Announcement

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

  • Stata output not reporting standard errors or p-values using PPMLHDFE command

    Hi everyone,

    I am busy running a regression using a PPML estimator with several fixed effects to estimate the impact of a civil unrest episode using a difference-in-differences estimator on export flows and decomposing these into the extensive (number of distinct products exported) and intensive (average value per product exported margins). As such, the difference-in-difference variable is an interaction between a dummy which takes the value of 1 if the observation is from an origin region that experienced a civil unrest episode (and 0 otherwise) and a dummy that takes the value of 1 if the observation is recorded during the unrest period (and 0 otherwise).The data is aggregated to the origin-differentiation level - destination - time level, where "differentiation level" simply refers to if the product is classified as differentiated or undifferentiated.

    I am trying to extend the analysis by estimating a triple differences specification to see if there was any difference in the impact on differentiated relative to undifferentiated products as follows:

    xi: ppmlhdfe numproducts DDD if t < 739, absorb(districttimeid districtproductid producttimeid destinationtimeid districtdestinationid) r cl(districtid destinationid)

    where:
    - numproducts refers to number of products
    - DDD is the triple differences variable (the difference in difference variable interacted with a dummy which takes a value of 1 if the product is differentiated)
    - districttimeid is origin x time fixed effects
    - districtproductid is origin x product fixed effects
    - producttimeid is group x time fixed effects ("group" referring to whether the product is differentiated versus undifferentiated)
    - destinationtimeid is destination x time fixed effects
    - districtdestinationid is origin x destination fixed effects

    and standard errors are clustered to the origin and destination level.

    However, the estimates I am getting do not include any standard errors or p-values (instead, the output just has dots where these values should be). Does anyone know why this might be? Have I included incorrect fixed effects, or am I clustering at the wrong level (issues go away when clustering at only the origin district level).


    I am using Stata 15, and my data looks as follows:

    Code:
    * Example generated by -dataex-. To install: ssc install dataex
    clear
    input float nproductsokdt str22 dof float DDifferentiated str4 d float t
      2 "Cape Town"    0 "ABW" 732
      0 "Cape Town"    . "ABW" 732
      0 "Cape Town"    . "ABW" 733
      2 "Cape Town"    0 "ABW" 734
      0 "Cape Town"    . "ABW" 734
      2 "Cape Town"    0 "ABW" 735
      0 "Cape Town"    . "ABW" 735
      1 "Cape Town"    0 "ABW" 736
      0 "Cape Town"    . "ABW" 736
      0 "Cape Town"    . "ABW" 737
      0 "Cape Town"    . "ABW" 738
      1 "Cape Town"    0 "ABW" 738
      0 "Cape Town"    . "ABW" 739
      2 "Cape Town"    0 "ABW" 739
      2 "Cape Town"    0 "ABW" 740
      0 "Cape Town"    . "ABW" 740
      0 "Cape Town"    . "ABW" 741
      0 "Cape Town"    . "ABW" 742
      0 "Cape Town"    . "ABW" 743
      0 "Durban"       . "AFG" 732
      0 "Durban"       . "AFG" 733
      0 "Durban"       . "AFG" 734
      0 "Durban"       . "AFG" 735
      0 "Durban"       . "AFG" 736
      0 "Durban"       . "AFG" 737
      0 "Durban"       . "AFG" 738
      0 "Durban"       . "AFG" 739
      0 "Durban"       . "AFG" 740
      0 "Durban"       . "AFG" 741
      1 "Durban"       0 "AFG" 742
      0 "Durban"       . "AFG" 743
     18 "Durban"       0 "AGO" 732
      9 "Cape Town"    0 "AGO" 732
     38 "Cape Town"    1 "AGO" 732
      1 "Richards Bay" 0 "AGO" 732
     74 "Durban"       1 "AGO" 732
      0 "Cape Town"    . "AGO" 732
      5 "Durban"       . "AGO" 732
    132 "Cape Town"    1 "AGO" 733
     11 "Cape Town"    . "AGO" 733
      0 "Richards Bay" . "AGO" 733
     22 "Durban"       0 "AGO" 733
     88 "Durban"       1 "AGO" 733
     21 "Cape Town"    0 "AGO" 733
      3 "Durban"       . "AGO" 733
      6 "Durban"       . "AGO" 734
     10 "Cape Town"    0 "AGO" 734
      1 "Richards Bay" 0 "AGO" 734
    143 "Durban"       1 "AGO" 734
      4 "Cape Town"    . "AGO" 734
     18 "Cape Town"    1 "AGO" 734
     36 "Durban"       0 "AGO" 734
     19 "Durban"       0 "AGO" 735
     13 "Cape Town"    0 "AGO" 735
      0 "Richards Bay" . "AGO" 735
     81 "Cape Town"    1 "AGO" 735
      4 "Durban"       . "AGO" 735
    156 "Durban"       1 "AGO" 735
      4 "Cape Town"    . "AGO" 735
      0 "Richards Bay" . "AGO" 736
      3 "Cape Town"    . "AGO" 736
     11 "Cape Town"    0 "AGO" 736
     56 "Cape Town"    1 "AGO" 736
     91 "Durban"       1 "AGO" 736
     21 "Durban"       0 "AGO" 736
      1 "Durban"       . "AGO" 736
     26 "Durban"       0 "AGO" 737
     14 "Cape Town"    0 "AGO" 737
      0 "Richards Bay" . "AGO" 737
      8 "Cape Town"    . "AGO" 737
     85 "Cape Town"    1 "AGO" 737
    150 "Durban"       1 "AGO" 737
      3 "Durban"       . "AGO" 737
     26 "Durban"       0 "AGO" 738
     61 "Cape Town"    1 "AGO" 738
      3 "Durban"       . "AGO" 738
      7 "Cape Town"    . "AGO" 738
     13 "Cape Town"    0 "AGO" 738
    121 "Durban"       1 "AGO" 738
      0 "Richards Bay" . "AGO" 738
    188 "Durban"       1 "AGO" 739
     30 "Cape Town"    0 "AGO" 739
     26 "Durban"       0 "AGO" 739
     14 "Cape Town"    . "AGO" 739
    246 "Cape Town"    1 "AGO" 739
      6 "Durban"       . "AGO" 739
      1 "Richards Bay" 0 "AGO" 739
      4 "Cape Town"    . "AGO" 740
     18 "Durban"       0 "AGO" 740
    153 "Durban"       1 "AGO" 740
      0 "Richards Bay" . "AGO" 740
     16 "Cape Town"    0 "AGO" 740
      6 "Durban"       . "AGO" 740
     50 "Cape Town"    1 "AGO" 740
      4 "Durban"       . "AGO" 741
      0 "Richards Bay" . "AGO" 741
    160 "Durban"       1 "AGO" 741
      6 "Cape Town"    . "AGO" 741
     31 "Durban"       0 "AGO" 741
     91 "Cape Town"    1 "AGO" 741
    end
    format %tm t

    Thanks!

  • #2
    Missing standard errors usually indicate insufficient degrees of freedom; either you have more parameters to be estimated than observations or with clustering, you have more parameters than the number of clusters. You do not show any actual output, so one cannot tell definitively whether this is the case based on what you have presented. Note that ppmlhdfe is from SSC, as you are asked to explain (FAQ Advice #12).

    Comment

    Working...
    X