beniaminogreen / cragg

Test for weak IV instruments in R.

• Hi, I'm Beniamino (Ben) Green
• I'm a predoctoral fellow at the Yale School of Public Heath
• I am interested in computational social science, record linkage, and semiparametric statistics

Please checkout my blog here!

cragg's Issues

Error in stock_yogo_reccomender

Hi,
I am having trouble running a toy example. How can I fix this problem?

> library(cragg)
> library(AER) # for dataaset
> data("WeakInstrument")
> stock_yogo_test(
+     # control variables
+     X = ~ 1, 
+     # endogeneous variables
+     D = ~ x, 
+     # instrument variables 
+     Z = ~ z, 
+     data = WeakInstrument
+ )
Error in stock_yogo_reccomender(K = cd$K2, N = cd$N, B, size_bias) : 
  Error: K must be greater than N+1 for this operation

Error: cannot allocate vector of size 741.2 Gb

Hello,
I ran into the error above when trying to implement the Stock-Yogo test using your cragg package. I'm working with a data set of about 1.4 million observations, 28 IVs, and one endogenous variable (I'm assuming that for size distortion results it's ok to have one endogenous variable but let me know if this is a problem for the package). The offending code is below:

stock_yogo_test(
  ~factor(region) + factor(yob) + Race + SMSA + Married + Age + Age2,  #Controls
  ~educ,            #Treatments
  ~factor(birthqtr):factor(yob),   #Instruments
  B=.1,       #Maximum Allowable Size Distortion
  size_bias="size", #Calculate critical value for size distortions
  data = df
)

I'm happy to provide the data. They are the Angrist and Krueger "quarter of birth-returns to education" data, which I'm using to create a problem set. Thanks in advance.

intercept handling

I have issues with the functions in the package when an intercept is included explicitly and when the model should not contain an intercept.

The following code models the intercept explicitly but throws an error. The reason is that it leads to two intercept columns in X_m that are of course perfectly collinear.

df <- iris
df$CONS <- 1
cragg_donald(X=~CONS+Sepal.Length, D=~Sepal.Width, Z=~Petal.Length + Petal.Width + Species, data = df)

Similarly, the code internally always adds an intercept column to X_m but some models might not have an intercept.

It would be good to simply have a consistent handling of an intercept. It is possibly easiest if users need to specify an intercept explicitly by creating a variable and add it to the ~X formula. Alternatively, a function argument could ask the user whether an intercept should be added or not.