What statistic method should I use for this situation?

flyforward_bin

New Member
Imagine if I have 3 types of parts A, B, C and there is one parameter to judge the quality of the parts under certain condition, but this parameter is not that reproducible. For example, under certain condition, A, B and C will have different value of electric surge. I understand if we collect large quantity of data, we can do t-test to compare among each two of them. But if under the condition, the electric surge on A, B and C is not that reproducible. Ex: there is percentage of chance for A, B, C to have electric surge. How can I combine this percentage of occurrence and absolute value of electric surge to determine which part is better. ( I would like to consider less occurrence rate and low electric surge value as the good one). What statistic method would be suitable for this situation?

Miner

TS Contributor
Are you referring to a surge test to detect insulation defects (such as in armature windings)? When you say the results are not reproducible, do you mean that there is a measurement error problem, or that the surge problem occurs infrequently in the product?

galegirly

New Member
So - putting this in psychology terms - A,B, and C are 3 different interventions. We want to know which performs best. But you cant find comparable samples to test them on?

flyforward_bin

New Member
Not every unit shows the surge. But we would like to consider both surge amplitude and occurrence of the surge together to validate which group is better. There is no measurement error in the problem. It is just the intrinsic property.

flyforward_bin

New Member
So - putting this in psychology terms - A,B, and C are 3 different interventions. We want to know which performs best. But you cant find comparable samples to test them on?
When you say " not comparable samples", did you mean not all the parts show the surge all the time? Is there a model in statistics can be used for this?

galegirly

New Member
I think i am understanding you correctly -you are saying you can't test ABandC in the exact same way - which is a bit like not have 3 nicely random samples to test your psychology intervention on - when this happens in psychology we still do the evaluation we just call it quasi-experimental - its not a true experiment but its the best we can do - its better than nothing as it were

Now, whether there is any value in doing a less than completely fair test of A, B and C in your case ...

If you can't run A, B and C through the exact same procedure, then any difference in results can be argued to be attributable to the procedure rather than widget A, B or C

galegirly

New Member
What about a correlational analysis (sorry, probably distracting you with my psychology analogy). If A has 10 performance data points, and, say 10 electric surge data points, and 10 amplitude points ... can you give us a better understanding of what the data looks like?

flyforward_bin

New Member
What about a correlational analysis (sorry, probably distracting you with my psychology analogy). If A has 10 performance data points, and, say 10 electric surge data points, and 10 amplitude points ... can you give us a better understanding of what the data looks like?
Thank you very much for your suggestions.
Actually, A, B and C they are going through the exact same procedure. Maybe I did not explain this well. Let me explain this in neuroscience ( might be closer to psychology).

There are A, B and C three types of neuron. Given the same electric stimulation, A, B and C will release neurotransmitter. We would like to study which type of neuron A, B or C is more resistant to electric stimulation, means given certain stimulation, the neuron releases min neurotransmitter or not at all. In reality, when we stimulate the neuron, we can see that some neurons within group A, B or C did not even release neural transmitter at all and this happens randomly. For the neurons that can release neural transmitter, we have ways like fastCV to record the quantity. Then the experiment results is the chance to release neurotransmiter is A>B>C( A has higher chance to release than B and C). Once the neurons release neurotransmitter, the quantity of neurotransmiter release A<B<C ( A release less than B, B less than C, we can get this through a t-test for mean).

Then if we consider both factors ( chance of release) and ( quantity of release), how do I pick which group is more resistant to electric stimulation, which means less chance to release and / or less volume released.

Karabiner

TS Contributor
non-release means "quantitity = zero". so you have only 1 outcome, as it seems,.

With kind regards

Karabiner

galegirly

New Member
It looks like 2 outcomes to me. Possibly 2 studies. Possibly some value in considering a categorical analysis? I might have a proper think later - having breakfast!

galegirly

New Member
It looks like 2 outcomes to me. Possibly 2 studies. Possibly some value in considering a categorical analysis? I might have a proper think later - having breakfast!