Probit and Logit Models (Page 586)
library(foreign)
data = read.dta("Dataset/msc_fail.dta")
data=na.omit(data)
probit = glm(Fail~Age+English+Female+WorkExperience+Agrade+BelowBGrade+PGDegree+
Year2004+Year2005+Year2006+Year2007,data,family=binomial("probit"))
summary(probit)
##
## Call:
## glm(formula = Fail ~ Age + English + Female + WorkExperience +
## Agrade + BelowBGrade + PGDegree + Year2004 + Year2005 + Year2006 +
## Year2007, family = binomial("probit"), data = data)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.2066 -0.5872 -0.4269 -0.2792 2.6302
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.287212 0.583782 -2.205 0.02746 *
## Age 0.005677 0.020910 0.272 0.78601
## English -0.093792 0.154685 -0.606 0.54429
## Female -0.194107 0.185278 -1.048 0.29480
## WorkExperience -0.318247 0.156118 -2.039 0.04150 *
## Agrade -0.538813 0.235001 -2.293 0.02186 *
## BelowBGrade 0.341802 0.215586 1.585 0.11286
## PGDegree 0.132957 0.229911 0.578 0.56306
## Year2004 0.349664 0.255411 1.369 0.17099
## Year2005 -0.108329 0.289379 -0.374 0.70814
## Year2006 0.673613 0.245478 2.744 0.00607 **
## Year2007 0.433786 0.257441 1.685 0.09199 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 393.92 on 499 degrees of freedom
## Residual deviance: 358.91 on 488 degrees of freedom
## AIC: 382.91
##
## Number of Fisher Scoring iterations: 6
logit = glm(Fail~Age+English+Female+WorkExperience+Agrade+BelowBGrade+PGDegree+
Year2004+Year2005+Year2006+Year2007,data,family=binomial)
summary(logit)
##
## Call:
## glm(formula = Fail ~ Age + English + Female + WorkExperience +
## Agrade + BelowBGrade + PGDegree + Year2004 + Year2005 + Year2006 +
## Year2007, family = binomial, data = data)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.2275 -0.5870 -0.4228 -0.2980 2.5579
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.25637 1.07300 -2.103 0.03548 *
## Age 0.01101 0.03813 0.289 0.77275
## English -0.16512 0.28295 -0.584 0.55952
## Female -0.33389 0.34923 -0.956 0.33902
## WorkExperience -0.56877 0.28847 -1.972 0.04865 *
## Agrade -1.08503 0.49110 -2.209 0.02715 *
## BelowBGrade 0.56235 0.37351 1.506 0.13217
## PGDegree 0.21208 0.41990 0.505 0.61350
## Year2004 0.65321 0.50092 1.304 0.19223
## Year2005 -0.18382 0.58794 -0.313 0.75454
## Year2006 1.24658 0.47365 2.632 0.00849 **
## Year2007 0.85042 0.49705 1.711 0.08710 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 393.92 on 499 degrees of freedom
## Residual deviance: 359.43 on 488 degrees of freedom
## AIC: 383.43
##
## Number of Fisher Scoring iterations: 5
Figure 12.4 Fitted values from the failure probit regression (Page 587)
par(mfcol = c(1,1), oma = c(0,0,1,0) + 0.2, mar = c(0,1,0,0) + 1, mgp = c(0, 0.2, 0))
plot(probit$fitted.values,type="l",las=1,xlab="",ylab="",main="",xaxs="i",yaxs="i",tck=.02,col="navy",ylim=c(0,0.6))
Marginal effects for logit and probit models (Page 588)
library(margins)
margins(logit)
## Age English Female WorkExperience Agrade BelowBGrade PGDegree Year2004 Year2005
## 0.001186 -0.01779 -0.03597 -0.06127 -0.1169 0.06058 0.02285 0.07036 -0.0198
## Year2006 Year2007
## 0.1343 0.09161
## Age English Female WorkExperience Agrade BelowBGrade PGDegree Year2004 Year2005
## 0.001118 -0.01847 -0.03822 -0.06267 -0.1061 0.0673 0.02618 0.06885 -0.02133
## Year2006 Year2007
## 0.1326 0.08542