1. In R formula speak, y ~ a + b + a:b is the same as y ~ a * b

  2. The need for centering predictors when model averaging across interactions:

http://seananderson.ca/2014/07/30/centering-interactions.html

  1. Attached is Schielzeth 2010 “Simple means to improve the interpretability of regression coefficients”, which comes up again and again. Definitely worth a read.

  2. The Zuur books cover selecting random effect structures with information criteria, making sure to fit using REML.

  3. We talked about options for modelling with ordinal predictors. Andrew Gelman has some interesting thoughts on the topic:

http://andrewgelman.com/2009/10/06/coding_ordinal/

  1. We talked about interpreting coefficients for log-transformed response data. I wrote a few notes on the subject last year:

https://dl.dropboxusercontent.com/u/254940/log-slope.pdf Essentially, remember that the exponentiated coefficient is the expected percent change in the (untransformed) response for one unit change in the predictor.

  1. Rolling your own hurdle model to model continuous-positive data with zeros:

http://seananderson.ca/2014/05/18/gamma-hurdle.html

  1. The last object displayed in the R console is stored in an object named .Last.value This is useful if you forgot to capture some output in an object or, say, if you want to look at the output again with head(), View(), or dplyr::glimpse(). E.g.
1 + 1
## [1] 2
.Last.value
## [1] 2