model selection Flashcards
three types of logistic regression
binary, multinomial, ordinal
what does logic function do
transforms an s shaped curve into a straight line
assumptions of logistic regression
- dependent variable must be binary
- each observation is independent
- little or no collinearity
what’s the advantage of logistic regression over chi square
-0can add multiple predictor variables
what is stepwise regression
examines the impact of each variable to the model
- variables that cannot contribute much to the variance explained are thrown out
describe information-theoretic approach
develops linklihood of a particular model being correct, given the data
AIC is used for what
to discriminate between a series of candidate models baed on the principle of parsimony
(the simplest of two models should be preferred)
- compare and rank multiple competing models
- estimates which of them approximates the true process
how do you decide what variables to make available in your models
- use common sense
- use correlations, including partial correlations, use PCA
does AIC depend on N
AIC is weakly dependent on N
- if N is small, relatively little information contained in the data
- maximum number of variables to allow in the model should ne n/10
maximum number of variables to allow in the model is
n/10 (n=# samples)
the lower the AIC value,
the better the model
how to pick the best model with AIC
- run every model and calculate AIC for each
- determine the smallest
- calculate the change in AIC for each model
if change in AIC fr the model is 0-2,
highest support
if change in AIC for the model is 4-7,
there is considerably less supprt
if the change in AIC for the model is >10,
there is essentially no support for the model
