Digit Span Forward - Psychological Aggression
Model 1
DSFlog1 <- glm(PsychAggress ~ DigitSpanFwd + WRATtotal, data=Allvars,family = "binomial")
summary(DSFlog1)
##
## Call:
## glm(formula = PsychAggress ~ DigitSpanFwd + WRATtotal, family = "binomial",
## data = Allvars)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.2313 0.4682 0.5210 0.5643 0.7291
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 0.78702 0.64531 1.220 0.223
## DigitSpanFwd 0.07846 0.05929 1.323 0.186
## WRATtotal 0.01168 0.01645 0.710 0.478
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 505.40 on 640 degrees of freedom
## Residual deviance: 501.43 on 638 degrees of freedom
## AIC: 507.43
##
## Number of Fisher Scoring iterations: 4
exp(cbind(OR = coef(DSFlog1), confint(DSFlog1)))
## OR 2.5 % 97.5 %
## (Intercept) 2.196839 0.6382732 8.079733
## DigitSpanFwd 1.081621 0.9650907 1.218189
## WRATtotal 1.011748 0.9790101 1.044422
########Compare to null model
#Difference in Deviance
with(DSFlog1,null.deviance - deviance)
## [1] 3.968932
#Degrees of freedom for the difference between two models
with(DSFlog1,df.null - df.residual)
## [1] 2
#p-value
with(DSFlog1,pchisq(null.deviance-deviance,df.null-df.residual,lower.tail=FALSE))
## [1] 0.137454
#Pseudo R-Squared
nagelkerke(DSFlog1)
## $Models
##
## Model: "glm, PsychAggress ~ DigitSpanFwd + WRATtotal, binomial, Allvars"
## Null: "glm, PsychAggress ~ 1, binomial, Allvars"
##
## $Pseudo.R.squared.for.model.vs.null
## Pseudo.R.squared
## McFadden 0.00785303
## Cox and Snell (ML) 0.00617265
## Nagelkerke (Cragg and Uhler) 0.01131650
##
## $Likelihood.ratio.test
## Df.diff LogLik.diff Chisq p.value
## -2 -1.9845 3.9689 0.13745
##
## $Number.of.observations
##
## Model: 641
## Null: 641
##
## $Messages
## [1] "Note: For models fit with REML, these statistics are based on refitting with ML"
##
## $Warnings
## [1] "None"
Model 3
DSFlog3 <- glm(PsychAggress ~ (DigitSpanFwd + Sex + PovStat)^3 + Age + WRATtotal, data = Allvars, family = "binomial")
summary(DSFlog3)
##
## Call:
## glm(formula = PsychAggress ~ (DigitSpanFwd + Sex + PovStat)^3 +
## Age + WRATtotal, family = "binomial", data = Allvars)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.2630 0.4174 0.4864 0.5671 0.8936
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 2.05209 1.11257 1.844 0.0651 .
## DigitSpanFwd 0.04840 0.09789 0.494 0.6210
## SexMen -0.95692 0.94536 -1.012 0.3114
## PovStatBelow 0.19631 1.31460 0.149 0.8813
## Age -0.02016 0.01320 -1.527 0.1267
## WRATtotal 0.01412 0.01657 0.852 0.3939
## DigitSpanFwd:SexMen 0.05684 0.12434 0.457 0.6476
## DigitSpanFwd:PovStatBelow -0.01234 0.17920 -0.069 0.9451
## SexMen:PovStatBelow 0.20818 1.84272 0.113 0.9100
## DigitSpanFwd:SexMen:PovStatBelow 0.02781 0.25559 0.109 0.9134
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 505.40 on 640 degrees of freedom
## Residual deviance: 491.88 on 631 degrees of freedom
## AIC: 511.88
##
## Number of Fisher Scoring iterations: 5
exp(cbind(OR = coef(DSFlog3), confint(DSFlog3)))
## OR 2.5 % 97.5 %
## (Intercept) 7.7841672 0.91152064 72.181868
## DigitSpanFwd 1.0495926 0.87188396 1.282729
## SexMen 0.3840725 0.05929822 2.438312
## PovStatBelow 1.2169073 0.09333419 16.695902
## Age 0.9800401 0.95480585 1.005630
## WRATtotal 1.0142231 0.98123115 1.047252
## DigitSpanFwd:SexMen 1.0584906 0.82721892 1.349796
## DigitSpanFwd:PovStatBelow 0.9877360 0.69974951 1.421807
## SexMen:PovStatBelow 1.2314393 0.03214253 45.408760
## DigitSpanFwd:SexMen:PovStatBelow 1.0281966 0.62385173 1.711342
########Compare to null model
#Difference in Deviance
with(DSFlog3,null.deviance - deviance)
## [1] 13.52276
#Degrees of freedom for the difference between two models
with(DSFlog3,df.null - df.residual)
## [1] 9
#p-value
with(DSFlog3,pchisq(null.deviance-deviance,df.null-df.residual,lower.tail=FALSE))
## [1] 0.1403431
#Pseudo R-Squared
nagelkerke(DSFlog3)
## $Models
##
## Model: "glm, PsychAggress ~ (DigitSpanFwd + Sex + PovStat)^3 + Age + WRATtotal, binomial, Allvars"
## Null: "glm, PsychAggress ~ 1, binomial, Allvars"
##
## $Pseudo.R.squared.for.model.vs.null
## Pseudo.R.squared
## McFadden 0.0267565
## Cox and Snell (ML) 0.0208754
## Nagelkerke (Cragg and Uhler) 0.0382715
##
## $Likelihood.ratio.test
## Df.diff LogLik.diff Chisq p.value
## -9 -6.7614 13.523 0.14034
##
## $Number.of.observations
##
## Model: 641
## Null: 641
##
## $Messages
## [1] "Note: For models fit with REML, these statistics are based on refitting with ML"
##
## $Warnings
## [1] "None"
Compare Models 1 & 3
anova(DSFlog1,DSFlog3,test = "LR")
## Analysis of Deviance Table
##
## Model 1: PsychAggress ~ DigitSpanFwd + WRATtotal
## Model 2: PsychAggress ~ (DigitSpanFwd + Sex + PovStat)^3 + Age + WRATtotal
## Resid. Df Resid. Dev Df Deviance Pr(>Chi)
## 1 638 501.43
## 2 631 491.88 7 9.5538 0.2153
Digit Span Forward - Physical Assault
Model 1
DSFlog1 <- glm(PhysAssault ~ DigitSpanFwd + WRATtotal, data=Allvars,family = "binomial")
summary(DSFlog1)
##
## Call:
## glm(formula = PhysAssault ~ DigitSpanFwd + WRATtotal, family = "binomial",
## data = Allvars)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -0.5618 -0.5322 -0.5148 -0.4907 2.1702
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -2.532311 0.732701 -3.456 0.000548 ***
## DigitSpanFwd 0.001784 0.058017 0.031 0.975469
## WRATtotal 0.013030 0.018186 0.716 0.473706
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 482.53 on 640 degrees of freedom
## Residual deviance: 481.85 on 638 degrees of freedom
## AIC: 487.85
##
## Number of Fisher Scoring iterations: 4
exp(cbind(OR = coef(DSFlog1), confint(DSFlog1)))
## OR 2.5 % 97.5 %
## (Intercept) 0.07947516 0.01776384 0.3163918
## DigitSpanFwd 1.00178564 0.89231857 1.1207414
## WRATtotal 1.01311505 0.97850173 1.0509656
########Compare to null model
#Difference in Deviance
with(DSFlog1,null.deviance - deviance)
## [1] 0.6824734
#Degrees of freedom for the difference between two models
with(DSFlog1,df.null - df.residual)
## [1] 2
#p-value
with(DSFlog1,pchisq(null.deviance-deviance,df.null-df.residual,lower.tail=FALSE))
## [1] 0.7108906
#Pseudo R-Squared
nagelkerke(DSFlog1)
## $Models
##
## Model: "glm, PhysAssault ~ DigitSpanFwd + WRATtotal, binomial, Allvars"
## Null: "glm, PhysAssault ~ 1, binomial, Allvars"
##
## $Pseudo.R.squared.for.model.vs.null
## Pseudo.R.squared
## McFadden 0.00141436
## Cox and Snell (ML) 0.00106413
## Nagelkerke (Cragg and Uhler) 0.00201180
##
## $Likelihood.ratio.test
## Df.diff LogLik.diff Chisq p.value
## -2 -0.34124 0.68247 0.71089
##
## $Number.of.observations
##
## Model: 641
## Null: 641
##
## $Messages
## [1] "Note: For models fit with REML, these statistics are based on refitting with ML"
##
## $Warnings
## [1] "None"
Model 3
DSFlog3 <- glm(PhysAssault ~ (DigitSpanFwd + Sex + PovStat)^3 + Age + WRATtotal, data = Allvars, family = "binomial")
summary(DSFlog3)
##
## Call:
## glm(formula = PhysAssault ~ (DigitSpanFwd + Sex + PovStat)^3 +
## Age + WRATtotal, family = "binomial", data = Allvars)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -0.8189 -0.5530 -0.4756 -0.3929 2.4095
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -1.19141 1.18123 -1.009 0.3132
## DigitSpanFwd -0.01788 0.09490 -0.188 0.8505
## SexMen -0.47145 1.09257 -0.432 0.6661
## PovStatBelow 0.30840 1.14433 0.270 0.7875
## Age -0.03343 0.01381 -2.421 0.0155 *
## WRATtotal 0.01844 0.01896 0.973 0.3306
## DigitSpanFwd:SexMen 0.03636 0.13301 0.273 0.7846
## DigitSpanFwd:PovStatBelow 0.01956 0.14872 0.131 0.8954
## SexMen:PovStatBelow 0.55778 1.74829 0.319 0.7497
## DigitSpanFwd:SexMen:PovStatBelow -0.06336 0.22405 -0.283 0.7773
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 482.53 on 640 degrees of freedom
## Residual deviance: 469.30 on 631 degrees of freedom
## AIC: 489.3
##
## Number of Fisher Scoring iterations: 5
exp(cbind(OR = coef(DSFlog3), confint(DSFlog3)))
## OR 2.5 % 97.5 %
## (Intercept) 0.3037925 0.02850390 2.9521760
## DigitSpanFwd 0.9822779 0.81000265 1.1779150
## SexMen 0.6240948 0.07160496 5.2782498
## PovStatBelow 1.3612487 0.14165469 12.8327101
## Age 0.9671185 0.94088741 0.9933439
## WRATtotal 1.0186144 0.98239582 1.0583396
## DigitSpanFwd:SexMen 1.0370278 0.79841304 1.3478636
## DigitSpanFwd:PovStatBelow 1.0197475 0.75939339 1.3652236
## SexMen:PovStatBelow 1.7467977 0.05645784 54.5444613
## DigitSpanFwd:SexMen:PovStatBelow 0.9386075 0.60233209 1.4547645
########Compare to null model
#Difference in Deviance
with(DSFlog3,null.deviance - deviance)
## [1] 13.23718
#Degrees of freedom for the difference between two models
with(DSFlog3,df.null - df.residual)
## [1] 9
#p-value
with(DSFlog3,pchisq(null.deviance-deviance,df.null-df.residual,lower.tail=FALSE))
## [1] 0.1521644
#Pseudo R-Squared
nagelkerke(DSFlog3)
## $Models
##
## Model: "glm, PhysAssault ~ (DigitSpanFwd + Sex + PovStat)^3 + Age + WRATtotal, binomial, Allvars"
## Null: "glm, PhysAssault ~ 1, binomial, Allvars"
##
## $Pseudo.R.squared.for.model.vs.null
## Pseudo.R.squared
## McFadden 0.0274327
## Cox and Snell (ML) 0.0204391
## Nagelkerke (Cragg and Uhler) 0.0386412
##
## $Likelihood.ratio.test
## Df.diff LogLik.diff Chisq p.value
## -9 -6.6186 13.237 0.15216
##
## $Number.of.observations
##
## Model: 641
## Null: 641
##
## $Messages
## [1] "Note: For models fit with REML, these statistics are based on refitting with ML"
##
## $Warnings
## [1] "None"
Compare Models 1 & 3
anova(DSFlog1,DSFlog3,test = "LR")
## Analysis of Deviance Table
##
## Model 1: PhysAssault ~ DigitSpanFwd + WRATtotal
## Model 2: PhysAssault ~ (DigitSpanFwd + Sex + PovStat)^3 + Age + WRATtotal
## Resid. Df Resid. Dev Df Deviance Pr(>Chi)
## 1 638 481.85
## 2 631 469.30 7 12.555 0.08373 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1