how to plot actual and predicted values?

1
votes

Im a student of statistics and i would like kindly request for some assistance. I would like to plot predicted values together with actual values of the course of 100 days in my dataset:

Sample Data:

             set.seet(1010)
             count<-rpois(100, lambda = 5)
             mood<- rbinom(100, size = 1, prob = .7)
             temp<-rnorm(100, mean = 20, sd = 5)
             wind<-rbinom(100, size = 3, prob = .7)
             days<-seq(1,100,by=1)
             df<-data.frame(count,mood,temp,wind,days)

Plotting actual values during 100 days:

             plot(count~days,type="l")

Regression:

             poisson <- glm(count ~mood+wind+temp)

Condition on my predictors and otaining predicted values:

             hyp<- c(1,1,3,20)
             coeff.p1 <- poisson$coefficients
             XB <- hyp%*%coeff.p1
             predv.y <- exp(XB)
             predv.y

May be there is a way to predict values for all observations as for example:

              coeff.p1 <- poisson$coefficients
              XB <- c(,2:4)%*%t(coeff.p1)

I intend to multiply with columns 2:4 by always get 

              Error in c(, 2:4) : argument 1 is empty

Here Im stuck. As a result I would like to obtain predicted values and actuall values for 100 day on one plot.

Thank you

2
r

2 Answers

2
votes

Your object poisson is of class glm, so it has a predict method when given any data. 

poisson <- glm(count ~mood+temp+ wind)
df$pred<-predict(poisson,df[,2:4])
plot(df$days,df$count)
lines(df$days, df$pred,type='l',col='blue')
0
votes

Try this:

set.seed(1010)
count<-rpois(100, lambda = 5)
mood<- rbinom(100, size = 1, prob = .7)
temp<-rnorm(100, mean = 20, sd = 5)
wind<-rbinom(100, size = 3, prob = .7)
days<-seq(1,100,by=1)
df<-data.frame(count,mood,temp,wind,days)
poisson <- glm(count ~ mood+wind+temp
               , family = poisson() #specify your model type
               , data=df)

# Calculate the predicted
phat.poisson <- predprob(poisson)  # for each subj, prob of observing each value
phat.poisson.mn <- apply(phat.poisson, 2, mean) # mean predicted probs

#your plot of observed vs. predicted
hist(count, prob = TRUE, col = "grey60", breaks=seq(-0.5, 12.5, 1), xlab = "Counts",main = NULL, ylim=c(0, .20))
lines(x = seq(0, 12, 1), y = phat.poisson.mn, lty=2, lwd=2, col="red")
points(x = seq(0, 12, 1), y = phat.poisson.mn, cex=1, pch=16, col="red")