Coming full circle and working on my BRMS, this is the Bayesian model alongside the regular old frequentist one. I am still of the opinion that based solely on this data, there is no prediction to be made about the midterms.

But the range of predictions is wrong too. To get that you need not just the std err of the fit, but the std err of the residuals, in particular the prediction is distributed as T(df=10, mu=.fitted, se=sqrt(se(fit)^2 +se(resid)^2))).

With the correct model it's anywhere from -88 to +6.

The CI he uses is the CI for the estimate of the slope and intercept. Here I use the ggdist package (ty @mjskay.com) to visualize the model. Note even for a generous 90% (or even 80%) CI this is non-significant so the statement about lower approval <-> lost seats is not supported by the model.

Saw this post from @gelliottmorris.com and felt something was off.

www.gelliottmorris.com/p/trumps-app...

Reproducing his chart, I had to use a confidence interval of 90% to match his. My guess is this is just a typo or some other trivial mistake. But that isn't my main gripe.

Though I bounce through their whole catalogue.

A favorite of mine is this one showing cumulative new vulnerabilities year over year. Shows how the current year is out pacing others.

This was used to diagnose a couple of data changes and errors in several years. Made by Jay Jacobs of EPSS fame.

If you see this, post a photo taken in the mountains.

Because I am an absolute nervous wreck, it's Friday I did the same chart, fixed it up with MD flag colors, and included a binomial test to see if all the districts were significantly different than the overall population breakdown.

gist.github.com/bjedwards/b0...

I get how this is the most consistent way to set up a new tool so it "works", but jeez....

When I was a kid my grandpa lived in Hawaii and gave me this book and it was/is one of my favorites.

So many stories similar to this one of spirits out in the jungle messing with people, or sometimes posing real danger.

I have the softest of spots for useless, hairball network visualizations.

That has some pretty specific specs. I looked around and it probably uses the same movement as in this watch(never heard of that brand).

ba111od.com/collections/...

Manual wind, 105 hour power reserve, 3hz. The back on the ba111od looks like a better finished version of the trump watch.

I appreciate this your confidence in my js/svg. :D

I solved this with an ugly hack. I used a text mark, without a filter, but change the text channel to only render the middle points. I have render a " ", as having the text channel function return null seems to operate the same as filter?

I had an excess of milk, eggs, and a stale loaf of bread I so made this last night/this morning. I was halfway through when I realized this is bread pudding. It's pretty good if you like bread pudding. It's also massive. Brunch party massive.

First Wednesday. First Wednesday. First Wednesday.

Not bad, MN. Not bad at all.