Analyzing a 7-variable data set where multiple variables changed at once

[swooshfactory]

Hi,

I have some experimental data from a colleague which has been collected with different values for 7 parameters. It isn't nice, as in it's not as if 6 of the parameters were held fixed while the seventh was adjusted. Beyond that, I don't have much data. I'm not up on my statistical significance, but this data will probably prove to not be significant.

I'm wondering if there is some software that would take a shot and analyzing this data and maybe indicate some trends.

Thanks for reading!

 

[Stephen Tashi]

You haven't explained enough about the data for anyone to know what hypothesis about it you wish to test.

 

[swooshfactory]

Ok, my apologies. The data is for growing thin films with different growth conditions. Some of the sets of variables produce "good" films, some "mediocre" and some "bad". I cannot formulate a hypothesis based on the data - the conditions seem so different from one data set to the next.

I would like to be able to predict which conditions will produce good films. If more information is needed, I will gladly supply it.

Matt

 

[Stephen Tashi]

I cannot formulate a hypothesis based on the data - the conditions seem so different from one data set to the next.

I would like to be able to predict which conditions will produce good films.

There are two different goals in your previos posts. The question of determining "statistical signifcance" is much less ambitious that the goal of making good predictions. In the usual sort of statistical testing, one often assumes a "null hypothesis" that essentially says "The conditions don't affect the outcome". One analyzes the probability of the data based on that very unpredictive model. If the probability is low then one may subjectively "reject" the null hypothesis. However, this procedure merely decides that the conditions do affect the outcome without specifying exactly how you can use the conditions to make a prediction.

As to making predictions, I think there are forum members who could advise you if present the data. (I'm not sure that I'm one of them.) It would also be helpful to know any basic physics that underlies the process.

 

[viraltux]

I would like to be able to predict which conditions will produce good films. If more information is needed, I will gladly supply it.

Matt

Hi Matt,

This seems an optimization problem but Mr. Tashi is right; since there are a myriad of optimization techniques more details about your project would be helpful to determine which one better suits you.

Is there an underlying physical model you can adjust your data to? Are there economical constraints upon the value of the seven parameters (e.g. the lower one parameter the cheaper the process)? Would you be able to gather more data if necessary and, if you could, there are limitations in time or money to do it?

But anyway, at first sight you may try to fit some linear model with quadratic higher order terms and then, if the model fits nicely, simply check for the optimal solution in that model.

And about software to do these analysis, I would recommend you R (http://cran.r-project.org/). R is a very powerful language for statistical analysis and you'll be able to analyze your data in many different ways.

 

[swooshfactory]

Thank you both for your replies. Overall, there must be some pattern (I would think), since other people in my group have said that they can reproduce good films using certain conditions. The data I have, though, is from someone exploring many different growth conditions.

Some of the physical principles would be something like high temperatures destroy order.

The price is assumed to be the same for each process, but I will consider this in my review.

More data could be collected, but I'm not trained on this machine, so it's possible it won't be for a while.

Thanks again...

Matt

 

[viraltux]

Thank you both for your replies. Overall, there must be some pattern (I would think), since other people in my group have said that they can reproduce good films using certain conditions. The data I have, though, is from someone exploring many different growth conditions.

Some of the physical principles would be something like high temperatures destroy order.

The price is assumed to be the same for each process, but I will consider this in my review.

More data could be collected, but I'm not trained on this machine, so it's possible it won't be for a while.

Thanks again...

Matt

Well, then you could give a try to the lineal model and see how it goes. It will also give you hints on what parameters are more important for the quality which, by the way, you need a consistent way to measure it. So good luck!