Post-estimation commands for regression models for ... › meeting › 1nasug › freese.pdf ·...
Transcript of Post-estimation commands for regression models for ... › meeting › 1nasug › freese.pdf ·...
Post-estimation commands forregression models forcategorical & count outcomes
Jeremy FreeseUniversity of Wisconsin-Madison
J. Scott LongIndiana University
Models for categorical andcount outcomes
n Stata makes estimating these modelseasy
n Interpretation is more complicatedn Our SPost suite of commands is
designed to facilitate interpretation andother tasks with these models
n Type net search spost to download
SPost command: fitstat
n Computes goodness-of-fit statisticsn Both Pseudo-R2s and information
measures (i.e., AIC and BIC)
n Can be used with saving() and using()options
Why are results fromcategorical and count modelsoften difficult to interpret?
Nonlinearities mean thatinterpretation depends on the
values of all independentvariables.
SPost command: prvalue
n Produces predicted values for specifiedset of values of the independentvariables
n Specific values are set with x() optionn All other values set with rest() optionn save and dif options to calculate
differences between two sets of values
SPost command: prtab
n Predicted probabilities for a cross-classification of 2-4 categoricalindependent variables
n Values of other variables specified byx() and rest()
SPost command: prgen
n Adds pseudovariables to data that canbe used to generate plots of howpredicted probability changes overrange of continuous independentvariable
n from() and to() options specifyrange of independent variable
SPost command: praccum
n More flexible (but harder to use) syntaxthat works with more complex modelspecifications
SPost command: prchange
n Computes marginal and discretechange
n Discrete change from min->max, 0->1,as x increases by 1 unit, and as xincreases by 1 sd
SPost command: prcounts
n Akin to predict, but generates newvariables that contain the predictedprobabilities of observing counts 0through specified value (default=9)
n prcounts dog will generate dogp0,dogp1 ... dogp9, as well as cumulativeprobabilities dogs0 ... dogs9
SPost command: listcoef
n listcoef, std – standardizedcoefficients (x-standardized, y-standardized, fully standardized)
n listcoef, factor – factor changein the odds/expected count
n listcoef, percent – percentchange in the odds/expected count
SPost command: mlogtest
n Wald or LR test whether the effect of anindependent variable is zero across allequations
n Wald or LR test whether a pair ofoutcomes is indistinguishable
n Hausman or Small-Hsiao tests of the IIAassumption
SPost command: mlogview
n Dialog box interface for generatingdiscrete change plots or odds ratioplots for mlogit models