glance.ivreg fails with multiple endogenous regressors

[lukesonnet]

glance::ivreg(., diagnostics = TRUE) will fail if the 2SLS has multiple endogenous regressors. This is due to the fact that summary.ivreg(., diagnostics = TRUE) will return a diagnostics matrix that has no row named "Weak instruments", which is required in the following line:

broom/R/aer-tidiers.R

Line 123 in 66411c4

statistic.weakinst = diagnostics["Weak instruments", "statistic"],

library(broom)
library(AER)
# Works
glance(ivreg(mpg ~ hp | qsec + am, data = mtcars), diagnostics = TRUE)
# Fails
glance(ivreg(mpg ~ hp + wt | qsec + am, data = mtcars), diagnostics = TRUE)

As you can see in the latter case if you do summary(ivreg(mpg ~ hp + wt | qsec + am, data = mtcars)), the diagnostics have two rows named "Weak instruments (hp)" and "Weak instruments (wt)" instead of "Weak instruments", because the first-stage fstat is separate for each endogenous regressor.

This crash should be avoided by either a workaround that fails to return the f-stats or a solution that returns them all. Of course, returning them all means having an unexpected number of columns in the return, which is undesirable.

Thoughts? I encountered this while looking at mimicking glance.ivreg in estimatr::glance.iv_robust.

[IndrajeetPatil]

Here is a reprex for the issue mentioned above-

set.seed(123)
library(broom) 
library(AER) 
#> Loading required package: car
#> Loading required package: carData
#> Loading required package: lmtest
#> Loading required package: zoo
#> 
#> Attaching package: 'zoo'
#> The following objects are masked from 'package:base':
#> 
#>     as.Date, as.Date.numeric
#> Loading required package: sandwich
#> Loading required package: survival

# Works 
glance(ivreg(mpg ~ hp | qsec + am, data = mtcars), diagnostics = TRUE) 
#> # A tibble: 1 x 13
#>   r.squared adj.r.squared sigma statistic p.value    df statistic.Sargan
#>       <dbl>         <dbl> <dbl>     <dbl>   <dbl> <int>            <dbl>
#> 1     0.592         0.578  3.92      38.5 7.92e-7     2             12.8
#> # ... with 6 more variables: p.value.Sargan <dbl>,
#> #   statistic.Wu.Hausman <dbl>, p.value.Wu.Hausman <dbl>,
#> #   statistic.weakinst <dbl>, p.value.weakinst <dbl>, df.residual <int>

# Fails 
glance(ivreg(mpg ~ hp + wt | qsec + am, data = mtcars), diagnostics = TRUE)
#> Error in diagnostics["Weak instruments", "statistic"]: subscript out of bounds

And traceback-

> traceback()
9: eval_tidy(xs[[i]], unique_output)
8: lst_quos(xs, transform = expand_lst)
7: tibble(statistic.Sargan = diagnostics["Sargan", "statistic"], 
       p.value.Sargan = diagnostics["Sargan", "p-value"], statistic.Wu.Hausman = diagnostics["Wu-Hausman", 
           "statistic"], p.value.Wu.Hausman = diagnostics["Wu-Hausman", 
           "p-value"], statistic.weakinst = diagnostics["Weak instruments", 
           "statistic"], p.value.weakinst = diagnostics["Weak instruments", 
           "p-value"])
6: eval(substitute(expr), data, enclos = parent.frame())
5: eval(substitute(expr), data, enclos = parent.frame())
4: with.default(s, tibble(statistic.Sargan = diagnostics["Sargan", 
       "statistic"], p.value.Sargan = diagnostics["Sargan", "p-value"], 
       statistic.Wu.Hausman = diagnostics["Wu-Hausman", "statistic"], 
       p.value.Wu.Hausman = diagnostics["Wu-Hausman", "p-value"], 
       statistic.weakinst = diagnostics["Weak instruments", "statistic"], 
       p.value.weakinst = diagnostics["Weak instruments", "p-value"]))
3: with(s, tibble(statistic.Sargan = diagnostics["Sargan", "statistic"], 
       p.value.Sargan = diagnostics["Sargan", "p-value"], statistic.Wu.Hausman = diagnostics["Wu-Hausman", 
           "statistic"], p.value.Wu.Hausman = diagnostics["Wu-Hausman", 
           "p-value"], statistic.weakinst = diagnostics["Weak instruments", 
           "statistic"], p.value.weakinst = diagnostics["Weak instruments", 
           "p-value"]))
2: glance.ivreg(ivreg(mpg ~ hp + wt | qsec + am, data = mtcars), 
       diagnostics = TRUE)
1: glance(ivreg(mpg ~ hp + wt | qsec + am, data = mtcars), diagnostics = TRUE)

[weiyangtham]

I'm interested in taking this issue on. After talking with @alexpghayes, one possible solution is to not return the F-stat as part of glance() and have it returned from tidy instead. Let's have tidy return either the endogenous variables or the instruments by including an argument tidy(instruments = TRUE).

[alexpghayes]

Closed in #641!

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.