With thanks and apologies to Dr Bronner...


The project I'm working on right now is a 2-day course on multiple imputation of missing data. Way-nerdy stuff, even if you're a statistician. I am trying to overcome a dilemma in how to write the lecture portion of the course material.


Well. We have standards. For instance, we're not supposed to use bold, italics, or underlining for emphasis. There are very specific circumstances where these tools are used, such as defining a new term for the first time, and that is all. Which means that if I want to draw attention to one or two points on a page that are particularly important, there's not a good way to do so.


So in the shower this morning, it occurred to me- I'm adding a new style standard, known henceforth as the Dr Bronner method.


For example, instead of:

Most data analysis situations involve the estimation of unknown parameters from a sample drawn from the population of interest, typically with some assumptions being made about that sampling process.


You'd see:

Assumptions are being made about that sampling process!! OK!

Observations provide information for estimating the parameters corresponding to the variables used in the analysis!!!!! OK!


And instead of:

Nonignorable missingness depends upon either of two conditions. The first condition states that if the mechanism that leads to the missing observations (PHI) is unrelated to the (unknown) parameters for the model, THETA, missingness is nonignorable.

The second condition is more likely: if missingness for a particular variable is related to the true value of that variable, then we say that missingness is not ignorable, and furthermore, that it is not random.


You'd see:

Understand nonignorable missingnes!! Two conditions!! Two! Two!! OK!

First Condition: Phi is unrelated to Theta!! Unrelated! Unrelated!!!

Second Condition: Missingness is not related to the True value of the variable! Not related! Not related! Not random!!!


And finally, instead of:

Sophisticated methods of handling unobserved data can perform reasonably well even with nonrandom missingness. Specifically, it may be possible to model the missing data mechanism before performing certain types of imputation. The more information you have about your observations, the more likely it is that you can adequately model the missing data mechanism PHI and make the missingness ignorable.


you'd see:

We can handle the nonignorable missingess!! AllOne!! AllOne!!

Model the missing mechanism before the analysis!! Before the analysis!! Before the analysis!!

Find the model for Phi!! OK!!

We must protect ourselves from nonrandom missingness to protect Spaceship Earth!! OK!


This approach, I think, would really reach students where they live. Way more effective than long, boring paragraphs. What do you think? Will it work?