Minitab 17 Discrete X Discrete Y Question

#1
I am a little rusty with stats, and have a question. I have taken a survey for the Boy Scouts for an upcoming adventure. There are 24 scouts and I need to break them up into 3 groups of 8. I have a survey out where each of the scouts can select up to four other scouts they would like to be paired (grouped) with (ie, first preference, second preference, third preference and fourth preference). So there are 24 lines in minitab (one for each scout), and 5 columns (first column is scout providing their preferences, and the next 4 columns are the preferences (second column is first preference, third column is second preference, fourth column is third preference and 5th column is fourth preference. The scouts actual name is the data inside the minitab worksheet. Is there a tool in minitab which can help me decide statistically who I should group with who based on their input?

Minitab Example.JPG

Thanks in advance....
 
#3
I am running a 2^5-1 half-factorial design (with 16 runs and no replication). k=5, P=1 with a generator E=ABCD. in the result from Minitab the P-value and f-value appear as "*" without value. How can I resolve that problem. In Term I've chosen main and 2nd order interactions.
 

Attachments

Last edited:

Dason

Ambassador to the humans
#4
You have as many terms in the model as you have observations. You can get estimates but you need more data to do any significance tests if you want to stick with that model. With these types of designs sometimes methods other than pvalues are used to 'test' though.
 
#5
Just drop a few of the interaction terms. Then you will get tests and p-values.

And/or keep all main effects. Then also include those inteaction terms who's main effect are large or significant.

With these types of designs sometimes methods other than pvalues are used to 'test' though.
Yes, sometimes the parameter estimates (skip the intersept) are plotted on a PP-plot or a QQ-plot. What would be random numbers will look like a streight line. The "significant" will deviate from a streight line.
 
#6
Just drop a few of the interaction terms. Then you will get tests and p-values.

And/or keep all main effects. Then also include those inteaction terms who's main effect are large or significant.



Yes, sometimes the parameter estimates (skip the intersept) are plotted on a PP-plot or a QQ-plot. What would be random numbers will look like a streight line. The "significant" will deviate from a streight line.
I keep main and 2nd order interactions.You mean I need to keep more than this?like 3rd order?
 
#8
Frankly, I did not read the first post so well. I just saw the third post and that it was a half fractional factorial. So I don't understand how the matching problem turned to fractional factorial.

So I guess that Miners comment is most relevant. (But maybe the user learnt something about factorials.)
 

Dason

Ambassador to the humans
#9
Frankly, I did not read the first post so well. I just saw the third post and that it was a half fractional factorial. So I don't understand how the matching problem turned to fractional factorial.
It really isn't related and should probably be a separate thread.
 
#10
Just drop a few of the interaction terms. Then you will get tests and p-values.

And/or keep all main effects. Then also include those inteaction terms who's main effect are large or significant.



Yes, sometimes the parameter estimates (skip the intersept) are plotted on a PP-plot or a QQ-plot. What would be random numbers will look like a streight line. The "significant" will deviate from a streight line.
Thank you for replying. I keep main and 2nd order interactions.You mean I need to keep more than this?like 3rd order?
 

Miner

TS Contributor
#12
As previously stated, your model is fully saturated. There are only enough degrees of freedom to estimate 15 terms. Your main effects and all 2-way interactions total 15. You need at least one degree of freedom to estimate the error term for a significance test. The best way of doing that is to add replicates, which add degrees of freedom. Another (next best) way is to add 3-5 center points to your design. The last (least effective) way is to remove at least one 2-way interaction (with the smallest effect size) from the model.
 
#13
As previously stated, your model is fully saturated. There are only enough degrees of freedom to estimate 15 terms. Your main effects and all 2-way interactions total 15. You need at least one degree of freedom to estimate the error term for a significance test. The best way of doing that is to add replicates, which add degrees of freedom. Another (next best) way is to add 3-5 center points to your design. The last (least effective) way is to remove at least one 2-way interaction (with the smallest effect size) from the model.
Thank you Minor,