Inferential Statistics Hypothesis Testing Confidence Intervals 32 min read July 30, 2026
BY: Statistics Fundamentals Team
Reviewed By: Minsa A (Senior Statistics Editor)

Inferential Statistics: Definition, Methods, Examples & Applications

A clinical researcher tests whether a new drug lowers blood pressure in all patients — using data from 200. A pollster predicts an election outcome from 1,000 responses. A manufacturer decides whether an entire production batch meets quality standards by inspecting 50 units. All three are doing the same thing: drawing conclusions about a population from a sample. That is inferential statistics.

This guide covers the complete inferential statistics framework — from the foundational logic of sampling and estimation to hypothesis testing, confidence intervals, choosing the right statistical test, and interpreting results. Worked examples, comparison tables, a test-selection decision guide, and an interactive calculator are included throughout.

What You'll Learn
  • ✓ The precise definition of inferential statistics and how it differs from descriptive statistics
  • ✓ Population vs. sample, and parameters vs. statistics
  • ✓ The role of sampling distributions and the Central Limit Theorem
  • ✓ Estimation: point estimates and confidence intervals with worked examples
  • ✓ Hypothesis testing: the full procedure, p-values, and decision rules
  • ✓ Every major inferential test: z-test, t-test, chi-square, ANOVA, regression, correlation
  • ✓ A decision guide for choosing the right test based on your data and research question
  • ✓ Real-world applications in medicine, business, psychology, and data science

What Is Inferential Statistics?

Definition — Inferential Statistics
Inferential statistics is the branch of statistics that uses probability theory to draw conclusions about a population based on data collected from a sample. It goes beyond describing the data at hand — it generalizes from the observed sample to the unobserved population.
Inference: Sample Data → Population Conclusions

The word "infer" means to derive a conclusion from evidence. In statistics, that evidence is a sample — a subset of the population you actually measure. Because measuring every individual in a population is usually impossible (or prohibitively expensive), inferential statistics gives researchers a rigorous, probability-based method for making general claims from limited data.

Inferential statistics stands on two main legs: estimation (using a sample to estimate population parameters like the mean or proportion) and hypothesis testing (using a sample to decide whether a specific claim about the population is supported by the data). Both methods produce results that come with a quantified level of uncertainty — expressed as confidence levels, p-values, or margins of error.

The mathematical foundation comes from probability theory and the behavior of sampling distributions. Researchers at the Pew Research Center use these tools to generalize from a few thousand survey responses to the views of entire national populations. The same logic appears in clinical trials, A/B testing, quality control, and machine learning validation. For the underlying mathematical framework, the NIST Statistical Engineering Division maintains authoritative reference material.

⚡ Quick Reference — Key Inferential Statistics Concepts
  • Population: The complete set of individuals or observations you want to draw conclusions about
  • Sample: A subset of the population, actually measured or observed
  • Parameter: A numerical characteristic of the population (e.g., population mean μ) — usually unknown
  • Statistic: A numerical characteristic calculated from the sample (e.g., sample mean x̄) — used to estimate parameters
  • Sampling error: The natural difference between a sample statistic and the population parameter it estimates
  • Inference: A probability-based conclusion about the population drawn from sample data

Inferential vs. Descriptive Statistics

The distinction between these two branches is one of the most frequently tested concepts in introductory statistics. Both use the same data, but they answer different questions. The descriptive statistics section of Statistics Fundamentals covers summarizing data in detail; this page focuses on what comes next.

Feature Descriptive Statistics Inferential Statistics
PurposeSummarize the data you haveGeneralize beyond the data you have
ScopeDescribes only the sample (or population if fully measured)Makes claims about the population from a sample
Key toolsMean, median, mode, SD, charts, frequency tablesConfidence intervals, p-values, hypothesis tests
UncertaintyNone — it describes exactly what is in the dataAlways present — expressed through probability
Question answered"What does my data look like?""What does my data tell me about the population?"
ExampleThe average score in this class is 74.2The average score for all students nationwide is likely between 71 and 77 (95% CI)

Descriptive and inferential statistics work together. You always start with description — compute the sample mean, check the distribution, identify outliers — and then use inferential methods to generalize. Skipping the descriptive step before running inferential tests is one of the most common mistakes in applied statistics.

Population, Sample, Parameters, and Statistics

Inferential statistics rests on a precise distinction between what you can measure and what you want to know. Understanding this four-way vocabulary is necessary before any inferential procedure makes sense.

Population vs. Sample

A population is the entire group you want to study — every adult in a country, every unit produced by a factory, every patient with a specific diagnosis. A sample is the subset you actually observe. The key requirement: the sample must be drawn randomly so that every member of the population has a known probability of being selected. Non-random samples produce biased estimates that inferential methods cannot correct. The full treatment of this distinction is covered on the population vs. sample page.

Parameters vs. Statistics

A parameter is a fixed (but usually unknown) numerical characteristic of the population. The population mean is written μ (mu); the population standard deviation is σ (sigma); the population proportion is π or p. A statistic is computed from the sample and used to estimate the corresponding parameter. Sample mean: x̄; sample standard deviation: s; sample proportion: p̂.

Concept Population (Parameter) Sample (Statistic)
Meanμ (mu)x̄ (x-bar)
Standard deviationσ (sigma)s
Proportionπ or pp̂ (p-hat)
Varianceσ²
Correlationρ (rho)r
Regression slopeβ (beta)b

This notation matters because statistics are calculated from data (they are known), while parameters are what the statistics estimate (they are unknown). All inferential methods are ultimately about using statistics to draw conclusions about parameters. For a full breakdown of this distinction, see the statistic vs. parameter page.

Sampling Distributions and the Central Limit Theorem

The mathematical bridge between a sample and a population is the sampling distribution — the probability distribution of a statistic (like the sample mean) across all possible samples of the same size from the same population. You never actually collect all possible samples; the sampling distribution is a theoretical construct that describes how sample statistics would vary if you did.

Standard Error

The standard deviation of the sampling distribution of the mean is called the standard error (SE). It measures how much sample means typically vary from the population mean:

Standard Error of the Mean
SE = σ / √n
σ = population standard deviation n = sample size SE = standard error of the mean

When σ is unknown (the typical case), we estimate SE using the sample standard deviation: SE = s / √n. A larger sample produces a smaller standard error — meaning sample means cluster closer to the true population mean, which is why larger samples yield more precise estimates. The full derivation and examples are on the sampling distributions page.

The Central Limit Theorem (CLT)

The Central Limit Theorem is arguably the most important result in applied statistics. It states: regardless of the shape of the population distribution, the sampling distribution of the sample mean approaches a normal distribution as sample size n increases, with mean μ and standard deviation σ/√n.

📐
Central Limit Theorem — The Practical Rule

For most populations, n ≥ 30 is sufficient for the sampling distribution of x̄ to be approximately normal. For strongly skewed populations or those with heavy tails, n ≥ 50 or more may be needed. For normally distributed populations, the CLT applies at any sample size.

Why does this matter? Because virtually all classical inferential procedures — z-tests, t-tests, confidence intervals — assume the sampling distribution of the statistic is approximately normal. The CLT is what makes those procedures valid even when the raw data are not normally distributed. For deeper coverage, see the Central Limit Theorem page.

Estimation: Point Estimates and Confidence Intervals

Estimation uses sample data to assign values to unknown population parameters. There are two forms: a point estimate (a single number) and an interval estimate (a range of plausible values).

Point Estimates

A point estimate is a single value computed from the sample that serves as the best single guess for the population parameter. The sample mean x̄ is the point estimate for μ; the sample proportion p̂ estimates the population proportion π. Point estimates are useful but incomplete — they give no indication of how precise the estimate is. That's where confidence intervals come in.

Confidence Intervals

A confidence interval (CI) is an interval computed from sample data that is likely to contain the true population parameter. A 95% CI means: if you repeated the sampling process many times and built a CI each time, 95% of those intervals would contain the true parameter.

Confidence Interval for a Population Mean
CI = x̄ ± z* × (σ / √n)
= sample mean z* = critical z-value for confidence level σ = population SD (use s when unknown) n = sample size

The critical value z* depends on the confidence level: 1.645 for 90%, 1.960 for 95%, and 2.576 for 99%. When σ is unknown and n is small, replace z* with the appropriate t* from the t-distribution table.

Worked Example — 95% Confidence Interval

A hospital samples 64 patients and records a mean recovery time of x̄ = 8.4 days, with a sample standard deviation of s = 2.1 days. Construct a 95% confidence interval for the population mean recovery time.

1

Identify values: x̄ = 8.4, s = 2.1, n = 64, confidence level = 95% → z* = 1.960

2

Calculate SE: SE = s / √n = 2.1 / √64 = 2.1 / 8 = 0.2625

3

Calculate margin of error: ME = z* × SE = 1.960 × 0.2625 = 0.515

4

Build the interval: CI = x̄ ± ME = 8.4 ± 0.515 → (7.885, 8.915)

✅ Conclusion: We are 95% confident that the true mean recovery time for the population lies between 7.89 and 8.92 days. The margin of error is ±0.52 days.

Confidence interval methodology follows Neyman, J. (1937). "Outline of a Theory of Statistical Estimation Based on the Classical Theory of Probability." Philosophical Transactions of the Royal Society A, 236, 333–380. Also see the NIST Confidence Interval reference.

For a more detailed treatment — including intervals for proportions, differences between means, and the relationship between confidence intervals and hypothesis tests — see the confidence intervals guide and the confidence interval for the mean page.

Hypothesis Testing: The Core Framework

Hypothesis testing is the other major branch of inferential statistics. Where estimation asks "what is the population parameter?", hypothesis testing asks "is a specific claim about the population supported by the data?" The answer is always probabilistic — never certain.

📋
One-Sentence Definition

Hypothesis testing is a formal procedure that uses sample data to decide whether there is enough evidence to reject a default assumption (the null hypothesis) in favor of an alternative claim.

The 5-Step Procedure

1

State the Hypotheses

Write the null hypothesis (H₀) — the default claim, typically "no effect" or "equals a specific value" — and the alternative hypothesis (H₁ or Hₐ) — what you're trying to show. Example: H₀: μ = 100 vs. H₁: μ ≠ 100 (two-tailed) or H₁: μ > 100 (one-tailed right).

2

Set the Significance Level (α)

α is the maximum probability of rejecting H₀ when it is actually true (Type I error rate). The conventional choice is α = 0.05 (5%). Medical and pharmaceutical research often uses α = 0.01; exploratory research may use α = 0.10. Set α before collecting data.

3

Calculate the Test Statistic

Convert the sample data into a single number (z, t, F, or χ²) that measures how far the observed result is from what H₀ predicts, measured in units of standard error. A larger absolute value means the data is further from the null hypothesis prediction.

4

Find the p-value

The p-value is the probability of observing a test statistic as extreme as the one calculated, assuming H₀ is true. Use statistical tables or software. For a z-test, use the z-table; for a t-test, use the t-distribution table.

5

Make a Decision and State the Conclusion

If p < α: reject H₀ — the result is statistically significant. If p ≥ α: fail to reject H₀ — insufficient evidence. Always write the conclusion in plain language referencing the original research question. Never write "we accept H₀."

p-values, Significance, and What They Mean

The p-value is one of the most misunderstood numbers in statistics. What it is: the probability of getting your sample result (or a more extreme one) if the null hypothesis were true. What it is not: the probability that H₀ is true, the probability that your result occurred by chance, or a measure of how important the effect is. Statistical significance only means the result is unlikely under H₀ — not that it is practically meaningful. For a dedicated treatment, see the p-values page.

Type I and Type II Errors

Decision H₀ Is Actually True H₀ Is Actually False
Reject H₀ Type I Error (α) Correct Decision (Power)
Fail to Reject H₀ Correct Decision (1 − α) Type II Error (β)

A Type I error is rejecting H₀ when it is true — a false positive. Its rate is α. A Type II error is failing to reject H₀ when it is actually false — a false negative. Its rate is β. The power of a test (1 − β) is the probability of correctly detecting a real effect. Power increases with larger sample sizes, larger effect sizes, and higher α values. For more, see the Type I and Type II errors page and the statistical power guide.

Common Inferential Statistical Tests

Different research questions and data types require different inferential tests. The table below summarizes the most commonly used tests. Use the decision guide in the next section to select the right one for your situation. For the full methodology behind selecting tests, see the statistical test selector.

Test What It Tests When to Use Key Statistic
One-Sample Z-TestWhether a population mean equals a specified valueσ known; n ≥ 30z = (x̄ − μ₀) / (σ/√n)
One-Sample T-TestWhether a population mean equals a specified valueσ unknown; any nt = (x̄ − μ₀) / (s/√n); df = n−1
Two-Sample T-TestWhether two population means differTwo independent groups; σ unknownt = (x̄₁ − x̄₂) / SE_pooled
Paired T-TestWhether means differ across matched pairsBefore/after; matched pairst = d̄ / (s_d/√n)
Chi-Square TestWhether observed and expected frequencies differCategorical data; counts in cellsχ² = Σ (O−E)² / E
One-Way ANOVAWhether 3+ group means are equalOne categorical predictor; continuous outcomeF = MS_between / MS_within
Pearson CorrelationStrength of linear relationship between two continuous variablesBoth variables continuous; linear relationshipr ranges from −1 to +1
Simple Linear RegressionWhether one continuous variable predicts anotherOne predictor; continuous outcomeŷ = b₀ + b₁x
Mann-Whitney UWhether two groups differ in rank distributionNon-parametric alternative to independent t-testU statistic
Wilcoxon Signed-RankWhether paired differences are symmetric around zeroNon-parametric alternative to paired t-testW statistic

Z-Test

The one-sample z-test is used when you want to know whether a sample mean differs from a hypothesized population mean and when the population standard deviation σ is known. In practice, σ is rarely known, so the z-test is most common when n > 30 and the sample standard deviation is a reliable estimate of σ. Detailed instructions and tables: one-sample z-test guide.

T-Tests

T-tests are the workhorse of inferential statistics. Use the one-sample t-test when testing whether a sample mean equals a hypothesized value and σ is unknown. Use the two-sample t-test (independent samples) to compare the means of two separate groups. Use the paired t-test when the same subjects are measured twice (before/after designs or matched pairs). The key difference from the z-test is that t-tests use the t-distribution with n−1 degrees of freedom, which has heavier tails to account for the additional uncertainty of estimating σ from s. See: one-sample t-test, two-sample t-test, paired t-test.

Worked Example — Two-Sample T-Test

A school tests two teaching methods. Group A (n=25) scores x̄=78, s=10; Group B (n=25) scores x̄=83, s=11. Do the groups differ at α = 0.05?

1

Hypotheses: H₀: μ_A = μ_B  |  H₁: μ_A ≠ μ_B (two-tailed)

2

Pooled SE: SE = √(s₁²/n₁ + s₂²/n₂) = √(100/25 + 121/25) = √(4 + 4.84) = √8.84 ≈ 2.973

3

Test statistic: t = (78 − 83) / 2.973 = −5 / 2.973 ≈ −1.682; df ≈ 48

4

p-value: For t = −1.682, df = 48, two-tailed: p ≈ 0.099. The t-table gives the critical value t* ≈ ±2.011.

5

Decision: p = 0.099 > α = 0.05 → Fail to reject H₀. |t| = 1.682 < 2.011.

✅ Conclusion: At α = 0.05, there is insufficient evidence that the two teaching methods produce different mean scores. The 5-point difference is not statistically significant at this sample size.

Chi-Square Test

The chi-square (χ²) test applies to categorical data. The chi-square goodness-of-fit test checks whether observed category frequencies match expected frequencies. The chi-square test of independence checks whether two categorical variables are related in a contingency table. Unlike t-tests, chi-square tests use counts, not means — making them the right tool for survey data with categorical responses, genetics studies, and quality control checks. For full coverage: chi-square test guide and chi-square table.

ANOVA (Analysis of Variance)

When the research question involves three or more group means, the one-way ANOVA is the appropriate test. Running multiple t-tests between every pair of groups would inflate the Type I error rate (a problem called multiple comparisons). ANOVA tests the null hypothesis that all group means are equal using a single F-statistic that compares between-group variance to within-group variance. If ANOVA rejects H₀, a post-hoc test like Tukey's HSD identifies which specific groups differ. See: ANOVA guide and Tukey HSD.

Correlation and Regression

Pearson correlation (r) measures the direction and strength of the linear relationship between two continuous variables. Inferential statistics tests whether r is significantly different from zero — that is, whether there is a real relationship in the population, not just a coincidental pattern in the sample. Simple linear regression goes further: it estimates a linear equation (ŷ = b₀ + b₁x) that predicts the outcome variable from the predictor, and tests whether the slope b₁ is non-zero. For details, see the Pearson correlation and simple linear regression pages. The t-test vs. ANOVA comparison and ANOVA vs. regression pages cover how these methods relate.

Choosing the Right Inferential Test

Selecting the wrong statistical test is one of the most consequential errors in applied research — it can produce invalid p-values and wrong conclusions. The decision depends on your research question, data type, number of groups, and whether the data meet parametric assumptions. Use the guide below, then verify with the interactive test selector.

Statistical Test Selection Guide

Comparing a sample mean to a known value — σ known, n ≥ 30
One-Sample Z-Test
Comparing a sample mean to a known value — σ unknown
One-Sample T-Test
Comparing means of 2 independent groups — continuous outcome
Two-Sample T-Test (independent)
Comparing 2 measurements from the same subjects
Paired T-Test
Comparing means of 3 or more independent groups
One-Way ANOVA + post-hoc tests
Testing relationship between 2 categorical variables
Chi-Square Test of Independence
Testing whether 1 variable predicts another (continuous)
Simple Linear Regression
Testing strength of linear association between 2 continuous variables
Pearson Correlation
2 independent groups — normality violated or small n
Mann-Whitney U Test
Paired data — normality violated or small n
Wilcoxon Signed-Rank Test

Additional comparisons: t-test vs. z-test, chi-square vs. t-test, parametric vs. nonparametric tests, and one-tailed vs. two-tailed tests.

Assumptions of Inferential Statistics

Every inferential test rests on a set of assumptions. Violating them does not always invalidate the analysis, but it does change how results should be interpreted. Checking assumptions before running any test is not optional — it is standard practice.

Assumption What It Means How to Check
Random samplingThe sample was drawn randomly from the populationStudy design review; examine sampling method
IndependenceEach observation is independent of othersStudy design; no repeated measures without accounting for them
NormalityThe sampling distribution of the statistic is approximately normalShapiro-Wilk test, Q-Q plot, histogram; CLT covers this for large n
Homogeneity of varianceGroups have equal population variances (for t-tests, ANOVA)Levene's test; if violated, use Welch's t-test or Welch's ANOVA
Adequate sample sizen is large enough for the CLT to apply (n ≥ 30 for most cases)Power analysis prior to data collection
Scale of measurementData meets the measurement level required by the test (interval/ratio for t-tests; nominal/ordinal for chi-square)Review measurement scale of each variable

For normality testing, see the normality tests page. For variance assumptions, see the equal vs. unequal variance page. For overall assumption checking, see the statistical assumptions guide.

Real-World Applications of Inferential Statistics

Inferential statistics is not confined to academic textbooks. It drives decisions in every field that works with data — which is most of them.

🏥

Medicine & Clinical Trials

Drug efficacy is established through randomized controlled trials. A two-sample t-test or ANOVA determines whether a treatment group recovers faster than a control group. The FDA requires statistically significant evidence before approving a drug.

🗳️

Political Polling

Election polls sample 1,000–2,000 voters and use confidence intervals to report results with a ±margin of error, generalizing to an electorate of millions. Sample size determines the precision of the estimate.

💻

A/B Testing in Tech

Product teams test two versions of a feature (control vs. treatment) and use a two-proportion z-test to determine whether the difference in conversion rate is due to the change or random variation.

🏭

Manufacturing & Quality Control

A one-sample t-test checks whether the mean weight of sampled products meets specification. Control charts use inferential logic to flag processes that have shifted out of control.

🧠

Psychology & Education

Researchers use paired t-tests to measure whether an intervention changes test scores, ANOVA to compare multiple teaching methods, and regression to model how study time predicts GPA.

📈

Finance & Economics

Economists test whether policy changes affect unemployment rates. Financial analysts use regression to model the relationship between interest rates and equity returns.

Confidence Interval Calculator

Enter your sample data below to compute a confidence interval for the population mean. The calculator uses the z-distribution when n ≥ 30 and the t-distribution for smaller samples. Results include the standard error, margin of error, and the interval itself.

📊 Confidence Interval for the Population Mean

Entity & Formula Glossary

The table below defines the core terms, symbols, and formulas used in inferential statistics. This glossary is structured for quick reference and is organized to be useful both as a learning resource and as a lookup guide during analysis.

Term / Symbol Definition Formula / Notes
μ (mu)Population mean — the true average of all values in the populationUnknown; estimated by x̄
x̄ (x-bar)Sample mean — the average of observed sample valuesx̄ = (Σxᵢ) / n
σ (sigma)Population standard deviation — spread of population valuesUnknown; estimated by s
sSample standard deviation — estimated spread from samples = √[Σ(xᵢ−x̄)² / (n−1)]
SEStandard error — standard deviation of the sampling distributionSE = σ/√n ≈ s/√n
nSample size — number of observations in the sampleLarger n → smaller SE → more precise inference
NPopulation size — total number of members in the populationUsually unknown or very large
H₀Null hypothesis — the default claim being testede.g., H₀: μ = 50 or H₀: μ₁ = μ₂
H₁ / HₐAlternative hypothesis — the claim you're testing forCan be one-tailed or two-tailed
α (alpha)Significance level — the pre-set Type I error rateConventionally 0.05, 0.01, or 0.10
β (beta)Probability of a Type II error — false negative ratePower = 1 − β
p-valueProbability of the observed data (or more extreme) under H₀Reject H₀ when p < α
CIConfidence interval — range of plausible values for a parameterCI = x̄ ± z* × SE
MEMargin of error — half-width of the confidence intervalME = z* × SE
dfDegrees of freedom — free values in a calculationOne-sample t: df = n−1; chi-square: df = (r−1)(c−1)
zZ-statistic — test statistic for z-testz = (x̄ − μ₀) / (σ/√n)
tT-statistic — test statistic for t-testt = (x̄ − μ₀) / (s/√n)
FF-statistic — ratio of variances; used in ANOVAF = MS_between / MS_within
χ²Chi-square statistic — used for categorical data testsχ² = Σ(O−E)²/E
rPearson correlation coefficientr ∈ [−1, +1]; r = 0 means no linear relationship
CLTCentral Limit Theorem — x̄ is approximately normal for large nApplies when n ≥ 30 for most populations

Common Mistakes in Inferential Statistics

Errors in interpreting inferential results are common even in published research. The table below addresses the most consequential ones, drawing on guidance from the American Statistical Association's statement on p-values.

Common Mistake Why It's Wrong The Correct Interpretation
"p = 0.03 means there's a 3% chance H₀ is true" p-values say nothing about the truth probability of H₀ p = 0.03 means: if H₀ were true, you'd see this result only 3% of the time by chance
"We accept H₀" after a non-significant result Failing to reject H₀ is not the same as proving it true "Fail to reject H₀ — insufficient evidence against the null"
"Statistically significant = practically important" With large n, trivially small effects become significant Always report effect size (Cohen's d, η², r²) alongside the p-value
"p > 0.05 means no effect exists" A non-significant result may reflect low power, not absence of effect The test may lack power to detect the effect; check power and sample size
Running many tests without correction Multiple comparisons inflate Type I error (false positive rate) Apply Bonferroni correction or control the false discovery rate. See Bonferroni correction
Selecting α after seeing the data (p-hacking) Invalidates the test; inflates Type I error Set α before data collection; pre-register the analysis plan

Frequently Asked Questions

Inferential statistics uses information from a sample to make conclusions about a larger population. Instead of measuring every individual, researchers analyze a representative sample and use statistical methods to estimate population characteristics and test hypotheses. Common applications include opinion polls, clinical trials, and scientific research.
Descriptive statistics summarize the data you have collected using measures such as the mean, median, standard deviation, and charts. Inferential statistics go one step further by using sample data to make conclusions or predictions about an entire population. Descriptive statistics describe the sample, while inferential statistics generalize beyond it.
Inferential statistics has two primary branches: estimation and hypothesis testing. Estimation includes point estimates and confidence intervals that estimate population parameters. Hypothesis testing includes methods such as z-tests, t-tests, chi-square tests, ANOVA, and regression, which determine whether observed results are statistically significant.
Most inferential methods assume that the sample is randomly selected, observations are independent, and the sample size is adequate for reliable analysis. Many parametric methods also assume approximately normal data and, for group comparisons, equal variances. Checking these assumptions helps ensure that statistical conclusions are valid.
A p-value of 0.05 means that if the null hypothesis were true, there would be a 5% chance of obtaining results at least as extreme as those observed due to random sampling alone. It does not mean there is a 5% probability that the null hypothesis is true. A result with p ≤ 0.05 is commonly considered statistically significant.
The Central Limit Theorem states that the sampling distribution of the sample mean becomes approximately normal as the sample size increases, regardless of the population's original distribution. This principle forms the foundation for many inferential techniques, including confidence intervals and hypothesis tests, because it allows normal distribution methods to be applied to sample means.
Use a z-test when the population standard deviation is known or when the sample size is sufficiently large for the normal approximation to be appropriate. Use a t-test when the population standard deviation is unknown, particularly with smaller samples. In most real-world studies, the t-test is the preferred choice because the population standard deviation is rarely known.
To calculate a confidence interval, first compute the sample estimate and its standard error. Multiply the standard error by the appropriate critical value (z or t) to obtain the margin of error, then add and subtract that margin from the sample estimate. The general formula is Confidence Interval = Point Estimate ± Margin of Error. Larger samples produce narrower and more precise confidence intervals.

Sources and Further Reading

Fisher, R.A. (1925). Statistical Methods for Research Workers. Oliver and Boyd. — The foundational text that established the framework of significance testing still used today.
Neyman, J. & Pearson, E.S. (1933). "On the Problem of the Most Efficient Tests of Statistical Hypotheses." Philosophical Transactions of the Royal Society A, 231, 289–337. — Introduced the decision-theoretic framework of hypothesis testing.
American Statistical Association (2016). "ASA Statement on Statistical Significance and P-Values." The American Statistician. Official PDF
NIST/SEMATECH. e-Handbook of Statistical Methods. https://www.itl.nist.gov/div898/handbook/ — A comprehensive free reference for all major inferential procedures with worked examples.
OpenStax (2022). Introductory Statistics. OpenStax. Free online textbook covering probability, estimation, and hypothesis testing at the introductory level.
Penn State STAT 200. Elementary Statistics. https://online.stat.psu.edu/stat200/ — Free course materials including confidence intervals and hypothesis testing modules.