Design of Experiments

The Basics

Molecular, Cellular, & Tissue Bioengineering

 Overview of Talk
Design of Experiments - What is DOE?
Terminology
Why use DOE Method?
Benefits of DOE Process
Basic setup of a DOE
Discovering the seven basics steps in designing
a DOE
Some Doug Theorems
 What is DOE?
Design of Experiments (DOE) is a process that
starts with careful planning prior to a study. It
allows for maximum interpretation of data with
minimal experimentation.
First, the process begins with discussions of parameters
that may affect the system under study.
The next step is to design the experiment properly.
Finally comes the statistical analysis of the experiment.
 Terminology
Factorial experiment: 2n experiment where n factors are varied
with respect to one another
Factor: a variable in an experiment that may effect the system
under study
Response: the output under study
Level: Upper or lower limit of a factor
Centerpoint: Middle condition of a factor
Replication: Number of times an experiment is repeated
OFAT: One factor at a time method of research
Best Guess: Attempt to run experiment at a best guess
Residues: Enable to determine model adequacy
 Terminology cont.
Blocking: Method to separate a lengthy or immobile series
of experiments without losing statistical analysis
ANOVA: Analysis of Variance method to statistically analyze
the results of an experiment using t-test or other methods
Randomization: Allows for error analysis: in operator,
equipment, experiment design, etc
t-test (prob>|t|): Measure of error, effect, or interaction by
seeing if prob>|t| is significant in a t-distribution (>95 or
99%).
Factor interaction: When two factors are certain levels
cause a significant change in response together
 Why use DOE?
OFAT Design: DOE Design:
3 factors 3 factors
3 levels 3 levels
3 replicates 3 replicates
Total # of expts: 3 centerpoints
Total # of expts:
3x3x3x3 = 81
23x3+3 = 27
 Why use DOE?
OFAT Design: DOE Design:
4 factors 4 factors
3 levels 3 levels
3 replicates 3 replicates
Total # of expts: 3 centerpoints
Total # of expts:
4x3x3x3x3 = 324
24x3+3 = 51
 Benefits of DOE
Allows for correlation of data, statistically
easier verses OFAT or best guess designs
Reduces the total number of experiments
Allows for good, thought out experimental
designs
Allows for error to be quantified
Can distinguish if factor(s) have any to no
effect or if interaction occurs
 7 Steps of DOE

1) Recognition and statement of problem

2) Choice of factors, levels, and ranges
3) Selection of response variable
4) Choice of experimental design
5) Perform the experiment
6) Statistical analysis of data
7) Conclusions and recommendations
 Example
Photosynthetic bacterial growth
Determine which factors effect growth of strain
Factors/levels(- 0 +):
[A] Light Intensity (100 500 1000 lux)
[B] Temperature (25 40 55C)
[C] Media/Air volume (1/3 1/2 2/3)
Response is concentration (by absorbance
spectroscopy) after 3 days growth
Design: 23 with 3 replicates and 3 centerpoints
 Results of Analysis
Factor Coeff Est Std Error Prob>|t|

Intercept 2.12 0.013

A-Temp -0.5 0.013 <.0001

B-Intensity 0.72 0.013 <.0001

AC -1.18 0.013 <.0001

ABC -0.77 0.013 <.0001

 Factor Effects

Off all factors and

possible interactions,
the most significant
were A, B, AC, and
ABC.
The interaction of
factors AC can be seen
in the graph to the
right.
 Conclusions

Growth of photosynthetic bacteria is highly

dependent upon its A (temperature), B (light
intensity), interaction between A and C
(media/air volume ratio), and A, B, and C.
Highest concentration (in millions of cells) is
when:
conc = 2.12-.5A+.72B-1.18AC-.77ABC

is maximized when: A(-), B(+), and C(+)

 Doug
Montgomery
Theorems
Theorem 1. If something can go wrong in
conducting an experiment, it will.
Theorem 2. The probability of successfully
completing an experiment is inversely
proportional to the number of runs.
Theorem 3. Never let one person design and
conduct an experiment alone, particularly if
that person is a subject-matter expert in the
field of study.
 Doug
Montgomery
Theorems
Theorem 4. All experiments are designed experiments;
some of them are designed well, and some of them are
designed really badly. The badly designed ones often tell
you nothing.
Theorem 5. About 80 percent of your success in
conducting a designed experiment results directly from
how well you do the pre-experimental planning (steps 1-3
in the 7- step procedure in the textbook).
Theorem 6. It is impossible to overestimate the logistical
complexities associated with running an experiment in a
complex setting, such as a factory or plant.
 Final Word

Finally, my friend Stu Hunter has for many years

said that without good experimental design, we
often end up doing PARC analysis. This is an
acronym for:

Planning After the Research is Complete

What does PARC spell backwards?