Written and maintained by CASRAI Editorial Board
Last updated
In inferential statistics, an alternate hypothesis (denoted as Ha or H1, and interchangeably called an alternative hypothesis) is the operational claim that a genuine effect, group difference, or parameter relationship exists in the target population. Formally introduced within the Neyman-Pearson decision-theoretic framework, the alternative hypothesis constitutes the exhaustive mathematical complement to the null hypothesis (H0). When researchers collect empirical observations, statistical tests evaluate whether sample data provide compelling evidence to reject the default assumption of zero difference. Establishing a clear alternative hypothesis specifies the parameter space, determines statistical power requirements, and establishes whether a design requires a non-directional (two-tailed) or directional (one-tailed) test.
Definition: What Is an Alternate Hypothesis?
An alternative hypothesis is the formal proposition that observed data reflect a systematic underlying mechanism or treatment effect rather than random sampling error. Rather than serving as an informal speculation, a statistical alternative hypothesis defines a precise mathematical subset of the parameter space (θ ∈ Θ1) that excludes the baseline value specified by the null condition (θ ∈ Θ0). In scientific research, investigators formulate the alternative hypothesis before data collection to outline the empirical divergence required to overturn existing baselines or establish novel interventions.
What Are Null and Alternative Hypotheses? (Formulation and Symbols)
Understanding what null and alternative hypotheses represent requires viewing statistical testing as a decision between two mutually exclusive assertions. In any formal statistical test, researchers evaluate sample evidence against two structured positions:
- Null Hypothesis (H0): The benchmark statement positing no true difference, association, or effect across the population (e.g., μ1 − μ2 = 0).
- Alternative Hypothesis (Ha or H1): The competing proposition asserting that the population parameter deviates from the null benchmark, either non-directionally or in a specified direction.
Mathematical Symbols and Parameter Partitions
Statistical hypothesis testing uses standardized mathematical symbols to define parameter partitions:
- H0: Null hypothesis statement.
- Ha or H1: Alternative hypothesis statement.
- μ: Population mean (or difference between two population means, μ1 − μ2).
- μ0: Fixed reference value or null parameter constant.
- ≠: Inequality operator for a non-directional (two-tailed) alternative hypothesis.
- > or <: Directional inequality operators for upper-tailed or lower-tailed alternative hypotheses.
- α: Significance level or Type I error rate (probability of rejecting H0 when H0 is true).
- β: Type II error rate (probability of failing to reject H0 when Ha is true).
- 1 − β: Statistical power (probability of correctly rejecting H0 under a specific alternative effect size).
Two-Tailed vs. One-Tailed Alternative Hypotheses
The formulation of the alternative hypothesis determines how rejection probability is distributed across the test distribution:
- Two-Tailed Alternative: Formulated as H0: μ = μ0 versus Ha: μ ≠ μ0. The rejection region α is divided equally between both tails (α/2 in each tail), guarding against unexpected differences in either direction.
- One-Tailed Directional Alternative: Formulated as H0: μ ≤ μ0 versus Ha: μ > μ0 (upper tail) or H0: μ ≥ μ0 versus Ha: μ < μ0 (lower tail). The full rejection area α is concentrated in one tail. For criteria governing directional tests, see our guide on one-tailed vs. two-tailed tests.
What Is the Difference Between Null and Alternative Hypothesis?
The primary difference between null and alternative hypotheses lies in their analytical roles. The null hypothesis serves as the mathematical anchor used to construct the sampling distribution and calculate the p-value. The alternative hypothesis defines the substantive research claim that gains probabilistic support when sample observations are incompatible with the null model.
Worked Example: Testing an Alternative Hypothesis
Consider an experiment assessing whether an active training program improves cognitive task scores compared to a control group. Researchers recruit 32 participants randomized into two equal cohorts of 16 participants each.
1. Stating the Hypotheses
At significance level α = .05, the investigators specify a non-directional test:
- H0: μ1 − μ2 = 0
- Ha: μ1 − μ2 ≠ 0
2. Sample Statistics
- Intervention Group: n1 = 16, mean x̄1 = 24.50, variance s12 = 16.00 (s1 = 4.00).
- Control Group: n2 = 16, mean x̄2 = 20.50, variance s22 = 12.96 (s2 = 3.60).
- Mean Difference: x̄1 − x̄2 = 24.50 − 20.50 = 4.00.
- Degrees of Freedom: df = 16 + 16 − 2 = 30.
3. Step-by-Step Calculation
Pooled variance (sp2):
sp2 = [ (15 × 16.00) + (15 × 12.96) ] / 30 = [ 240.00 + 194.40 ] / 30 = 434.40 / 30 = 14.4800
Pooled standard deviation (sp):
sp = √(14.4800) ≈ 3.8053
Standard error of the difference (SE):
SE = 3.8053 × √( 1/16 + 1/16 ) = 3.8053 × √(0.1250) ≈ 1.3454
Calculate the t-statistic:
t = ( 4.00 − 0 ) / 1.3454 ≈ 2.973
4. Evaluation and Effect Size
For a two-tailed Student’s t-distribution with df = 30, the critical value at α = .05 is tcrit = ±2.042. Because calculated t = 2.973 exceeds 2.042, the two-tailed p-value is p = .0058 (reported as p = .006). Because p < .05, the null hypothesis is rejected in favor of the alternative hypothesis.
Standardized effect size using Cohen’s d:
d = 4.00 / 3.8053 ≈ 1.051
For guidelines on quantifying substantive impact, consult our guide on effect size in research.
Interpretation: Reading Guidance and Decision Matrix
Interpreting statistical outcomes requires recognizing the four decision outcomes defined by Type I and Type II errors:
| Empirical Outcome | Evidentiary Criterion | Statistical Decision | Substantive Research Implication |
|---|---|---|---|
| Statistically Significant | p ≤ α (e.g., p < .05) | Reject H0 in favor of Ha | Sample evidence supports the alternative hypothesis; report effect size and confidence interval. |
| Not Significant | p > α (e.g., p ≥ .05) | Fail to reject H0 | Insufficient evidence to dismiss the null model; does not prove H0 is true. |
| Type I Error (α) | False Positive Risk | Incorrect rejection of true H0 | Asserting an effect when none exists; bounded by significance level α. |
| Type II Error (β) | False Negative Risk | Failure to reject false H0 | Missing a genuine effect, often due to small sample size or high measurement variance. |
Assumptions and Common Mistakes
Proper application of alternative hypotheses in experimental design relies on sound methodological safeguards. For distribution considerations, see our comparison on parametric vs. non-parametric tests.
Core Methodological Assumptions
- A Priori Formulation: Directionality and parameters must be defined before inspecting data. Switching from two-tailed to one-tailed post hoc doubles Type I error risk.
- Exhaustive and Disjoint Partition: H0 and Ha must cover all possible parameter values without overlap.
- Independence: Data points must be independently sampled to avoid distorted standard errors.
- Distributional Fit: Parametric tests assume continuous measurement, normality of residuals, and equal variances.
Common Conceptual Errors
- Claiming Absolute Proof: Hypothesis testing provides probabilistic evidence against H0; it never proves Ha with deductive certainty.
- Including Equality in Ha: Equality (=, ≤, ≥) belongs strictly in the null hypothesis. The alternative hypothesis strictly uses inequalities (≠, >, <).
- Accepting the Null: Finding p ≥ .05 means “failing to reject” H0, not confirming zero effect.
- Confusing p-values with Effect Size: A statistically significant result can correspond to a clinically negligible difference if sample sizes are large.
How to Report an Alternative Hypothesis (APA 7th Format)
Under APA 7th edition rules, reporting inferential tests requires the test statistic, degrees of freedom, exact p-value (to two or three decimal places, without leading zero), and effect size.
APA Narrative Examples
Significant Result Supporting Ha:
“An independent-samples t-test evaluated the alternative hypothesis that cognitive training enhances retention. Participants in the intervention condition (M = 24.50, SD = 4.00) scored significantly higher than those in the control condition (M = 20.50, SD = 3.60), t(30) = 2.97, p = .006, 95% CI [1.25, 6.75], Cohen’s d = 1.05. The data support the alternative hypothesis.”
Non-Significant Result:
“Scores did not differ significantly between intervention (M = 21.10, SD = 3.20) and control groups (M = 20.40, SD = 3.10), t(30) = 0.62, p = .538, 95% CI [−1.60, 3.00], Cohen’s d = 0.22. Evidence was insufficient to reject the null hypothesis in favor of the alternative hypothesis.”
How to Calculate and Test It in Excel, R, and Python
Statistical software facilitates direct testing of alternative hypotheses.
1. Python (SciPy)
from scipy import stats
group1 = [25.1, 23.4, 27.2, 21.8, 26.5, 24.1, 20.9, 28.0, 22.4, 25.7, 24.8, 23.9, 26.1, 21.5, 27.6, 21.0]
group2 = [19.8, 21.2, 18.5, 22.1, 20.4, 19.1, 24.0, 18.9, 21.7, 20.8, 22.5, 17.8, 23.1, 19.5, 20.2, 18.4]
# Independent t-test testing two-sided alternative hypothesis (mu1 != mu2)
result = stats.ttest_ind(group1, group2, equal_var=True, alternative='two-sided')
print(f"t-statistic: {result.statistic:.3f}, p-value: {result.pvalue:.4f}")
2. R Syntax
group1 <- c(25.1, 23.4, 27.2, 21.8, 26.5, 24.1, 20.9, 28.0, 22.4, 25.7, 24.8, 23.9, 26.1, 21.5, 27.6, 21.0)
group2 <- c(19.8, 21.2, 18.5, 22.1, 20.4, 19.1, 24.0, 18.9, 21.7, 20.8, 22.5, 17.8, 23.1, 19.5, 20.2, 18.4)
# Two-sided independent samples t-test
test_result <- t.test(group1, group2, var.equal = TRUE, alternative = "two.sided")
print(test_result)
3. Microsoft Excel Function
=T.TEST(A2:A17, B2:B17, 2, 2)
Excel syntax arguments:
array1(A2:A17): Cell range for Group 1.array2(B2:B17): Cell range for Group 2.tails(2): Two-tailed test (Ha: μ1 ≠ μ2). Enter 1 for a directional one-tailed test.type(2): Two-sample equal variance (homoscedastic) test. Enter 3 for unequal variance (Welch’s test).
Frequently Asked Questions
What is null and alternative hypothesis in simple terms?
In simple terms, the null hypothesis is the default presumption that no effect or difference exists. The alternative hypothesis is the claim that a real effect, difference, or relationship is present in the population.
Can an alternative hypothesis ever include an equality sign?
No. An alternative hypothesis cannot contain an equality sign (=, ≤, ≥). Mathematical equality defines the null baseline. The alternative hypothesis strictly uses inequalities (≠, >, <) to represent departures from the null condition.
When should researchers choose a one-tailed over a two-tailed alternative hypothesis?
A one-tailed alternative hypothesis is chosen only when investigators have strong theoretical justification that an effect in the opposing direction is physically impossible or of zero scientific interest. In general research, two-tailed tests are preferred to detect unexpected effects.
Does rejecting the null hypothesis prove that the alternative hypothesis is true?
No. Rejecting the null hypothesis indicates that sample observations are improbable if the null hypothesis were true. It provides probabilistic support for the alternative hypothesis within the selected α threshold, but does not provide deductive certainty.
Sources and Standards
- NIST/SEMATECH. (2012). e-Handbook of Statistical Methods. Section 7.1.3: What are statistical tests? National Institute of Standards and Technology. https://www.itl.nist.gov/div898/handbook/prc/section1/prc13.htm (accessed 2026-11-07).
- International Organization for Standardization. (2006). Statistics — Vocabulary and symbols — Part 1: General statistical terms and terms used in probability (ISO Standard No. 3534-1:2006). ISO. (accessed 2026-11-07).
- Neyman, J., and Pearson, E. S. (1933). On the problem of the most efficient tests of statistical hypotheses. Philosophical Transactions of the Royal Society of London. Series A, 231(694-706), 289–337. https://doi.org/10.1098/rsta.1933.0009 (accessed 2026-11-07).
- American Psychological Association. (2020). Publication Manual of the American Psychological Association (7th ed.). Section 6.43–6.48: Statistical and mathematical copy. APA. (accessed 2026-11-07).
