- 14.1 cbind and rbind
- 14.2 Joins
- 14.3 reshape2
- 14.4 Conclusion
This chapter is from the book
This chapter is from the book
This chapter is from the book
14.2 Joins
Data do not always come so nicely aligned for combing using cbind and need to be joined together using a common key. This concept should be familiar to SQL users. Joins in R are not as flexible as SQL joins, but are still an essential operation in the data analysis process.
The three most commonly used functions for joins are merge in base R, join in plyr and the merging functionality in data.table. Each has pros and cons with some pros outweighing their respective cons.
To illustrate these functions we have prepared data originally made available as part of the USAID Open Government initiative.1 The data have been chopped into eight separate files so that they can be joined together. They are all available in a zip file at http://jaredlander.com/data/US_Foreign_Aid.zip. These should be downloaded and unzipped to a folder on our computer. This can be done a number of ways (including using a mouse!) but we show how to download and unzip using R.
> download.file(url="http://jaredlander.com/data/US_Foreign_Aid.zip",
+ destfile="data/ForeignAid.zip")
> unzip("data/ForeignAid.zip", exdir="data")
To load all of these files programatically, we utilize a for loop as seen in Section 10.1. We get a list of the files using dir, and then loop through that list, assigning each dataset to a name specified using assign. The function str_sub extracts individual characters from a character vector and is explained in Section 16.3.
> library(stringr)
> # first get a list of the files
> theFiles <- dir("data/", pattern="\\.csv")
> ## loop through those files
> for(a in theFiles)
+ {
+ # build a good name to assign to the data
+ nameToUse <- str_sub(string=a, start=12, end=18)
+ # read in the csv using read.table
+ # file.path is a convenient way to specify a folder and file name
+ temp <- read.table(file=file.path("data", a),
+ header=TRUE, sep=",", stringsAsFactors=FALSE)
+ # assign them into the workspace
+ assign(x=nameToUse, value=temp)
+ }
14.2.1 merge
R comes with a built-in function, called merge, to merge two data.frames.
> Aid90s00s <- merge(x=Aid_90s, y=Aid_00s,
+ by.x=c("Country.Name", "Program.Name"),
+ by.y=c("Country.Name", "Program.Name"))
> head(Aid90s00s)
Country.Name Program.Name
1 Afghanistan Child Survival and Health
2 Afghanistan Department of Defense Security Assistance
3 Afghanistan Development Assistance
4 Afghanistan Economic Support Fund/Security Support Assistance
5 Afghanistan Food For Education
6 Afghanistan Global Health and Child Survival
FY1990 FY1991 FY1992 FY1993 FY1994 FY1995 FY1996 FY1997 FY1998
1 NA NA NA NA NA NA NA NA NA
2 NA NA NA NA NA NA NA NA NA
3 NA NA NA NA NA NA NA NA NA
4 NA NA NA 14178135 2769948 NA NA NA NA
5 NA NA NA NA NA NA NA NA NA
6 NA NA NA NA NA NA NA NA NA
FY1999 FY2000 FY2001 FY2002 FY2003 FY2004 FY2005
1 NA NA NA 2586555 56501189 40215304 39817970
2 NA NA NA 2964313 NA 45635526 151334908
3 NA NA 4110478 8762080 54538965 180539337 193598227
4 NA NA 61144 31827014 341306822 1025522037 1157530168
5 NA NA NA NA 3957312 2610006 3254408
6 NA NA NA NA NA NA NA
FY2006 FY2007 FY2008 FY2009
1 40856382 72527069 28397435 NA
2 230501318 214505892 495539084 552524990
3 212648440 173134034 150529862 3675202
4 1357750249 1266653993 1400237791 1418688520
5 386891 NA NA NA
6 NA NA 63064912 1764252
The by.x specifies the key column(s) in the left data.frame and by.y does the same for the right data.frame. The ability to specify different column names for each data.frame is the most useful feature of merge. The biggest drawback, however, is that merge can be much slower than the alternatives.
14.2.2 plyr join
Returning to Hadley Wickham’s plyr package, we see it includes a join function, which works similarly to merge but is much faster. The biggest drawback, though, is that the key column(s) in each table must have the same name. We use the same data used previously to illustrate.
> library(plyr)
> Aid90s00sJoin <- join(x=Aid_90s, y=Aid_00s,
+ by=c("Country.Name", "Program.Name"))
> head(Aid90s00sJoin)
Country.Name Program.Name
1 Afghanistan Child Survival and Health
2 Afghanistan Department of Defense Security Assistance
3 Afghanistan Development Assistance
4 Afghanistan Economic Support Fund/Security Support Assistance
5 Afghanistan Food For Education
6 Afghanistan Global Health and Child Survival
FY1990 FY1991 FY1992 FY1993 FY1994 FY1995 FY1996 FY1997 FY1998
1 NA NA NA NA NA NA NA NA NA
2 NA NA NA NA NA NA NA NA NA
3 NA NA NA NA NA NA NA NA NA
4 NA NA NA 14178135 2769948 NA NA NA NA
5 NA NA NA NA NA NA NA NA NA
6 NA NA NA NA NA NA NA NA NA
FY1999 FY2000 FY2001 FY2002 FY2003 FY2004 FY2005
1 NA NA NA 2586555 56501189 40215304 39817970
2 NA NA NA 2964313 NA 45635526 151334908
3 NA NA 4110478 8762080 54538965 180539337 193598227
4 NA NA 61144 31827014 341306822 1025522037 1157530168
5 NA NA NA NA 3957312 2610006 3254408
6 NA NA NA NA NA NA NA
FY2006 FY2007 FY2008 FY2009
1 40856382 72527069 28397435 NA
2 230501318 214505892 495539084 552524990
3 212648440 173134034 150529862 3675202
4 1357750249 1266653993 1400237791 1418688520
5 386891 NA NA NA
6 NA NA 63064912 1764252
join has an argument for specifying a left, right, inner or full (outer) join.
We have eight data.frames containing foreign assistance data that we would like to combine into one data.frame without hand coding each join. The best way to do this is put all the data.frames into a list, and then successively join them together using Reduce.
> # first figure out the names of the data.frames
> frameNames <- str_sub(string=theFiles, start=12, end=18)
> # build an empty list
> frameList <- vector("list", length(frameNames))
> names(frameList) <- frameNames
> # add each data.frame into the list
> for(a in frameNames)
+ {
+ frameList[[a]] <- eval(parse(text=a))
+ }
A lot happened in that section of code, so let’s go over it carefully. First we reconstructed the names of the data.frames using str_sub from Hadley Wickham’s stringr package, which is shown in more detail in Chapter 16. Then we built an empty list with as many elements as there are data.frames, in this case eight, using vector and assigning its mode to “list”. We then set appropriate names to the list.
Now that the list is built and named, we looped through it, assigning to each element the appropriate data.frame. The problem is that we have the names of the data.frames as characters but the <- operator requires a variable, not a character. So we parse and evaluate the character, which realizes the actual variable. Inspecting, we see that the list does indeed contain the appropriate data.frames.
> head(frameList[[1]])
Country.Name Program.Name
1 Afghanistan Child Survival and Health
2 Afghanistan Department of Defense Security Assistance
3 Afghanistan Development Assistance
4 Afghanistan Economic Support Fund/Security Support Assistance
5 Afghanistan Food For Education
6 Afghanistan Global Health and Child Survival
FY2000 FY2001 FY2002 FY2003 FY2004 FY2005 FY2006
1 NA NA 2586555 56501189 40215304 39817970 40856382
2 NA NA 2964313 NA 45635526 151334908 230501318
3 NA 4110478 8762080 54538965 180539337 193598227 212648440
4 NA 61144 31827014 341306822 1025522037 1157530168 1357750249
5 NA NA NA 3957312 2610006 3254408 386891
6 NA NA NA NA NA NA NA
FY2007 FY2008 FY2009
1 72527069 28397435 NA
2 214505892 495539084 552524990
3 173134034 150529862 3675202
4 1266653993 1400237791 1418688520
5 NA NA NA
6 NA 63064912 1764252
> head(frameList[["Aid_00s"]])
Country.Name Program.Name
1 Afghanistan Child Survival and Health
2 Afghanistan Department of Defense Security Assistance
3 Afghanistan Development Assistance
4 Afghanistan Economic Support Fund/Security Support Assistance
5 Afghanistan Food For Education
6 Afghanistan Global Health and Child Survival
FY2000 FY2001 FY2002 FY2003 FY2004 FY2005 FY2006
1 NA NA 2586555 56501189 40215304 39817970 40856382
2 NA NA 2964313 NA 45635526 151334908 230501318
3 NA 4110478 8762080 54538965 180539337 193598227 212648440
4 NA 61144 31827014 341306822 1025522037 1157530168 1357750249
5 NA NA NA 3957312 2610006 3254408 386891
6 NA NA NA NA NA NA NA
FY2007 FY2008 FY2009
1 72527069 28397435 NA
2 214505892 495539084 552524990
3 173134034 150529862 3675202
4 1266653993 1400237791 1418688520
5 NA NA NA
6 NA 63064912 1764252
> head(frameList[[5]])
Country.Name Program.Name
1 Afghanistan Child Survival and Health
2 Afghanistan Department of Defense Security Assistance
3 Afghanistan Development Assistance
4 Afghanistan Economic Support Fund/Security Support Assistance
5 Afghanistan Food For Education
6 Afghanistan Global Health and Child Survival
FY1960 FY1961 FY1962 FY1963 FY1964 FY1965 FY1966 FY1967 FY1968
1 NA NA NA NA NA NA NA NA NA
2 NA NA NA NA NA NA NA NA NA
3 NA NA NA NA NA NA NA NA NA
4 NA NA 181177853 NA NA NA NA NA NA
5 NA NA NA NA NA NA NA NA NA
6 NA NA NA NA NA NA NA NA NA
FY1969
1 NA
2 NA
3 NA
4 NA
5 NA
6 NA
> head(frameList[["Aid_60s"]])
Country.Name Program.Name
1 Afghanistan Child Survival and Health
2 Afghanistan Department of Defense Security Assistance
3 Afghanistan Development Assistance
4 Afghanistan Economic Support Fund/Security Support Assistance
5 Afghanistan Food For Education
6 Afghanistan Global Health and Child Survival
FY1960 FY1961 FY1962 FY1963 FY1964 FY1965 FY1966 FY1967 FY1968
1 NA NA NA NA NA NA NA NA NA
2 NA NA NA NA NA NA NA NA NA
3 NA NA NA NA NA NA NA NA NA
4 NA NA 181177853 NA NA NA NA NA NA
5 NA NA NA NA NA NA NA NA NA
6 NA NA NA NA NA NA NA NA NA
FY1969
1 NA
2 NA
3 NA
4 NA
5 NA
6 NA
Having all the data.frames in a list allows us to iterate through the list, joining all the elements together (or applying any function to the elements iteratively). Rather than using a loop, we use the Reduce function to speed up the operation.
> allAid <- Reduce(function(...){
+ join(..., by=c("Country.Name", "Program.Name"))},
+ frameList)
> dim(allAid)
[1] 2453 67
> library(useful)
> corner(allAid, c=15)
Country.Name Program.Name
1 Afghanistan Child Survival and Health
2 Afghanistan Department of Defense Security Assistance
3 Afghanistan Development Assistance
4 Afghanistan Economic Support Fund/Security Support Assistance
5 Afghanistan Food For Education
FY2000 FY2001 FY2002 FY2003 FY2004 FY2005 FY2006
1 NA NA 2586555 56501189 40215304 39817970 40856382
2 NA NA 2964313 NA 45635526 151334908 230501318
3 NA 4110478 8762080 54538965 180539337 193598227 212648440
4 NA 61144 31827014 341306822 1025522037 1157530168 1357750249
5 NA NA NA 3957312 2610006 3254408 386891
FY2007 FY2008 FY2009 FY2010 FY1946 FY1947
1 72527069 28397435 NA NA NA NA
2 214505892 495539084 552524990 316514796 NA NA
3 173134034 150529862 3675202 NA NA NA
4 1266653993 1400237791 1418688520 2797488331 NA NA
5 NA NA NA NA NA NA
> bottomleft(allAid, c=15)
Country.Name Program.Name FY2000 FY2001 FY2002
2449 Zimbabwe Other State Assistance 1341952 322842 NA
2450 Zimbabwe Other USAID Assistance 3033599 8464897 6624408
2451 Zimbabwe Peace Corps 2140530 1150732 407834
2452 Zimbabwe Title I NA NA NA
2453 Zimbabwe Title II NA NA 31019776
FY2003 FY2004 FY2005 FY2006 FY2007 FY2008 FY2009
2449 NA 318655 44553 883546 1164632 2455592 2193057
2450 11580999 12805688 10091759 4567577 10627613 11466426 41940500
2451 NA NA NA NA NA NA NA
2452 NA NA NA NA NA NA NA
2453 NA NA NA 277468 100053600 180000717 174572685
FY2010 FY1946 FY1947
2449 1605765 NA NA
2450 30011970 NA NA
2451 NA NA NA
2452 NA NA NA
2453 79545100 NA NA
Reduce can be a difficult function to grasp, so we illustrate it with a simple example. Let’s say we have a vector of the first ten integers, 1:10, and want to sum them (forget for a moment that sum(1:10) will work perfectly). We can call Reduce(sum, 1:10), which will first add 1 and 2. It will then add 3 to that result, then 4 to that result and so on, resulting in 55.
Likewise, we passed a list to a function that joins its inputs, which in this case was simply ..., meaning that anything could be passed. Using ... is an advanced trick of R programming that can be difficult to get right. Reduce passed the first two data.frames in the list, which were then joined. That result was then joined to the next data.frame and so on until they were all joined together.
14.2.3 data.table merge
Like many other operations in data.table, joining data requires a different syntax, and possibly a different way of thinking. To start, we convert two of our foreign aid datasets’ data.frames into data.tables.
> library(data.table)
> dt90 <- data.table(Aid_90s, key=c("Country.Name", "Program.Name"))
> dt00 <- data.table(Aid_00s, key=c("Country.Name", "Program.Name"))
Then, doing the join is a simple operation. Note that the join requires specifying the keys for the data.tables, which we did during their creation.
> dt0090 <- dt90[dt00]
In this case dt90 is the left side, dt00 is the right side and a left join was performed.