l2_reg() miscellaneous

[federicovegetti]

These are possible features that can be implemented in the level-2 function l2_reg()

• [ ] Write l2_reg() in a way that it accepts both

• (A) the full data, including level-1 and level-2 predictors.

• (B) the coefficients resulting from l1_reg(). To do so, it should behave differently based on the input. We can save the output of l1_reg() with a specific object class, so it will recognize it right away.

• [ ] How to handle the level-2 predictors?

• In version (A), l2_reg() should take a usual dataset with level-1 and level-2 predictors, as you would feed into lmer().

• How about version (B)? Should it ask for a data.frame containing the level-2 predictors and the key variable(s) to merge only if the input is an output from l1_reg()?

• [ ] Possibility to use the pipe %>% to concatenate l2_reg() right after l1_reg(). Is that useful?

[cmbosancianu]

It's a weak preference, but I thought the easiest way in the beginning is to actually never bring in l1_reg() to the surface. Rather, it becomes a background "helper" function for l2_reg() and plot_l1().

So a user would specify in l2_reg() a data frame, a l1_model, a l2_model, and a l1_outcome (kind of like focal_var does in l1_reg()). And then l2_reg() would pass the l1_model to l1_reg() in the background, and take back from l1_reg() the data frame with coefficients and SEs (for the l1_outcome), and group-level indicator.

Then, based on the l2_model specification, it would subset only country-level aggregates, merge with the data frame of coefficients and SEs and, run the estimation.

Maybe because it allows things to proceed quickly, I would suggest that l2_reg() takes B (the coefficients), and that it accepts as input a data frame with both level-1 and level-2 predictors, like lmer().

How does that sound for you?

[cmbosancianu]

@federicovegetti , what do you think of maybe piggybacking on a function that Lewis and Linzer (2005) already wrote for this? see here