Interpretation of DOE interaction plot

I don't do DOE's often/ever and wanted some help interpreting my results. I have concluded the following

a) Power/Seal time and Power/Shim height have no interactions. the rest do.

Here are my questions.
1) How can I rank the strength of the interactions? So far I would say
Pre Seal/Seal then a close call between Pre seal/Shim & Seal/Shim
and last Power/Pre Seal. Is this correct? Is there some metric in Minitab I can use to rank the interactions?

2) Does the position of the cross or the fact that they don't cross have any weight on the strength of the interaction? For example Pre seal/Shim & Seal/Shim. Pre seal/Shim crosses close to the -1 but Seal/Shim doesn't cross at all. Do you just look at the difference in slope?

3) Is there an idiots guide to DOE interpretation?

debun said:

a) Power/Seal time and Power/Shim height have no interactions. the rest do.

Click to expand...

I'd be very careful about this. just because the mean connectors aren't parallel doesn't mean that an interaction actually exists. It could simply be sampling error of the mean. the p value should help with this, but the REAL test is to plot the individual values about the means. remember statistical significance doesn't mean practical importance.

debun said:

Here are my questions.
1) How can I rank the strength of the interactions? So far I would say
Pre Seal/Seal then a close call between Pre seal/Shim & Seal/Shim
and last Power/Pre Seal. Is this correct? Is there some metric in Minitab I can use to rank the interactions?

Click to expand...

again plot the individual points about the mean to visualize the effect size. you can start with the adjusted r-square value to help you 'rank' them.

debun said:

2) Does the position of the cross or the fact that they don't cross have any weight on the strength of the interaction? For example Pre seal/Shim & Seal/Shim. Pre seal/Shim crosses close to the -1 but Seal/Shim doesn't cross at all. Do you just look at the difference in slope?

Click to expand...

where they cross is irrelevant. the most common interactions across industry are those that look like < or > and not X...

debun said:

2) Does the position of the cross or the fact that they don't cross have any weight on the strength of the interaction? For example Pre seal/Shim & Seal/Shim. Pre seal/Shim crosses close to the -1 but Seal/Shim doesn't cross at all. Do you just look at the difference in slope?

3) Is there an idiots guide to DOE interpretation?

Click to expand...

2) Converging lines will eventually cross. It just happens outside of your experimental design space.

3) While not an Idiot's Guide, this is a decent overview. Another from Minitab.

Miner (or Statistical Steven if you are out there): I seem to vaguely recall that there are two types of interactions: the one where the detail looks like X and the one where it looks like: /. I believe they had names? I can't find the reference right now. Some X interactions will look like / in a constrained design space, but some / interactions will never cross (X) because they are physically different types of mechanisms...

Bev D said:

Miner (or Statistical Steven if you are out there): I seem to vaguely recall that there are two types of interactions: the one where the detail looks like X and the one where it looks like: /. I believe they had names? I can't find the reference right now. Some X interactions will look like / in a constrained design space, but some / interactions will never cross (X) because they are physically different types of mechanisms...

Click to expand...

Hey Bev, sorry I missed connecting with you again after your talk at WCQI. I am not familiar with different names for the interaction plot patterns, though your point is spot on about looking at the individual values and see how the different runs impact the interaction!

BevD

Do you mean qualitative vs. quantitative? Qualitative interactions are those that change slope and direction, while quantitative interactions change slope, but not direction.

You can also have interactions between continuous variables as well as between continuous/discrete variables.

Miner - THANK YOU! that's exactly what I was thinking of...I need to 5S my brain...

Lots of good info in here. I was able to use the minitab help and they have a good write up on interpreting the results. The pretty pictures are cool but the p values are where the real info is.

A plot with individual values will beat a p value every day. From a stats standpoint all the value is in the adjusted r square number...