unbalanced panel error when using sdid (the package by Daniel-Pailanir)

[Laurence Jones]

Hi, I am trying to run a staggered synthetic diff-in-diff framework, using the sdid package created by Daniel-Pailanir available at https://github.com/Daniel-Pailanir/sdid.

When I define my panel it is strongly balanced:

. xtset id year
panel variable: id (strongly balanced)
time variable: year, 1968 to 2003
delta: 1 unit

I then get the error:

. sdid total_hours_per_year id year treatment if total_hours_per_year != ., vce(placebo) method(sdid)
Panel is unbalanced.
r(451);

This is in contradiction to stata's earlier assessment. Any ideas what might be causing the issue? Many thanks and much appreciated.

[Daniel PV]

Hello Laurence Jones ! Yes, you have a balanced panel by construction, but you are actually missing values in your dependent variable. This means you do not have a balanced panel for that particular variable.

The methodology requires that you observe your dependent variables in every period.

[Laurence Jones]

Hello Daniel PV ! Many thanks for the response. I did suspect that at first, but hoped I could get around it with the ignore missing values addition "if total_hours_per_year != .". I take it this is not sufficient?

Will I need to redefine my panel after dropping any missing observations? What should my next step be?

Many thanks and much appreciated!

[Daniel PV]

One way to handle this is to estimate SDID over the units you observe in your entire data panel, sort of like counting and ensuring that only those "ids" that are not missing are included in your estimate. Another option, if you are only missing a few values, is perhaps to impute some values (this should be carefully considered).

[Loic Guerin]

Hi,

I am running in a similar issue.
I balanced my panel in a "simple" way:

*deleting missing values:
    reg $yit $tit
    capture gene sample2 =e(sample)
    drop if sample2 != 1
      
*Treatment group :
        capture drop streated
        gen streated = 0
        //replace streated=1 if oldght==2 & i_year>$date    
        //replace streated=1 if oldght==1 & i_year>$date  
        replace streated=1 if oldght==3 & i_year>$date            
  
* balancing the panel:
        count
        bysort id (i_year): gene nb = _N
            sum nb
            keep if nb==`r(max)'
            drop nb
        count
        xtset id i_year
        
*checking for missing values in yit and xit
        foreach v in id i_year $yit $tit{
            sum `v' if `v' == .    
}
        xtreg $yit $tit, fe
        
        sdid $yit id i_year streated if oldght!=1 & oldght!=2, method(sdid) vce(bootstrap) reps(50) covariates($tit)

A first dataset passed all those balancing program and tests. SDID runs normaly.
But a second dataset, with a change in level of data agregation do not let me use sdid, with the infamous

Panel is unbalanced.

, even though the

xtreg $yit $tit, fe

outputs:

Number of obs = 9,740
Number of groups = 974

Obs per group:
min = 10
avg = 10.0
max = 10

Suggesting a highly balanced panel

and

      
 foreach v in id i_year $yit $tit{
            sum `v' if `v' == .    
}

seems to suggest that there are 0 missing values for id, i_year, $yit and $tit.

As it should with a balanced panel.

Do you see any improvement possible in my code? Or any SDID functionality that I am not aware of ?

[Jared Greathouse]

If you can simulate a dataset that has this, I'll look at it. Thus far I can't tell what the issue is