Skip to main content

Inferential Statistics, Estimation, and Hypothesis Testing

Introduction​

Descriptive statistics answers the question: "What does this data look like?" But science rarely stops there. The deeper question is: "Does what I observed in my sample tell me something true about the wider population?" This leap from sample to population is the territory of inferential statistics.

This file covers the entire inferential pathway: estimation (point and interval), hypothesis testing, the logic of null and alternative hypotheses, significance levels, one-tailed and two-tailed tests, Type I and Type II errors, statistical power, and the step-by-step procedure for testing hypotheses.

📄 Source PDF: Block-1/Unit-2.pdf — Pages 29–35
📚 MPC-006 Statistics in Psychology, IGNOU

What is Inferential Statistics?​

Statistical inference is the process of drawing conclusions about an unknown population from a known sample drawn from it.

In most psychological investigations, studying the entire population is impossible or impractical. A researcher studying depression in Indian adults cannot test every Indian adult. Instead, a representative sample is tested, and conclusions are generalised.

Classic example from the IGNOU text: Before Lok Sabha or Vidhan Sabha election results, media houses conduct exit polls — testing a small sample of voters and inferring the likely outcome for the entire electorate. This is inferential statistics in everyday action.

Two Types of Inferential Procedures​

ProcedureQuestionOutput
EstimationWhat is the population value?A number or range
Hypothesis TestingIs a specific hypothesis about the population supported?Reject or retain H₀

Estimation​

In estimation, inference is made about population characteristics (parameters) on the basis of sample characteristics (statistics).

Characteristics of a Good Estimator​

A good statistical estimator must be:

PropertyMeaningExample
UnbiasedMean of the statistic equals the parameter across infinite samplesSample mean is an unbiased estimator of population mean
ConsistentAs sample size increases, estimate gets closer to parameterMean from N=200 is closer to true value than from N=20
AccurateThe estimate is close to the true parameter valueDepends on sample representativeness, N, and variability

Point Estimation​

A point estimate uses a single sample statistic to estimate the population parameter:

μ^=xˉ\hat{\mu} = \bar{x}

Example: A researcher tests 100 students on an adjustment scale and obtains xˉ=45.0\bar{x} = 45.0. This single value is used as the point estimate of the population mean.

Limitation: A point estimate is almost certainly wrong — the sample mean will rarely equal the exact population mean. There is no information about how far off the estimate might be.

Interval Estimation (Confidence Interval)​

To address the limitation of point estimation, we construct a confidence interval (CI): a range of values expected to contain the true population parameter.

CI=xˉ±z⋅σNCI = \bar{x} \pm z \cdot \frac{\sigma}{\sqrt{N}}

Key concepts:

  • Confidence limits: The lower and upper bounds of the interval
  • Confidence level: The probability that the interval contains the true parameter (commonly 95% or 99%)
  • Interpretation: A 95% CI means that if we drew 100 samples and constructed 100 intervals, approximately 95 of them would contain the true population mean

🔗 Further Reading:

Hypothesis Testing​

Hypothesis testing is the statistical decision-making process for evaluating whether an observed result is due to the variable of interest or merely to chance.

The key question: Could random sampling alone explain what I observed?

Statement of Hypotheses​

A statistical hypothesis is a statement — which may or may not be true — about a population parameter or probability distribution.

Null Hypothesis (H₀)​

The null hypothesis is a statement of no difference (or no effect, no relationship).

Common formulations:

  • "The two samples come from the same population"
  • "There is no significant difference between group means"
  • "The correlation is zero in the population"

H₀ is testable — it can be accepted or rejected based on data. It assumes the population is normally distributed with equal means and standard deviations across groups.

Alternative Hypothesis (H₁)​

The alternative hypothesis proposes a difference (or effect, or relationship). It states that:

  1. The two samples belong to different populations
  2. The means are estimates of two different parametric means
  3. There is a significant difference between sample means

Critical logic: H₁ is never directly tested. It is accepted only when H₀ is rejected.

Level of Significance (α)​

The level of significance (α) is the probability threshold below which the null hypothesis is rejected.

LevelMeaningUse
α = .05 (5%)Results expected by chance in only 5 out of 100 similar experimentsStandard for most psychological research
α = .01 (1%)Results expected by chance in only 1 out of 100 experimentsMore stringent; for serious consequences of error

Interpreting p < .05: The null hypothesis is rejected. The results are statistically significant. The probability that the observed difference occurred by chance alone is 5% or less.

One-Tailed and Two-Tailed Tests​

TestH₁ SpecificationUse When
Two-tailedμ₁ ≠ μ₂ (non-directional)You predict a difference but not its direction
One-tailedμ₁ > μ₂ or μ₁ < μ₂ (directional)You predict the specific direction of the effect

One-tailed tests are more powerful but require stronger theoretical justification for directionality.

🔗 Further Reading:

Errors in Hypothesis Testing​

No decision-making procedure is perfect. When deciding to accept or reject H₀ based on sample data, two types of error are possible:

RealityDecision: Retain H₀Decision: Reject H₀
H₀ is TRUE✅ Correct Decision❌ Type I Error (α) — False Positive
H₀ is FALSE❌ Type II Error (β) — False Negative✅ Correct Decision — POWER

Type I Error (α Error — False Positive)​

Definition: Rejecting H₀ when it is actually true.
Probability: α (the significance level set by the researcher)
Example: Concluding a new therapy is effective when it actually isn't
Consequence: Adopting ineffective treatments; false scientific claims

Type II Error (β Error — False Negative)​

Definition: Retaining H₀ when it is actually false.
Probability: β
Example: Concluding a therapy has no effect when it actually does
Consequence: Abandoning effective treatments; missed discoveries

The α–β Trade-off​

For a fixed sample size, α and β are inversely related:

ChangeEffect on αEffect on β
Lower α (more stringent)DecreasesIncreases
Higher α (more lenient)IncreasesDecreases
Increase sample size NCan decrease both simultaneously

Key principle: You cannot simultaneously minimise both errors with a fixed sample size. The only solution is to increase N.

Power of a Test (1 − β)​

Statistical power = the probability of correctly rejecting H₀ when it is false = 1−β1 - \beta

Power measures how well the test detects real effects. Factors that increase power:

  • Larger sample size (N)
  • Larger effect size (bigger true difference between groups)
  • Higher α level (but this increases Type I error risk)
  • Lower measurement error

Convention: Power ≥ 0.80 (80% chance of detecting a real effect) is the standard target in psychology.

🔗 Further Reading:

General Procedure for Testing a Hypothesis​

The IGNOU text provides a six-step systematic procedure:

Applied Example​

A researcher tests whether a new mindfulness programme reduces anxiety in college students:

StepContent
H₀The programme has no effect on anxiety — μ_pre = μ_post
H₁The programme reduces anxiety — μ_pre > μ_post (one-tailed)
Test statisticPaired-samples t-test
dfN − 1
p valueObtained from t-distribution table
DecisionIf p < .05, reject H₀; mindfulness significantly reduces anxiety

Indian Context: Hypothesis Testing in Indian Psychological Research​

The Indian Council of Social Science Research (ICSSR) and University Grants Commission (UGC) fund numerous studies that follow the hypothesis-testing framework. Common applications include:

  • Testing whether rural vs urban populations differ on mental health indices (two-sample t-tests)
  • Examining gender differences in stress among IT professionals (ANOVA)
  • Evaluating the effectiveness of yoga interventions on psychological well-being (paired t-tests)

Understanding the hypothesis-testing framework is essential for interpreting research published in Indian journals like Psychological Studies (Springer India) and the Journal of the Indian Academy of Applied Psychology.

🔗 Further Reading:

Recent Research Spotlight​

Rethinking Statistical Significance (2019–2024)

  • Wasserstein, R. L. et al. (2019). Moving to a World Beyond "p < 0.05." The American Statistician, 73(S1), 1–19. Argues for richer reporting beyond binary significance testing. DOI: 10.1080/00031305.2019.1583913

  • Lakens, D. (2022). Improving Your Statistical Inferences. Free online course via Coursera covering power analysis, confidence intervals, and effect sizes as complements to traditional hypothesis testing.

  • Benjamin, D. J. et al. (2018). Redefine statistical significance. Nature Human Behaviour, 2(1), 6–10. Proposed lowering the α threshold from .05 to .005 to reduce false positives in psychological research. DOI: 10.1038/s41562-017-0189-z

Self-Assessment Questions​

Level 1 — Knowledge​

  1. Fill in the blanks:

    • a) Alternative hypothesis is a statement of __________ difference.
    • b) Null hypothesis is denoted by __________.
    • c) Alternative hypothesis is __________ directly tested statistically.
    • d) Level of significance is the probability of __________ occurrence of observed results.
    • e) When the null hypothesis is true, a decision to reject it is known as __________.
    • f) When a null hypothesis is false, a decision to accept it is known as __________.

  2. What are the three characteristics of a good estimator?

  3. Distinguish between point estimation and interval estimation.

Level 2 — Comprehension​

  1. A researcher sets α = .01 instead of α = .05. What effect does this have on (a) Type I error probability, (b) Type II error probability, and (c) power of the test?

  2. A study finds p = .03. Should H₀ be rejected at α = .05? At α = .01? Explain what each decision means.

Level 3 — Application/Analysis​

  1. Research Design: A clinical psychologist is testing a new cognitive-behavioural intervention for social anxiety. The consequences of claiming effectiveness when the therapy doesn't work are serious — but missing a genuinely effective treatment is also costly. How should these considerations influence the researcher's choice of α level, sample size, and which error type they most want to minimise?

Memory Aids​

Mnemonic 1: Type I vs Type II Error​

"Alpha Alarm, Beta Blind"

  • Alpha = Type I = False Alarm (rejected a true H₀ — cried wolf)
  • Beta = Type II = Blind miss (failed to detect a real effect)

Mnemonic 2: The α–β Trade-off​

"Squeeze one, the other grows" — for fixed N, reducing α inflates β; the only way to reduce both is to increase sample size.

Mnemonic 3: Six Steps of Hypothesis Testing​

"State → Define → Calculate → Determine → Find → Decide"

Mnemonic 4: Point vs Interval Estimation​

  • Point: One sharp dart at the target — precise but probably misses the bullseye
  • Interval: A net cast around the target — wider but much more likely to catch the true value

Key Terms Glossary​

TermDefinition
Statistical inferenceProcess of drawing conclusions about a population from a sample
EstimationInferring population parameter value from sample statistic
Point estimationEstimating a parameter as a single value
Interval estimationEstimating a parameter as a range with a stated confidence level
Confidence intervalRange expected to contain the true population parameter
Confidence limitsLower and upper bounds of a confidence interval
Statistical hypothesisA testable statement about a population parameter
Null hypothesis (H₀)Statement of no difference; tentatively held true
Alternative hypothesis (H₁)Statement of a significant difference; accepted when H₀ is rejected
Level of significance (α)Probability threshold for rejecting H₀
One-tailed testDirectional test; H₁ specifies direction of effect
Two-tailed testNon-directional test; H₁ specifies only that a difference exists
Type I error (α)Rejecting H₀ when it is true — false positive
Type II error (β)Retaining H₀ when it is false — false negative
Power (1 − β)Probability of correctly rejecting a false H₀
ParameterCharacteristic of a population
StatisticCharacteristic of a sample

Video Resources​

Cross-References​