The relationship between alpha and the type one error.

Understanding the relationship between an alpha level and the subsequent type 1 error is vital when it comes to designing an experiment. This is because the significance of research, to some extent, is dependent on how these two concepts interact. Having a poorly designed alpha level can leave your results without meaning. You can also make errors when extrapolating your results, which may be worse. These are known as type 1 errors.

The alpha level is the point at which research becomes statistically significant. In other words, your design has shown sufficient evidence to support your proposed theory. However, when designing an experiment, researchers have the freedom to set their alpha level to what they want. The most common value is 5%. This is saying that there is a 5 in 100 probability that your result is obtained by chance. The lower the alpha level, lets say 1% or 1 in every 100, the higher the significance your finding has to be to cross that hypothetical boundary.

On the other hand, there are also type 1 errors. These are errors made from rejecting a true null hypothesis (Hubery & Morris, 1989). There are many possible explanations for rejecting a true null hypothesis. One of them, however, relates to the alpha level the researcher has decided to use. When you have a high alpha level, lets use 50% as an extreme example, the chances of obtaining an error is also raised. This is because the result you need to cross that hypothetical barrier of significance doesn’t have to be as high. So you may have a result deemed significant, but is not really because your alpha level is too high. Furthermore, this result doesn’t mean much if 50% of the time you find data appropriate enough to pass this alpha level.

So the question is; what is the correct alpha level to use? Well there is not really a correct level as such. It is up to the discretion of the researcher to decide how strict a design is needed. However, there are bonuses for having a high alpha level as well as having a low alpha level. High alpha levels provide a much better opportunity to find significant results. Just to make things clear, not all high alpha levels lead to type 1 errors. So finding a genuine result is still possible with higher alpha levels. On the other hand, having a low alpha level also has its bonuses. As well as stating the obvious in saying that it reduces the chance of obtaining a type 1 error, it also makes sure that research is significant enough to benefit society. Drug trials are a good example of being strict in the use of its alpha level, whilst producing tangible benefits (University of the Sciences in Philadelphia, 2005).

In conclusion, there is a fine line in the balance between setting the alpha level and the possibility of obtaining a type 1 error. This means that a researcher needs to take into consideration multiple factors when designing an experiment. The level at which alpha is set can have its bonuses, but the consequences may be much worse if the alpha level is not set at the correct level.

Hubery, C., J. & Morris, J., D. (1989). Multivariate Analysis Versus Multiple Univariate Analyses. Psychological Bulletin, 105(2), 302-308.

University of the Sciences in Philadelphia. (2005). Remington: The Science and Practise of Pharmacy (21st ed.). Philadelphia, PA: Lippincott Williams and Wilkins.