What does a p-value mean in statistics?

A p-value is the probability of observing results at least as extreme as your data if the null hypothesis were true. A p-value below 0.05 means you reject the null hypothesis at the 5% significance level. It does not mean there is a 95% chance your hypothesis is true, nor does it measure practical significance. Always pair p-values with effect sizes and confidence intervals.

📐 Math ResourcesLast updated May 3, 2026

Statistics & Data Analysis Guide 2026: Standard Deviation, Probability, Hypothesis Testing & More

Master core statistical concepts with formulas, real examples, and free calculators

U0001F9EE

Free Calculators

U0001F4CA

20+

Concepts Covered

U0001F514

σ Coverage

68-95-99.7 Rule

U0001F3AF

95% CI

Confidence Level

⏱️

16 min

Read Time

💡

Key Takeaways

Standard deviation (SD) measures data spread: σ = √[Σ(xᵢ – μ)² / N] for population; s = √[Σ(xᵢ – x̅)² / (N-1)] for samples.
Mean is the average, median is the middle value, mode is the most frequent — always compare all three before drawing conclusions from data.
The 68-95-99.7 rule: 68% of normally distributed data falls within 1 SD of the mean, 95% within 2 SDs, 99.7% within 3 SDs.
A p-value below your significance threshold (commonly 0.05) means you reject the null hypothesis — but it does NOT measure effect size or practical significance.
Correlation does not imply causation: r = 0.95 between two variables does not mean one causes the other.
The central limit theorem guarantees that sample means follow a normal distribution when sample size is large (n ≥ 30), regardless of population distribution.
Type I error (false positive) = rejecting a true null hypothesis; Type II error (false negative) = failing to reject a false null hypothesis.
Confidence intervals are more informative than p-values: a 95% CI means if you repeated the experiment 100 times, 95 CIs would contain the true parameter.
Bayes’ theorem: P(A|B) = [P(B|A) × P(A)] / P(B) — the foundation of Bayesian statistics and many machine learning algorithms.
Outlier detection: values beyond Q1 – 1.5×IQR or Q3 + 1.5×IQR are statistical outliers (Tukey’s fence rule).

Statistics is the language that turns raw data into decisions. From Bureau of Labor Statistics reports to medical trials and machine learning models, statistical reasoning underpins almost every evidence-based field. This 2026 guide — aligned with AP Statistics, college introductory statistics, and Khan Academy curriculum — covers every foundational concept from descriptive statistics through hypothesis testing, with worked examples and our free statistics calculators to practice each skill.

U0001F4CA

Descriptive vs. Inferential Statistics: What’s the Difference?

Statistics divides into two main branches. Descriptive statistics summarizes and describes data you have: mean, median, mode, range, standard deviation, charts, and frequency tables. It tells you “what is.” Inferential statistics uses sample data to make predictions or inferences about a larger population: hypothesis tests, confidence intervals, regression, and ANOVA. It tells you “what probably is” for data you haven’t collected. Most real data analysis uses both: first describe, then infer. Use our mean/median/mode calculator for descriptive work, and our standard deviation calculator to quantify spread.

U0001F3AF

Mean, Median & Mode: Central Tendency Explained

Mean (arithmetic average): Sum of all values ÷ count. Sensitive to outliers. Best for symmetric distributions. Median: Middle value when sorted. If even count, average of two middle values. Resistant to outliers — use for incomes, house prices, wait times. Mode: Most frequent value(s). A dataset can be unimodal, bimodal, or multimodal. Best for categorical data or identifying peaks. Key insight: In a skewed right distribution (e.g., income), mode < median < mean. In a skewed left distribution, mean < median < mode. When mean and median diverge significantly, your data is likely skewed — the median is usually the more honest central measure. Use our mean median mode calculator to compute all three instantly.

U0001F4CF

Standard Deviation, Variance & IQR: Measuring Spread

Range: Max − Min. Simple but very sensitive to outliers. Variance (σ²): Average of squared deviations from the mean. Units are squared — hard to interpret directly. Standard deviation (σ or s): Square root of variance. Same units as data — the most commonly used spread measure. IQR: Q3 − Q1. The middle 50% spread. Robust to outliers. Coefficient of variation (CV): (SD / Mean) × 100%. Useful for comparing variability across datasets with different means (e.g., comparing risk across investments of different prices). Rule of thumb: CV < 15% = low variability, 15–30% = moderate, > 30% = high. Use our standard deviation calculator for population and sample calculations.

U0001F514

Normal Distribution & Z-Scores: The Bell Curve

The normal (Gaussian) distribution is the most important probability distribution in statistics. Properties: Perfectly symmetrical bell curve, described entirely by mean (μ) and standard deviation (σ), mean = median = mode. 68-95-99.7 empirical rule: 68% of data within ±1σ; 95% within ±2σ; 99.7% within ±3σ. Z-score formula: z = (x − μ) / σ. A z-score of +2 means the value is 2 standard deviations above the mean (top 2.3% of a normal distribution). Practical uses: Standardized tests (SAT, GRE scores), manufacturing quality control (Six Sigma), financial returns, natural measurements (heights, weights). Standard normal table (Z-table): Converts z-scores to probabilities. z = 1.645 → 90th percentile; z = 1.96 → 97.5th; z = 2.576 → 99.5th.

U0001F3B2

Probability Fundamentals: Rules, Events & Bayes’ Theorem

Probability is a number between 0 (impossible) and 1 (certain). Basic rules: P(A) = favorable outcomes / total equally-likely outcomes. Complement rule: P(not A) = 1 − P(A). Addition rule: P(A or B) = P(A) + P(B) − P(A and B). For mutually exclusive events: P(A or B) = P(A) + P(B). Multiplication rule: P(A and B) = P(A) × P(B|A). For independent events: P(A and B) = P(A) × P(B). Conditional probability: P(A|B) = P(A and B) / P(B). Bayes’ theorem: P(A|B) = [P(B|A) × P(A)] / P(B). Updates prior beliefs with new evidence — foundational in machine learning, spam detection, and medical testing. Use our probability calculator for any event scenario.

U0001F52C

Hypothesis Testing: p-Values, Type I & II Errors

Hypothesis testing is the formal statistical procedure for making decisions from data. Steps: (1) State null hypothesis H₀ (no effect, no difference) and alternative H₁. (2) Choose significance level α (commonly 0.05). (3) Collect data and calculate test statistic (z, t, F, χ²). (4) Find p-value: probability of data this extreme if H₀ true. (5) Decision: p < α → reject H₀; p ≥ α → fail to reject H₀. Error types: Type I error (α) = false positive = rejecting a true H₀. Type II error (β) = false negative = not rejecting a false H₀. Power = 1 − β = probability of correctly rejecting a false H₀. Critical insight: A non-significant result (p ≥ 0.05) is not proof the null is true — you may simply lack statistical power.

U0001F4D0

Confidence Intervals: Estimating Population Parameters

A confidence interval gives a range of plausible values for a population parameter based on sample data. 95% CI for a mean: x̅ ± 1.96 × (s/√n). Interpretation: If we repeat this experiment 100 times, about 95 of the resulting intervals will contain the true population mean. NOT “95% probability the true value is in this specific interval” — a common misconception. Width factors: Larger sample → narrower CI; higher confidence level → wider CI; less variability → narrower CI. Why CIs are better than p-values alone: CIs show magnitude and precision of the estimate, not just whether an effect exists. A drug trial showing “p = 0.03” is much more informative when paired with “95% CI: 2.1 to 8.4 mmHg reduction in blood pressure.”

U0001F4C9

Correlation & Regression: Measuring & Modeling Relationships

Pearson’s r (correlation coefficient): Measures linear relationship strength between two continuous variables, ranging −1 to +1. |r| > 0.7 = strong; 0.3–0.7 = moderate; < 0.3 = weak. Always plot a scatter plot before interpreting r — Anscombe’s quartet shows four datasets with identical r but completely different shapes. Simple linear regression: y = β₀ + β₁x + ε. β₁ (slope) = change in y per unit change in x. R² (coefficient of determination) = proportion of variance in y explained by x. R² = 0.80 means 80% of y variance is explained. Key reminder: Correlation ≠ causation. Always check for confounding variables, reverse causality, and spurious correlations before concluding a causal relationship.

U0001F465

Sampling Methods: How to Collect Representative Data

The quality of your statistical conclusions depends on how data is collected. Simple random sampling: Every individual has an equal chance of selection. Gold standard for eliminating selection bias. Stratified sampling: Divide population into subgroups (strata) and randomly sample from each. Better representation of subpopulations. Cluster sampling: Divide population into clusters (e.g., cities), randomly select clusters, sample all members in chosen clusters. Cost-effective for geographically dispersed populations. Systematic sampling: Select every k-th element from a list. Easy to implement but can introduce bias if list has a periodic pattern. Convenience sampling: Use whoever is available. Cheap but often biased — avoid for inferential conclusions. Sample size rule of thumb: n ≥ 30 for CLT to apply; n ≥ 100 for stable estimates; power analysis for formal studies.

U0001F4C8

Data Visualization: Choosing the Right Chart

The right chart makes patterns obvious; the wrong chart hides or distorts them. Histogram: Distribution of a single continuous variable — shows shape, skew, outliers. Box plot: Shows Q1, median, Q3, IQR, and outliers — great for comparing distributions across groups. Scatter plot: Relationship between two continuous variables — always use before correlation/regression. Bar chart: Comparing frequencies or means across categories — use for categorical data. Line chart: Changes over time — start y-axis at zero to avoid misleading visuals. Heat map: Correlation matrices or geographic data. Principles: Label axes clearly, include units, avoid 3D charts (distort perception), don’t truncate y-axes to exaggerate effects.

🧮

What is standard deviation and how do you calculate it?

Standard deviation measures how spread out data points are from the mean. Steps: (1) Calculate the mean (μ). (2) Subtract the mean from each value and square the result. (3) Sum all squared differences. (4) Divide by N (population) or N-1 (sample). (5) Take the square root. Example: dataset {2, 4, 4, 4, 5, 5, 7, 9} → mean = 5, SD = 2. A small SD means data clusters near the mean; a large SD means high variability. Use our standard deviation calculator for any dataset.

When should I use mean vs. median vs. mode?

Mean: best for symmetric distributions with no extreme outliers (test scores, heights). Median: best for skewed distributions or data with outliers — house prices, incomes, wait times. Median is not affected by a billionaire in your salary dataset. Mode: best for categorical data (most popular color, most common shoe size) or finding peaks in multimodal distributions. Always report all three for a complete picture of central tendency.

What does a p-value of 0.05 mean?

A p-value is the probability of observing results at least as extreme as your data IF the null hypothesis were true. P = 0.05 means there is a 5% chance you'd see these results by random chance alone. It does NOT mean: there is a 95% chance the hypothesis is true, or that the effect is practically important. Always pair p-values with effect sizes (Cohen's d, R²) and confidence intervals for meaningful interpretation.

What is a confidence interval?

A 95% confidence interval (CI) is a range of values where, if you repeated your study 100 times with different random samples, approximately 95 of those intervals would contain the true population parameter. It is NOT a 95% probability that the true value is in this particular interval. Wider CIs indicate more uncertainty (smaller samples, higher variability). Formula for 95% CI on mean: x̅ ± 1.96 × (s / √n).

What is the difference between correlation and causation?

Correlation measures the linear relationship between two variables (r ranges from −1 to +1). A strong correlation (r = 0.90) only means the variables tend to move together — it does not establish that one causes the other. Classic example: ice cream sales and drowning rates are correlated (both increase in summer) but neither causes the other (confounding variable: hot weather). Establishing causation requires controlled experiments or careful causal inference methods.

How do I choose between a z-test and a t-test?

Z-test: use when population standard deviation (σ) is known AND sample size n ≥ 30. T-test: use when σ is unknown (most real-world cases) or n < 30. Types of t-tests: one-sample t-test (compare sample mean to known value), independent two-sample t-test (compare two group means), paired t-test (compare before/after in same subjects). For most practical work, use the t-test — it converges to the z-distribution at large sample sizes anyway.

What is the central limit theorem (CLT)?

The central limit theorem states that the distribution of sample means approaches a normal distribution as sample size increases, regardless of the shape of the original population distribution. Key requirements: samples are independent and identically distributed (i.i.d.); n ≥ 30 is the common rule of thumb (less for symmetric distributions, more for heavily skewed). CLT is why so many statistical tests assume normality — they work on sample means, not raw data.

What is the interquartile range (IQR) and how is it used?

IQR = Q3 − Q1 (the range of the middle 50% of data). It measures spread while being resistant to outliers. Outlier detection (Tukey’s fence): Lower fence = Q1 − 1.5 × IQR; Upper fence = Q3 + 1.5 × IQR. Values outside these fences are suspected outliers. Box plots visualize Q1, median, Q3, and the fences. IQR is preferred over standard deviation when data is skewed or has outliers.

When do I use a chi-square test?

Chi-square tests analyze categorical data (counts/frequencies). Two main types: (1) Chi-square goodness-of-fit: tests if observed frequencies match expected frequencies for one variable (e.g., is a die fair?). (2) Chi-square test of independence: tests if two categorical variables are associated (e.g., is smoking status independent of gender?). Requirements: all expected cell counts ≥ 5, data is a random sample, observations are independent. Calculate test statistic: χ² = Σ[(O − E)² / E].

What is Bayes’ theorem and when is it used?

Bayes’ theorem: P(A|B) = [P(B|A) × P(A)] / P(B). It updates the probability of a hypothesis given new evidence. Classic example: A medical test is 99% accurate; a disease affects 1% of the population. If you test positive, what’s the probability you have the disease? Using Bayes: P(disease|positive) = (0.99 × 0.01) / [(0.99 × 0.01) + (0.01 × 0.99)] ≈ 50% — far less than most people expect. Bayes’ theorem is the foundation of spam filters, medical diagnosis, and Bayesian machine learning.

Ready to Calculate?

Browse our complete math calculator suite — free, instant, and trusted by professionals worldwide.

Browse Math Calculators All Guides