Analysis code for looking at the relationship between page views and dollars raised for #SciFund

pageviews.R

source("./sciFundFunctions.r")
library(gridExtra)

############
#Begin with a plot of both with and without the outlier
############

viewsAll<-qplot(Pageviews, total, data=projects)+theme_bw(base_size=24) + ylab("Total Dollars Raised\n") + 
  xlab("\nPage Views") +
  scale_x_continuous(breaks=seq(0,40000, 10000))
viewsNoOut<-qplot(Pageviews, total, data=projects[-outIDX,])+theme_bw(base_size=24) + ylab("") + 
  xlab("\nPage Views")+geom_text(aes(x=2000, y=8500, label="No Outlier"), size=7)
grid.arrange(viewsAll, viewsNoOut, ncol=2)


#model the sucker!
viewGlm <- glm(total ~ Pageviews+0, data=projects, family=poisson(link="identity"))
viewGlmLog <- glm(total ~ Pageviews+0, data=projects, family=poisson(link="log"))

#linear or not so linear
AIC(viewGlm) - AIC(viewGlmLog)

#fit stats
Anova(viewGlm)
summary(viewGlm)
r2.corr(viewGlm)

############
#ok, now rinse and repeat, but without the outlier
############

viewGLM2 <- glm(total ~ Pageviews+0, data=projects[-outIDX,], family=poisson(link="identity"))
viewGLM2Log<- glm(total ~ Pageviews+0, data=projects[-outIDX,], family=poisson(link="log"))

#check linearity again
AIC(viewGLM2) -AIC(viewGLM2Log)

#and fit..
summary(viewGLM2)
r2.corr(viewGLM2)

############
#plot the fitted models
############

grid.arrange(viewsAll+stat_smooth(method="glm", family=poisson(link="identity")),
             viewsNoOut+stat_smooth(method="glm", family=poisson(link="identity")), ncol=2)


############
#Split pre and post success page views
############
viewTotalSplit<-qplot(preGoalPageviews, total, data=projects, size=postGoalPageviews)+theme_bw(base_size=24)+ 
  ylab("Total Dollars Raised") + 
  xlab("\nPage Views Before\nReaching Goal") +
  scale_size("Views After\n Goal")+
  scale_x_continuous(breaks=seq(0,40000, 10000))

grid.arrange(viewTotalSplit, viewTotalSplitNoOut, ncol=2)
viewTotalSplitNoOut<-qplot(preGoalPageviews, total, data=projects[-outIDX,], size=postGoalPageviews)+
  theme_bw(base_size=24)+ ylab("") + 
  xlab("\nPage Views Before\nReaching Goal")+geom_text(aes(x=2000, y=8500, label="No Outlier"), size=7) +
  scale_size("Views After\n Goal")+
  scale_x_continuous(breaks=seq(0,10000, 2000))

grid.arrange(viewTotalSplit, viewTotalSplitNoOut, ncol=2)

############
#Analysis
############

viewGLMSplit <- glm(total ~ preGoalPageviews+postGoalPageviews+0, data=projects, family=poisson(link="identity"))
Anova(viewGLMSplit)
summary(viewGLMSplit)
r2.corr(viewGLMSplit)

#witout outlier
viewGLMSplitNoOut <- glm(total ~ preGoalPageviews+postGoalPageviews+0, data=projects[-outIDX,], family=poisson(link="identity"))
anova(viewGLMSplitNoOut)
summary(viewGLMSplitNoOut)
r2.corr(viewGLMSplitNoOut)

#test for correlations between pre and post
cor(projects$preGoalPageviews, projects$postGoalPageviews)
cor(projects[-outIDX,]$preGoalPageviews, projects[-outIDX,]$postGoalPageviews)