Exploratory Data Analysis; coined by Tukey 1977

1
Exploratory Data Analysis coined by Tukey 1977
-Illuminate underlying pattern in noisy
data -Predecessor to formal analysis -May lead to
different analysis than originally planned
Data visualization (The first thing you do with
your data!!)
2
Important functions of exploratory data
visualization

• Spot outliers
• Discriminate clusters
• Check distributional and other assumptions
• Examine relationships
• Compare mean differences
• Observe a time-based process

http//seamonkey.ed.asu.edu/alex/teaching/WBI/EDA
.html
3
Univariate data (one variable) frequency
distributions
Distributions of height, biomass, etc. often
used to describe populations

• How are the data distributed (including
  summary/descriptive statistics)
• Are the data normal? (required to meet
  assumptions of many statistical techniques- more
  later)
• If not normal, can they be transformed?

4

• Histograms
• Raw data hidden
• Division to categories arbitrary
• Excel, many programs

Identify skew, non-normality
Identify outliers
5
(No Transcript)
6
Stem-leaf plots -show original data -division to
categories arbitrary -easier to order data
first -a histogram on its side (sort of)
Stem
leaves
2
0
0
1
5
9
3
2
6
7
8
4
1
4
6
5
0
3
8
7
Box (box-whisker) plots

• -calculate median, draw horizontal line
• -draw a box with ends at the quartiles Q1 (25)
  and Q3 (75)
• extend the "whiskers" to the farthest points that
  are not outliers
• outliers are outside 3/2 times the interquartile
  range (Q3-Q1)
• Draw a dot for every outlier

Can be done for a single distribution or
comparing several
http//mathworld.wolfram.com/Box-and-WhiskerPlot.h
tml
8
Normal probability plots will be covered later
9

• Bivariate (2 variable) data
• -Relationship between the 2 variables
• Are there outliers?
• Examined by Scatterplots

negative
none
10
Non-linear
Graphing helps you see relationships. Formal
analysis guided by a priori knowledge that one
variable causes change in the other (more later)
11
Classified Data often result from an ecological
experiment

• - Bar chart
• Shows means and variance
• - shows treatment differences magnitude

15
10
5
Epilithon NPP (mg O2/m2/hr)
0
-5
high light
low light
-10
Mean ? one S.E.
12
List things that are wrong with this graph.
15
10
5
Epilithon NPP
0
-5
-10
13
Graphing Exercise
Obtain a dataset, preferably your own or a
colleagues, but can be anything Choose a
graphing style that best illustrates the
message of your data Use Excel or other
program to make a graph Print on an overhead to
show in class