Tools for analyzing 3 factors at the same time

[DavidB]

Hi!
I am searching for a tool to analys 3 factors at the same time. I have a sample with 50 parts where I have taken 3 different measures. Now I need to know if there is any interaction between these measures. Someone that knows a good software?

 

[Rob Nix]

A simple starting point may be to put the data into an Excel spreadsheet (3 columns [factors] and 50 rows [samples]). Then use the CORREL (for correlation coefficient) function, comparing column A with B, B with C, and A with C.

A correlation coefficient close to 1 or -1 shows strong "interaction" and close to zero means very little or none.

Of course, some of the experts here at the cove are better at this than me, so I look forward to their ideas.

p.s. Look at the graphs of the data too, for any interesting patterns.

 

[Bev D]

are the 3 measures input factors or output factors for the 50 parts?
...and as a quick 'note to self', it is usually mor eeffective to understand how the data will be analyzed before collecting it - this ensures that we collect the right data in the right way...

 

[DavidB]

Thanks!
I found Minitab wich I am noe using

 

[Tim Folkerts]

I just want to reiterate what Rob said about plottting the data. Plot x1 vs x2, x1 vs x3 , and x2 vs x3 and see if there appears to be any relationships. (Minitab can do this in one easy graph using a matrix plot.)

Calculating the correlation or doing a linear regression tells you how well the data fits a straight line. Even if the data perfectly fits a parabola or other curve, the correlation may be close to 0. The caclulation would tell you there is little or no relationship, but you eyeball might spot the actual relationship.

Tim F

 

[wmarhel]

Hi!
I am searching for a tool to analys 3 factors at the same time. I have a sample with 50 parts where I have taken 3 different measures. Now I need to know if there is any interaction between these measures. Someone that knows a good software?

Something to keep in mind is that "correlation" doesn't always equate to "causation". The trick is to figure out why the interactions occur and what variables are to blame.

Not to muddy the waters or anything, but it is something to keep in mind.

wayne

 

[Jim Wynne]

Something to keep in mind is that "correlation" doesn't always equate to "causation". The trick is to figure out why the interactions occur and what variables are to blame.

Not to muddy the waters or anything, but it is something to keep in mind.

wayne

Good point. Height is related to weight in that in general, taller people are heavier than shorter ones, but tallness doesn't cause heaviness, or vice-versa. Before an experiment is designed, there needs to be a hypothesis to be confirmed or supported by the results. Simply examining relationships between variables might not be productive. Examples of hypotheses might be, "A causes B, independently of the influence of C" or "A, in combination with B, always results in C."

 

[Rob Nix]

Yes, and if A and B once dated, and A and C both smoked, and B and C are allergic to pears, then A is likely to be taller than D.

I used the illustration in some training materials that at a certain beach, ice cream cone sales were measured each day during the summer, and so was the number of drownings. A correlation study showed that as ice cream sales went up, so did drownings. Therefore, does eating ice cream at the beach cause drownings?? A third variable is temperature, which relates to both.

So, LOOK at the raw data, the graphs, for relationships (not for causation) and do enough research to find other contributing variables and interactions until something reasonable and provable (reproducible) appears.

 

[Jim Wynne]

Yes, and if A and B once dated, and A and C both smoked, and B and C are allergic to pears, then A is likely to be taller than D.

I used the illustration in some training materials that at a certain beach, ice cream cone sales were measured each day during the summer, and so was the number of drownings. A correlation study showed that as ice cream sales went up, so did drownings. Therefore, does eating ice cream at the beach cause drownings?? A third variable is temperature, which relates to both.

So, LOOK at the raw data, the graphs, for relationships (not for causation) and do enough research to find other contributing variables and interactions until something reasonable and provable (reproducible) appears.

This is part of where the general distrust of statistics comes from--the deceitful use of relationships to suggest causal influence. I recall reading about a study several years ago about the relationship between eating red meat and incidence of heart disease. IIRC, the researchers published findings that they claimed proved a causal link; people who ate x amount of red meat per week were y times more likely to develop cardiac problems than people who ate less than x per week. Of course, what they didn't tell anyone, and what beef producers were quick to point out, was that the researchers didn't control for other potentially significant factors, and that in general, people who eat less red meat have a tendency to live healthier lifestyles in general--they exercise more, drink less alcohol, aren't as likely to smoke, etc.

 

[wmarhel]

This is part of where the general distrust of statistics comes from--the deceitful use of relationships to suggest causal influence.

There is a short book, "How to Lie With Statistics" that sums this up nicely. It isn't a bad little read at only 142 pages.

The sad reality is that we occasionally find something we can point at and say, "Voila" and run with it. Unfortunately, if we don't take the time to verify the results we may have ended up with a "False Positive", and a little bit of egg of on our face at the same time.

Wayne