Department of Statistics

COURSE STATS 330

Model answers for Assignment 1, 2007.

The data set in the file titanic.txt (available on the course web page) contains some
data on 633 passengers on the liner Titanic, which sank in the North Atlantic on 15th
April 1912 after striking an iceberg.

The data set has 5 variables and 633 cases. The variables are

age.group: The age group of the passenger (0-9, 10-19, 20-29, 30-39, 40-49, 50-
59, 60+), treated as a factor;

age: The age of the passenger, treated as a continuous variable;

survived: 0 = died, 1 = survived. This is a numeric variable.

pclass: The passenger class (1st, 2nd, 3rd), treated as a factor;

sex: The gender (female, male) of the passenger.

Questions

1. Load the data into R, and make a data frame titanic.df to contain the data.
Check for any typographical errors. [5 marks]

There are several ways to do this. You can download the file titanic.txt onto your
computer, and set the R directory to point to the folder containing the data. You set
the R directory by pulling down the File menu in R, choosing Change dir, and
navigating to the correct folder. Having set the directory, type

titanic.df = read.table(titanic.txt, header=T)

Another way that is more convenient is to load the data directly from the web
site. (You have to be connected to the internet to do this.) Type

titanic.df =
read.table("http://www.stat.auckland.ac.nz/~lee/330/datasets.dir/titanic.
txt", header=T)

You can cut and paste the URL for the data directly from the browser. To check for
typographical errors, we can just inspect the values of the variables to see if there are
any typos. You can proofread the file, but a simpler way is just to inspect the unique
(i.e. distinct) values of the variables. Type

> unique(titanic.df$age.group)
[1] 20-29 0-9 30-39 40-49 60-69 50-59 70-79 10-19 100-19
Levels: 0-9 10-19 100-19 20-29 30-39 40-49 50-59 60-69 70-79

1
Here we see that there is a typo : 100-19 is not a valid age group. It probably should
be 10-19, so we will correct it to that. Which case is the offending one?

By typing

> titanic.df[titanic.df$age.group=="100-19",]
pclass survived age sex age.group
633 3rd 0 19 male 100-19

we see that it is case number 633. Correct the original file and read the data in again
Alternatively, we can correct it by

titanic.df[633,5]="10-19"

We also have to readjust the factor to eliminate the incorrect level:

titanic.df$age.group = factor(titanic.df$age.group)

since age.group is the 5th variable in the data frame. The other data can be checked
similarly, but are OK.

2. Make an additional variable by turning the numeric variable survived into a

factor with levels survived and died. [5 marks]

Lets call the new variable survival. Make it by typing

survival = factor(titanic.df$survived, labels=c(Died, Survived))

Add the new variable to the data frame, calling the result titanic2.df

titanic2.df = data.frame(survival, titanic.df)

3. What is the relationship between survival and age? Does it depend on class
and gender? Draw a suitable trellis plot to answer this question. Dont try and
fit any models. [10 marks, 5 for the plot and 5 for the discussion]

We will draw a trellis plot of survival versus age, using gender and class as
conditioning factors.

library(lattice) # need to load the lattice library

trellis.par.set(background = list(col = "white"),

dot.symbol=list(col="darkblue"))

# trellis.par.set is the method for changing the background

# colours and plot symbol colours (this is optional, but these colours
are easier to see when printed)

2
 0 20 40 60

pclass : 3rd pclass :3rd

sex: female sex:male

Survived

Died

pclass : 2nd pclass :2nd

sex: female sex:male

Survived
Survival

Died

pclass : 1st pclass :1st

sex: female sex:male

Survived

Died

0 20 40 60

Age

Interpretation: For the third class passengers, the ages seem similar for those who
died and those who survived. Note that relatively few third class passengers survived.
For the second and first class passengers, the survivors tended to be younger than
those who died, with the exception of female first class passengers. (There are too
few of these who died for any pattern to emerge.)

4. For each combination of age group, class and gender, calculate the fraction of
passengers that survived. Present your results in a table. [10 marks]

The R function table is useful for this.

my.table =
table(titanic2.df$age.group,titanic2.df$pclass,titanic2.df$sex,
titanic2.df$survival)

> my.table

3
, , = female, = Died

1st 2nd 3rd

0-9 1 0 5
10-19 0 1 7
20-29 1 4 8
30-39 1 3 5
40-49 0 1 4
50-59 1 1 0
60-69 1 0 0
70-79 0 0 0

, , = male, = Died

1st 2nd 3rd

0-9 0 0 7
10-19 2 10 21
20-29 10 43 53
30-39 18 30 23
40-49 22 13 14
50-59 16 8 1
60-69 11 1 1
70-79 3 1 0

, , = female, = Survived

1st 2nd 3rd

0-9 0 9 4
10-19 13 11 11
20-29 20 23 4
30-39 19 19 8
40-49 19 9 1
50-59 18 4 0
60-69 7 0 0
70-79 0 0 0

, , = male, = Survived

1st 2nd 3rd

0-9 3 11 6
10-19 3 1 2
20-29 10 4 6
30-39 12 4 3
40-49 10 1 1
50-59 4 0 0
60-69 1 0 0
70-79 0 0 0

The object my.table is an array this is like a matrix but in this case has 4
dimensions. We can make separate tables of just the survivors and just those

4
who survived by subsetting:

> survivors=my.table[,,,2] # survivors are second level

> died = my.table[,,,1] # died are first level
> fraction = survivors/(died + survivors)
> round(fraction,3) # rounds to 3 decimal places

, , = female

1st 2nd 3rd

0-9 0.000 1.000 0.444
10-19 1.000 0.917 0.611
20-29 0.952 0.852 0.333
30-39 0.950 0.864 0.615
40-49 1.000 0.900 0.200
50-59 0.947 0.800 NaN
60-69 0.875 NaN NaN
70-79 NaN NaN NaN

, , = male

1st 2nd 3rd

0-9 1.000 1.000 0.462
10-19 0.600 0.091 0.087
20-29 0.500 0.085 0.102
30-39 0.400 0.118 0.115
40-49 0.312 0.071 0.067
50-59 0.200 0.000 0.000
60-69 0.083 0.000 0.000
70-79 0.000 0.000 NaN

An alternative way to make the table is to use the fact that, for binary
(0/1) data like the variable survived, the proportion of ones is just
the mean. We can calculate the mean (ie the proportion surviving) for
each age group/class./sex combination by using the R function tapply:

> survival.frac = tapply(titanic2.df$survived,

list(titanic2.df$age.group,titanic2.df$pclass,titanic2.df$sex), mean)
> round(survival.frac,3)

, , female

1st 2nd 3rd

0-9 0.000 1.000 0.444
10-19 1.000 0.917 0.611
20-29 0.952 0.852 0.333
30-39 0.950 0.864 0.615
40-49 1.000 0.900 0.200
50-59 0.947 0.800 NA
60-69 0.875 NA NA
70-79 NA NA NA

5
, , male

1st 2nd 3rd

0-9 1.000 1.000 0.462
10-19 0.600 0.091 0.087
20-29 0.500 0.085 0.102
30-39 0.400 0.118 0.115
40-49 0.312 0.071 0.067
50-59 0.200 0.000 0.000
60-69 0.083 0.000 0.000
70-79 0.000 0.000 NA

5. How does the fraction surviving depend on age group, gender and class? Draw
another Trellis plot to explore this. [10 marks, 5 for the plot and 5 for the
discussion]

First, we need to convert the table sentries into a data frame, with extra variables
indicating the factor levels. The R function expand.grid does this job nicely:

survival.frac.df = data.frame(frac = as.vector(survival.frac), expand.grid(

age.group=c("0-9","10-19","20-29","30-39","40-49","50-59","60-69","70-79"),
pclass=c("1st","2nd","3rd"),
sex=c("female", "male")))

> survival.frac.df
frac age.group pclass sex
1 0.000 0-9 1st male
2 1.000 10-19 1st male
3 0.952 20-29 1st male
4 0.950 30-39 1st male
5 1.000 40-49 1st male
6 0.947 50-59 1st male
7 0.875 60-69 1st male
8 NA 70-79 1st male
9 1.000 0-9 2nd male
10 0.917 10-19 2nd male
11 0.852 20-29 2nd male
12 0.864 30-39 2nd male
13 0.900 40-49 2nd male
14 0.800 50-59 2nd male
15 NA 60-69 2nd male
16 NA 70-79 2nd male
17 0.444 0-9 3rd male
. 48 lines in all

Then we draw the dot plot

dotplot(frac~age.group|sex*pclass,
data=survival.frac.df,
xlab="Age",
ylab="Survival fraction",
strip=function(...)strip.default(..., strip.names=T))

6
 pclass:3rd pclass :3rd
sex:female sex:male
1.0

0.8

0.6

0.4

0.2

0.0
pclass:2nd pclass :2nd
sex:female sex:male
1.0
Survival fraction

0.8

0.6

0.4

0.2

0.0
pclass:1st pclass :1st
sex:female sex:male
1.0

0.8

0.6

0.4

0.2

0.0
0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79 0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79

Age

Discussion: It is clear from the plot that survival was much higher for females than
males for all classes, although the higher the class, the higher the survival. Survival
was more likely for younger persons. This trend is particularly pronounced for the first
class males. Age did not have such a strong effect on survival for the other
categories.

Sounds like the movie got it right!