• When investigating the characteristics of a numerical variable, you can use the following:
  1. Summary statistics
  2. Box plots
  3. Cleveland dotplots
  4. Histograms
  5. Cumulative distribution functions (CDFs)
  6. Rank-order plots
• Comparing differences across categorical variables can lead to insights
  1. Bar charts
  2. Dot plots
  3. Mosaic charts
• Grouping numerical variables together for comparison

Get it from http://cs.wellesley.edu/~qtw/code/r_cat.r

> BFCases <- read.table(file="BirdFluCases.txt", header = TRUE,sep="\t")
> names(BFCases)
 [1] "Year"       "Azerbaijan" "Bangladesh" "Cambodia"   "China"     
 [6] "Djibouti"   "Egypt"      "Indonesia." "Iraq"       "LaoPDR"    
[11] "Myanmar"    "Nigeria"    "Pakistan"   "Thailand"   "Turkey"    
[16] "VietNam"   
> str(BFCases)
'data.frame':   6 obs. of  16 variables:
 $ Year      : int  2003 2004 2005 2006 2007 2008
 $ Azerbaijan: int  0 0 0 8 0 0
 $ Bangladesh: int  0 0 0 0 0 1
 $ Cambodia  : int  0 0 4 2 1 0
 $ China     : int  1 0 8 13 5 3
 $ Djibouti  : int  0 0 0 1 0 0
 $ Egypt     : int  0 0 0 18 25 7
 $ Indonesia.: int  0 0 20 55 42 18
 $ Iraq      : int  0 0 0 3 0 0
 $ LaoPDR    : int  0 0 0 0 2 0
 $ Myanmar   : int  0 0 0 0 1 0
 $ Nigeria   : int  0 0 0 0 1 0
 $ Pakistan  : int  0 0 0 0 3 0
 $ Thailand  : int  0 17 5 3 0 0
 $ Turkey    : int  0 0 0 12 0 0
 $ VietNam   : int  3 29 61 0 8 5

> BFDeaths <- read.table(file="BirdFluDeaths.txt", header = TRUE, sep="\t")
> names(BFDeaths)
 [1] "Year"       "Azerbaijan" "Bangladesh" "Cambodia"   "China"     
 [6] "Djibouti"   "Egypt"      "Indonesia." "Iraq"       "LaoPDR"    
[11] "Myanmar"    "Nigeria"    "Pakistan"   "Thailand"   "Turkey"    
[16] "VietNam"   
> str(BFDeaths)
'data.frame':   6 obs. of  16 variables:
 $ Year      : int  2003 2004 2005 2006 2007 2008
 $ Azerbaijan: int  0 0 0 5 0 0
 $ Bangladesh: int  0 0 0 0 0 0
 $ Cambodia  : int  0 0 4 2 1 0
 $ China     : int  1 0 5 8 3 3
 $ Djibouti  : int  0 0 0 0 0 0
 $ Egypt     : int  0 0 0 10 9 3
 $ Indonesia.: int  0 0 13 45 37 15
 $ Iraq      : int  0 0 0 2 0 0
 $ LaoPDR    : int  0 0 0 0 2 0
 $ Myanmar   : int  0 0 0 0 0 0
 $ Nigeria   : int  0 0 0 0 1 0
 $ Pakistan  : int  0 0 0 0 1 0
 $ Thailand  : int  0 12 2 3 0 0
 $ Turkey    : int  0 0 0 4 0 0
 $ VietNam   : int  3 20 19 0 5 5
>

> Cases  <- rowSums(BFCases[,2:16])
> names(Cases) <- BFCases[,1]
> Cases
2003 2004 2005 2006 2007 2008 
   4   46   98  115   88   34 
> #Let's also compare deaths by year
> Deaths <- rowSums(BFDeaths[,2:16])
> names(Deaths) <- BFDeaths[,1]
> Deaths
2003 2004 2005 2006 2007 2008 
   4   32   43   79   59   26 
> #We want to plot Cases and Deaths in a bar chart
> Counts <- cbind(Cases, Deaths)
> Counts
     Cases Deaths
2003     4      4
2004    46     32
2005    98     43
2006   115     79
2007    88     59
2008    34     26

> Cases
2003 2004 2005 2006 2007 2008 
   4   46   98  115   88   34 
> barplot(Cases , main = "Bird flu cases")

> barplot(Counts)

> t(Counts)
       2003 2004 2005 2006 2007 2008
Cases     4   46   98  115   88   34
Deaths    4   32   43   79   59   26
> barplot(t(Counts), col = gray(c(0.5,1)))

> barplot(t(Counts), beside = TRUE)

par(mfrow = c(2,2), mar = c(3, 3, 2, 1))
barplot(Cases , main = "Bird flu cases")
barplot(Counts)
barplot(t(Counts), col = gray(c(0.5,1)))
barplot(t(Counts), beside = TRUE)

• While we all know about bar plots, other graph types often work better
• Mosaic plots are good for comaparing two categorical variables, particularly if you have a natural sorting or want to sort by size.

> Counts
     Cases Deaths
2003     4      4
2004    46     32
2005    98     43
2006   115     79
2007    88     59
2008    34     26
> BFProp<-cbind(Survivors=Cases-Deaths,Deaths)
> BFProp
     Survivors Deaths
2003         0      4
2004        14     32
2005        55     43
2006        36     79
2007        29     59
2008         8     26

mosaicplot(BFProp,col=c('green','red'),main="Bird flu survival rates by year")

• So far we have examined the bird flu data grouped by year across all countries. For this exercise you will create bar and mosaic plots grouped by country across all years.
• Your task:
  1. Download starter code from http://cs.wellesley.edu/~qtw/code/r_cat_ex1.r
  2. Compute vectors of bird flu cases and deaths grouped by country
  3. Create a bar chart that plots cases and deaths by country that looks like chart on next slide. Hint: you will need to use the las parameter to get the labels right.
  4. Create a mosaic plot of survivors and deaths by country that loooks like chart on 2 slides down.

Download the solution code from http://cs.wellesley.edu/~qtw/code/r_cat_ex1_sol.r

• The mosaic plot and bar chart for the exercise is hard to interpret, since the x-axis has no natural ordering
• We can impose order, which can make drawing inferences from the graph much easier
• A common ordering is to sort from largest to smallest in the category

> #here's a sorted list of countries
> names(sort(CCases,decreasing=T))
 [1] "Indonesia." "VietNam"    "Egypt"      "China"      "Thailand"  
 [6] "Turkey"     "Azerbaijan" "Cambodia"   "Iraq"       "Pakistan"  
[11] "LaoPDR"     "Bangladesh" "Djibouti"   "Myanmar"    "Nigeria"   
> #we can then reorder the matrix rows by the sorted names
> BFCountryProp<-BFCountryProp[names(sort(CCases,decreasing=T)),]
> 
> BFCountryProp
           Survivors CDeaths
Indonesia.        25     110
VietNam           54      52
Egypt             28      22
China             10      20
Thailand           8      17
Turkey             8       4
Azerbaijan         3       5
Cambodia           0       7
Iraq               1       2
Pakistan           2       1
LaoPDR             0       2
Bangladesh         1       0
Djibouti           1       0
Myanmar            1       0
Nigeria            0       1

> rs<-read.table('regSuperCensusMod.csv',header=T,sep=',')
> rsi<-rs[rs$IndOrg=="IND",]
> head(rsi)
> head(rsi)
      Candidate RegularSuper   CNumber             CName CAmount     Date
20 barack obama      Regular C00431445 Obama for America   35800 20110525
26 barack obama      Regular C00431445 Obama for America   25800 20110510
39 barack obama      Regular C00431445 Obama for America    5000 20110404
40 barack obama      Regular C00431445 Obama for America    5000 20110628
41 barack obama      Regular C00401224           ActBlue    5000 20110420
43 barack obama      Regular C00431445 Obama for America    5000 20110421
   State   ZIP IndOrg    PopSt  FrWhite  FrBlack USADiversity
20    AR 72207    IND  2915918 0.745381 0.154289     0.417706
26    CA 94115    IND 37253956 0.401468 0.061713     0.728676
39    ME  4110    IND  1328361 0.944244 0.011824     0.110012
40    CA 94708    IND 37253956 0.401468 0.061713     0.728676
41    MA  2111    IND  6547629 0.761314 0.066344     0.405299
43    CA 94904    IND 37253956 0.401468 0.061713     0.728676

Contingency tables count categorical variables matching each value of all combinations of categorical variables:

> ct<-table(rsi[,c('Candidate','RegularSuper')])
> ct
               RegularSuper
Candidate       Regular Super
  barack obama    28864    41
  mitt romney     10470   159
  newt gingrich    1266    18

barplot(ct,beside=T,col=rainbow(3),main='# of contributions by type and candidate')
legend("topright",rownames(ct),fill=rainbow(3))

mosaicplot(ct,main='# of contributions by type and candidate')

mosaicplot(~Candidate+RegularSuper,data=rsi,
    main='# of contributions by type and candidate')

Let's say we've got a numerica variable that we'd like to perform a function on but grouped by a categorical variable (e.g., total campaign contributions, not the number):

> ct.cont<-aggregate(CAmount~Candidate+RegularSuper,data=rsi,sum)
> ct.cont
      Candidate RegularSuper  CAmount
1  barack obama      Regular 23111899
2   mitt romney      Regular 17243855
3 newt gingrich      Regular  1101165
4  barack obama        Super   445200
5   mitt romney        Super 12167953
6 newt gingrich        Super  2080250
> cta<-array(data=ct.cont$CAmount,
+             dim=c(length(levels(ct.cont$Candidate)),
+                     length(levels(ct.cont$RegularSuper))),
+             dimnames=list(levels(ct.cont$Candidate),
+                     levels(ct.cont$RegularSuper)))
> 
> cta
               Regular    Super
barack obama  23111899   445200
mitt romney   17243855 12167953
newt gingrich  1101165  2080250

> r.state<-table(rsi[rsi$RegularSuper=='Regular',c('State','Candidate')])
> head(r.state)
     Candidate
State barack obama mitt romney newt gingrich
   99            0          35             0
   AK           75          10             5
   AL          101          52            13
   AR           76          13             6
   AS            2           0             0
   AZ          321         203            19

mosaicplot(r.state,main='# of contributions by type and state')

> stateCont<-table(rsi[rsi$RegularSuper=='Regular','State'])
> stateCont

  99   AK   AL   AR   AS   AZ   CA   CO   CT   DC   DE   FL   GA   GU   HI   IA   ID   IL   IN   KS 
  35   90  166   95    2  543 7545  726 1092  994   94 2302 1023    4  165  187  205 2138  277  120 
  KY   LA   MA   MD   ME   MI   MN   MO   MS   MT   NC   ND   NE   NH   NJ   NM   NV   NY   OH   OK 
 143  176 2429 1329  181 1023  375  572   70  101  579   20   70  219 1021  187  246 4497  541  145 
  OR   PA   PR   RI   SC   SD   TN   TX   UT   VA   VI   VT   WA   WI   WV   WY   ZZ 
 407 1244  101   88  166   27  371 2439  928 1299   21  285  929  350   54   66   28 
> r.state.srt<-r.state[names(sort(stateCont,decreasing=T)),]
> r.state.srt
     Candidate
State barack obama mitt romney newt gingrich
   CA         5995        1453            97
   NY         3295        1162            40
   TX         1488         837           114
   MA         1629         790            10
   FL         1221         939           142
   IL         1881         231            26

mosaicplot(r.state.srt[1:20,],col=rainbow(3),
    main='# of contributions (top 20 states sorted by # contributions)')