M.A.

PSYCHOLOGY (Regular Mode)

AY 2025-2026 | Semester II | Statistics in Psychology (Course Code-201)

MODULE – II
Inferential Statistics, Sampling and Hypothesis Testing
Detailed Study Notes
Source: Statistical Methods for Psychology, 7th Ed. — David C. Howell

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 1

 ■ MODULE II — OVERVIEW
Module II covers the core inferential statistical methods used in psychological research. Moving beyond
descriptive statistics, this module focuses on making decisions and drawing conclusions about
populations based on sample data. Six major areas are examined:
• Inferential Statistics: Concept of distribution, normal and binomial distributions, properties and area
under the normal curve.
• Sampling Distribution: Central Limit Theorem, sampling distribution of means, standard error.
• Parameter, Statistic, and Sampling: Concept of parameter and statistic, methods of sampling
(probabilistic and non-probabilistic), sampling error.
• Hypothesis Testing: Types of hypotheses, one- and two-tailed tests, one-sample t-test,
independent samples t-test, levels of significance, confidence intervals, Type I and Type II errors.
• Power of a Test: Factors affecting power, effect size.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 2

 ■ SECTION 1: Inferential Statistics and Types of Distributions
1.1 Descriptive vs. Inferential Statistics
Descriptive Statistics Used to summarise, organise, and describe the characteristics of a dataset.
Examples: mean, median, mode, standard deviation, frequency distributions.
These describe only the sample at hand.

Inferential Statistics Used to draw conclusions (inferences) about a population on the basis of
sample data. Because we rarely have access to an entire population, we must
use samples and statistical logic to make probabilistic statements about the
population.

The Central Role of The critical link in inferential statistics is between distributions and
Distributions probabilities. If we know the distribution of a statistic, we know the probability
that any particular value of that statistic will occur. This is the foundation for all
hypothesis testing.

1.2 The Normal Distribution

The normal distribution is the single most important distribution in statistics because (1) many
psychological variables are approximately normally distributed in the population, (2) it allows us to
make inferences about variable values, and (3) the sampling distribution of the mean is approximately
normal under a wide variety of conditions.

■ Properties of the Normal Distribution

Mathematical Definition f(X) = (1 / σ√2π) × e^[-(X-µ)²/2σ²] where π = 3.1416, e = 2.7183, µ = population
mean, σ = population standard deviation.

Bell-Shaped and The curve is perfectly symmetric around its mean. The left and right halves are
Symmetric mirror images. The mean, median, and mode all coincide at the centre of a
normal distribution.

Unimodal There is a single peak (mode) at the centre of the distribution.

Asymptotic The tails of the distribution extend infinitely in both directions but never actually
touch the X-axis (limits are ±∞).

Defined by µ and σ The exact shape and location of the normal distribution is completely determined
by its mean (µ) and standard deviation (σ). Different values of µ and σ produce
different normal distributions.

Total Area = 1.0 The total area under the normal curve equals 1.0 (representing 100% of all
possible values). Areas under the curve directly correspond to probabilities.

■ Area Under the Normal Curve — Key Percentages

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 3

 Range Area / Probability Interpretation

µ ± 1σ 68.26% About 2 out of 3 scores fall within 1

SD of the mean

µ ± 2σ 95.44% About 95 out of 100 scores fall

within 2 SDs of the mean

µ ± 3σ 99.74% Almost all scores (>99%) fall within

3 SDs of the mean

µ ± 1.96σ 95.00% Exactly 95% of scores fall within

1.96 SDs — basis for 95% CI

µ ± 2.58σ 99.00% Exactly 99% of scores fall within

2.58 SDs — basis for 99% CI

■ The Standard Normal Distribution (z-distribution)

Problem With Multiple Because every combination of µ and σ produces a different normal distribution,
Tables we would need an infinite number of tables. The solution: convert all normal
distributions to one standard form.

Standard Normal A normal distribution with mean = 0 and standard deviation = 1, designated as
N(0,1). A single table (Appendix z) works for all normal distributions by
transforming raw scores to z-scores.

z-Score Formula z = (X − µ) / σ Subtracting µ shifts the mean to 0; dividing by σ scales the SD to

1. The z-score tells how many standard deviations a score is above (+) or below
(−) the mean.

Interpretation A score of 60 from N(50,10) becomes z = (60-50)/10 = 1.0. This means the
score is exactly 1 SD above the mean. The shape of the distribution is
unchanged — only the numerical scale on the X-axis changes.

Using z-Tables The z-table gives the area (probability) above or below any z-value. To find the
probability that a randomly chosen score exceeds some value X: (1) convert X
to z, (2) look up the area in the tail from the z-table.

z = (X − µ) / σ OR z = (X■ − µ) / (σ/√n) [for sample means]

■ Application of the Normal Curve in Psychology

• Diagnosis and Norms: Test scores (IQ, anxiety, depression inventories) are often standardised to a
normal distribution. Clinicians use z-scores to identify scores that are unusually high or low (e.g.,
top or bottom 5% or 10%).
• Probability Statements: Given a normally distributed variable, we can calculate the probability that
a randomly chosen person will score above or below any given value.
• Foundation for Hypothesis Testing: The assumption that many variables are normally distributed
(or that sampling distributions are normal) underpins most parametric statistical tests used in
psychological research.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 4

 • Setting Cut-Off Scores: The normal curve allows researchers to identify scores that are so extreme
they occur less than 5% or 1% of the time in a given population — the basis for statistical
significance thresholds.

1.3 The Binomial Distribution

The binomial distribution describes situations where each of N independent trials results in one of
exactly two mutually exclusive outcomes (called 'success' or 'failure'). Such a trial is called a Bernoulli
trial.

Conditions for Use (1) Fixed number of trials N. (2) Each trial has exactly two outcomes
(success/failure). (3) Probability of success (p) is constant across trials. (4)
Trials are independent of each other.

Formula p(X) = [N! / X!(N−X)!] × p^X × q^(N−X) where X = number of successes, p =

P(success), q = 1 − p = P(failure), N = number of trials.

Mean and Variance Mean = Np | Variance = Npq | SD = √(Npq) Example: N=10, p=0.50 → Mean =
5, SD = √2.5 = 1.58

Discrete Distribution Unlike the normal distribution (which is continuous), the binomial is discrete —
outcomes are whole numbers only (3 heads, not 3.5 heads).

Approaches Normal As N increases and when both Np > 5 and Nq > 5, the binomial distribution
approaches the normal distribution. This allows us to use z-tests as
approximations for large-N binomial problems.

Shape When p = q = 0.50 the distribution is symmetric. As p and q diverge from 0.50,
the distribution becomes increasingly skewed. Positive skew when p < 0.50;
negative skew when p > 0.50.

■ Example: If the true probability of a correct response is p = 0.50, and a participant gets 9 correct out of 10
trials, p(X ≥ 9) = 0.011. This is less than 0.05, so we reject the null hypothesis that the person is guessing
randomly.

1.4 Normal vs. Binomial Distributions — Comparison

Feature Normal Distribution Binomial Distribution

Type Continuous Discrete

Shape Bell-shaped, symmetric Symmetric only when p = 0.5

Parameters µ (mean) and σ (SD) N (trials) and p (success prob.)

Range −∞ to +∞ 0 to N (whole numbers only)

Probabilities Calculated as areas under curve Calculated directly for each X

Approaches Normal Always normal When Np > 5 and Nq > 5

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 5

 Feature Normal Distribution Binomial Distribution

Psychology use IQ, personality scores, most Yes/No data, symptom

measurement data present/absent, survival analysis

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 6

 ■ SECTION 2: Sampling Distribution and Central Limit Theorem
2.1 The Concept of Sampling Distribution
Sampling Error The expected, natural variability from sample to sample when drawing samples
from the same population. It is NOT due to mistakes — it simply reflects the fact
that different samples of the same size from the same population will give
different sample statistics. A sample mean may be 49 or 52 purely by chance
even if the true population mean is 50.

Sampling Distribution The distribution of values we would obtain for a given statistic if we drew an
(general) infinite number of samples of the same size from the same population and
calculated the statistic for each sample. Sampling distributions are almost
always derived mathematically, not empirically.

Sampling Distribution The specific distribution of all possible sample means (X■) of size n drawn from
of the Mean a population. This distribution tells us what values of X■ to expect by chance,
and how likely different values are.

Standard Error (SE) The standard deviation of a sampling distribution. For the sampling distribution
of the mean: SE = σ/√n. Reflects the typical amount by which a sample mean
deviates from the population mean µ. Larger n → smaller SE → sample means
cluster more tightly around µ.

Why We Need They allow us to evaluate whether an obtained sample statistic is likely or
Sampling Distributions unlikely to occur if the null hypothesis is true. Without sampling distributions,
statistical hypothesis testing would not be possible.

2.2 The Central Limit Theorem (CLT)

The Central Limit Theorem is one of the most important theorems in all of statistics. It provides all the
information needed about the sampling distribution of the mean.

The Central Limit Theorem (Howell, 2010)

• Given a population with mean µ and variance σ², the sampling distribution of the mean will have:

• (1) Mean equal to µ (i.e., µ_X■ = µ)

• (2) Variance equal to σ²/n (i.e., σ²_X■ = σ²/n)

• (3) Standard deviation (Standard Error) equal to σ/√n (i.e., σ_X■ = σ/√n)

• (4) A shape that approaches the NORMAL DISTRIBUTION as n increases — regardless of the shape of
the parent population.

■ Key Implications of the CLT

Shape → Normal Even if the population is NOT normally distributed (e.g., uniform, skewed), the
sampling distribution of the mean will be approximately normal for sufficiently
large n. This is why the normal distribution is so central to inferential statistics.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 7

 How Large is 'Large For symmetric distributions: approximately normal even for small n. For
Enough'? moderately skewed distributions: n ≈ 30 usually sufficient. For very skewed
distributions: may need n > 30. Rule of thumb: n ≥ 30 is 'large enough' for most
practical purposes.

If Population is Normal If the parent population is itself normal, the sampling distribution of the mean is
perfectly normal for any sample size, including n = 1.

Mean of Sampling The mean of the sampling distribution of X■ always equals the population mean
Distribution µ. Therefore, X■ is an unbiased estimator of µ — on average, sample means hit
the target.

SE Decreases with n As sample size increases, the standard error decreases (SE = σ/√n). Larger
samples produce sample means that cluster more tightly around µ and are
therefore more precise estimates of the population mean.

Standard Error of the Mean: SE = σ_X■ = σ / √n

2.3 Sampling Distribution of the Mean — When σ Is Unknown

In practice, we almost never know the population standard deviation σ. When we estimate σ using the
sample standard deviation s, the resulting test statistic follows a t distribution (not the z distribution).
This is the basis for the t-test, discussed fully in Section 4.

When σ Known Use z-test: z = (X■ − µ) / (σ/√n)

When σ Unknown Estimate σ with s; use t-test: t = (X■ − µ) / (s/√n)

Student's t Distribution Derived by William Gosset ('Student', 1908). When s is substituted for σ, the test
statistic follows a t distribution with n−1 degrees of freedom. The t distribution is
wider and heavier in the tails than z, reflecting the extra uncertainty from
estimating σ. As n → ∞, t → z.

Degrees of Freedom For one-sample t: df = n − 1. One degree of freedom is lost because we

(df) compute s using X■ (the sample mean), and the deviations must sum to zero.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 8

 ■ SECTION 3: Parameter, Statistic, and Methods of Sampling
3.1 Parameter vs. Statistic
Population The entire collection of observations, cases, or measurements that the
researcher is interested in. Usually very large or theoretically infinite. We almost
never have access to the full population.

Sample A subset of the population that is actually measured. Samples are used because
measuring entire populations is usually impractical or impossible.

Parameter A numerical characteristic of a population. Parameters are the 'true' values that
we ultimately want to know about. Denoted by Greek letters: µ (population
mean), σ² (population variance), σ (population SD), ρ (population correlation).

Statistic A numerical characteristic calculated from a sample, used to estimate the

corresponding population parameter. Denoted by Roman/Latin letters: X■
(sample mean), s² (sample variance), s (sample SD), r (sample correlation).
Because the sample is random, statistics vary from sample to sample.

Estimation The goal of inferential statistics is to use statistics to make accurate inferences
about parameters. A good statistic should be an unbiased estimator of its
parameter (e.g., X■ is an unbiased estimator of µ; s² is an unbiased estimator of
σ²).

Population (Parameter) Sample (Statistic)

Mean µ (mu) X■ (X-bar)

Variance σ² (sigma squared) s²

Standard Deviation σ (sigma) s

Correlation ρ (rho) r

Proportion π or P p or p■

Size N n

3.2 Methods of Sampling

The method by which a sample is drawn from a population determines the extent to which the sample
accurately represents the population and the extent to which findings can be generalised.

■ A. Probabilistic (Random) Sampling Methods

In probabilistic sampling, every element in the population has a known, non-zero probability of being
selected. Results are generalisable to the population.

Simple Random Every member of the population has an equal and independent chance of being
Sampling selected. Achieved by using a random number table or computer random
number generator. The gold standard for producing representative samples.
Example: choosing 50 students from a list of 500 by lottery.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 9

 Systematic Random Starting from a randomly chosen point, every k-th element is selected (where k =
Sampling population size / desired sample size). Example: selecting every 10th name from
a school register. Efficient but can be biased if there is a periodic pattern in the
population list.

Stratified Random The population is divided into mutually exclusive strata (subgroups) based on a
Sampling relevant characteristic (e.g., age, gender, diagnosis). Random samples are then
drawn independently from each stratum. Ensures that all important subgroups
are represented. Example: sampling 50 men and 50 women separately to study
gender differences.

Cluster Sampling The population is divided into naturally occurring clusters (e.g., schools,
hospitals, neighbourhoods). A random sample of clusters is selected, and all or
a random sample of members within chosen clusters are included. Efficient for
geographically dispersed populations. Example: randomly selecting 5 schools
and testing all students in those schools.

Multi-Stage Sampling A combination of sampling methods applied in stages. For example: first
randomly select states, then randomly select cities within states, then randomly
select households within cities. Common in large national surveys.

■ B. Non-Probabilistic (Non-Random) Sampling Methods

In non-probabilistic sampling, elements do not have a known or equal probability of selection. Results
may not be generalisable to the full population.

Convenience Sampling Participants are selected because they are easily accessible. Example: testing
students in one's own psychology class. Highly practical but potentially very
unrepresentative. The most common method in psychology research, despite its
limitations.

Purposive (Judgement) The researcher deliberately selects participants who they judge to represent the
Sampling population or who have specific characteristics needed for the study. Example:
selecting expert therapists for a study of treatment decision-making.

Snowball Sampling Existing participants recruit further participants from their networks. Useful for
hard-to-reach or stigmatised populations (e.g., people with rare disorders, illicit
drug users). Can produce highly biased samples.

Quota Sampling The researcher specifies quotas for different subgroups and fills them using
convenience sampling within each group. Resembles stratified sampling but
lacks randomness within strata — still non-probabilistic.

3.3 Sampling Error

Definition The difference between a sample statistic and the population parameter it is
estimating, due solely to chance variation in sampling. Sampling error = X■ − µ
(for means). It is not a mistake — it is the expected and inevitable variability from
sample to sample.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 10

 Sources Every random sample will differ somewhat from the population because it
captures only part of the population. An unrepresentative cluster of observations
in a random sample produces sampling error even when sampling procedures
are perfect.

Reducing Sampling Sampling error decreases as sample size increases (SE = σ/√n). With n → ∞,
Error the sample mean X■ → population mean µ exactly. This is why larger samples
give more precise estimates.

Sampling Error vs. Bias Sampling error is random — sometimes the sample mean is above µ,
sometimes below — and averages to zero over many samples. Bias is
systematic error that consistently pushes the estimate in one direction; it is
caused by poor sampling design, not by chance.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 11

 ■ SECTION 4: Hypothesis Testing
4.1 Logic and Framework of Hypothesis Testing
Hypothesis testing is a formal decision-making procedure for evaluating whether sample data are
consistent with a specific claim about a population. The logic: if our data would be very unlikely to occur
if the null hypothesis were true, we have reason to reject the null hypothesis.

■ Steps in Hypothesis Testing

Step 1: Research State the research hypothesis (H■) — the substantive claim you want to
Hypothesis support. Example: 'Hospitalised children have higher behaviour problem scores
than the general population.'

Step 2: Null Hypothesis State the null hypothesis (H■) — the specific, testable opposite of the research
hypothesis. It always takes a precise form. Example: H■: µ = 50 (children's
mean does not differ from the population norm).

Step 3: Sampling Construct the sampling distribution of the test statistic assuming H■ is true. This
Distribution Under H■ specifies what values of the statistic we would expect if H■ were correct.

Step 4: Collect Data Draw a random sample and calculate the relevant sample statistics.

Step 5: Compare Calculate the probability (p-value) of obtaining a test statistic as extreme as (or
Statistic to Distribution more extreme than) the one obtained, given that H■ is true.

Step 6: Decision Reject H■ if p ≤ α (significance level). Retain H■ if p > α. Note: we never 'prove'
H■ true; we merely fail to reject it.

4.2 Types of Hypotheses

Null Hypothesis (H■) The hypothesis of 'no difference' or 'no effect'. States that any observed
difference is due to sampling error alone. Example: H■: µ = 100; H■: µ■ = µ■.
Always stated precisely so we can construct the sampling distribution under it.

Alternative Hypothesis The research hypothesis — the statement we hope the data will support. In
(H■ or H■) practice, H■ is often just the negation of H■: H■: µ ≠ 100; H■: µ■ ≠ µ■. A
specific alternative hypothesis is required for power calculations.

Simple vs. Composite A simple hypothesis specifies a single population value (H■: µ = 50). A
composite hypothesis specifies a range (H■: µ > 50). H■ is usually simple
(allowing us to compute the sampling distribution); H■ is usually composite.

4.3 One-Tailed and Two-Tailed Tests

Two-Tailed The rejection region is divided equally between both tails of the sampling
(Nondirectional) Test distribution. Used when the alternative hypothesis is H■: µ ≠ µ■ (could be
higher or lower). For α = .05: reject if z ≤ −1.96 or z ≥ 1.96. Most common in
psychological research. Protects against findings in the unexpected direction.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 12

 One-Tailed The rejection region is entirely in one tail. Used only when the researcher has a
(Directional) Test strong, justified prior reason to predict the direction of the effect. H■: µ > µ■
(upper tail) or H■: µ < µ■ (lower tail). For α = .05: reject if z ≥ 1.645 (upper) or z
≤ −1.645 (lower).

Why Two-Tailed Tests (1) Researchers often cannot be certain of the direction in advance. (2)
Are Preferred Unexpected findings in the other direction would be missed. (3) Changing from
one-tailed to two-tailed after seeing the data is scientifically improper — it
inflates the true α level. Most journals expect two-tailed tests.

Critical Values at α = Two-tailed: z = ±1.96 (for z-test); t value depends on df. One-tailed: z = ±1.645
.05 (for z-test); t value depends on df.

■ Important: The choice between one-tailed and two-tailed must be made BEFORE collecting data, based on
theoretical considerations. Switching after seeing results is equivalent to running an uncontrolled Type I error
rate of 7.5%.

4.4 Level of Significance (α)

Level of Significance The predetermined probability threshold below which the p-value must fall for us
(α, alpha) to reject H■. It defines the rejection region and sets the maximum tolerable
probability of a Type I error. Most commonly α = .05.

Significance Level of We reject H■ if there is less than a 5% probability that we would obtain these
.05 data (or more extreme data) if H■ were true. Equivalently, we would make a
Type I error (falsely rejecting a true H■) only 5% of the time.

Significance Level of More conservative threshold. Reject H■ only if p ≤ .01. Reduces Type I error
.01 probability to 1% but increases Type II error probability β.

Rejection Region The area(s) in the tails of the sampling distribution where test statistic values
lead to rejection of H■. Determined by α and the number of tails.

Critical Value The boundary of the rejection region. For a two-tailed z-test at α = .05: critical
values are ±1.96. For a t-test, critical values depend on df.

p-Value The exact probability of obtaining a test statistic as extreme as the one
observed, assuming H■ is true. We reject H■ when p ≤ α.

4.5 Type I and Type II Errors

Whenever we make a decision with a statistical test, there is always a chance of error. Statistical
inference specifies the exact conditional probabilities of each type of error.

Decision Made H■ is Actually TRUE H■ is Actually FALSE

Reject H■ TYPE I ERROR (p = α) 'False positive' — CORRECT DECISION (p = 1 − β = Power)

Rejecting a true H■ 'True positive'

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 13

 Decision Made H■ is Actually TRUE H■ is Actually FALSE

Retain H■ CORRECT DECISION (p = 1 − α) 'True TYPE II ERROR (p = β) 'False negative' —

negative' Failing to reject a false H■

Type I Error (α) Rejecting H■ when it is actually true. A 'false positive.' The probability of Type I
error equals α (the significance level). Set by the researcher in advance.
Example: concluding a drug works when it actually does not.

Type II Error (β) Failing to reject H■ when it is actually false. A 'false negative.' Denoted β (beta).
Example: concluding a drug does not work when it actually does. β depends on
α, effect size, sample size, and σ².

Relationship Between Increasing α (making it easier to reject H■) decreases β and increases power —
α and β but at the cost of more Type I errors. Setting α = .05 typically produces β
somewhere between .20 and .80, depending on n and effect size. There is
always a trade-off between α and β.

Power = 1 − β The probability of correctly rejecting a false H■. A high-power test rarely makes
Type II errors. Recommended standard: power ≥ .80 (Cohen, 1988), meaning β
≤ .20.

Practical Importance Most researchers focus on minimising α but neglect β. Because conducting an
experiment that has only a 30% chance of detecting a real effect (power = .30) is
a waste of resources, power analysis should be conducted before data
collection.

■ α and β cannot both be simultaneously minimised unless sample size is increased. This is why power
analysis is conducted before running a study — to ensure adequate sample size for detecting the expected
effect at acceptable error rates.

4.6 Hypothesis Testing with One Sample — The One-Sample t-Test

Used when we have one sample and want to test whether the population mean µ from which the
sample was drawn equals a specific hypothesised value µ■. The population standard deviation σ is
unknown and must be estimated from s.

■ Assumptions
• The sample is a random sample from the population of interest.
• The dependent variable is measured on at least an interval scale.
• The population is normally distributed, OR n is large enough (≥30) for the CLT to ensure a normal
sampling distribution of X■.

■ Formula and Procedure

t Statistic t = (X■ − µ■) / (s/√n) where X■ = sample mean, µ■ = hypothesised population
mean, s = sample standard deviation, n = sample size.

Degrees of Freedom df = n − 1

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 14

 Decision Rule Find the critical value of t from the t-table for the given df and α (two-tailed or
one-tailed). Reject H■ if |t_obtained| > t_critical, or if p-value ≤ α.

Standard Error of the SE = s/√n — this is the denominator of the t-formula. It estimates how much
Mean sample means vary around µ due to sampling error alone.

t = (X■ − µ■) / (s/√n) df = n − 1

■ Worked Example
Nurcombe et al. studied Psychomotor Development Index (PDI) scores in low-birthweight (LBW)
infants. Population norm: µ = 100, σ unknown.
Sample of n = 56 LBW infants: X■ = 104.125, s = 12.584
H■: µ = 100 H■: µ ≠ 100 α = .05 (two-tailed)
t = (104.125 − 100) / (12.584/√56) = 4.125 / 1.682 = 2.45
Critical value: t.025(55 df) ≈ 2.009. Since 2.45 > 2.009, reject H■.
Conclusion: LBW infants scored significantly differently from the normative population mean on the PDI
(t(55) = 2.45, p < .05, two-tailed).

4.7 Confidence Intervals

A confidence interval (CI) provides a range of plausible values for a population parameter, based on
sample data. It conveys the precision of the estimate and is more informative than a hypothesis test
alone.

Point Estimate A single value used to estimate a population parameter. Example: X■ = 1.463
as an estimate of µ for the moon illusion study. Provides no information about
precision or uncertainty.

Confidence Interval A range of values calculated from sample data that will, with a specified
probability (confidence level), contain the true population parameter. It captures
both the estimate and its precision.

95% Confidence CI■■ = X■ ± t.025(df) × (s/√n) Interpretation: If this procedure were repeated
Interval many times (with new random samples), 95% of the intervals calculated would
contain the true µ. Example from moon illusion data: CI■■ = 1.219 ≤ µ ≤ 1.707.

99% Confidence CI■■ = X■ ± t.005(df) × (s/√n) Wider interval because we require higher
Interval certainty of capturing µ. Moon illusion example: CI■■ = 1.112 ≤ µ ≤ 1.814.

Link to Hypothesis If the hypothesised µ■ falls outside the 95% CI, then a two-tailed test at α = .05
Testing would reject H■: µ = µ■. The CI inverts the hypothesis test to show all values of
µ■ that would not be rejected. A CI includes far more information than a single
hypothesis test.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 15

 Correct Interpretation The CORRECT interpretation: 'We place 95% confidence in the PROCEDURE
— i.e., in the long run, 95% of intervals constructed this way will contain µ.'
INCORRECT: 'There is a 95% probability that µ lies between 1.219 and 1.707.'
(µ is fixed; it either is or is not in the interval — probability is 0 or 1).

CI■■α = X■ ± t_{α/2}(df) × (s/√n)

■ Wider confidence intervals indicate less precision (more uncertainty). Precision increases with larger n and
smaller s. The width of a CI is: 2 × t_{α/2} × s/√n.

4.8 Hypothesis Testing with Two Samples — Independent Samples t-Test

Used when comparing the means of two independent (unrelated) groups. 'Independent' means that
being in one group provides no information about the other group — scores are not paired or matched.

■ Sampling Distribution of Differences Between Means

The Question When we observe a difference between two sample means (X■■ − X■■), is
this difference large enough to conclude that the population means differ, or is it
merely sampling error?

Variance Sum Law When two variables are independent, the variance of their sum OR difference
equals the SUM of their variances: σ²(X■■ − X■■) = σ²■/n■ + σ²■/n■

Sampling Distribution The sampling distribution of (X■■ − X■■) is approximately normal for
Shape reasonable sample sizes (by CLT). Its mean = µ■ − µ■; under H■, mean = 0.

■ The t Statistic for Independent Samples

Pooled Variance (s²_p) When n■ ≠ n■, we pool the two sample variances, weighting each by its
degrees of freedom: s²_p = [(n■-1)s²■ + (n■-1)s²■] / (n■+n■-2)

t Formula t = (X■■ − X■■) / √[s²_p(1/n■ + 1/n■)] Under H■: µ■ − µ■ = 0 (this term drops
out of the formula).

Degrees of Freedom df = n■ + n■ − 2

Assumption of Equal Pooling assumes σ²■ = σ²■ (homogeneity of variance). If equal, s²_p is a better
Variances estimate than either sample variance alone. If variances are unequal (use
Levene's test), use Welch's correction which adjusts the df.

t = (X■■ − X■■) / √[s²_p(1/n■ + 1/n■)] df = n■ + n■ − 2

■ Assumptions
• Both samples are independently and randomly drawn.
• The dependent variable is measured on at least an interval scale.
• Both populations are normally distributed (or n is large enough).

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 16

 • Both populations have equal variances — σ²■ = σ²■ (homogeneity of variance). Robustly holds
when sample sizes are approximately equal.

■ Worked Example — Adams et al. (1996)

Homophobic (n=35) and non-homophobic (n=29) heterosexual males were measured on sexual
arousal to a homosexual video.
Homophobic: X■■ = 24.00, s²■ = 148.87 Non-homophobic: X■■ = 16.50, s²■ = 139.16
H■: µ■ = µ■ H■: µ■ > µ■ α = .05 (one-tailed)
s²_p = [(34)(148.87) + (28)(139.16)] / (35+29−2)
= [5061.58 + 3896.48] / 62 = 8958.06/62 = 144.49
t = (24.00 − 16.50) / √[144.49(1/35 + 1/29)]
= 7.50 / √[144.49 × 0.0629] = 7.50 / 3.014 ≈ 2.49
Critical value t.05(62 df) ≈ 1.67. Since 2.49 > 1.67, reject H■.
Conclusion: Homophobic men showed significantly greater arousal to the homosexual video than
non-homophobic men (t(62) = 2.49, p < .05, one-tailed).

■ Confidence Interval for Two Independent Means

CI■■α = (X■■ − X■■) ± t_{α/2}(df) × √[s²_p(1/n■ + 1/n■)]

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 17

 ■ SECTION 5: Power of a Test
5.1 Definition of Power
Power The probability of correctly rejecting a false null hypothesis when a specific
alternative hypothesis H■ is true. Power = 1 − β (where β is the Type II error
probability). A high-power test rarely misses a real effect.

Why Power Matters Most researchers focus on controlling Type I errors (α) but neglect Type II
errors. An experiment with power = .30 has only a 30% chance of detecting a
real effect — a costly waste of time and resources. Cohen recommended a
minimum power of .80 for psychological research.

Power = 1 − β If β = .20 (acceptable standard), then power = .80. This means that when H■ is
false (a real effect exists), we correctly reject H■ 80% of the time and make a
Type II error only 20% of the time.

5.2 Factors Affecting Power

1. Alpha Level (α) Increasing α (e.g., from .01 to .05) moves the critical value closer to the centre of
the H■ distribution, allowing more of the H■ distribution to exceed it → power
increases. But Type I error probability also increases. Setting α higher is rarely
acceptable; increasing n is better.

2. Effect Size (d) The larger the true difference between µ■ and µ■ (relative to σ), the further
apart the H■ and H■ distributions are, and the less overlap there is → greater
power. Power depends on the ACTUAL magnitude of the effect, which the
researcher cannot directly control.

3. Sample Size (n) Increasing n decreases the standard error (σ/√n), which shrinks both sampling
distributions → less overlap → greater power. This is the most practical way to
increase power. Doubling n does not double power but increasing n from 10 to
100 dramatically increases power.

4. Population Variance A smaller σ² produces a smaller standard error → smaller, less overlapping
(σ²) sampling distributions → greater power. Researchers can sometimes reduce σ²
by using more precise measurement instruments or tightly controlled laboratory
conditions.

5. The Statistical Test Parametric tests (t, F) are generally more powerful than non-parametric
Used equivalents when their assumptions are met. When assumptions are violated,
non-parametric or resampling tests may be more powerful.

5.3 Effect Size (Cohen's d)

Effect size (d) is a standardised measure of the magnitude of an effect, independent of sample size. It
separates the 'size of the phenomenon' from 'the precision of the estimate.'

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 18

 Formula (one-sample) d = (µ■ − µ■) / σ The difference between the true population mean (H■) and the
null hypothesis mean (H■), divided by the population standard deviation.
Analogous to a z-score for the separation between distributions.

Formula (two-sample) d = (µ■ − µ■) / σ (when equal variances are assumed)

Estimating d Three methods: (1) Prior research — use means and SDs from published
studies. (2) Practical significance — decide the minimum difference that would
be scientifically meaningful. (3) Cohen's conventions.

Cohen's (1988) Conventions for Effect Size d

• Small effect: d = 0.20 (85% overlap between H■ and H■ distributions)

• Medium effect: d = 0.50 (67% overlap between H■ and H■ distributions)

• Large effect: d = 0.80 (53% overlap between H■ and H■ distributions)

• Note: Use these conventions only as a last resort when no prior knowledge is available. Using them
mindlessly instead of estimating d from the substantive context simply replaces one form of thoughtlessness
with another (Thompson, 2000).

5.4 Combining Effect Size and Sample Size: Delta (δ)

Delta (δ) Combines the effect size (d) with the sample size (n) to produce the
noncentrality parameter used to look up power in a table: δ = d × √n (for
one-sample t-test).

How to Use (1) Estimate d from prior research or practical importance. (2) Calculate δ = d ×
√n. (3) Look up power for the obtained δ and chosen α in a power table.

Solving for Required n To find n needed for a desired power level: (1) Specify desired power (e.g., .80)
and α (e.g., .05). (2) From the power table, find the required δ. (3) Solve: n =
(δ/d)²

δ = d × √n (one-sample t-test)

■ Example: A clinical psychologist expects d = 0.33 (IQ difference of 5 points, σ = 15). For n = 25: δ = 0.33 ×
√25 = 1.65. Power table gives power ≈ .38. For power = .80 with α = .05, required δ = 2.80, so n = (2.80/0.33)² ≈
72.

5.5 Power Analysis in Practice

• A priori power analysis: Conducted BEFORE data collection to determine the required sample size
for a desired level of power. The correct approach.
• Post hoc (retrospective) power analysis: Conducted after a non-significant result to estimate the
power of the completed study. Controversial — primarily useful for interpreting null results.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 19

 • G*Power software: A free program that automates power calculations for a wide range of statistical
tests (t-tests, ANOVA, regression, chi-square). Recommended for accurate power analysis.
• Recommended minimum power: 0.80 (Cohen, 1988). This makes Type II error 4 times as likely as
Type I error at α = .05, which many regard as a reasonable relative weighting of the two error
types.

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 20

 ■ MODULE II — SUMMARY: Key Hypothesis Tests
Test When to Use Formula df Key Assumption

One-sample z σ known; testing X■ z = (X■ − µ■) / N/A Population normal

vs. µ■ (σ/√n) OR n ≥ 30

One-sample t σ unknown; testing t = (X■ − µ■) / (s/√n) n−1 Population

X■ vs. µ■ approximately
normal OR n ≥ 30

Independent t Comparing means of t = (X■■−X■■) / n■+n■−2 Normal populations;

2 unrelated groups √[s²_p(1/n■+1/n■)] equal variances
(σ²■=σ²■)

■ QUICK REVISION — Key Terms and Definitions

Key Term Definition

Normal Distribution Bell-shaped, symmetric, continuous distribution defined by µ and σ. Area

under curve = 1.0; ~68% within ±1σ, ~95% within ±2σ.

Standard Normal (z) Normal distribution with µ=0 and σ=1. All normal distributions can be
converted to z using z = (X−µ)/σ.

Binomial Distribution Discrete distribution for number of successes in N independent Bernoulli

trials: p(X) = [N!/X!(N−X)!] × p^X × q^(N−X).

Sampling Distribution Distribution of all possible values of a statistic computed from an infinite
number of same-sized samples from the same population.

Standard Error (SE) Standard deviation of a sampling distribution. For means: SE = σ/√n.

Central Limit Theorem As n increases, the sampling distribution of X■ approaches normal with
mean µ and variance σ²/n, regardless of the population shape.

Parameter A numerical characteristic of a population (µ, σ², ρ). The 'true' value that
statistics are used to estimate.

Statistic A numerical value calculated from a sample (X■, s², r) used to estimate the
corresponding population parameter.

Sampling Error Natural, chance variation in a statistic from sample to sample. Not a
mistake; decreases as n increases.

Simple Random Sampling Every population member has an equal, independent chance of selection.
Gold standard for producing representative samples.

Null Hypothesis (H■) Hypothesis of no difference/effect. Always stated precisely so the sampling
distribution can be constructed under it.

Alternative Hypothesis (H■) Research hypothesis; the substantive claim the researcher hopes to
support.

Type I Error (α) Rejecting H■ when it is true. Probability = α (significance level). 'False
positive.'

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 21

 Key Term Definition

Type II Error (β) Failing to reject H■ when it is false. 'False negative.' Probability = β; related
to power by: Power = 1 − β.

Power Probability of correctly rejecting a false H■. Power = 1 − β. Recommended

minimum: .80 (Cohen, 1988).

One-Sample t-Test Tests whether a single sample mean equals a specified population mean
µ■ when σ is unknown. t = (X■ − µ■) / (s/√n), df = n−1.

Independent Samples t-Test Tests whether means of two unrelated groups differ. t = (X■■ − X■■) /
√[s²_p(1/n■ + 1/n■)], df = n■+n■−2.

Confidence Interval Range of plausible values for a parameter. CI■■ = X■ ± t.025 × s/√n. '95%
of intervals calculated this way contain µ.'

Level of Significance (α) Pre-set probability threshold for rejecting H■. Typically .05 or .01. Defines
the maximum acceptable Type I error rate.

Effect Size (d) Cohen's d = (µ■ − µ■) / σ. Standardised measure of effect magnitude
independent of sample size. Small=.20; Medium=.50; Large=.80.

One-Tailed Test Rejection region in one tail only; used for directional hypotheses. Must be
decided before data collection.

Two-Tailed Test Rejection region split between both tails; tests H■: µ ≠ µ■. Most commonly
used in psychological research.

Degrees of Freedom Number of independent pieces of information used to estimate a

parameter. One-sample t: df = n−1. Independent t: df = n■+n■−2.

Pooled Variance Weighted average of two sample variances: s²_p = [(n■-1)s²■ + (n■-1)s²■]
/ (n■+n■−2). Used when σ²■ = σ²■.

— End of Module II Notes —

Source: Statistical Methods for Psychology, 7th Ed. — David C. Howell | Osmania University M.A. Psychology AY 2025-26 |
Course Code-201

M.A. Psychology | Statistics in Psychology | Module II: Inferential Statistics Page 22