Add a couple of diagnostics for variance

[aeturrell]

See specr for some inspiration.

• bar chart of the variance in coefficient explained by each type of variable (eg y vs controls)
• box plot of the variation in coefficient

[aeturrell]

For the variance decomposition, cannot use an approach based on mixed effects models via pymer as has non-Python dependencies. Should be in principle possible to do with statsmodels, see last example in page on variance component analysis.