Power Project Home
Intro to nQuery Glossary of Terms Links & Resources nQuery website
Intro to Power
Evaluation
Paired Samples T-test Independent Samples T-test Multiple Regression One Way Analysis of Variance

 

 

Welcome to the nQuery tutorial.

This tutorial is designed to increase your knowledge of the statistical concept of power through real-world research examples using the nQuery Advisor Ò software.

The maximization of statistical power is heavily discussed in statistics and research methodology courses. But what is power? Power is defined as the probability of detecting a true significant difference. There are several factors that affect power. These include alpha level, direction of test (one or two-tailed), sample size, effect size, and the actual statistical test. The present tutorial will use these factors in the determination of power.

Our tutorial is based on published research. The research articles serve as scenarios for learning how to use the power analysis software. As a result, you will be working with actual data to determine the power of a research study. However, we will use the study data to imagine we are planning the research study and conducting an a-priori or prospective power analysis, rather than a retrospective or post-hoc power analysis. The tutorial covers five scenarios based on various statistical analyses. These include

  1. Independent samples t-test
  2. Paired sample t-test
  3. Chi-square
  4. Multiple regression
  5. One-way analysis of variance.

Effect size, sample size, alpha level, and other factors are used in the determination of power in these analyses.

There are also design scenarios that are a priori power analyses. A priori power analyses are conducted prior to data collection. In these exercises, you will determine how many subjects you will need to obtain a particular power and effect size at a certain alpha level.

As you proceed through the scenarios, keep in mind two issues regarding power. When power is low, meaningful effect sizes are difficult to detect. A possible explanation could be due to a small sample. However, an extremely large sample can pose a problem for power. Too much data can result in high power, even if the magnitude is small. Therefore, you detected very small effect sizes that may or may not be substantively meaningful.