data.table issue with SE computing in panel.qtet

[effedp]

Hello Brantly,

I am experiencing an issue when computing the SEs with panel.qtet.

With the option se=FALSE everything runs smoothly except for a couple of warnings, but when I specify se=TRUE, iters=10 the following error appears:

Error in `[.data.table`(data, , idname) : j (the 2nd argument inside [...]) is a single symbol but column name 'idname' is not found. Perhaps you intended DT[, ..idname] 

Do you have any idea what's going on? Sounds like a potential bug to me!

Thank you very much for your help and time!

[Rothj]

Hi Brantly,

I am experiencing the same issue and error with the ddid2 function. The error replicates with the lalonde data and your sample code. As the function did still work a year ago, I tried to run it with R version 4.0.5 and data.table version 1.13.6, but the same error appeared.
What seems strange to me is that I do not see where you are using the data.table package in your code, and the package is also not included in your namespace. However loading your package without data.table installed throws a warning as well.
My guess would be that the error originates from the follwoing section in the function bootiter (file bootstrap):

  if (qteparams$panel) {

        ids <- sample(unique(data[,idname]), length(unique(data[,idname])), replace=T)
        dta1 <- data[data[,tname] == t, ]
        dta2 <- data[data[,tname] == tmin1, ]
        n <- nrow(dta1)
        if (nrow(dta1) != nrow(dta2)) {
            warning("unexpected unbalanced panel")
        }
        boot.dta1 <- dta1[ids2rownum(ids, dta1, idname),]
        boot.dta2 <- dta2[ids2rownum(ids, dta2, idname),] ##c(ids, n+ids),]
        newids <- sample(seq(1, n), n)
        boot.dta1[,idname] <- newids
        boot.dta2[,idname] <- newids
        boot.dta <- rbind(boot.dta1, boot.dta2)
        if (!is.null(tmin2)) {
            dta3 <- data[data[,tname] == tmin2, ]
            boot.dta3 <- dta3[ids2rownum(ids, dta3, idname),]
            boot.dta3[,idname] <- newids
            boot.dta <- rbind(boot.dta, boot.dta3)
        }

Thank you very much for your help. Your package is very useful to me!

[bcallaway11]

Hi Jakob,

Sorry about this. I get the same error that you and @effedp are pointing out. I need to do a big update of the qte package --- hopefully this summer! Anyway, I think that we made some changes to BMisc for the did package that have caused this code to break. I'll try to come up with a real fix to this issue eventually, but something that should get it back going again is to use an older version of BMisc. In particular, I replaced the current version (1.4.4) with version 1.4.2, and that seems to get everything working again. Some code attached below.

Thanks,
Brant

remove.packages("BMisc") # might have to restart R after running this line
library(remotes)
install_version("BMisc", version="1.4.2")
packageVersion("BMisc")
#> [1] '1.4.2'

library(qte)
#> Warning in .recacheSubclasses(def@className, def, env): undefined subclass
#> "numericVector" of class "Mnumeric"; definition not updated
library(BMisc)

data(lalonde)

## Run the ddid2 method on the observational data with no covariates
d1 <- ddid2(re ~ treat, t=1978, tmin1=1975, tname="year",
            data=lalonde.psid.panel, idname="id", se=TRUE,
            probs=seq(0.05, 0.95, 0.05))
#> Warning in panelize.data(data, idname, tname, t, tmin1): dropping 2675
#> observations that are not in period: 1978, 1975, ...
#> Warning in panel.checks(qp): covariates appear to vary over time...
#>   only conditioning on first period covariates...
#>   this is recommended practice, but worth noting...
summary(d1)
#> 
#> Quantile Treatment Effect:
#>      
#> tau  QTE Std. Error
#> 0.05 10616.61      886.17
#> 0.1   5019.83      427.75
#> 0.15  2388.12      284.51
#> 0.2   1033.23      294.62
#> 0.25   485.23      364.90
#> 0.3    943.05      428.87
#> 0.35   931.45      590.50
#> 0.4    945.35      782.08
#> 0.45  1205.88      791.51
#> 0.5   1362.11      715.79
#> 0.55  1279.05      744.11
#> 0.6   1618.13      896.46
#> 0.65  1834.30      900.27
#> 0.7   1326.06      841.37
#> 0.75  1586.35      856.15
#> 0.8   1256.09      884.83
#> 0.85   723.10     1057.87
#> 0.9    251.36     1841.00
#> 0.95 -1509.92     2460.93
#> 
#> Average Treatment Effect:    2326.51
#>   Std. Error:        606.84

[Rothj]

Hi Brantly,

thank you very much for replying so fast! Changing BMisc to 1.4.2 fixed the issue for me as well, thank you!

Best,
Jakob

[PaoloZacchia]

Hello Brantly!

Thanks for the great method and package!

I am also experiencing this issue, but I cannot downgrade the package as I am working on a restricted version of R (I cannot update/install/reinstall packages on the fly). So, I decided to manually code up the bootstrap on my own, by doing block bootstrap on 'idname'. Any more advice while a new version of the package is in the making?

My best regards!

Paolo