Manuscript, code, and data for "Compression and Conditional Effects." [Paper]

Previous research in political methodology argues that researchers do not need to include a product term in a logistic regression model to test for interaction if they suspect interaction due to compression alone. I disagree with this claim and offer analytical arguments and simulation evidence that when researchers incorrectly theorize interaction due to compression, models without a product term bias the researcher, sometimes heavily, toward finding interaction. However, simulation studies also show that models with a product term fit a broad range of non-interactive relationships surprisingly well, enabling analysts to remove most of the bias toward finding interaction by simply including a product term.

If you have any comments or suggestion, please open an issue or just e-mail me.

To replicate all the results, simply download the directory, and run the files

• do-all.R, which performs all the Monte Carlo simulations, creates the plots, and does the Oneal and Russet replication.

Note that the simulation take a very long time to run. If you like, you can tradeoff the speed and precision of the simulations by changing the lines

# Set simulation parameters
n.sims1 <- 2000
n.sims2 <- 2000

in the file do-all.R. n.sims1 controls the number of times to compute each confidence interval to compute the probability of covering the true value. n.sims2 controls the number of simulations used to construct each confidence interval. Also, in the file sims-generic.R, the parameter niter controls the number of random DGPs to compute.

Note that I did some post-editing on the figures in order to clearly annotate relevant information. I did this in Inkscape.