Estimate the statistical significance of variables in a model .-
^bic^ [^,^ ^s^mallest(integer) ^l^argest(integer) ^minp^rob(real) ^best^(intege > r) ^dots^ ]
Description -----------
^bic^ is a post-estimation command that uses the Bayesian Information Criterion > (BIC) for estimating the probability that a variable is part of a model, the equivalent of the statistical significance of a variable’s effect. bic estimate > s models based on all possible combinations of the independent variables - it is computationally intensive. For each of these models it calculates the BIC statistic. It then calculates the probability of each of these models, based on Bayesian principles as first proposed by Schwarz (1978) and further develope > d by Raftery (1995).
The probability that a given independent variable is in the "true" model is based on the sum of the probabilities for the models containing that variable, divided by the sum of the probabilities for all models.
Basyesian based methods, such as BIC-based methods, are an alternative to the traditional t-test for statistical significance. The BIC is not as sensitive to N unlike the traditional null hypothesis t-test and some would argue it is a better method of determining the likelihood that a slope is non-zero.
Note that bic uses bicdrop1 probabilities to limit the models considered, rather than the leaps and bounds algorithm used by Raftery.
bic works after the following estimation commands: ^regress^, ^logistic^, ^logit^, ^ologit^, ^oprobit^, ^mlogit^, ^poisson^, ^nbre > g^.
bic requires the installation of the following commands: matsort, bicdrop1, pre
Options ------- ^s^mallest(integer) (default 1) The smallest number of explanatory variables allowed in the models Used to limit the models to be considered
^l^argest(integer) (default is all variables) The largest number of explanatory variables allowed in the models Used to limit the models to be considered
^minp^rob(real) (default .05) The minimum probability, based on bicdrop1, to be considered Used to limit the models to be considered to likely ones Setting this to zero forces all the models to be considered
^best^(integer) (default 10) The number of "best" models to list at the end
^dots^ prints out dots to show the program is working
Examples -------- . sysuse auto . regress price mpg rep78 foreign disp length turn gear_ratio headroom . bic
Author: Paul Millar www.paulmillar.ca paulmi@@nipissingu.ca
Acknowledgements ---------------- This program was developed with the support of: Augustine Brannigan, University of Calgary See also: --------- Online: help for @bicdrop1@, @pre@ (if installed)
References: Raftery, Adrian (1995) Bayesian Model Selection in Social Research Sociological Methodology, V. 25, p. 111-163 > . Schwarz, Gideon (1978) Estimating the Dimension of a Model The Annals of Statistics v. 6, p.461-64.