Garrett Grolemund Phd Student / Rice University Department of Statistics Data cleaning
Garrett GrolemundPhd Student / Rice University
Department of Statistics
Data cleaning
1. Intro to data cleaning
2. What you can’t fix
3. What you can fix
4. Intro to reshape
Your turn
Do you think men or women leave a larger tip when dining out? What data would you collect to test this belief? What would prompt you to change your belief?
Data Analysis
Data
Residuals
ModelCompare
Visualize
Transform
Data Analysis
Data
Residuals
ModelCompare
Visualize
Transform
Data Analysis
Data
Residuals
ModelCompare
Visualize
Transform
Data Analysis
Data
Residuals
ModelCompare
Visualize
Transform
Data Analysis
Data
Residuals
ModelCompare
Visualize
Transform
Data Analysis
Data
Residuals
ModelCompare
Visualize
Transform
10 - 20% of an analysis
Data Cleaning
Data
Residuals
ModelCompare
VisualizeTransform
Data cleaning
“Happy families are all alike; every unhappy family is unhappy in its own way.”—Leo Tolstoy
“Clean datasets are all alike; every messy dataset is messy in its own way.”—Hadley Wickham
Clean data is:
Complete
Correct(factual and internally consistent)
Concise
Compatible(required variables: observations in rows, one column per variable)
What you can’t fix:
Complete
Correct
Correct
Can’t restore incorrect values without original data but can remove clearly incorrect values
Options:
Remove entire row
Mark incorrect value as missing (NA)
When two rows present the same information with different values, at least one row is wrong.
Whenever there is inconsistency, you are going to have to make some tradeoff to ensure concision.
Detecting inconsistency is not always easy.
Inconsistency = incorrect
General strategy
To find incorrect values you need to be creative, combining graphics and data processing.
Tipping data
One waiter recorded information about each tip he received over a period of a few months
244 records
Do men or women tip more?
Your turn
Subset the tipping data to include only rows without NA’s. Judge whether you think all of the data points are correct. How will you make your decision?
tips <- read.csv("tipping.csv", stringsAsFactors = FALSE)
summary(tips)
tips <- subset(tips, !is.na(smoker) & !is.na(non_smoker))
qplot(tip, data = tips, binwidth = .5)qplot(total_bill, data = tips, binwidth = 2)qplot(total_bill, tip, data = tips)
nrow(tips)
sum(tips$male)
sum(tips$female)
subset(tips, male != female)
What you can fix:
Concise(each fact represented once)
Repeating facts: 1. wastes memory 2. creates opportunities for inconsistency
Compatible(Data is compatible with your analysis
in both form and fact)
1. Do you have the relevant variables for your analysis?
This often requires some type of calculation. For example,
proportion = sucesses / attempts
Avg score per game per team = ?
join(), transform(), summarise(), ddply(), plyr address this need
Compatible(Data is compatible with your analysis
in both form and fact)
2. Is the data in the right form for your analysis and visualization tools? (reshape)
Rectangular
Observations in rows
Variables in columns
(1 column per variable)
Your turn
What are the variables in tipping.csv? How are they arranged in rows and columns? Can you form the variables into two groups?
Reshape
install.packages("reshape")library(reshape)library(stringr)head(tips)
Molten data
We can use melt to put each variable into its own column.“Protect” the good columns. “Melt” the offending columns.Then subset.
1. ID variables - identify the object that measurements will take place on (we know these before the experiment)
2. Measured variables - the features of the object that will be measured (we have to do an experiment to observe these)
Two types of variables
object
ID Variables
Bruce Wayne
Batman
SSN: 555-89-3000
Measured Var.
Height (6’1’’)
IQ (180)
Age (71)
ID Variables
Gotham City +
male +
Top 1% tax bracket
Identifier variable Measured variable
Index of random variable
Random variable
Dimension Measure
Experimental design Measurement
predictors (Xi) response (Y)
Molten data
Molten data collapses all the measured variables into two columns: 1) the variable being measured and 2) the value. Sometimes called “long” form.
To protect a column from being melted, label it as an id variable.
reshape::melt(data, id)
tips1 <- melt(tips, id = c("customer_ID", "total_bill", "tip", "smoker", "non_smoker"))
# assign an appropriate variable namenames(tips1)[6] <- "sex"
# subset out unwanted rowstips1 <- subset(tips1, value == 1)tips1 <- tips1[ , c(1,2,6,4,5,3)]
Use melt to fix the smoking variable. One column should be enough to record whether a person smokes or not.
Your turn
Rectangular data are much easier to work with!
qplot(total_bill, tip, data = tips1, color = sex)
# vs.
qplot(total_bill, tip, data = tip, colour = ?)
qplot(total_bill, tip, data = tips1, color = sex) + geom_smooth(method = lm)
Clean data is:
Complete
Correct(factual and internally consistent)
Concise
Compatible(required variables: observations in rows, one column per variable)
Resource
Wickham, H. (2007) Reshaping data with the reshape package. Journal of Statistical Software. 22 (12)
http://www.jstatsoft.org/v21/i12
Summary
Clean data is:
Rectangular(observations in rows, one column per variable)
Consistent
Concise
Complete
Correct
Data
Residuals
ModelCompare
VisualizeTransform
Data
Residuals
ModelCompare
VisualizeTransform
ggplot2
Data
Residuals
ModelCompare
VisualizeTransform
ggplot2plyr
Data
Residuals
ModelCompare
VisualizeTransform
ggplot2plyr
reshape
Data
Residuals
ModelCompare
Visualize
Transform
most statistics classes
This work is licensed under the Creative Commons Attribution-Noncommercial 3.0 United States License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc/3.0/us/ or send a letter to Creative Commons, 171 Second Street, Suite 300, San Francisco, California, 94105, USA.