chapter on placebo groups

[djnavarro]

PR to add a chapter on placebo data in the book. [Work in progress: description to be updated later]

[djnavarro]

Hi @yoshidk6 -

I've started work on adapting the placebo handling code to work as a book chapter. It's very minimal so far (literally just shows basic univariate ER models where placebo data are present). However, it has made me think that there is a case to be made for a much more minimal PR to the package than the rather elaborate one we considered earlier.

I think the plot_er() method for binary models should support a bin_breaks option within the options_orig_data list argument. Currently we support n_bins for binary ER data, but that won't be sufficient when placebo samples are included: almost always the user will want placebo samples in a distinct group regardless of how many observations there are. The easiest way to handle this would be to let the user manually specify the bin edges. I could imagine that option being useful in other situations too. If that makes sense to you I could put together a quick PR that adds this option (and perhaps also adds the d_sim_placebo data set to the package) and then showcase it in this chapter.

As usual, let me know what you think :)

Best, Danielle

[yoshidk6]

Thanks a lot Danielle!

I agree that binning can get complicated rather quickly.
Actually for the cases where we do this (binary ER), more often than not I think people don't want to include placebo data in ER modeling, and that can result in different workflow than when the placebo data is included in the modeling.

In any case I like what you are proposing (enabling the use of manual breaks) for the case outlined in the draft, especially considering that xgxr::xgx_stat_ci() already handles breaks natively.
I assume you will show how we can calculate breaks manually and use that as input?

I think it's also helpful to show the workflow in which the placebo data was not included in the model building but people want to visualize data. Probably we can show how the users can add box & whiskers (another xgxr::xgx_stat_ci()?) as well as datapoints as jitter. Using grey/half-transparent colors might be a good way to clarify that this part of data was not used in model development.