DOE (Design of Experiments) or Hypothesis test? Applicability and use of each?

During Analyze Phase of Six Sigma Improvements, there are different tools and techniques. Some of them are DOE (design of Experiment) and Hypothesis Testing.

I wish to know how do the two differ with respect to applicability and use?

Thanks,

DOE vs. Hypothesis testing

patkim said:

During Analyze Phase of Six Sigma Improvements, there are different tools and techniques. Some of them are DOE (design of Experiment) and Hypothesis Testing.

I wish to know how do the two differ with respect to applicability and use?

Click to expand...

DOE is a statistical tool to examine the input factors and any response variables and their relative importance in affecting the process.
https://www.isixsigma.com/tt/doe/

Once you have acted on the process (i.e. made some change to hopefully improve), a hypothesis test is used to determine if what you believed would happen actually did happen. This compares samples to population parameters.

(broken link removed)

DOEs are typically used to evaluate the effects of multiple factors and their potential interactions.

Hypothesis tests are more typically used to evaluate the effect of either changing a single factor or the effect of a single change of multiple factors.

Of the two, DOE is much more effective in modeling how the process responds to changes in the input factors.

Let me second what has been said and add a few comments.

First of all, at its most basic level, "DOE" simply means a designed experiment. (As opposed to what - a random experiment???) Any experiment that has been planned to accomplish a specific purpose is a DOE

Traditionally, "DOE" has become synonymous with the use of a particular set of standard designs - 2 level factorials, central composite, Box-Behnken, Taguchi, ... These designs were created by experts so that they can efficiently accomplish specific purposes. When used properly, these designs are quite effective.

At some level, the standard designs are just sets of hypothesis tests performed simultaneously. For example, a 1 factor, 2 level factorial experiment tests 1 hypothesis: does changing Factor A make a difference? A 2 factor, 2 level full factorial tests 3 hypotheses: does changing Factor A make a difference, does changing Factor B make a difference, does changing both factors at the same time make a difference? As you add more factors and more levels, you have more hypothesis tests. When you do a frctional factorial, you remove some of the hypothesis tests.


I suppose you coud say that choosing the proper design allows testing of all the hypotheses you are interested in without too much extra work.


Tim F

I agree with the comments made by Tim, and would add a few additional comments and observations.

1. Hypothesis testing is a subset of DOE. All DOEs are setup based on a hypothesis (Is level 1 different than level 2 for a given factor, does Factor 1 and Factor 2 interact,etc.).

2. It is ironic how when asked to explain statistical concepts, the links quoted are from isixsigma.com. There are far better resources available to understand statistics.

3. When you get your green belt or black belt are these concepts reviewed?

4. I assume the tools you would use are DOE and statistical tests of significance (comapring means, variances, etc.).

Do you know of any inernet sites which are statistical resources?

Try these:


https://www.itl.nist.gov/div898/handbook/index.htm
https://www.math.com/homeworkhelp/Statistics.html
https://mathworld.wolfram.com/topics/ProbabilityandStatistics.html
https://www.sqconline.com/
https://www.statsoft.com/textbook/stathome.html

Concept is nicely explained by all the members, I just want to summarize it as Hypothesis testing is part of DOE . DoE is based on hypothesis (what we are looking for) and we design our experiment such that we can able to learn from data about our hypothesis.