Hacker theme

Hacker is a theme for GitHub Pages.

Download as .zip Download as .tar.gz View on GitHub
8 August 2025

Mind on Statistics (6th. Ed) Chapter 13 - Testing Hypotheses About Means

by Arpon Sarker

Introduction

Hypothesis Testing for 1 Population Mean

Conditions:

\[H_0: \mu = \mu_0\]

This is a one-sample t-test because the t-distribution is used to determine the p-value.

\[t = \frac{\bar{x}-\mu_0}{s/\sqrt{n}}\]

with df=n-1

Hypothesis Testing for Population Mean of Paired Differences

Same conditions as above

\[t = \frac{\bar{d} - 0}{s_d/\sqrt{n}}\]

paired t-test

Hypothesis Testing for Difference in 2 Population Means

Same conditions as above

Unpooled (General) Case

two-sample t-test \(t = \frac{(\bar{x}_1 - \bar{x}_2) - 0}{\sqrt{\frac{s_1^2}{n_1}+\frac{s_2^2}{n_2}}}\)

Use Welch’s Approximation for df or min($n_1-1, n_2-1$)

Pooled Case

More precise than unpooled method. If we assumed two populations have the same/similar variance, then there is a procedure for which the t-distribution is the correct distribution of the t-statistic when the null hypothesis is true. pooled two-sample t-test.

\[\textrm{pooled sample variance } = s_p^2 = \frac{(n_1-1)s_1^2 + (n_2-1)s_2^2}{n_1+n_2-2}\\ \textrm{pooled t } = \frac{(\bar{x}_1 - \bar{x}_2)-0}{\textrm{pooled std. error}}\\ \textrm{pooled s.e.} = \textrm{pooled s.e.}(\bar{x}_1-\bar{x}_2) = \sqrt{\frac{s_p^2}{n_1}+\frac{s_p^2}{n_2}}\]

with df = $n_1 + n_2 - 2$

Guidelines:

Randomisation Test

The Relationship Between Significance Tests and Confidence Intervals

Confidence intervals should be used to supplement significance testing.

Effect Size

Effect Size: How much the truth differs from chance or from a control condition. Helps evaluate magnitude of condition using a measure that allows researchers to compare results across studies.

Single Sample Effect Size for Means \(\hat{d} = \frac{\bar{x}-\mu_0}{s}\)

Two Sample Effect Size for Means \(\hat{d} = \frac{\bar{x}_1 - \bar{x}_2}{s}\)

Relationship between test statistic t and estimated effect size: \(t = \frac{\bar{x}-\mu_0}{s\sqrt{n}}, \textrm{ so } \hat{d}=\frac{t}{\sqrt{n}} \\ \textrm{for two samples,}\\ \textrm{pooled t } = \frac{\bar{x}_1 - \bar{x}_2}{s\sqrt{\frac{1}{n_1}+\frac{1}{n_2}}} = \frac{\hat{d}}{s\sqrt{\frac{1}{n_1}+\frac{1}{n_2}}}\)

\[\textrm{test statistic} = \textrm{effect size} \cdot \textrm{size of study}\]

Evaluating Statistical Results in Research Reports

Multiple Testing: With each test and interval there is a chance that an erroneous conclusion will be made, so if multiple tests and intervals are considered, the chance of making at least one erroneous conclusion will increase. For 20 95% confidence intervals 1 will miss the truth value 20(0.05) = 1 or a type 1 error.

The simplest and most conservative method to fix this is the Bonferroni Method which divides significance level across tests.

tags: mathematics - statistics