Can someone explain how the mean and standard deviation are defined for normpdf from the following matlab example?

I m looking at the following example, http://fr.mathworks.com/help/stats/examples/bayesian-analysis-for-a-logistic-regression-model.html

The code

 prior1 = @(b1) normpdf(b1,0,20);    % prior for intercept
 prior2 = @(b2) normpdf(b2,0,20);

The mean is defined as 0 and the standard deviation is defined as 20. I know for a standard normal distribution the mean is defined as 0 and the standard deviation is 1. However it isn't explained how these the inputs 0 and 20 are obtained which is frustrating.

[Torsten]

They are not obtained, they are only arbitrarily taken.

Best wishes

Torsten.

[Ronan]

But if they are arbitrarily taken is that relative to the data used. Eg if I m using a completely different data set would it be appropriate to just use 20?

[Adam]

They are specific to that pdf, as you can see in the graph shown in that example. You can see the 20 standard deviation coming through in the halfwidth of the pdf. If you put 50 in then your half width would be around 50.

[Ronan]

ok thank you.

Answer 1

[the cyclist]

Ronan, I think that example assumes a certain amount of knowledge of Bayesian analysis. Maybe you have some of that knowledge, and maybe you don't.

The values you ask about parameterize the prior probability distribution. In this example, these values indicate that the analyst had evidence (maybe a prior experiment or literature) that the distribution with mean 0 and standard deviation of 20. They are a bit "out of the blue" here, but in a real application they would not be arbitrary. They would be driven by prior knowledge.

[Ronan]

My knowledge of Bayesian analysis is pretty limited. I know its a bit of a random question. There were one or two things holding me back from applying this example and it is very beneficial to me if I can. Thank you.