Hypothesis Testing

1. Introduction to Hypothesis Testing

A hypothesis is a formal, testable statement predicting the relationship between two or more variables. In pediatric biomedical research, hypothesis testing is the core statistical method used to make inferences about a population based on sample data, allowing clinicians to differentiate true physiological effects from random chance.

2. The Null Hypothesis ( $H_{0}$ )

Definition: The Null Hypothesis postulates that there is no true difference, no association, or no treatment effect between the groups being compared. It represents the "status quo" or the assumption of baseline equality.
Core Concept: In statistical testing, $H_{0}$ is assumed to be true until the sample data provides sufficient evidence to reject it. Researchers do not seek to prove the null hypothesis; they attempt to disprove or reject it.
Statistical Expression: $H_{0} : μ_{1} = μ_{2}$ (Mean of group 1 equals Mean of group 2).
Pediatric Example: "There is no difference in the mean duration of fever reduction between pediatric patients receiving Drug A (new antipyretic) and those receiving Paracetamol."

3. The Alternative Hypothesis ( $H_{1}$ or $H_{A}$ )

Definition: The Alternative Hypothesis is the logical opposite of the null hypothesis. It asserts that a real difference, association, or treatment effect does exist in the population. It is the actual research question the investigator aims to prove.
Core Concept: If statistical analysis shows that the data is highly unlikely under the null hypothesis (typically indicated by a $p$ -value < $0.05$ ), $H_{0}$ is rejected in favor of $H_{1}$ .
Pediatric Example: "There is a significant difference in the mean duration of fever reduction between pediatric patients receiving Drug A and those receiving Paracetamol."

4. Types of Alternative Hypotheses

Based on the direction of the expected outcome, $H_{1}$ can be formulated in two ways:

A. Two-Tailed (Non-Directional) Hypothesis

Predicts a difference exists but does not specify the direction (greater or lesser).
Usage: Standard in most clinical trials to account for an intervention being either beneficial or harmful.
Expression: $H_{1} : μ_{1} \neq μ_{2}$ .

B. One-Tailed (Directional) Hypothesis

Predicts a difference in one specific direction.
Usage: Used only when an outcome in the opposite direction is biologically impossible or clinically irrelevant (e.g., a non-inferiority trial).
Expression: $H_{1} : μ_{1} > μ_{2}$ OR $H_{1} : μ_{1} < μ_{2}$ .

5. Statistical Interpretation and The P-Value

The decision to reject $H_{0}$ depends on the $p$ -value, which is the probability of obtaining the observed results (or more extreme) assuming $H_{0}$ is entirely true.

$p$ < $α$ (Level of Significance, usually 0.05): The results are statistically significant. Reject $H_{0}$ and accept $H_{1}$ .
$p$ $\geq$ $α$ : The results are not statistically significant. Fail to reject $H_{0}$ (Note: We "fail to reject" rather than "accept" $H_{0}$ , as lack of evidence of a difference is not proof of equivalence).

6. Errors in Hypothesis Testing

Because hypothesis testing is based on probabilities derived from samples, errors can occur when extrapolating to the true population.

A. Type I Error ( $α$ Error / False Positive)

Definition: Rejecting the Null Hypothesis ( $H_{0}$ ) when it is actually true in reality.
Clinical Consequence: Concluding a drug is effective when it is actually useless. This is considered the more dangerous error in medicine, as it can lead to the adoption of ineffective or harmful treatments.
Control: Controlled by setting the significance level ( $α$ ), typically at 5% (0.05).

B. Type II Error ( $β$ Error / False Negative)

Definition: Failing to reject the Null Hypothesis ( $H_{0}$ ) when the Alternative Hypothesis ( $H_{1}$ ) is actually true in reality.
Clinical Consequence: Missing a true therapeutic breakthrough. Concluding a life-saving drug has no effect when it actually does.
Control: Minimized by increasing the sample size, which increases the Statistical Power ( $1 - β$ ) of the study (typically set at 80% or 0.80).

7. Importance in Evidence-Based Pediatrics

Standardization: Provides a universally understood, rigorous mathematical framework for evaluating clinical trials (e.g., vaccine efficacy studies, new antibiotic regimens).
Bias Reduction: Forces the researcher to define success criteria ( $α$ and $β$ ) a priori (before data collection begins), preventing the manipulation of data to fit a desired narrative.