Question 9: Hypothesis Testing Errors and Procedures
a. Two Types of Errors in Hypothesis Testing
The two types of errors in hypothesis testing are Type I error (false positive) and Type II error (false negative). 1, 2
- Type I error (α): Rejecting the null hypothesis when it is actually true—concluding a difference exists when it does not
- Type II error (β): Failing to reject the null hypothesis when it is actually false—concluding no difference exists when one actually does exist
b. Description of Type I Error
Type I error occurs when you incorrectly conclude that a difference exists between groups being compared when, in reality, no difference exists. 2
This is analogous to a false positive result. The probability of committing a Type I error is denoted by α (alpha), conventionally set at 0.05 or 5%. This means you accept a 5% risk of falsely claiming a significant difference. 1, 3, 2
When performing multiple comparisons, the chance of Type I error increases substantially, which is why post hoc tests include corrections (such as Tukey's HSD) to maintain the initial significance level. 1 For example, if you perform 20 independent tests at α = 0.05, you would expect one false positive result purely by chance.
c. Steps for Hypothesis Testing Comparing Two Group Means
The systematic approach to comparing two group means follows these algorithmic steps: 1, 4
Step 1: State the Hypotheses
- Null hypothesis (H₀): The two group means are equal (no difference)
- Alternative hypothesis (H₁): The two group means are different (two-sided test) or one is greater/less than the other (one-sided test)
Step 2: Choose the Appropriate Test
- For normally distributed data with equal variances: Use unpaired t-test
- For normally distributed data with unequal variances: Use Welch's t-test
- For non-normally distributed data or presence of outliers: Use Mann-Whitney U test (non-parametric alternative) 1
Step 3: Set the Significance Level
Step 4: Calculate the Test Statistic
- For t-test: Calculate the t-statistic as the scaled difference between sample means, which follows a Student's t-distribution under the null hypothesis 1
- For Mann-Whitney U test: Rank all observations, calculate rank sums, and compute the U-statistic 1
Step 5: Determine the P-value
- Compare the calculated test statistic to its theoretical distribution
- Report exact P-values (e.g., P = 0.03) rather than using symbols like P < 0.05 5
Step 6: Make a Decision
- If P ≤ α: Reject the null hypothesis (statistically significant difference exists)
- If P > α: Fail to reject the null hypothesis (insufficient evidence of difference)
- Critical caveat: Absence of evidence is not evidence of absence—a non-significant result does not prove groups are identical 5
Step 7: Report Results with Effect Size and Confidence Intervals
- Always report the mean difference with 95% confidence intervals alongside the P-value
- Include group means, standard deviations, and sample sizes 5, 6
Question 10: Confidence Intervals and Statistical Measures
a. Interpretation of 95% Confidence Interval (6.0 kg to 7.2 kg)
This statement means that we are 95% confident that the true population mean weight of children in Agogo lies between 6.0 kg and 7.2 kg. 7, 6
More precisely, if we repeated this study many times with different random samples of 100 children from the same population, approximately 95% of the calculated confidence intervals would contain the true population mean. 6 The confidence interval provides both statistical significance information (since it doesn't include extreme values that would suggest measurement error) and clinically meaningful information about the actual range of plausible values for the population mean. 6
b. Calculating Standard Deviation from Confidence Interval
To calculate the standard deviation, use the relationship between the confidence interval width and the standard error:
Given:
- Sample mean (x̄) = 6.4 kg
- 95% CI: 6.0 kg to 7.2 kg
- Sample size (n) = 100
Calculation steps:
Margin of error = (Upper limit - Lower limit) / 2 = (7.2 - 6.0) / 2 = 0.6 kg
- Alternatively: Upper limit - Mean = 7.2 - 6.4 = 0.8 kg (using upper bound)
- Or: Mean - Lower limit = 6.4 - 6.0 = 0.4 kg (using lower bound)
- Note: The CI appears asymmetric, but using the average: (0.8 + 0.4)/2 = 0.6 kg
Standard error (SE) = Margin of error / 1.96 (for 95% CI)
- SE = 0.6 / 1.96 = 0.306 kg
Standard deviation (SD) = SE × √n
- SD = 0.306 × √100 = 0.306 × 10 = 3.06 kg
The estimated standard deviation of children's weights is approximately 3.06 kg.
c. Statistic for Comparing Dispersion of Height vs. Weight
The coefficient of variation (CV) is the appropriate statistic to compare the spread of height against weight.
The coefficient of variation is calculated as:
- CV = (Standard Deviation / Mean) × 100%
Rationale: Height and weight are measured on completely different scales (centimeters vs. kilograms) with different means and variances. You cannot directly compare their standard deviations because a 5 cm variation in height has entirely different meaning than a 5 kg variation in weight. 5
The CV standardizes the variability by expressing it as a percentage of the mean, making it dimensionless and allowing meaningful comparison of dispersion across variables measured in different units. This addresses the fundamental principle that you cannot rank variables in terms of strength of association or variability when they are on different scales without standardization. 5