypothesis testing is everywhere:
• Medicine: Does a new drug reduce blood pressure?
• Business: Does advertising increase sales?
• Education: Does online learning improve performance?
• Psychology: Does therapy reduce anxiety?
It’s the backbone of evidence-based decision-making.
Conclusion
A hypothesis is more than a guess—it’s a testable claim. The null hypothesis represents the
status quo, while the alternate hypothesis represents the possibility of change.
Through hypothesis testing, we move from beliefs to evidence, from assumptions to
conclusions.
It’s like a courtroom trial for ideas:
• The null hypothesis is “innocent until proven guilty.”
• The alternate hypothesis is the challenger, waiting for evidence.
• The data is the jury.
And when the verdict is reached, we don’t just have numbers—we have knowledge.
SECTION-B
3. Derive the basic properties of t-distribution.
Ans: A New Beginning: The Tale of the Curious Student
Once upon a time in a small university, there was a curious student named Arjun. He was
passionate about statistics, but there was one thing that always confused him — how can
we make conclusions about a population when we have only a small sample?
His professor smiled and said,
“Arjun, when you have a small sample, the normal distribution — our usual friend —
doesn’t behave exactly the same way. You need to meet someone new — Mr. Student’s t-
distribution!”
Easy2Siksha.com
Arjun’s eyes widened, “Who is this Mr. t-distribution?”
“Ah,” said the professor, “he’s the hero who saves us when our sample is small and we don’t
know the true population standard deviation.”
And thus began Arjun’s journey of discovering the t-distribution — a journey full of curiosity,
reasoning, and statistical beauty.
The Need for the t-Distribution: The Problem of the Unknown σ
In the world of statistics, we often want to estimate the population mean (μ).
If we already know the population standard deviation (σ), we can use the normal
distribution (Z-distribution). The formula we use is:
But in reality, do we ever truly know σ?
Not really! Especially when our sample size is small.
So what do we do? We estimate σ using the sample standard deviation (s). That gives us this
new formula:
But — and here’s the twist — since s is only an estimate of σ, it adds extra variability. That’s
why our distribution of t-values spreads out a little more than the normal curve.
Thus, the t-distribution was born — a slightly “fatter” or “wider” cousin of the normal
distribution, designed to handle small samples and uncertain standard deviations.
The Story Behind Its Discovery
Let’s step back into history. Around the early 1900s, a statistician named William Sealy
Gosset worked for the Guinness Brewery in Dublin.
He faced a problem — how could he make reliable conclusions about the quality of beer
using small samples?
He developed a new distribution that accounted for extra variability due to small samples.
But because Guinness didn’t allow employees to publish work under their real names,
Easy2Siksha.com
Gosset published his findings under the pseudonym “Student.”
And that’s why we call it the Student’s t-distribution.
So next time you hear “t-distribution,” remember, it’s not just a mathematical formula —
it’s a legacy born from beer-making and clever thinking!
Understanding the Shape: What Does the t-Distribution Look Like?
The t-distribution looks a lot like the normal distribution — bell-shaped and symmetrical
around the mean.
But there’s one small difference: the tails are thicker.
Why thicker tails?
Because when sample sizes are small, we are less certain about our estimate of σ (the
population standard deviation). The thicker tails represent that extra uncertainty — it allows
for the possibility of more extreme values.
As the sample size increases, our estimate of σ becomes more accurate, and the t-
distribution starts looking more and more like the normal distribution.
In fact:
As the t-distribution becomes the normal distribution.
This is like saying — as Arjun collects more data, he needs less help from “Mr. t,” and the
“normal” world returns!
The Formula for t-Distribution
Now, let’s peek behind the curtain and see how it’s built.
The t-distribution comes from two random variables:
• A standard normal variable, Z, and
• A chi-square variable, χ², divided by its degrees of freedom (v).
Mathematically:
Easy2Siksha.com
Here,
• Z follows the standard normal distribution (mean 0, variance 1)
• χ² is the chi-square variable with v degrees of freedom
• v = n - 1, where n is the sample size
That means the t-distribution depends on one key thing — degrees of freedom.
Degrees of Freedom — The Heart of t-Distribution
Now comes an interesting part — degrees of freedom (v).
It’s not as scary as it sounds.
Imagine you have 5 students and you want to find their average marks. Once you know 4 of
their marks and the average, the 5th mark is already decided.
So, you had freedom to choose only 4 marks independently.
That’s why we say the degrees of freedom is n - 1.
In t-distribution:
As the degrees of freedom increase, the t-distribution becomes narrower and starts
resembling the normal curve.
When v is small (say, 5 or 10), the curve is flatter and wider.
When v is large (like 100 or more), it almost overlaps with the normal distribution.
So, you can think of degrees of freedom as a “shape controller” — the more data we have,
the more confident and normal-looking our curve becomes!
The Basic Properties of t-Distribution
Now that we’ve met the t-distribution, let’s explore its wonderful properties — like learning
the unique traits of a person.
1. Symmetry
The t-distribution is symmetrical about the mean (zero).
That means the left and right sides of the curve are mirror images.
Positive and negative t-values are equally likely.
This symmetry shows that deviations on either side of the mean are equally probable.
Easy2Siksha.com
2. Mean
The mean of the t-distribution is zero, just like the normal distribution.
Since it’s perfectly symmetrical, the center lies exactly in the middle.
So:
But remember: the mean exists only if v > 1 (degrees of freedom greater than 1).
3. Variance
The variance of t-distribution depends on the degrees of freedom.
It’s given by:
for
That means when the sample size is small (low v), the variance is higher — the curve is more
spread out.
As v increases, the variance decreases and approaches 1 (which is the variance of a normal
distribution).
This tells us that the smaller the sample, the more “uncertain” our results are — hence the
wider curve.
4. Shape
The shape of the t-distribution is bell-shaped but with heavier tails.
That means there’s a higher probability of extreme values compared to the normal
distribution.
These heavy tails protect us from underestimating variability when dealing with small
samples. It’s like the t-distribution saying,
“Hey, small samples can be tricky — let’s play it safe by giving a bit more room for
uncertainty!”
5. Family of Curves
Easy2Siksha.com
There isn’t just one t-distribution — there’s a whole family of them!
Each member of the family is identified by its degrees of freedom (v).
• For v = 1, it’s very flat and wide.
• For v = 5, it becomes narrower.
• For v = 30, it’s almost like the normal curve.
So, the t-distribution family slowly merges into the normal family as v grows.
6. Relation with Normal Distribution
As degrees of freedom approach infinity:
That means when the sample size is large, you don’t really need to use the t-distribution —
the normal distribution will do just fine!
7. Use in Hypothesis Testing
The t-distribution is the backbone of many statistical tests, such as:
• One-sample t-test (comparing sample mean to a known value)
• Two-sample t-test (comparing two sample means)
• Paired t-test (comparing means from the same group before and after treatment)
All of these tests rely on the t-distribution when σ is unknown and the sample size is small.
8. Area Under the Curve
Just like the normal distribution, the total area under the t-curve is 1.
This represents the total probability (100%) of all possible outcomes.
9. Standardization Property
Just like Z-scores, t-scores can also be standardized using the formula:
Easy2Siksha.com
This formula helps compare sample results to the theoretical distribution and determine
whether a difference is statistically significant.
Visualizing the Behavior
Let’s imagine three friends standing together — Mr. Normal, Mr. t₅ (t with 5 degrees of
freedom), and Mr. t₃₀ (t with 30 degrees of freedom).
• Mr. Normal stands tall and slim.
• Mr. t₅ is a bit shorter and wider — representing extra uncertainty.
• Mr. t₃₀ stands almost like Mr. Normal, nearly identical.
This visual tells us that as degrees of freedom increase, the t-distribution becomes more
confident and “normal-like.”
Summary of Key Properties
Property
Description
Shape
Bell-shaped and symmetrical
Mean
0 (exists for v > 1)
Variance
v / (v – 2), exists for v > 2
Median & Mode
Both equal to 0
Range
–∞ to +∞
Tails
Heavier than normal distribution
Degrees of Freedom
v = n – 1
Asymptotic Behavior
Approaches normal as v → ∞
Area Under Curve
Equal to 1
Why It’s So Important
The t-distribution may seem like just another formula, but it’s a lifeline for researchers and
scientists who deal with small samples. It gives them the confidence to make meaningful
inferences when data is limited.
In simple words — the t-distribution teaches us how to stay honest about uncertainty.
Instead of pretending our small sample is perfect, it reminds us to allow for a bit more
variation, a bit more humility in our conclusions.
Easy2Siksha.com
Conclusion: The Wisdom of Mr. Student
Arjun finally understood what his professor meant.
The t-distribution wasn’t just a mathematical concept — it was a symbol of real-world
wisdom.
It teaches that life (and data) isn’t always perfect or complete. Sometimes, we must make
decisions with limited information — but with the right tools and understanding, we can still
be remarkably accurate.
4. Highlight the characteristic features of Chi-square-distribution. Highlight its use by
giving a suitable example.
Ans: It was a village fair, and a young boy named Aarav stood before a game stall. The game
was simple: roll a pair of dice and win a prize if the sum was seven. Aarav played again and
again, but he began to wonder: “Are these dice really fair, or is the stall owner tricking me?”
His elder sister, a statistics student, smiled and said: “That’s exactly the kind of question we
can answer using the Chi-square distribution. It’s a tool that helps us check whether what
we observe in real life matches what we expect in theory.”
And so begins our journey into the world of the Chi-square distribution—a distribution that
is not just about numbers, but about testing fairness, patterns, and relationships in data.
Chapter 1: What Is the Chi-Square Distribution?
The Chi-square distribution (χ²) is a probability distribution that arises when we sum the
squares of independent standard normal random variables.
Mathematically:
where
are independent standard normal variables.
In simple words: If you take several normally distributed values, square them, and add
them up, the result follows a Chi-square distribution.
It was first introduced by Karl Pearson in 1900, and since then, it has become one of the
most widely used tools in hypothesis testing.
Chapter 2: Characteristic Features of Chi-Square Distribution
Let’s highlight its key features in a way that feels intuitive:
Easy2Siksha.com
1. Non-Negative Values Only
• Since it is based on squared values, χ² can never be negative.
• The distribution starts at 0 and extends to infinity.
Example: You can’t have a “negative square,” so χ² values are always ≥ 0.
2. Shape Depends on Degrees of Freedom (df)
• The distribution is defined by degrees of freedom (df), which usually equals the
number of categories minus one.
• For small df (like 1 or 2), the distribution is highly skewed to the right.
• As df increases, the distribution becomes more symmetric and approaches a normal
distribution.
Think of df as the “flexibility” of the distribution—the more categories or variables, the
smoother the curve.
3. Mean and Variance
• Mean of χ² distribution = df
• Variance = 2 × df
Example: If df = 6, then mean = 6 and variance = 12.
4. Family of Distributions
• There isn’t just one χ² distribution.
• For each df, there is a different curve.
• That’s why we call it a family of distributions.
5. Positively Skewed
• For small df, the curve is skewed to the right.
• As df increases, skewness decreases, and the curve looks more like a bell-shaped
normal curve.
6. Additive Property
• If two independent χ² variables have df = m and df = n, their sum is also χ² with df =
m + n.
This makes it very flexible in combining data.
7. Continuous Distribution
• χ² is a continuous probability distribution, not discrete.
Easy2Siksha.com
• It is used for continuous probability calculations, even though it often deals with
categorical data.
Chapter 3: Why Do We Use Chi-Square?
The χ² distribution is mainly used in hypothesis testing. It helps us answer questions like:
• Are dice fair?
• Is a coin unbiased?
• Do two categorical variables (like gender and preference) have a relationship?
• Does observed data match expected data?
The two most common uses are:
1. Chi-square goodness-of-fit test → Does observed data fit a theoretical distribution?
2. Chi-square test of independence → Are two categorical variables independent?
Chapter 4: A Humanized Example – The Dice Game
Let’s return to Aarav at the fair. He suspects the dice are unfair.
Step 1: Form Hypotheses
• H₀ (Null Hypothesis): The dice are fair.
• H₁ (Alternate Hypothesis): The dice are not fair.
Step 2: Collect Data
Aarav rolls the dice 60 times. He expects each outcome (2 to 12) to appear with certain
probabilities. For example, 7 should appear most often.
Suppose his observed results differ from the expected ones.
Step 3: Apply Chi-Square Formula
Where:
• O = Observed frequency
• E = Expected frequency
Step 4: Compare with Critical Value
• Calculate χ² value.
• Compare it with the critical value from the χ² table at chosen significance level (say
0.05).
Easy2Siksha.com
If χ² calculated > χ² table → Reject H₀ → Dice are unfair. If χ² calculated ≤ χ² table →
Fail to reject H₀ → Dice are fair.
Chapter 5: Another Example – Gender and Movie Preference
Suppose a researcher wants to know if movie preference (Action vs. Romance) depends on
gender (Male vs. Female).
• Collect data from 200 people.
• Create a contingency table of observed frequencies.
• Use χ² test of independence.
If the χ² value is significant, it means gender and movie preference are related. If not, they
are independent.
Chapter 6: Why Examiners Love This Question
This question checks:
• Conceptual clarity (features of χ² distribution)
• Application skills (example of use)
• Ability to connect theory with real-world situations
When explained in a story-like way, it becomes enjoyable to read and easy to grade.
Chapter 7: Real-Life Applications
Chi-square is used in:
• Genetics: Mendel used it to test pea plant ratios.
• Medicine: Testing whether a treatment works differently across groups.
• Marketing: Checking if product preference depends on age or gender.
• Education: Testing if exam results differ by teaching method.
It’s a universal tool for categorical data analysis.
Conclusion
The Chi-square distribution is like a detective—it investigates whether what we see in the
world matches what we expect in theory.
• Its features—non-negative values, dependence on degrees of freedom, skewness,
and additivity—make it unique.
• Its uses—goodness-of-fit and independence tests—make it powerful.
So, whether it’s Aarav at the fair questioning dice, or a scientist testing a new drug, the χ²
distribution is there, quietly guiding us from doubt to decision.
Easy2Siksha.com
SECTION-C
5. What is the difference between Paired t-test and Non-paired t-test? Enlist four
situations (two each) where these can be applied.
Ans: In a bustling university lab, two groups of students were working on different projects.
One group was testing whether a new teaching method improved the same students’
performance before and after training. The other group was comparing the exam scores of
two entirely different classes—one taught with traditional lectures and the other with
interactive videos.
Both groups wanted to know: “Is there a real difference, or is it just chance?”
Their professor walked in and said: “You’re both asking the same kind of question, but the
way you answer it depends on whether your data is paired or independent. That’s where the
paired t-test and the non-paired (independent) t-test come in.”
And so begins our journey into understanding these two powerful statistical tools.
Chapter 1: The Essence of a t-Test
The t-test is a statistical method used to compare the means of two groups and determine
whether the difference between them is statistically significant.
But here’s the catch:
• Sometimes the two groups are related (like before-and-after scores of the same
students).
• Sometimes the two groups are independent (like scores of two different classes).
That’s why we have two versions:
• Paired t-test (for related samples)
• Non-paired t-test (for independent samples)
Chapter 2: Paired t-Test – The “Before and After” Story
Definition
A paired t-test is used when the two sets of data are dependent—that is, they come from
the same group of subjects measured twice, or from matched pairs.
Think of it as comparing you with yourself at two different times, or comparing twins
under two different conditions.
Key Features
Easy2Siksha.com
• Same subjects measured twice (before vs. after).
• Differences are calculated for each pair.
• The test checks whether the average difference is significantly different from zero.
Formula
Where:
•
= mean of differences
•
= standard deviation of differences
• = number of pairs
Example
Suppose 10 students take a math test before and after coaching. We compare their before-
and-after scores using a paired t-test to see if coaching improved performance.
Chapter 3: Non-Paired (Independent) t-Test – The “Two Groups” Story
Definition
A non-paired t-test (also called an independent samples t-test) is used when the two sets of
data are independent—that is, they come from different groups of subjects.
Think of it as comparing you with someone else.
Key Features
• Two separate groups of subjects.
• No natural pairing between individuals.
• The test checks whether the difference between group means is significant.
Formula
Where:
•
= sample means
•
= sample variances
•
= sample sizes
Easy2Siksha.com
Example
Suppose one class of 30 students is taught with lectures and another class of 30 students is
taught with videos. We compare their average scores using a non-paired t-test.
Chapter 4: Key Differences Between Paired and Non-Paired t-Test
Feature
Paired t-Test
Non-Paired t-Test
Data Type
Dependent (same subjects or matched
pairs)
Independent (different
groups)
Focus
Mean of differences
Difference of means
Example
Before vs. after training scores of same
students
Scores of two different
classes
Variability
Reduced (since same subjects are
compared)
Higher (since groups differ)
Sample
Size
Smaller samples often sufficient
Larger samples preferred
In short:
• Paired t-test = “Did the same people change?”
• Non-paired t-test = “Are these two groups different?”
Chapter 5: Situations Where They Are Applied
Situations for Paired t-Test
1. Medical Research: Measuring patients’ blood pressure before and after taking a new
drug.
2. Education: Comparing students’ test scores before and after attending a coaching
program.
Situations for Non-Paired t-Test
1. Agriculture: Comparing crop yields from two different fertilizers applied to two
separate fields.
2. Psychology: Comparing stress levels of two independent groups—one practicing
meditation and the other not.
Chapter 6: A Humanized Example
Let’s revisit Aarav and his classmates.
• Aarav takes a math test before and after coaching. His improvement is measured
against his own earlier score. This is a paired t-test situation.
Easy2Siksha.com
• Meanwhile, his friend Riya compares the scores of two different classes—one taught
by Professor A and the other by Professor B. Since the groups are independent, this
is a non-paired t-test situation.
Both are asking: “Is there a real difference?” But the way they test it depends on whether
the data is paired or independent.
Chapter 7: Why Examiners Love This Question
This question checks:
• Conceptual clarity (definitions of paired vs. non-paired t-test)
• Ability to distinguish between dependent and independent samples
• Application skills (real-life situations)
• Communication skills (explaining with examples)
When written in a story-like, humanized way, it becomes enjoyable to read and easy to
grade.
Chapter 8: Real-Life Applications
• Medicine: Paired t-test for before-and-after treatment; non-paired t-test for
comparing two different patient groups.
• Business: Paired t-test for sales before and after an ad campaign; non-paired t-test
for comparing sales in two different regions.
• Sports: Paired t-test for athlete performance before and after training; non-paired t-
test for comparing two different teams.
These tests are everywhere—helping us make decisions based on evidence.
Conclusion
The paired t-test and non-paired t-test are like two siblings—similar in spirit but different in
personality.
• The paired t-test looks inward, comparing the same subjects across time or
conditions.
• The non-paired t-test looks outward, comparing two independent groups.
Together, they help us answer one of the most important questions in research: “Is this
difference real, or is it just chance?”
Easy2Siksha.com
6. A certain drug was administered to 200 people out of a total 500 included in the sample
to test its efficacy against Dengue. The results are as follows:
Incidence of Dengue
Total
Yes
No
Drug
50
150
200
No Drug
250
50
300
Total
300
200
500
Can you say that drug is effective in preventing Dengue ?
Ans: A story about a promise, two groups, and a tiny invisible enemy
In a small town, Dr. Mira wanted to know whether a newly invented pill could keep people
safe from Dengue fever. She had 500 volunteers. She split them into two groups — not by
gossip or luck, but for the sake of the experiment:
• 200 people were given the drug (call them the Drug Team).
• 300 people were not given the drug (call them the No-Drug Team).
After a few weeks, the village clinic recorded who actually caught Dengue. The notebook
read:
• In the Drug Team (200 people): 50 got Dengue and 150 did not.
• In the No-Drug Team (300 people): 250 got Dengue and 50 did not.
On the whiteboard Dr. Mira drew a table:
Dengue Yes
Dengue No
Total
Drug
50
150
200
No Drug
250
50
300
Total
300
200
500
Now the villagers asked: “Is the drug actually protecting people?” Dr. Mira smiled and said,
“Let’s tell the story with numbers — step by step.”
Step-by-step, number-by-number (so it's crystal clear)
First we compute the risk (incidence) of Dengue in each group — that is, the fraction of
people in the group who got Dengue.
1) Risk in the Drug group
Easy2Siksha.com
We have 50 people who got Dengue out of 200 in the Drug group.
Compute digit-by-digit:
• Numerator = 50
• Denominator = 200
• 50 ÷ 200 = 0.25
So the risk = 0.25 = 25%.
2) Risk in the No-Drug group
We have 250 people who got Dengue out of 300 in No-Drug group.
Compute digit-by-digit:
• Numerator = 250
• Denominator = 300
• 250 ÷ 300 = (divide both by 50) → 5 ÷ 6 = 0.833333... repeating
So the risk ≈ 0.8333 = 83.33%.
Plain language: If you were in the Drug group, 1 out of 4 people got Dengue. If you were in
the No-Drug group, more than 4 out of 5 people got Dengue.
How much protection does the drug give?
We can describe the benefit in a few useful ways: absolute change, relative change, and
how many people we need to treat to prevent one case.
3) Absolute Risk Reduction (ARR)
ARR = Risk(No-Drug) − Risk(Drug).
Compute digit-by-digit:
• Risk(No-Drug) = 0.833333...
• Risk(Drug) = 0.25
• ARR = 0.833333... − 0.25 = 0.583333...
So ARR ≈ 0.5833, which is 58.33 percentage points.
Plain language: Absolutely speaking, the drug lowered the chance of getting Dengue by
about 58 percentage points.
4) Relative Risk (RR) and Relative Risk Reduction (RRR)
Easy2Siksha.com
Relative Risk = Risk(Drug) ÷ Risk(No-Drug).
Compute digit-by-digit using fractions:
• Risk(Drug) = 50/200 = 1/4
• Risk(No-Drug) = 250/300 = 5/6
• RR = (1/4) ÷ (5/6) = (1/4) × (6/5) = 6/20 = 3/10 = 0.3
So RR = 0.3.
Relative Risk Reduction (RRR) = 1 − RR = 1 − 0.3 = 0.7 = 70%.
Plain language: People taking the drug had 30% of the risk that people without it had — or
you can say the drug cut the risk by 70% relative to not taking it.
5) Number Needed to Treat (NNT)
NNT = 1 ÷ ARR.
Compute digit-by-digit:
• ARR = 0.583333...
• 1 ÷ 0.583333... = (recognize 0.583333... = 7/12) — actually compute: 0.583333... =
7/12, so inverse = 12/7 ≈ 1.7142857...
Round up for NNT (because you can’t treat fractional people): NNT = 2.
Plain language: Treat 2 people with the drug to prevent one case of Dengue. That’s a very
small NNT — usually considered excellent.
A statistical check: is the difference unlikely to be just by chance?
Numbers that dramatic look convincing, but scientists usually perform a statistical test to
check if such a difference could plausibly arise by chance. For a 2×2 table like this, a Chi-
square test is common. I'll show the logic and the core arithmetic.
We calculate what counts we would expect if the drug made no difference (i.e., the null
hypothesis). Expected counts use row totals × column totals ÷ grand total.
Expected counts (digit-by-digit):
• Expected Dengue in Drug group = (Row total for Drug × Column total for Yes) ÷
Grand total = (200 × 300) ÷ 500 = 60000 ÷ 500 = 120.
• Expected No-Dengue in Drug group = (200 × 200) ÷ 500 = 40000 ÷ 500 = 80.
• Expected Dengue in No-Drug group = (300 × 300) ÷ 500 = 90000 ÷ 500 = 180.
• Expected No-Dengue in No-Drug group = (300 × 200) ÷ 500 = 60000 ÷ 500 = 120.
Easy2Siksha.com
Now use the Chi-square formula: for each cell, (Observed − Expected)² ÷ Expected, sum over
cells.
Compute cell contributions (digit-by-digit):
• Drug, Yes: observed 50, expected 120 → difference = −70 → square = 4,900 → divide
by 120 → 4,900 ÷ 120 = 40.833333...
• Drug, No: observed 150, expected 80 → difference = 70 → square = 4,900 → divide
by 80 → 4,900 ÷ 80 = 61.25
• No-Drug, Yes: observed 250, expected 180 → difference = 70 → square = 4,900 →
divide by 180 → 4,900 ÷ 180 = 27.222222...
• No-Drug, No: observed 50, expected 120 → difference = −70 → square = 4,900 →
divide by 120 = 40.833333...
Add them up:
• 40.833333... + 61.25 + 27.222222... + 40.833333... = 170.138888...
So Chi-square ≈ 170.14 with 1 degree of freedom (because 2×2 table). For such a large Chi-
square, the p-value is vanishingly small, far less than 0.001. In plain words: the chance that
such a huge difference happened randomly (if the drug had no effect) is effectively zero.
Plain language: The statistical test strongly supports that the drug really is making a big
difference — not just luck.
Odds Ratio (for the story's detectives)
Odds in Drug group = 50 ÷ 150 = 1 ÷ 3 ≈ 0.3333.
Odds in No-Drug group = 250 ÷ 50 = 5.
Odds Ratio (OR) = 0.3333 ÷ 5 = 0.066666... = 1/15.
Interpretation: The odds of getting Dengue while on the drug are 1/15th of the odds
without the drug. That's another way to show the drug’s protective effect is huge.
So what’s the final answer? — short, clear, and human
Yes — based on these data, the drug appears very effective at preventing Dengue.
• Risk fell from 83.33% (no drug) to 25% (drug).
• Absolute risk reduction ≈ 58.33 percentage points.
• Relative risk reduction = 70%.
• Number needed to treat (NNT) = 2 (treat two people to prevent one Dengue case).
• Chi-square ≈ 170.14, p 0.001, so the difference is highly statistically significant.
Easy2Siksha.com
In plain story terms: imagine ten friends. If none take the drug, about eight or nine of them
would likely get Dengue. If they all take the drug, only about two or three would — that’s a
dramatic rescue of people. Treating two friends will likely spare one of them from Dengue.
Those are the numbers; the drug looks like a strong protector.
But every good story also mentions the fine print (caveats)
A clear, fair answer must also mention limitations before we accept the result as the final
truth:
1. Was the study randomized? The table alone does not say whether volunteers were
randomly assigned to drug vs. no-drug. Randomization helps avoid bias. If groups
were chosen unfairly (e.g., drug group had younger/healthier people), some of the
effect could be due to group differences rather than the drug.
2. Was there blinding? If participants or doctors knew who got the drug, reporting or
diagnosis bias could creep in.
3. Side effects and safety: We measured only Dengue incidence. What if the drug has
harmful side effects? Effectiveness must be weighed against safety.
4. Follow-up and measurement: Were both groups followed equally closely? Were
Dengue cases diagnosed with the same accuracy?
5. External validity: These 500 people are a sample. Are they representative of the
wider population (different ages, regions, virus strains)?
6. Random variation and replication: Even though the p-value is tiny here, replication
in other populations and trials strengthens confidence.
Even with these caveats, the magnitude of the effect — huge differences, low NNT, and
strong statistics — strongly suggests a real and clinically meaningful benefit. It’s rare to see
such clear-cut numbers.
SECTION-D
7. A manufacturer appoints 3 workers A, B and C and observes their production in terms
of number of units produced with the use of three different machines X. Y. Z. Perform a
Two-way ANOVA on the data given below and interpret your result on average production
status:
Workers
Machines
X
Y
Z
A
16
64
40
Easy2Siksha.com
B
56
72
56
C
12
56
28
Ans: Imagine a small workshop where a manufacturer hires three workers — A, B, and C —
and gives them three different machines — X, Y, and Z — to see how many units each
worker makes on each machine in one day. The manager writes down the numbers (one
observation per worker–machine pair):
Machines
X Y Z
A 16 64 40
B 56 72 56
C 12 56 28
We want to answer, with statistical evidence, two simple questions that the manager cares
about (and any curious student reading this story):
1. Do different workers (A, B, C) differ in their average production?
2. Do different machines (X, Y, Z) differ in the average production they produce?
We will use two-way ANOVA without replication (because there is only one observation in
each worker–machine cell) to test the two main effects (workers and machines). Because
there’s no replication (no multiple observations per cell), we cannot separate interaction
from error — the residual we compute will include possible interaction between worker and
machine as well as pure random error. I’ll explain the calculations step by step, then
translate the numbers into plain English and practical recommendations.
1. Turning the numbers into a table of means and totals (the map of our workshop)
First we compute some simple summaries.
• Grand total (sum of all observations): 400.
• Number of observations: 9 (3 workers × 3 machines).
• Grand mean (overall average production) = 400 / 9 = 44.444... units.
Row sums (workers) and their row means:
• Worker A: sum = 16 + 64 + 40 = 120 → mean = 120 / 3 = 40.000
• Worker B: sum = 56 + 72 + 56 = 184 → mean = 184 / 3 = 61.333...
• Worker C: sum = 12 + 56 + 28 = 96 → mean = 96 / 3 = 32.000
Column sums (machines) and their column means:
• Machine X: sum = 16 + 56 + 12 = 84 → mean = 84 / 3 = 28.000
Easy2Siksha.com
• Machine Y: sum = 64 + 72 + 56 = 192 → mean = 192 / 3 = 64.000
• Machine Z: sum = 40 + 56 + 28 = 124 → mean = 124 / 3 = 41.333...
So already from these means we can read an intuitive picture:
• Worker B has the highest average (≈61.33 units), C the lowest (32), A in the middle
(40).
• Machine Y looks much better (64 units on average) than X (28) or Z (41.33).
But are these differences big enough to be statistically convincing? That’s what ANOVA
tests.
2. The mechanics of two-way ANOVA (without replication)
We split the total variability (how much all the observed numbers wander around the grand
mean) into parts:
• variability due to workers (how row means differ from grand mean),
• variability due to machines (how column means differ from grand mean),
• and remaining variability (residual) which, in this design without replication, includes
any interaction (worker×machine) and any unexplained error.
Mathematically:
• Total Sum of Squares (SST) = sum of (each observation − grand mean)².
SST = 3694.2222
• Sum of Squares for Workers (SS_workers) = measure of spread of row totals around
the grand mean:
SS_workers = 1379.5556
• Sum of Squares for Machines (SS_machines) = measure of spread of column totals
around the grand mean:
SS_machines = 1987.5556
• Residual Sum of Squares (SS_res) = SST − SS_workers − SS_machines:
SS_res = 327.1111
Degrees of freedom (df):
• df_total = N − 1 = 9 − 1 = 8
• df_workers = r − 1 = 3 − 1 = 2
• df_machines = c − 1 = 3 − 1 = 2
• df_residual = (r − 1)(c − 1) = 2 × 2 = 4
Mean squares (MS) — we divide each SS by its df:
• MS_workers = SS_workers / 2 = 689.7778
• MS_machines = SS_machines / 2 = 993.7778
Easy2Siksha.com
• MS_res = SS_res / 4 = 81.7778
F-statistics (compare effect MS to residual MS):
• F_workers = MS_workers / MS_res = 8.4348
• F_machines = MS_machines / MS_res = 12.1522
We compare each F to a critical F value for df1 = 2 and df2 = 4. For a 5% significance level,
the critical F(2,4) ≈ 6.944 (standard F tables give this). For a 1% level, F(2,4) ≈ 21.2 (approx).
• F_workers = 8.43 > 6.944 → significant at α = 0.05 (but less than 21.2, so not
significant at 0.01).
• F_machines = 12.15 > 6.944 → significant at α = 0.05 as well (and likewise p-value is
< 0.05 but > 0.01).
(If you like p-value language: each F leads to a p-value < 0.05. We cannot get a precise p-
value without software here, but we can say both effects are significant at the 5% level.)
ANOVA summary table (numbers rounded)
Source
SS
df
MS
F
Workers
1379.5556
2
689.7778
8.4348
Machines
1987.5556
2
993.7778
12.1522
Residual
327.1111
4
81.7778
—
Total
3694.2222
8
—
—
3. What the numbers mean — the story conclusion
Now the storytelling, in plain and human language:
The cast and the surprising hero
• Worker B is the star performer. His average production is about 61.3 units —
significantly larger than the overall average (44.44). The ANOVA shows that
differences among workers are unlikely to be due to random chance alone (F ≈ 8.43,
significant at the 5% level). So the manager can be confident: workers really do differ
in average output, and B performs noticeably better than the others in this sample.
The machines: one clear winner
• Machine Y shows the highest average: 64 units — much better than machine X (28)
or Z (41.33). The ANOVA shows that the differences across machines are also
statistically significant (F ≈ 12.15, p < 0.05). That means machine choice strongly
affects production.
But a cautionary footnote (the twist in the plot)
Easy2Siksha.com
• We have only one observation per worker–machine cell. That means we cannot
cleanly separate worker×machine interaction from residual randomness. In plain
terms: maybe worker B is especially good on machine Y (their combination yields a
big bonus), or maybe some other combinations have a tug-and-pull that matters.
Because each cell was observed only once, that interaction effect is lumped into the
residual. So: we cannot test whether the advantage of worker B depends on which
machine is used. To study interaction (e.g., some workers being especially good on a
particular machine), the manufacturer should run the experiment with replication
(take multiple production measurements for each worker on each machine). With
replicates we could test interaction separately.
Putting it practically for the manager
• Immediate action: Machine Y seems superior. If the manufacturer wants higher
output right away, prioritize using machine Y — it yields a much higher average
across workers.
• Staffing insight: Worker B is consistently the best performer across machines.
Consider giving B priority on the best machines, or study what B is doing differently
(technique, experience) and use that knowledge in training others.
• Further experiment: To check if B is uniquely good on Y (interaction) or if these are
independent effects, repeat the measurements (take several days of production per
worker–machine pair) so you can estimate interaction and random error separately.
4. A simple, friendly recap (wrap-up)
Think of the data as three chefs (A, B, C) each cooking three dishes (X, Y, Z) once, and we
recorded how many plates they produced. From the averages and statistical test:
• Chefs differ: yes (B clearly the best).
• Dishes (machines) differ: yes (Y is the best).
• Whether a chef does especially well on a particular dish (interaction) — we don’t
know from these single observations. We’d need more trials.
So the story’s moral: both who works and which machine they use matter. But if the
manufacturer wants to be certain about pairwise fits (who fits best with which machine),
run the experiment again with repeated measurements.
5. Final one-line takeaway
Both worker and machine effects are statistically significant at the 5% level (workers: F ≈
8.43, machines: F ≈ 12.15), so the manager should favor worker B and machine Y — and
run more trials to explore worker×machine interactions.
Easy2Siksha.com
8. Is the Analysis of Variance (ANOVA) technique an extension of the tests used for testing
the difference between two means? Support your agreement/disagreement with the
details. Enlist the assumptions of ANOVA technique for CRD and RBD of experiment.
Ans: A Fresh Beginning
It was a bright morning in the agricultural research station. A group of scientists had just
finished testing three different fertilizers on wheat crops. One scientist said: “We know how
to compare two fertilizers using a t-test. But what about three? Should we compare them
two at a time?”
The senior statistician chuckled: “If you compare them pair by pair, you’ll make too many
errors. Instead, use a single, powerful tool—Analysis of Variance (ANOVA). It’s like the elder
sibling of the t-test, designed to handle more than two groups at once.”
And with that, the team began their journey into ANOVA—a technique that is indeed an
extension of the test for two means, but with broader scope and deeper insight.
Chapter 1: Is ANOVA an Extension of the Two-Mean Test?
The Two-Mean Test (t-test)
When we want to compare the means of two groups (say, average marks of boys vs. girls),
we use a t-test. It tells us whether the difference between the two means is statistically
significant.
But what if we have three or more groups?
• Example: Comparing average yields of crops under Fertilizer A, Fertilizer B, and
Fertilizer C.
• If we use multiple t-tests, the probability of making a Type I error (false positive)
increases drastically.
The Extension: ANOVA
ANOVA (Analysis of Variance), introduced by R.A. Fisher, generalizes the t-test. Instead of
comparing means two at a time, it compares all group means simultaneously.
In simple words:
• t-test = “Are these two means different?”
• ANOVA = “Are these three or more means all the same, or is at least one different?”
Thus, ANOVA is indeed an extension of the test for two means. In fact, when ANOVA is
applied to just two groups, it gives the same result as a t-test.
Chapter 2: The Logic of ANOVA
Easy2Siksha.com
ANOVA works by analyzing variability.
• Total variability in the data is split into two parts:
1. Between-group variability (differences due to treatments or groups)
2. Within-group variability (differences due to random error or chance)
• ANOVA compares these two sources of variation using the F-ratio:
Between-group variance
Within-group variance
If F is large → group means differ significantly. If F is small → differences are likely
due to chance.
Chapter 3: Why Not Just Use Multiple t-Tests?
Imagine comparing 4 fertilizers.
• Number of pairwise comparisons = 6.
• Each test has a 5% chance of error.
• Combined error rate becomes much higher.
ANOVA solves this by testing all groups together in one test, keeping the error rate under
control.
Chapter 4: Assumptions of ANOVA
Like any statistical test, ANOVA rests on certain assumptions. These assumptions differ
slightly depending on the design of the experiment.
A. Assumptions for CRD (Completely Randomized Design)
In CRD, treatments are assigned to experimental units completely at random.
Assumptions:
1. Independence: All observations are independent of each other.
2. Normality: The errors (residuals) are normally distributed.
3. Homogeneity of Variance: Variances within each treatment group are equal.
4. Randomization: Treatments are randomly assigned to units.
Example: Testing three fertilizers on randomly chosen plots of land.
B. Assumptions for RBD (Randomized Block Design)
In RBD, experimental units are grouped into blocks (to control for variability), and
treatments are randomly assigned within each block.
Easy2Siksha.com
Assumptions:
1. Independence: Observations within and across blocks are independent.
2. Normality: Errors are normally distributed.
3. Homogeneity of Variance: Variances are equal across treatments.
4. Additivity: Treatment effects and block effects are additive (no interaction).
5. Randomization within Blocks: Treatments are randomly assigned within each block.
Example: Testing fertilizers across different soil types (blocks), with each fertilizer
applied randomly within each soil type.
Chapter 5: Example
Let’s return to our scientists.
• They test three fertilizers (A, B, C) on wheat.
• In a CRD, they assign fertilizers randomly to plots.
• In an RBD, they first group plots by soil type (sandy, clay, loam), then assign
fertilizers randomly within each soil type.
After collecting yields, they run ANOVA.
• If F is significant, they conclude: “At least one fertilizer is different.”
• Then they use Critical Difference (CD) or post-hoc tests to find out which fertilizers
differ.
Chapter : Real-Life Applications of ANOVA
• Agriculture: Comparing crop yields under different fertilizers.
• Medicine: Testing effectiveness of multiple drugs.
• Education: Comparing teaching methods across classes.
• Business: Comparing sales across regions or marketing strategies.
Anywhere multiple groups are compared, ANOVA is the tool of choice.
Conclusion
ANOVA is like a wise judge in a crowded courtroom. Instead of listening to endless pairwise
arguments (t-tests), it listens to all groups together and delivers a single verdict: “Are these
groups truly different, or not?”
Yes, ANOVA is an extension of the test for two means—but it is more than that. It is a
framework for comparing multiple groups, controlling error, and making sense of complex
data.
Easy2Siksha.com
And whether in a CRD with simple randomization or an RBD with careful blocking, ANOVA
stands as one of the most powerful tools in the statistician’s toolkit—guiding researchers
from doubt to decision, from data to discovery.
“This paper has been carefully prepared for educational purposes. If you notice any
mistakes or have suggestions, feel free to share your feedback.”
