Hey folks,

Can you let me know if you use substack to read and comment on newsletters? I’m thinking about changing platforms for this newsletter and would love to get your opinion on what the experience on substack is like for you as an audience member versus through an email newsletter like this one.

Something amazing happened on Wednesday morning. At some point earlier in the week, I updated the software I use to do my Wednesday’s streams, which is called OBS. Upon launching OBS I got a spinning ball of death. Of course, I was launching it with about 10 minutes to spare before the start of the stream. After quitting and restarting OBS a few times, I decided to restart the computer. Of course, earlier in the week I had also downloaded an update to the operating system but kept putting off restarting the computer. By the time the computer rebooted it was time to go live. Only the spinning ball of death was still there when I opened OBS. Argh. After doing some googling, I learned there was another update to OBS. That did the trick. Only I was now about 10 minutes late to the stream. As I have the past two weeks, once I started the stream I pushed the wrong button and killed the stream.

All through this 30 minute frenzy I kept asking, why am I doing this?! I could just record videos, edit, and post whenever I want. Since I’m someone who understands the Sunk Cost Fallacy, but I refuse to live by it, I decided to go ahead and start a new stream rather than move on with my morning. I kept trying to remind myself that when I streamed weekly last year it really gave me a lot of joy. So I launched the new stream and I could see that there were people there watching and commenting.

This was when the amazing thing happened. I was all set to recreate one of the panels from the figure I talked about in last week’s critique video (Figure 1b). A few minutes in, someone asked what I thought of plotting the same data as a scatterplot, but with marginal histograms. Uhhhh. I hadn’t thought of that. I spent the next minute or so talking about one thing, but thinking about how I would make that plot. This is why I love livestreaming. I spent the next hour and a half making the plot the viewer asked about. You should go watch the stream to see how I made it. The plot below on the left was what I had planned on making, the plot on the right is what I made.

At the end, I discussed why I didn’t really like the version we made with the marginal histograms. It’s harder to interpret compared to the original version.

I relate this story for two reasons. First, you really should tune in to my livestreams! They are Wednesday mornings at 9 AM Eastern US on my YouTube channel. If I could bottle the type of question we had this week and open it each week I might just livestream every day ;) Second, the try something different and see if it works is a pretty mature take to visualizing data. I’d like to explore this with you more today.

One of the challenges with critiquing data visualizations without the authors available to answer questions is we don’t get the full context of the visual. I’m genuinely curious what other approaches they pursued to visualize the same data and why they chose this approach. Panel d is a scatterplot, albeit without the histograms. Why one approach in b and another in d? One strategy we could try in critiques is to think of all the different approaches one could take, rank them, and then see where the one the authors gave us ranked.

What I think we learned in Wednesday’s livestream was that what we thought would be the best way to plot the data might be true in theory, but when we apply it to the actual data, it might not be that effective. This comes back to the fact that context matters!

Now, I’d like to try this with you using a different figure. Check out this panel from a paper published as part of the paper, “Prior therapy defines mutation profiles in childhood cancer at relapse” from Nature.

Briefly, this panel is made up of 5 facets. Each facet represents different types of mutations on the same samples. Across the x-axis within each facet, the samples are divided between the stage of tumors and stage of treatment. The y-axis is the rate of that mutation. The data are then depicted as a scatterplot where each sample is plotted. They also included a red line with a number to indicate the median for each condition.

Now, take a few minutes and think of all the ways you might represent the data.

What did you come up with? Here are some of my ideas…

• Why not use box and whisker plots rather than showing the individual data points?
• Why not use a violin plot?
• Why not use a horizontal jitter rather than rank ordering the samples on the x-axes?
• What if we flipped the axes in each of the facets so that the names across the tops of each facet were on the left side of the plot, leading the mutation rate to be across the x-axis?
• If we flip the axes, what would this look like as a ridgeline plot?

Do you have any other ideas? There are other things we might play with like the order of the sample types, whether to put the y-axes on the same scale, colors, etc, but the ideas I tried to describe represent big structural changes. I’d love it if you could send me any other ways you’d change the overall structure of the panel.

Tune in next Wednesday morning! I’ll do my best to try out these ideas and any others you suggest along the way.

​

​