Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • xtivreg2 drops most of observations

    Hi all
    I am tackling the effect of mobile money adoption (a binary variable) on firm-level energy intensity. Following the literature, I suspect that my model may be contaminated by endogeneity, so I run a fixed effect 2SLS model using xtivreg2. But what confuses me is that xtivreg2 runs on only 22 observations out of 800 under xtset firm_id year. When I read some posts here, I found a similar issue, and the proposed solution is to xtset only year. I followed the suggestion, and it worked perfectly for the whole sample, but I cant get the intuition behind xtivreg2 working with panel definition upon year variable only without accounting for firm time invariation? Should I include a firm fixed effect in the model in that case ?
    Here is the code that I used
    Code:
     xtivreg2 EE20152 finance_measure R_D  human_capital labprod2015 exporter  privatedomestic_ownership ln_age  firm_sizeWBES industry_type  (MM = new_var_2 ),  fe endog (MM)


  • #2
    have many singletons?

    Comment


    • #3
      649 observations are not used.

      Comment


      • #4
        did it tell you have 649 singletons?

        Comment


        • #5
          yes

          Comment


          • #6
            Then you've got variables that are saturating 649 observations. Need to stare at the data and find the source of that problem. May be the fixed effect (you've got one observation for the cross section identifier).

            look at iveghdfe/ivreg2 (search for singleton in help ivreg2).

            Comment


            • #7
              Thanks a lot George for you reply. I have tried both ivreg2 and ivreghdfe and it ran on my whole sample without any dropped singelton observations. Actually I am quite confused about the concept of singelton and why it has been dropped with xtivreg2. Another question, Can I rely on ivreg2 or ivreghdfe, I have concerns that the results might not be that reliable because it doesnt account for panel dimensions of my data. Thank you so much for your help

              Comment

              Working...
              X