Introduction to R for Epidemiologists

Course details

• Thursdays 4:00 PM-5:50 PM
• Room: GCR 119
• Course website www.jennakrall.com/IntrotoRepi

Instructor and TA details

• Instructor: Dr. Jenna Krall, PhD
• E-mail: jenna.krall@emory.edu
• Office: GCR 369
• Office hours: 10 AM-11 AM Wednesdays and by appointment
• Lead TA: Brooke Alhanti
• TA: Anran Liu

Office hours with Anran

• Monday 12-1, 1-2, 2-3, 3-4, 4-5
• Tuesday 10-11, 11-12, 12-1, 3-4, 4-5
• Thursday 1-2, 2-3, 3-4

Prerequisites

• Introductory-level statistics course covering regression
• No programming experience required

What you need

• R and RStudio installed
• No required textbook

Class format

• 60 minute lecture
• 50 minute lab

These two portions of the class provide two different ways of learning R

1. I tell you how to do things
2. Self-guided learning (with help from us)

Grading

• 20% lab assignments
  • Must work on during class to receive credit
• 45% homework assignments (15% each)
  • May work with others
  • Must submit your own code
• 35% final project
  • Epidemiological analysis of real data
  • You may consult me or the TAs
  • You may not consult other students
• No late assignments will be accepted

Course objectives

• Read in and clean data
• Compute summary statistics
• Apply and interpret statistical methods
• Create well-formatted output tables
• Create publication-quality figures
• Create reproducible reports
  

If there is time

• R packages
• Shiny applications

1. History of R
2. R vs. other statistical programming languages
3. Orientation to R
4. Some basics
5. Getting data into R
6. Rules about programming
7. Resources

• The R language

  • We can create objects (e.g. a vector of heights)
  • Objects can interact with other objects (e.g. comparing height and age)
  • Programs consist of applying methods to objects (e.g. what is the mean height?)

• R is interactive

• R and RStudio

• R is easily customizable

  • Customize figures
  • Make nicely formatted tables
  • Write functions

SAS

• No interactivity
• Easy to look at your data
• Difficult to create customized output
• Can be more stable for complex models
• Need a lot of code to perform simple tasks (e.g. to find a mean)

In SAS,

 proc means data = data1;
 var variable1;
 run;

In R,

mean(variable1)

• STATA

  • One dataset at a time
  • Intuitive and easy data cleaning

• MATLAB

  • Matrix computation
  • Powerful

• SPSS

  • Intuitive
  • No programming required

The console

• Type code interactively
• Do not plan to save this code

Editor

• Save code for later
• Can run in console from editor

R vs. RStudio

• R opens the console
• RStudio
  • Runs usual R
  • Nice layout with different panels for editor, console, etc.
  • Additional features
  • Can customize the arrangement of the windows

R version 3.1.2: codename “Pumpkin Helmet”

R version 3.0.2:

5 + 3

[1] 8

123/2 + (2 * 17.2)

[1] 95.9

Can store results using assignment operators

As an example:

nameofobject <- 5
print(nameofobject)

[1] 5

nameofobject2 = 5
print(nameofobject2)

[1] 5

Generally <- is preferred

Can calculate result as object:

nameofsum <- 5 + 3
print(nameofsum)

[1] 8

nameofsum

[1] 8

Objects can be functions of other objects:

nameofnew <- nameofobject + nameofsum
nameofnew

[1] 13

Objects can be reassigned

• Recall that we created an R object nameofobject

nameofobject

[1] 5

• Suppose we reassign the name

nameofobject <- 1500
nameofobject

[1] 1500

• So far, all the R objects we have dealt with have been numeric (number valued).
  
• There are different data classes that R objects can belong to
• The class function in R can tell us the class of an R object

class(nameofobject)

[1] "numeric"

• String or character objects that are non-numeric

nameofstring <- "IntrotoR"
nameofstring

[1] "IntrotoR"

class(nameofstring)

[1] "character"

We can also create vectors by combining numbers or text. The c function in R combines objects together.

vector1 <- c(6, 2, 3, 4)
vector1

[1] 6 2 3 4

class(vector1)

[1] "numeric"

length(vector1)

[1] 4

Vectors can also consist of strings

vectorstring <- c("R", "is", "awesome", "for", "Epidemiologists")
vectorstring

[1] "R"               "is"              "awesome"         "for"            
[5] "Epidemiologists"

class(vectorstring)

[1] "character"

length(vectorstring)

[1] 5

Mixing data classes

vectormix <- c("Intro", "to", "R", "version", 1)
vectormix

[1] "Intro"   "to"      "R"       "version" "1"      

class(vectormix)

[1] "character"

length(vectormix)

[1] 5

Factors in R

grades <- c("A", "B", "A", "A", "C", "F", "D", "B")
grades

[1] "A" "B" "A" "A" "C" "F" "D" "B"

class(grades)

[1] "character"

factorgrades <- factor(grades)
factorgrades

[1] A B A A C F D B
Levels: A B C D F

class(factorgrades)

[1] "factor"

Why does class matter?

• Functions will perform differently for different classes

  • Print function for factor shows levels
  • Mean of a character vector is meaningless

• There are many other classes

  • Next class: matrices, data frames, lists, dates
  • Classes for output (e.g. linear model class lm)
  • In R, you can create your own classes of R objects

Use brackets to select elements of a vector

vectorstring

[1] "R"               "is"              "awesome"         "for"            
[5] "Epidemiologists"

vectorstring[1]

[1] "R"

vectorstring[5]

[1] "Epidemiologists"

• Used as nameoffunction(Robject)

vector1

[1] 6 2 3 4

mean(vector1)

[1] 3.75

• Functions can take several arguments separated by commas

mean(vector1, trim = 0.25)

[1] 3.5

log(1)

[1] 0

length(vector1)

[1] 4

median(vector1)

[1] 3.5

summary(vector1)

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
   2.00    2.75    3.50    3.75    4.50    6.00 

ls()

 [1] "factorgrades"  "grades"        "nameofnew"     "nameofobject" 
 [5] "nameofobject2" "nameofstring"  "nameofsum"     "vector1"      
 [9] "vectormix"     "vectorstring" 

Packages include tools to perform many different analyses

• Some packages are preloaded

  • base (mean, print, length), stats (median, lm)
  • datasets
  • graphics

• Can install and load additional packages from CRAN

  • ggplot2, RColorBrewer (advanced graphics)
  • dplyr, reshape2 (data cleaning/manipulation tools)

Installing a package (may need to specify a CRAN mirror)

install.packages("ggplot2")

Loading a package

library(ggplot2)

Where R

• looks for data
• saves data, figures, etc.

getwd()

[1] "/Users/jennakrall/Dropbox/IntrotoREpi/data"

Change your working directory using setwd

setwd("/Users/jennakrall/Dropbox/")

• For Windows users, setwd(“C:/Users/yourusername/etc.”)
• For Mac users, setwd(“/Users/yourusername/etc.”)
• Can use home directory

setwd("~/")
getwd()

[1] "/Users/jennakrall"

R can read in data from many different sources

Types of data files

• R data files: (extension .RData or .rda)
• CSV file: comma separated values (extension .csv).
  
• Excel file: Microsoft Office (extension .xls or .xlsx). Can use excel to create a .csv file or read into R directly.
• SAS: datafiles (extension .sas7bdat) or Xport file (extension .xpt)

You must either specify the path to the data or set your working directory to the where the data are located

Google flu trends

• Estimates flu activity using google searches
• This dataset includes estimated flu activity (by day) for the whole United States, Georgia, Atlanta, and Health and Human Services Region 4 (which includes Alabama, Florida, Georgia, Kentucky, Mississippi, North Carolina, South Carolina, and Tennessee).
  
• Google's Nature paper: http://www.nature.com/nature/journal/v457/n7232/full/nature07634.html
  
• Science paper: http://www.sciencemag.org/content/343/6176/1203
• Data Source: Google Flu Trends (http://www.google.org/flutrends), accessed 12/3/2014

Example using .RData or .rda files

load("googleflu.RData")
ls()

[1] "flu"

head(flu)

        Date United.States Georgia Atlanta HHSRegion4
1 2003-09-28           902     514     519        631
2 2003-10-05           952     532     484        652
3 2003-10-12          1092     557     497        735
4 2003-10-19          1209     608     563        822
5 2003-10-26          1249     745     845        797
6 2003-11-02          1374     767     771        850

class(flu)

[1] "data.frame"

Example using .csv file

flu <- read.csv("googleflu.csv", stringsAsFactors = FALSE)
head(flu)

        Date United.States Georgia Atlanta HHSRegion4
1 2003-09-28           902     514     519        631
2 2003-10-05           952     532     484        652
3 2003-10-12          1092     557     497        735
4 2003-10-19          1209     608     563        822
5 2003-10-26          1249     745     845        797
6 2003-11-02          1374     767     771        850

class(flu)

[1] "data.frame"

Example using .xls or .xlsx file

library(XLConnect)
wkbook_flu <- loadWorkbook("googleflu.xlsx")
class(wkbook_flu)

[1] "workbook"
attr(,"package")
[1] "XLConnect"

flu <- readWorksheet(wkbook_flu, 1)

Example using .xls or .xlsx file

head(flu)

        Date United.States Georgia Atlanta HHSRegion4
1 2003-09-28           902     514     519        631
2 2003-10-05           952     532     484        652
3 2003-10-12          1092     557     497        735
4 2003-10-19          1209     608     563        822
5 2003-10-26          1249     745     845        797
6 2003-11-02          1374     767     771        850

class(flu)

[1] "data.frame"

Example using SAS

1. Save SAS data as .csv and read into R
2. Use SAS Xport Transport file
3. Read SAS data directly
   • Only if SAS is installed
   • Do not recommend

1) Save SAS data as .csv and read into R

• Save data (.sas7bdat) in SAS as .csv file

proc export data = googleflu 
    outfile = "googleflu.csv" 
    dbms = csv replace; 
putnames = yes;
run; 

• Then use read.csv function in R

flu <- read.csv("googleflu.csv", stringsAsFactors = FALSE)

2) Use SAS Xport Transport file

SAS xport files

• Save data in SAS as SAS xport file

* Set xport filepath ;
libname lib1 xport "H:\googleflu.xpt";

* Set the xport file ; 
data lib1.flu;
* This is your original data;
set flu;
run; 

2) Use SAS Xport Transport file

• Read xport file into R

library(Hmisc)
flu <- sasxport.get("googleflu.xpt")

Processing SAS dataset FLU   ..

head(flu)

        date   us georgia atlanta hhs
1 2003-09-28  902     514     519 631
2 2003-10-05  952     532     484 652
3 2003-10-12 1092     557     497 735
4 2003-10-19 1209     608     563 822
5 2003-10-26 1249     745     845 797
6 2003-11-02 1374     767     771 850

3) Read SAS data directly (only if SAS is installed)

• This command sometimes yields ambiguous errors from SAS
• Need Windows/SAS

• Use read.ssd function to read .sas7bdat file directly into R
  • read.ssd function in R creates xport file (using SAS)
  • reads the resulting dataset into R using read.xport.
    

library(foreign)
lib1 <- "c:/"
flu <- read.ssd(lib1, "flu", sascmd = "filepath/to/where/SAS/is")

May need to tell R where to find SAS using sascmd argument.

Other functions to read in data include

• read.table
• read.delim
• readLines
• scan
• many others

• Always comment your code– your grade will depend on it (and so will your future self)

# This is the mean of our numeric vector
mean(vector1)

[1] 3.75

Your closest collaborator is you six months ago but you don’t reply to email. – Erin Jonaitis (via Andrew Gelman)

• Use appropriate spacing

newvector <- c(5, 3, 4, 4)

Naming conventions

• The more descriptive, the better
• Names are case sensitive
• Capitalization and camelCase can make objects hard to type and remember
• Can use underscores to separate words in names

new_vector <- c(1, 2, 3, 4)
new_vector

[1] 1 2 3 4

• Don't overwrite existing R functions or objects (e.g. c or T)

Style guides:

• Hadley Wickham: http://r-pkgs.had.co.nz/style.html
• Google's: https://google-styleguide.googlecode.com/svn/trunk/Rguide.xml

• R help files (use ?function)

?mean
help.search("mean")

• Search engines
  • Google: Append CRAN onto your google search for R.
  • R seek: http://www.rseek.org/
• Stack Overflow: http://stackoverflow.com/questions/tagged/r
• UCLA R stats: http://www.ats.ucla.edu/stat/r/
• Other resources: www.jennakrall.com/IntrotoRepi/resources.html