#### CFA FOR UNWANTED TRAITS #### library(lavaan) #### ** cross-validation #### library(caret) # setting seed to generate a # reproducible random sampling set.seed(3) train.test <- createFolds(final.data$ID, k = 2) train.test.data <- lapply(train.test, function(ind, dat) dat[ind,], dat = final.data) train.test.data1 <- train.test.data$Fold1 train.test.data2 <- train.test.data$Fold2 write.table(train.test.data1,file="train.test.data1.csv") write.table(train.test.data2,file="train.test.data2.csv") train.test.data1 <- read.table(file="train.test.data1.csv", stringsAsFactors = T) summary(train.test.data1) ## ** fitting different models #### #### **** null model #### null.model <- ' noise_sensitivity_score ~~ noise_sensitivity_score fearfulness_score ~~ fearfulness_score barking_score ~~ barking_score stranger_aggression_score ~~ stranger_aggression_score owner_aggression_score ~~ owner_aggression_score dog_aggression_score ~~ dog_aggression_score surface_phobia_score ~~ surface_phobia_score separation_behavior_score ~~ separation_behavior_score inattention_score ~~ inattention_score impulsivity_score ~~ impulsivity_score' fit.null.model.1 <- cfa(null.model, estimator="ML", missing="ml", data = train.test.data1) summary(fit.null.model.1, fit.measures=TRUE) fit.null.model.2 <- cfa(null.model, estimator="ML", missing="ml", data = train.test.data2) summary(fit.null.model.2, fit.measures=TRUE) #### **** one general factor #### p.model <- ' one.p =~ noise_sensitivity_score + fearfulness_score + barking_score + stranger_aggression_score + owner_aggression_score + dog_aggression_score + surface_phobia_score + separation_behavior_score + inattention_score + impulsivity_score' fit.p.model.1 <- cfa(p.model, estimator="ML", missing="ml", data = train.test.data1) summary(fit.p.model.1, fit.measures=TRUE) fit.p.model.2 <- cfa(p.model, estimator="ML", missing="ml", data = train.test.data2) summary(fit.p.model.2, fit.measures=TRUE) #### **** one general factor + ADHD #### p.a.model <- ' one.p =~ noise_sensitivity_score + fearfulness_score + barking_score + stranger_aggression_score + owner_aggression_score + dog_aggression_score + surface_phobia_score + separation_behavior_score adhd =~ inattention_score + impulsivity_score' fit.pa.model.1 <- cfa(p.a.model, estimator="ML", missing="ml", data = train.test.data1) summary(fit.pa.model.1, fit.measures=TRUE) fit.pa.model.2 <- cfa(p.a.model, estimator="ML", missing="ml", data = train.test.data2) summary(fit.pa.model.2, fit.measures=TRUE) #### **** internalizing, externalizing & ADHD #### i.e.a.model <- ' internal =~ noise_sensitivity_score + fearfulness_score + surface_phobia_score + separation_behavior_score external =~ barking_score + stranger_aggression_score + owner_aggression_score + dog_aggression_score adhd =~ inattention_score + impulsivity_score' fit.iea.model.1 <- cfa(i.e.a.model, estimator="ML", missing="ml", data = train.test.data1) summary(fit.iea.model.1, fit.measures=TRUE) fit.iea.model.2 <- cfa(i.e.a.model, estimator="ML", missing="ml", data = train.test.data2) summary(fit.iea.model.2, fit.measures=TRUE) #### **** internalizing & externalizing #### i.e.model <- ' internal =~ noise_sensitivity_score + fearfulness_score + surface_phobia_score + separation_behavior_score external =~ barking_score + stranger_aggression_score + owner_aggression_score + dog_aggression_score + inattention_score + impulsivity_score' fit.ie.model.1 <- cfa(i.e.model, estimator="ML", missing="ml", data = train.test.data1) summary(fit.ie.model.1, fit.measures=TRUE) fit.ie.model.2 <- cfa(i.e.model, estimator="ML", missing="ml", data = train.test.data2) summary(fit.ie.model.2, fit.measures=TRUE) #### **** hierarchical, ADHD separate #### hierarchical.a.model <- ' internal =~ noise_sensitivity_score + fearfulness_score + surface_phobia_score + separation_behavior_score external =~ barking_score + stranger_aggression_score + owner_aggression_score + dog_aggression_score adhd =~ inattention_score + impulsivity_score one.p =~ internal + external' fit.hier.a.model.1 <- cfa(hierarchical.a.model, estimator="ML", missing="ml", data = train.test.data1) summary(fit.hier.a.model.1, fit.measures=TRUE) fit.hier.a.model.2 <- cfa(hierarchical.a.model, estimator="ML", missing="ml", data = train.test.data2) summary(fit.hier.a.model.2, fit.measures=TRUE) #### **** hierarchical #### hierarchical.model <- ' internal =~ noise_sensitivity_score + fearfulness_score + surface_phobia_score + separation_behavior_score external =~ barking_score + stranger_aggression_score + owner_aggression_score + dog_aggression_score + inattention_score + impulsivity_score one.p =~ internal + external' fit.hier.model.1 <- cfa(hierarchical.model, estimator="ML", missing="ml", start="simple", data = train.test.data1) summary(fit.hier.model.1, fit.measures=TRUE) fit.hier.model.2 <- cfa(hierarchical.model, estimator="ML", missing="ml", start="simple", data = train.test.data2) summary(fit.hier.model.2, fit.measures=TRUE) #### **** dog studies #### dog.model <- ' fearaggre =~ fearfulness_score + barking_score + stranger_aggression_score fear.related =~ noise_sensitivity_score + separation_behavior_score + surface_phobia_score + fearfulness_score aggression =~ owner_aggression_score + dog_aggression_score + stranger_aggression_score adhd =~ inattention_score + impulsivity_score' fit.dog.model.1 <- cfa(dog.model, estimator="ML", missing="ml", data = train.test.data1) summary(fit.dog.model.1, fit.measures=TRUE) standardizedSolution(fit.dog.model.1, output="pretty") fit.dog.model.2 <- cfa(dog.model, estimator="ML", missing="ml", data = train.test.data2) summary(fit.dog.model.2, fit.measures=TRUE) #### ** model comparison #### library(semTools) library(nonnest2) #### **** models vs null model #### # are the models nested? net(fit.null.model.1, fit.p.model.1) net(fit.null.model.1, fit.pa.model.1) net(fit.null.model.1, fit.iea.model.1) net(fit.null.model.1, fit.ie.model.1) net(fit.null.model.1, fit.hier.a.model.1) net(fit.null.model.1, fit.hier.model.1) net(fit.null.model.1, fit.dog.model.1) # all models nested, thus use nested=TRUE in vuong test # p.model fits better than null vuongtest(fit.null.model.1, fit.p.model.1, nested=TRUE) # pa.model fits better than null vuongtest(fit.null.model.1, fit.pa.model.1, nested=TRUE) # iea model fits better than null vuongtest(fit.null.model.1, fit.iea.model.1, nested=TRUE) # ie model fits better than null vuongtest(fit.null.model.1, fit.ie.model.1, nested=TRUE) # hier.a model fits better than null vuongtest(fit.null.model.1, fit.hier.a.model.1, nested=TRUE) # hier model fits better than null vuongtest(fit.null.model.1, fit.hier.model.1, nested=TRUE) # dog model fits better than null vuongtest(fit.null.model.1, fit.dog.model.1, nested=TRUE) vuongtest(fit.null.model.2, fit.p.model.2, nested=TRUE) vuongtest(fit.null.model.2, fit.pa.model.2, nested=TRUE) vuongtest(fit.null.model.2, fit.iea.model.2, nested=TRUE) vuongtest(fit.null.model.2, fit.ie.model.2, nested=TRUE) vuongtest(fit.null.model.2, fit.hier.a.model.2, nested=TRUE) vuongtest(fit.null.model.2, fit.hier.model.2, nested=TRUE) vuongtest(fit.null.model.2, fit.dog.model.2, nested=TRUE) #### **** other models vs p model #### # pa.model fits better than p vuongtest(fit.p.model.1, fit.pa.model.1, nested=TRUE) # iea model fits better than p vuongtest(fit.p.model.1, fit.iea.model.1, nested=TRUE) # ie model fits better than p vuongtest(fit.p.model.1, fit.ie.model.1, nested=TRUE) # hier.a model fits better than p vuongtest(fit.p.model.1, fit.hier.a.model.1, nested=TRUE) # hier model fits better than p vuongtest(fit.p.model.1, fit.hier.model.1, nested=TRUE) # dog model fits better than p vuongtest(fit.p.model.1, fit.dog.model.1, nested=TRUE) vuongtest(fit.p.model.2, fit.pa.model.2, nested=TRUE) vuongtest(fit.p.model.2, fit.iea.model.2, nested=TRUE) vuongtest(fit.p.model.2, fit.ie.model.2, nested=TRUE) vuongtest(fit.p.model.2, fit.hier.a.model.2, nested=TRUE) vuongtest(fit.p.model.2, fit.hier.model.2, nested=TRUE) vuongtest(fit.p.model.2, fit.dog.model.2, nested=TRUE) #### **** other models vs pa model #### # iea model fits better than pa vuongtest(fit.pa.model.1, fit.iea.model.1, nested=TRUE) # pa model fits better than ie vuongtest(fit.pa.model.1, fit.ie.model.1, nested=TRUE) # hier.a model fits better than pa vuongtest(fit.pa.model.1, fit.hier.a.model.1, nested=TRUE) # pa model fits better than hier vuongtest(fit.pa.model.1, fit.hier.model.1, nested=TRUE) # dog model fits better than pa vuongtest(fit.pa.model.1, fit.dog.model.1, nested=TRUE) vuongtest(fit.pa.model.2, fit.iea.model.2, nested=TRUE) vuongtest(fit.pa.model.2, fit.ie.model.2, nested=TRUE) vuongtest(fit.pa.model.2, fit.hier.a.model.2, nested=TRUE) vuongtest(fit.pa.model.2, fit.hier.model.2, nested=TRUE) vuongtest(fit.pa.model.2, fit.dog.model.2, nested=TRUE) #### **** other models vs iea model #### # iea model fits better than ie vuongtest(fit.iea.model.1, fit.ie.model.1, nested=TRUE) # hier.a and iea indistinguishable vuongtest(fit.iea.model.1, fit.hier.a.model.1, nested=TRUE) # iea model fits better than hier vuongtest(fit.iea.model.1, fit.hier.model.1, nested=TRUE) # dog model fits better than iea vuongtest(fit.iea.model.1, fit.dog.model.1, nested=TRUE) vuongtest(fit.iea.model.2, fit.ie.model.2, nested=TRUE) vuongtest(fit.iea.model.2, fit.hier.a.model.2, nested=TRUE) vuongtest(fit.iea.model.2, fit.hier.model.2, nested=TRUE) vuongtest(fit.iea.model.2, fit.dog.model.2, nested=TRUE) #### **** other models vs ie model #### # hier.a fits better than ie vuongtest(fit.ie.model.1, fit.hier.a.model.1, nested=TRUE) # hier and ie indistinguishable vuongtest(fit.ie.model.1, fit.hier.model.1, nested=TRUE) # dog model fits better than ie vuongtest(fit.ie.model.1, fit.dog.model.1, nested=TRUE) vuongtest(fit.ie.model.2, fit.hier.a.model.2, nested=TRUE) vuongtest(fit.ie.model.2, fit.hier.model.2, nested=TRUE) vuongtest(fit.ie.model.2, fit.dog.model.2, nested=TRUE) #### **** other models vs hier.a model #### # hier.a fits better than hier vuongtest(fit.hier.a.model.1, fit.hier.model.1, nested=TRUE) # dog model fits better than hier.a vuongtest(fit.hier.a.model.1, fit.dog.model.1, nested=TRUE) vuongtest(fit.hier.a.model.2, fit.hier.model.2, nested=TRUE) vuongtest(fit.hier.a.model.2, fit.dog.model.2, nested=TRUE) #### **** other models vs hier #### # dog model fits better than hier vuongtest(fit.hier.model.1, fit.dog.model.1, nested=TRUE) vuongtest(fit.hier.model.2, fit.dog.model.2, nested=TRUE) #### SEM BASED ON THE DOG MODEL #### library(lavaan) final.model<- ' # latent variables fearaggre =~ fearfulness_score + barking_score + stranger_aggression_score fear.related =~ noise_sensitivity_score + separation_behavior_score + surface_phobia_score + fearfulness_score aggression =~ owner_aggression_score + dog_aggression_score + stranger_aggression_score adhd =~ inattention_score + impulsivity_score # regressions fear.related ~ insecurity_score + aggressiveness_dominance_score + perseverance_score + training_focus_score + activity_playfulness_score + human_sociability_score + dog_sociability_score fearaggre ~ insecurity_score + aggressiveness_dominance_score + perseverance_score + training_focus_score + activity_playfulness_score + human_sociability_score + dog_sociability_score aggression ~ insecurity_score + aggressiveness_dominance_score + perseverance_score + training_focus_score + activity_playfulness_score + human_sociability_score + dog_sociability_score adhd ~ insecurity_score + aggressiveness_dominance_score + perseverance_score + training_focus_score + activity_playfulness_score + human_sociability_score + dog_sociability_score noise_sensitivity_score ~ sex.bin + mean.age + socialization + noise_sensitivity_breed fearfulness_score ~ sex.bin + mean.age + socialization + fearfulness_breed barking_score ~ sex.bin + mean.age + socialization + barking_breed owner_aggression_score ~ sex.bin + mean.age + socialization + owner_aggression_breed stranger_aggression_score ~ sex.bin + mean.age + socialization + stranger_aggression_breed dog_aggression_score ~ sex.bin + mean.age + socialization + dog_aggression_breed surface_phobia_score ~ sex.bin + mean.age + socialization + surface_phobia_breed separation_behavior_score ~ sex.bin + mean.age + socialization + separation_behavior_breed inattention_score ~ sex.bin + mean.age + socialization + inattention_breed impulsivity_score ~ sex.bin + mean.age + socialization + impulsivity_breed insecurity_score ~ sex.bin + mean.age + socialization + insecurity_breed aggressiveness_dominance_score ~ sex.bin + mean.age + socialization + aggressiveness_dominance_breed perseverance_score ~ sex.bin + mean.age + socialization + perseverance_breed training_focus_score ~ sex.bin + mean.age + socialization + training_focus_breed activity_playfulness_score ~ sex.bin + mean.age + socialization + activity_playfulness_breed human_sociability_score ~ sex.bin + mean.age + socialization + human_sociability_breed dog_sociability_score ~ sex.bin + mean.age + socialization + dog_sociability_breed # residual correlations insecurity_score ~~ training_focus_score insecurity_score ~~ aggressiveness_dominance_score insecurity_score ~~ dog_sociability_score insecurity_score ~~ human_sociability_score training_focus_score ~~ aggressiveness_dominance_score training_focus_score ~~ dog_sociability_score training_focus_score ~~ human_sociability_score activity_playfulness_score ~~ aggressiveness_dominance_score activity_playfulness_score ~~ dog_sociability_score activity_playfulness_score ~~ human_sociability_score aggressiveness_dominance_score ~~ dog_sociability_score aggressiveness_dominance_score ~~ human_sociability_score ' fit.final.model <- sem(final.model, estimator="MLR", missing="ml.x", fixed.x=T, data = final.data) # model fit fitmeasures(fit.final.model, c("cfi","tli","rmsea","srmr")) results <- standardizedSolution(fit.final.model, type="std.all", output="pretty") write.table(results,file="results.csv") summary(fit.final.model)