The content/main points of the course:

• Basic intro to R
• Reading in a Shapefile
• Doing a table join with a shapefile and a data.frame
• Generating random points
• Doing a point in polygon spatial join
• Reading and cropping raster data
• Doing a pixel in polygon spatial join (using extract() from the raster pacakge)
• Plotting and annotating a shapefile in ggplot2
• Making panel maps in ggplot2

The notes contain all the R-code and a lot of grammatical and spelling errors. I hope to improve and update this class over time. Please let me know if:

1. You find the material useful
2. You use it/modify for a class of your own (and please share the results)
3. You have any major suggestions for improvements (I'm aware of most of the minor issues)

I can never seem to get exactly what I want from an R text editor. Let me correct that, I can never seem to get exactly what I want from an R text editor on a MAC. I used to use Tinn-R  which met most  my needs:

• Free,lightweight with an easy install and a slim UI
• Good commenting and code folding tools
• Customizable short cuts
• Support for multiple monitors
• An R console that was separate from the editor
• A flexible, simple, and powerful Find and Replace Interface
•  Good SWeave support with some built in LaTeX tools

When I made the switch to Mac I went with StatET  on Eclipse   StatET has more or less all those things, but it is not simple or straightforward (or maybe just Eclipse is too complex for me). It also has somewhat lousy graphics support and does not allow for fuzzy help searches (??’help’). However, it worked so I’ve been using it. But I just upgraded to R 2.15 and the built in graphics driver (and a few other options) no longer work.  I was never that attached to StatET anyway because the Eclipse environment is just too complex for my needs.

I really like R Studio  but it’s missing some very key features:

• The ability to Find and Replace on a selection, rather than the whole document, plus general ‘fancy’ text editing find and replace tools
• Multiple monitor support (I’d like to be able to detach one of the panes and move them to another monitor)
• Customizable shortcuts
• The ability to restart the R console without restarting the editor (this I can live without, but it’s nice to have)

So what else is out there? Maybe it’s time to try TextMate? In general I prefer free and cross platform, but I can make exceptions.

Addendum: This one by Digithead is also pretty good

#----------Simulate an Arima (2,1,1) Process-------------
library(forecast)

set.seed(1234)  
y<-arima.sim(model=list(order=c(2,1,1),ar=c(0.5,.3),ma=0.3),n=144)
y<-ts(y,freq=12,start=c(2000,1))
 
#-- Extract Training Data, Fit the Wrong Model, and Forecast
yt<-window(y,end=2009.99)
 
yfit<-Arima(yt,order=c(1,0,1))
 
yfor<-forecast(yfit)
 
#---Extract the Data for ggplot using funggcast()
pd<-funggcast(y,yfor)
 
#---Plot in ggplot2 0.9
library(ggplot2)
library(scales)
 
 
p1a<-ggplot(data=pd,aes(x=date,y=observed)) 
p1a<-p1a+geom_line(col='red')
p1a<-p1a+geom_line(aes(y=fitted),col='blue')
p1a<-p1a+geom_line(aes(y=forecast))+geom_ribbon(aes(ymin=lo95,ymax=hi95),alpha=.25)
p1a<-p1a+scale_x_date(name='',breaks='1 year',minor_breaks='1 month',labels=date_format("%b-%y"),expand=c(0,0))
p1a<-p1a+scale_y_continuous(name='Units of Y')
p1a<-p1a+opts(axis.text.x=theme_text(size=10),title='Arima Fit to Simulated Data\n (black=forecast, blue=fitted, red=data, shadow=95% conf. interval)')
#p1a

Created by Pretty R at inside-R.org

The function below does all that and should work for any forecast object (though I've only tested it on Arima() outputs). The only arguments it takes are the original observations and the forecast object (whatever results from calling forecast()). In my next post I'll give some examples of plotting the results using ggplot and explain why I wanted more than the default plot.forecast() function.

#--Produces a data.frame with the Source Data+Training Data, Fitted Values+Forecast Values, forecast data Confidence Intervals
funggcast<-function(dn,fcast){ 
	require(zoo) #needed for the 'as.yearmon()' function
 
	en<-max(time(fcast$mean)) #extract the max date used in the forecast
 
	#Extract Source and Training Data
	ds<-as.data.frame(window(dn,end=en))
	names(ds)<-'observed'
	ds$date<-as.Date(time(window(dn,end=en)))
 
	#Extract the Fitted Values (need to figure out how to grab confidence intervals)
	dfit<-as.data.frame(fcast$fitted)
	dfit$date<-as.Date(time(fcast$fitted))
	names(dfit)[1]<-'fitted'
 
	ds<-merge(ds,dfit,all.x=T) #Merge fitted values with source and training data
 
	#Exract the Forecast values and confidence intervals
	dfcastn<-as.data.frame(fcast)
	dfcastn$date<-as.Date(as.yearmon(row.names(dfcastn)))
	names(dfcastn)<-c('forecast','lo80','hi80','lo95','hi95','date')
 
	pd<-merge(ds,dfcastn,all.x=T) #final data.frame for use in ggplot
	return(pd)
 
}

Created by Pretty R at inside-R.org

#---Simulate Data and Fit a linear model with an interaction term
y<-rnorm(100,5,1)
x<-rnorm(100,5,1)
d<-data.frame(y=y,x=x,fac=sample(letters[1:3],100,replace=T))
 
mod<-lm(y~x*fac,data=d)
 
#========================================================
 
#---Extract the Main Effects, including the baseline, into a data.frame
dusp<-funinteff(mod,'x') #returns a data.frame of the Estimate and Standard Error, row.names correspond to the variables
 
#----Now Set the data up to visualize in ggplot-----
library(ggplot2)
#------Quick ggplot (move into graph code later)
#quick convenience function to compute significance at .95
funsig<-function(d){
	tstat<-abs(d$b/d$se)
	sig<-ifelse(tstat>=1.96,'yes','no')
	return(sig)
}
 
 
names(dusp)[1:2]<-c('b','se') #change the names to to make typing easier
 
#Add confidence intervals and signficance test
dusp$hi<-dusp$b+1.96*dusp$se
dusp$lo<-dusp$b-1.96*dusp$se
dusp$sig95<-funsig(dusp)
 
dusp$var<-row.names(dusp)
 
 
pd<-dusp
 
p1<-ggplot(data=pd,aes(x=var,y=b,shape=sig95))
p1<-p1+geom_hline(yintercept=0,col='grey')+geom_line()
p1<-p1+geom_pointrange(aes(ymin=lo,ymax=hi)) #+coord_flip() #uncomment coord_flip to switch the axes
p1<-p1+scale_y_continuous(name='Marginal Effect of Interaction Terms')

Created by Pretty R at inside-R.org

 Note: The code below assumes you have already loaded car (for deltaMethod()), sandwich (for vcovHAC), and lmtest (for coeftest()).

##Calls this function which is a wrapper for coeftest from the sandwhich package
sehac<-function(fit,vcov=vcovHAC){ #Convenience function for HAC standard erros
	coeftest(fit,vcov)
}
 
funinteff<-function(mod,var,vcov=vcovHAC){ #mod is an lm() object, var is the name of the main effect that was interacted, vcov is the type of variance covariance method you want to use 
	#Extract Coefficient names create 'beta names' to feed to deltaMethod()
	cnames<-coef(mod)
	pnams<-data.frame('b'=paste('b',0:(length(cnames)-1),sep=""),'est'=cnames) #assign parameter names so that deltaMethod does not throw an error
 
	#Extract the specific parameters of interest
	vars<-grep(var,names(cnames),value=T)
	var1<-vars[1]
	intvars<-vars[2:length(vars)]
	bi<-pnams[var1,'b']
 
	#--Create Data Frame to store Main Effect
	int<-sehac(mod,vcov=vcov)[var1,c('Estimate','Std. Error')]
	int<-as.data.frame(t(int))
	names(int)<-c('Estimate','SE')
	row.names(int)<-var1
 
	#Loop through and store the results in a data.frame
	for(i in 1:length(intvars)){
		bint<-pnams[intvars[i],'b']
		eq<-paste(bi,bint,sep="+")
		interac<-deltaMethod(mod,eq,parameterNames=pnams[,1],vcov=vcov)
		row.names(interac)<-intvars[i]
 
		int<-rbind(int,interac)
	}
	return(int)
}

Created by Pretty R at inside-R.org

For example, unemployment often exhibits high spatial correlation. Does this mean that if your neighbor loses her job it will likely happen to you too, simply because you live next door? No, people with similar incomes, and often those who work in similar industries, tend to live in close proximity to one another. So the same forces that impacted your neighbors employer might also impact yours.

This may seem like an obvious argument, but I see this mistake alot. People, usually geographers, use the fact that many things are spatially correlated to argue that 'place matters' and 'spatial is special'. Well, sort of. In the above example, yes place does matter, but only as a special case of cross-sectional correlation, that if properly controlled for, will not impact whatever broader statistical inference you want to make with the data.

I think this problem stems from the back-seat that Geography had in academia for much of the 20th century. Prior to the quantitative revolution (in geography) and especially the availability of consumer accessible GIS packages, Geographers spent alot of time persuading others that what they did was relevant. This has created a great sense of camaderie within the community, but, especially after the rise of GIS and Remote Sensing, has created a sense that every other discipline is misguided until they see the spatial light.

I use to feel this way. As a PhD student in a prominent geography department I see (and sympathize) with this view alot. It was not until I really delved into statistics and started working with economists that I realized that yeah, spatial is special, but not all time, and the burden of proof lies with the geographers. And, even if spaital is not special, there may still be a lot of interesting questions to ask of spatially correlated data.