0

I run instrumental variable regression (2SLS) with ivreg package.

library(ivreg)
as_tibble(mtcars)

model_iv <- ivreg(formula = mpg ~ disp + drat |
                                         drat + carb,
              data = mtcars)

As many other people, I am interested in the diagnostics (e.g. weak instruments, wu-hausman and sargan) attached to my model reporting.

summary(model_iv, diagnostics = TRUE)

Diagnostic tests:
                 df1 df2 statistic p-value   
Weak instruments   1  29     8.286 0.00743 **
Wu-Hausman         1  28    11.594 0.00202 **
Sargan             0  NA        NA      NA   
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

There is a workaround to add those to stargazer:

R: Robust SE's and model diagnostics in stargazer table

And I found the very discussion in an issue for modelsummary:

https://github.com/vincentarelbundock/modelsummary/issues/123

I tried the modelsummary but it seems that the feature hasn't been implemented. Is there another package for model reporting including diagnostics for instrumental variable models?

Marco
  • 2,368
  • 6
  • 22
  • 48

1 Answers1

3

You can easily do this with modelsummary. Please see the detailed documentation on how to add new goodness-of-fit statistics on the website: https://vincentarelbundock.github.io/modelsummary/articles/modelsummary.html#new-models-and-custom-statistics

There are basically 3 options.

Option 1: add_rows

Build your own data frame with the same number of columns as your table, and use the add_rows argument to stick it at the bottom of the table. This allows you full control, but can be tedious. It is also quite self-explanatory, so I don’t give an example here. Just see ?modelsummary.

Option 2: metrics argument

Behind the scenes, modelsummary extracts goodness of fit statistics using the performance package. That package supports a metrics="all" argument which sometimes return more information. In this case, we can feed metrics directly to modelsummary and get the Wu-Haussman statistic:

library(ivreg)
library(tibble)
library(modelsummary)

model_iv <- ivreg(
    formula = mpg ~ disp + drat | drat + carb,
    data = mtcars)

modelsummary(model_iv, metrics = "all")
(1)
(Intercept) 51.687
(17.451)
disp -0.072
(0.020)
drat -4.169
(3.649)
Num.Obs. 32
R2 0.455
R2 Adj. 0.418
AIC 193.3
BIC 199.2
RMSE 4.38
wu.hausman 11.5944242731131
wu.hausman.p 0.00201604586133519

Of course, you’ll probably want to format this statistic with the gof_map argument to reduce the number of digits and clean up the names.

Option 3: glance_custom

Finally, you can define a glance_custom.ivreg() method to extract, format, and add the statistic automatically. There is a detailed tutorial at the link I posted above, so here’s a simple example:

glance_custom.ivreg <- function(x, ...) { # don't forget ...
    s <- summary(x)$diagnostics
    wi <- s[1, "statistic"]
    wh <- s[2, "statistic"]
    tibble(
        "Weak Instrument" = round(wi, 2),
        "Wu-Haussman" = round(wh, 3))
}

modelsummary(model_iv)
(1)
(Intercept) 51.687
(17.451)
disp -0.072
(0.020)
drat -4.169
(3.649)
Num.Obs. 32
R2 0.455
R2 Adj. 0.418
AIC 193.3
BIC 199.2
RMSE 4.38
Weak Instrument 8.29
Wu-Haussman 11.594

All subsequent ivreg models will automatically have this new statistic.

Vincent
  • 15,809
  • 7
  • 37
  • 39