Stats & Methodology

# Type 2 Errors | Statistics

Go to

## Type 2 Errors | Statistics

A type 2 or type II error occurs when the paper concludes that there is no effect when indeed there is. The researchers stick to the null-hypothesis when, in fact, it is false.

### Expressed in Beta

The likelihood of committing a type 2 error is indicated in statistical hypothesis testing as (beta). It stands for the possibility of incorrectly determining there is no difference or no impact between groups or treatments when in fact there is. It is generally set by researchers to 20% or 0.20, signifying a willingness to accept a 20% chance of making such an error over the long run. However, if a false negative result would have serious repercussions, researchers can choose to reduce it to 10% or even 5%. Deciding your beta level should be an informed decision, however, oftentimes researchers fall back on the 20% heuristic.

Note that these percentages speak of the long term. When doing a hundred similar studies, 20% of them might show a false negative result, and 80% of them not. One single study simply has a type 2 error, or it doesn’t.

### Decreasing Type 2 Errors

Due to the potentially harmful repercussions of type 2 errors, researchers must make every effort to reduce them. The objective is to increase a test’s statistical power to detect a true effect or difference. In other words, scientists aim to lessen the possibility that they may mistakenly accept/keep a null hypothesis that is actually wrong.

Do note that there is an intrinsic trade-off between type 1 and type 2 errors. The likelihood of a type 2 error increases when the likelihood of a type 1 error decreases, and vice versa. It takes a careful assessment of the specific research setting, risk mitigation, potential outcomes, and available resources to strike the proper balance between the two.

Researchers use a variety of tactics to reduce the likelihood of type 2 errors. They can include boosting statistical power by boosting sample size, improving study design and methodology, using more sensitive outcome measures, and taking other statistical methods into account. To learn more about statistical power, click here.

## References

Bower, D. (2019), Medical Statistics from Scratch: An Introduction for Health Professionals (4th edition), Wiley

Akobeng A. K. (2016). Understanding type I and type II errors, statistical power and sample size. Acta paediatrica (Oslo, Norway : 1992)105(6), 605–609. https://doi.org/10.1111/apa.13384

Like what you're learning?

#### Use the assessment app

• Over 300 orthopedic physical assessment tests
• Statistics, basic assessments, and screening tests included
• Direct links to PubMed references
• Concise test descriptions
• Video demonstration
• Easy search & favorites function
E-Book

Reviews