nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

How do I show evidence best?

Timo

Hi,

I'm working on my bachelor project with organic solar cells but I'm not sure what's the best way of presenting and analysing the data I gathered. You can see a typical plot in the attachment. On the x-axis there is the number of the solar cell, on the y-axis the efficiency that I got. Every device consists of 6 small solar cells on the same glass plate, the error bars show the standard deviation of these 'fingers' as we call them. The data point is just the mean of all six fingers. The different colours represent that I once used chloroform and once chlorobenzene for the solar cells.

Now I need to analyse the data, but that's a bit difficult. The main aim of my project is not to maximise the efficiency, but to mimize the standard deviation as of the six fingers on one device as of the devices itself. I thought about doing something like a hypothesis test where I use the standard deviations and try to minimize them. I'm extremely new to JMP and I could google hoe to make this hypothesis test but I'd especially like to know if you have some better ideas to show evidence in this case, what statistics I should consider in my analysis.

Cheers,

Timo

For_JMP_community.pdf

Accepted answers

dale_lehman

I think the hypothesis test is far less meaningful than the visualization (though probably some kind of t test for the means and the associated tests of the variances will serve the purpose). I think there are some features of the visualization (that you attached) worth thinking about. I am assuming that there is no importance to the ordering of the 12 devices and that these are not in any way matched samples (if either of these are incorrect assumptions, that will be important for any visualization). So, I would not bother ordering the X axis by the device number since the latter is not particularly relevant information. You might consider pooling together all the device data for each type (chloroform or chlorobenzene). It is evident from the graph that the chloroform samples are more variable - your visualizaion should make this clearer. So, I don't think you need to show each device separately. And, if you use the type (chloroform, chlorobenzene) in the overlay box in Graph Builder, that will be the most direct way to show the differences between the two.

What I would want to think about more carefully, is the variation within each device and the variation across devices (of a particular type). Can these two kinds of variation be put together (as with pooling the devices) or should they be looked at separately? Since I have no background whatsoever in this subject area, I have no opinion on that - but I think that is a critical matter to consider.

All comments

dale_lehman

To elaborate a bit further (and perhaps prompt someone else to give more definitive advice): I think a standard ANOVA would be the way to address whether the variability across devices is "significant" relative to the variability across the fingers. You could do this ANOVA for the two types and there is probably some appropriate test to compare the two ANOVA tests. However, I suspect your sample sizes are too small for this to be very fruitful, and I shy away from reliance on hypothesis tests for your conclusions - I think good visuallizations will be more meaningful. But you might start with ANOVA as a way to think about the variability within/between devices.

Timo

Thanks for your ideas. The order of the devices is the order in which they were made, I randomized this to avoid misleading trends. But you're right that it might be better to group this in the final plot, I will ask my supervisor how he thinks that is better. The idea of separating the two different variabilities is very good, I think I'm gonna try an ANOVA. The script that I wrote to plot the data already gives me an overall average and deviation for the whole group, but the deviation is the deviation of all the little fingers. In the context of separating the standard deviations it might be useful to add another errorbar that shows the standard deviation of the whole devices, I will implement that.