stdPriors: Standardisation and priors
In wiqid: Quick and Dirty Estimates for Wildlife Populations
Priors R Documentation
Standardisation and priors
Description
This page documents the priors used in the Bayesian analysis. For logistic models, sensible priors depend on the range of values and thus on the standardisation scheme used, which is also detailed here.
At present, this represents good intentions! It has not been implemented in all the functions in wiqid.
Standardisation
Continuous variables are standardised to facilitate writing models, and optimisation. Standardisation also means that the size of regression coefficients directly reflect the importance of the corresponding variables.
Binary variables coded as TRUE/FALSE and dummy variables are not changed. To make continuous variables comparable with these, they are centred by subtracting the mean, and then divided by their standard deviation.
Update: In versions prior to 0.2.x, continuous variables were centred by subtracting the mean, and then divided by two times their standard deviation (Gelman, 2008). With the new default, beta coefficients will be exactly half the size. There may be some rounding errors in the fourth decimal place for other estimates.
Note that all numerical inputs (ie, is.numeric == TRUE) that appear in the data argument will be standardised, including binary variables coded as 0/1. Variables coded as TRUE/FALSE or as factors are not affected.
The same standardisation strategy is used for both Bayesian and maximum likelihood functions.
Priors for logistic regression coefficients
Following Gelman et al (2008), we use independent Cauchy priors with centre 0 and scale 10 for the intercept and scale 2.5 for all other coefficients.
Priors for probabilities
We use independent Uniform(0, 1) priors for probabilities.
References
Gelman, A. (2008) Scaling regression inputs by dividing by two standard deviations. Statistics in Medicine, 27, 2865-2873
Gelman, Jakulin, Pittau and Su (2008) A weakly informative default prior distribution for logistic and other regression models. Annals of Applied Statistics 2, 1360-1383.
wiqid documentation built on Nov. 18, 2022, 1:07 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
| Priors | R Documentation |
Standardisation and priors
Description
This page documents the priors used in the Bayesian analysis. For logistic models, sensible priors depend on the range of values and thus on the standardisation scheme used, which is also detailed here.
At present, this represents good intentions! It has not been implemented in all the functions in wiqid.
Standardisation
Continuous variables are standardised to facilitate writing models, and optimisation. Standardisation also means that the size of regression coefficients directly reflect the importance of the corresponding variables.
Binary variables coded as TRUE/FALSE and dummy variables are not changed. To make continuous variables comparable with these, they are centred by subtracting the mean, and then divided by their standard deviation.
Update: In versions prior to 0.2.x, continuous variables were centred by subtracting the mean, and then divided by two times their standard deviation (Gelman, 2008). With the new default, beta coefficients will be exactly half the size. There may be some rounding errors in the fourth decimal place for other estimates.
Note that all numerical inputs (ie, is.numeric == TRUE) that appear in the data argument will be standardised, including binary variables coded as 0/1. Variables coded as TRUE/FALSE or as factors are not affected.
The same standardisation strategy is used for both Bayesian and maximum likelihood functions.
Priors for logistic regression coefficients
Following Gelman et al (2008), we use independent Cauchy priors with centre 0 and scale 10 for the intercept and scale 2.5 for all other coefficients.
Priors for probabilities
We use independent Uniform(0, 1) priors for probabilities.
References
Gelman, A. (2008) Scaling regression inputs by dividing by two standard deviations. Statistics in Medicine, 27, 2865-2873
Gelman, Jakulin, Pittau and Su (2008) A weakly informative default prior distribution for logistic and other regression models. Annals of Applied Statistics 2, 1360-1383.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.