Error in <fitstat (iv_tsls_fe, "kpr")> - non-compatible array dimensions in the expression involving "vcov" and "vcovClust"

This is my first time to your blog and I would need your kind help or advice, if possible to solve the problem I am about to describe.

Me and my team have estimated the following fixed effects OLS estimation using R:

iv_tsls_fe <- feols(mort_u65_pop ~ year + datacoverage + GDP + unemployment_rate + at_risk_poverty | clust2  | pm25 ~ rain_yr_mm + temp_c + wdsp, data = df2, vcov = vcov_cluster("clust2"))

Where we are estimating a fixed-effects instrumental variable (IV) model using the "feols" function. The dependent variable is "mort_u65_pop," and the independent variables include "year", "datacoverage", "GDP", "unemployment_rate" and "at_risk_poverty." The instrumental variables are "rain_yr_mm", "temp_c" and "wdsp". We use "vcov_cluster" to compute the clustered VCOV of fixest objects. In our case "clust2" is a variable inside the dataset "data = df2".

And then we have computed the fit statistics of fixed objects as follows:

fitstat(iv_tsls_fe, "ivwald")
fitstat(iv_tsls_fe, "cd")

fitstat(iv_tsls_fe, "kpr")

Although Wald and The Cragg-Donald tests for weak instrument work, for what concerns the The Kleibergen-Paap test for weak instruments I encounter the following error:

Error in vcov + (-1)^(i + 1) * vcovClust(index, bread, scores, adj = ssc$cluster.adj &&  : 
  non-conformable arrays

Unfortunately, we can't figure out where the problem lies, since the previous two tests work.

Would anyone have any suggestions on how to handle this problem? I would be very grateful.

I hope I have been clear in my presentation and remain available to provide further details in case I missed something.

Thank you very much,
Lorena

I think the issue here may be due to clustered errors not playing well with the Kleibergen-Paap test in this package. I have also had this error and noticed it only appeared when I was using clustered errors and would be interested if it is the same in your case.

Unfortunately I don't know of any direct fix for this within fixest. My current solution is to utilise ivDiag for my IV diagnostics. While in my case the number of fixed effects involved is on the lower end as to what one may have when using fixest, the backend of ivDiag is felm from lfe so it should be fine even for very large numbers of fixed effects. One last thing to note is that ivDiag doesn't actually report the KP test, however it does report the Montiel Olea & Pflueger effective f-statistic which reduces to the KP test in the just identified case and seems to be preferred to the kp test by at least the authors of this paper "https://scholar.harvard.edu/files/wirev_092218-_corrected_0.pdf".

Apologies for not being more help, however I hope that this is of some assistance.