Statistics and Statistical Packages in Experimental Design

1
Statistics and Statistical Packages in
Experimental Design

• Presented by
• Amelia Potter
• MEES 688K Experimental Design Seminar

2
What are statistics?

• Statistics - the mathematical comparison of data.
• Statistics test - either descriptive or
  comparative.
• Descriptive statistics describes the data.
• Mean.
• Variance.
• Standard deviation.
• Standard error.
• Comparative statistics calculates a mathematical
  probability that variations in the data set
  were, or were not, due to treatment effects.
• T-test
• Chi-Square test
• Pearsons
• etc.

3
Why Use Statistics ?

• When forming your hypothesis, you plan to study a
  treatment effect.
• Lets assume that you wish to compare fire
  effects on prairie plant diversity and growth.
• You collect treatment data for burned and
  unburned plots.

Brown, Jackie, 1999, Statistical Analysis in
Ecology and Evolution, for Biology 195 - Prairie
Restoration, Grinnell. College, Grinnell, Iowa.
4
Biomass (g) of prairie plants in 20 experimental
units.
Brown, Jackie, 1999, Statistical Analysis in
Ecology and Evolution, for Biology 195 - Prairie
Restoration, Grinnell. College, Grinnell, Iowa.
5
Mean biomass (g) of prairie plants in 20
experimental units.
Brown, Jackie, 1999, Statistical Analysis in
Ecology and Evolution, for Biology 195 - Prairie
Restoration, Grinnell. College, Grinnell, Iowa.
Conclusion based on descriptive statistics
(means) Burning is beneficial to plants!
6
Conclusion Should Evaluate Effectsof
Experimental Variations
Brown, Jackie, 1999, Statistical Analysis in
Ecology and Evolution, for Biology 195 - Prairie
Restoration, Grinnell. College, Grinnell, Iowa.

• Before making such a conclusion
• Consider the many environmental variables
  occurring during the experiment
• soil moisture
• nutrient levels
• amount of sunlight
• pollinators
• pesticide levels
• herbicide levels
• pollutants
• others?
• Predicting the probability that they had an
  effect requires comparative statistical tests.

7
Comparative Statistical Tests

• Comparative statistical tests are mathematical
  formulas which calculate the probability of an
  effect based on the data collected, or on
  recorded data.
• Selecting the particular statistical test to use
  requires some knowledge of the assumptions in the
  formulas.
• Parametric .vs. Nonparametric Data
• Does your data fit a normal distribution?
• When in doubt
• Use a text book
• Look at similar studies in the literature,
• and then use the text book.
• Consult your advisor, and then use the text book.
• Consult a statistician or specialist, and then
  use the text book.

8
Other reasons

• Everyone is using them and you are expected to
  know it.
• In order to read and interpret articles in
  Ecology, Ecological Applications and Ecological
  Monographs, as well as the environmental/chemistry
  journals, the reader must have a working
  knowledge of statistics.
• Examples of selected tests for Ecology
• ANOVA, t-test, one tailed, two tailed, Ryans Q
  test, stepwise multiple regression, nonparametric
  methods, Principle Component Analysis, Cannonical
  Discrimminant Analysis.

9
Statistical Tests and Use of Statistical
PackagesThisted, R.A., 1986, Computing
Environments for Data Analysis, Statistical
Science, 12,269-278.

• While you can still perform these tests by hand
  with a calculator, it is much easier to use a
  computer software package.
• Statistical packages have developed since the
  1950s with the development of personal computers.
• In 1940s, data analysis used pen and mechanical
  calculators.
• In 1950s, large computers calculated large data
  sets, and performed data calculations, never
  before possible.
• In 1960s, statistics packages were developed.
• In 1970s, nonprogrammers and nonstatisticians
  were able to do the complex calculations
  performed in 1940s.

10
Modern Day Statistical Software Packages

• 1980s brought together complex systems for
  statistical analysis, including software packing,
  terminals, programming language, editors,
  operating systems and output devices and data
  analysis.
• 1990s brought windows-based software packages, so
  easy that anyone with limited statistical
  knowledge can perform complex analysis.
• For a list of web resources
• HTTP//www.umes.edu/sciences/MEESProgram/Experime
  ntalDesign/StatisticalPackages/

11
Choosing a Statistical Software Package

• Features of a good statistical software package
• Easy to Use
• Easy to Enter and Rerun Data
• Multiple Data Input Routes
• Power
• Data Handling Ability
• How much data are you planning to process?
• Speed and Memory
• How quick do you need the answer?

12
Choosing a Statistical Software Package

• Features of a good statistical software package
  (cont)
• Statistical Tests/Statistical Knowledge
• What type of data do you expect?
• Do you know what tests you plan to run?
• Do you understand statistics and mathematical
  formulas?
• Many Built-In Analysis Routines
• Programming Capabilities
• Data History Log
• Cost and Availability
• Hardware Requirements
• UNIX, IBM, MAC, etc.

13
Computer Software Available for Facilitating
Statistical Tests

• Spreadsheets
• Excel, Quatro Pro, Lotus123
• Databases
• Access
• Commercial Statistical Packages
• Statistix, SPSS (SYSTAT), STATVIEW, MINITAB SAS,
  etc.
• Freeware Statistical Packages
• Geographic Information Packages
• User-Written Software
• Modeling Packages

14
Spreadsheets

• Pros
• Easy to Use
• Easy to Manipulate Data
• Some Built-in Statistical Tests (Analysis)
• Limited Programming Capability

• Cons
• Limited Data Capacity
• Limited Number of Analytical Options
• No Transaction Log.

15
Databases

• Pros
• Large Data Capacity.
• Efficient Data Manipulations and Queries.
• Good Transaction Log.
• Some Built-in Analysis.

• Cons
• Not designed for science.

16
Statistical Packages

• Pros
• Wide Range of Built-in Analysis.
• Pre-formated Output.
• Good Transaction Logs.

• Cons
• Hard to Learn to Operate.
• Easy to Generate Probabilities Without Knowledge
  of Statistics.

17
Types of Customer
Wetheril Curran, 1985, The Statistician, vol.
34, pp. 391-427

• Expert Statistician
• Industrial Statistician
• Statistical Novice
• Biologists
• Amateur Statistician
• Dont have a math degree but think they know
  statistics.

18
Statistical Packages and Their Intended Users

• Excel Spreadsheet
• Amateur Statistician
• Statistix
• Statistical Novice
• SPSS
• Novice/Industrial Statistician
• SAS
• Expert Statistician and Programmer

19
UMES Available Statistical Software Packages

• Now lets look at some manipulations with the
  software available free at UMES
• Statistix
• Statistical Novice
• Limited capabilities
• SPSS
• More advanced
• SAS
• Expert Statistician and Programmer

20
Conclusions

• Statistical Software Packages
• Fast, Easy to Use, and Require Knowledge of
  Statistics.
• UMES Statistical Software Packages
• Statistix
• Rapid, Easy-to-use tool for limited data analysis
• SPSS
• Rapid, Easy-to-use tool with more advanced data
  analysis. For more complicated data sets.
• SAS
• Hard to use, requires programming or prewritten
  programs, Should only be used for complex or
  large data analysis.