University:
Western Sydney University
Course:
MEDI 2003 | Clinical Sciences 2
Academic year:

2024
Views:

135

Pages:

2
Author:

Mariam A. A.

Statistical Power • Probability that a study will produce a statistically significant result if the research hypothesis is true (i.e. finding a significant difference when it does exist) • 1-β or 100(1- β)% • Studies should have at least 80% power (in some 90% is preferred) Power and the relationship to beta • Recall that Type 1 error = false positive i.e. rejecting the null hypothesis when it is actually true • The value of alpha (level of significance) has a direct bearing on type 1 error. If we set alpha at 5%, this means there is a 5% probability we will reject a true null hypothesis. • Type II error = false negative. This occurs when we fail to reject a null hypothesis which is false. • Type II error is denoted by β. This number is determined by power (1β). Remember that in research we aim for 80% or 90% power. Therefore if power is set at 80%, β = 20% and if power is set at 90% β = 10%. • Power depends on: o Sample size: Increasing sample size increases power 27 o Effect size of interest (mean difference): Increasing mean difference increases power o The variability (Standard deviation) among samples: using a population with little variation (i.e. small SD) will increase power o Increasing α will also increase power Statistical vs. clinical significance • You want to determine the effect of drug A on migraine pain relief in 200 trial candidates • A clinically significant pain reduction is considered 2- points on the (0-10) pain scale • To test this you collect data on pain at baseline and at six hours post administration • You calculate the mean difference = 1.1; p=0.03 • The result (MD) is statistically significant but is it clinically significant? Role of power in interpreting results • When result is significant: o Small sample size 🡪 effect is probably practically significant o Large sample size 🡪 effect may be too small to be useful • When result is insignificant: o Small sample size 🡪 study is inconclusive o Large sample size 🡪 research hypothesis is false Statistical tests ⮚ TAN: normally distributed (parametric data) ⮚ MKR: non-normally distributed (non-parametric) TAN • T-test (independent samples or paired samples) o Compare a normally distributed variable (e.g. age) between 2 groups o E.g. mean waiting time at 2 hospitals o Independent t-test (when you have two independent groups) o Paired sample t-test (repeated measure e.g. marks on a sample test before and after a revision lecture – the marks are coming from the same set of individuals therefore this is paired data) • Anova (analysis of variance) o Compare a normally distributed continuous variable (e.g. height) between 3 or more groups o E.g. mean height amount 3 different groups of healthcare providers – doctors, pharmacists, nurses • Normal distribution – continuous variable of interest must assume a normal distribution i.e. bell shaped, symmetrical about the mean