#Create scatterplot matrix
pairs(CO[,2:17])

#Fit Complete linear model
allterms <- lm(PercentYes ~ PercentObama + PercentUnder_18+Percent18_24+PercentOver65+MedianAge+AverageHouseholdSize+PercentPoorHealth+PercentSmoker+PercentObese+PercentUninsured+PercentGraduationRate+PercentPostHSEducation+MedianHouseholdIncome+PercentUnemployment+PercentChildrenPoverty,data=CO)
summary(allterms)

#Plot actual vs. predicted values
plot(PercentYes~predict(allterms,newdata=CO), data=CO)
abline(0,1)

#Computing R-Squared from plot
cor(CO$PercentYes,predict(allterms,newdata=CO))^2


#Getting rid of NA in orginal data.frame
CO_NoNA <- na.omit(CO)

allterms <- lm(PercentYes ~ PercentObama + PercentUnder_18+Percent18_24+PercentOver65+MedianAge+AverageHouseholdSize+PercentPoorHealth+PercentSmoker+PercentObese+PercentUninsured+PercentGraduationRate+PercentPostHSEducation+MedianHouseholdIncome+PercentUnemployment+PercentChildrenPoverty,data=CO_NoNA)
summary(allterms)

#Problems with NA avoided now...
cor(CO_NoNA$PercentYes,predict(allterms,newdata=CO_NoNA))^2

#Getting predictor importance information
importance_order <- order(abs(summary(allterms)$coefficients[,4]))
summary(allterms)$coefficients[importance_order,]

avg_error_original <- mean(abs(CO_NoNA$PercentYes - predict(allterms,newdata=CO_NoNA)))
avg_error_original




 CO_NoNA_Temp <- CO_NoNA
 View(CO_NoNA_Temp)
 CO_NoNA_Temp[,(2+i)] <- sample(CO_NoNA_Temp[,(2+i)])
 avg_error_temp[i] <- mean(abs(CO_NoNA_Temp$PercentYes - predict(allterms,newdata=CO_NoNA_Temp)))


avg_error_temp = data.frame(i=rep(0,15),error=rep(0,15))
#Looping through all predictors
for(i in 1:15){
  CO_NoNA_Temp <- CO_NoNA
  #View(CO_NoNA_Temp)
  avg_error_temp[i,1] <- i
  CO_NoNA_Temp[,(2+i)] <- sample(CO_NoNA_Temp[,(2+i)])
  avg_error_temp[i,2] <- mean(abs(CO_NoNA_Temp$PercentYes - predict(allterms,newdata=CO_NoNA_Temp)))
}