Home / Summaries / Class notes - ARMS / prior-bayesian-null

Grasple ANCOVA - Supporting the null hypothesis

5 important questions on Grasple ANCOVA - Supporting the null hypothesis

What is one limitation of NHST?

We can only test the null, and at best conclude that an effect is not in line with it (e.g., that an effect is not exactly zero).

Quantify evidence for the absence of an effect

If we predict that there is an effect, we can test the prediction (i.e., try to disprove or 'falsify' it) by asking whether there is more support for the null

Why is Bayesian more flexible?

We can compare multiple hypotheses that are more informative than just the null.
Instead of looking for evidence against a model, Bayesian statistics look for evidence supporting a model

One criticism is that Bayesian results are affected by the choice for the prior distribution. Bayesians reply:
For estimation

you can (1) choose a non-informative (diffuse) prior, which reduces the impact of the prior to almost nothing; or (2) include prior knowledge in your analysis and consider this a strength of the Bayesian approach (accumulating knowledge instead of starting from scratch with every new study).

One criticism is that Bayesian results are affected by the choice for the prior distribution. Bayesians reply Sensitivity analyses

Sensitivity analyses can be used, which means that the same analysis is done multiple times with different priors, so you can check if the conclusion is stable under these different choices. Instead of determining which prior would be an 'objective' choice, this can solve the problem through transparency (other can see how much the results are affected by different priors and see which results occur with the priors they would choose).

The question on the page originate from the summary of the following study material:

ARMS

View summary