Announcement

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

  • Controlling for time specific effect

    I have a dataset with 800 crowdfunded technological projects (each of which crowdfunded in a particular year), plus a nunmber of variables per project. The time window is 2009-2018. One of the reviewers of the paper asked us to control for time-specific effects. We used robust regression by Verardi & Croux (2009). Is it reasonable/feasible?

  • #2
    Francesco:
    welcome to this forum.
    I guess the reviewer is asking for a new categorical predictor in the right-hand side of your regression equation:
    Code:
    i.year
    See -help fvvarlist- on how to create categorical variables and interactions in an effective way.
    After the regression, you can test the joint statistical signifcance of -i.year- typing:
    Code:
    testparm(i.year)
    Kind regards,
    Carlo
    (Stata 16.0 SE)

    Comment


    • #3
      Thanks a lot, Carlo!

      Comment


      • #4
        Dear, I implemented the robust regression (as per Verardi and Croux, 2009) with MM-estimators. Then, I included the time dummies (g1 to g9) and run the regression with an efficiency level of 95% and some thresholds to do not account for outliers. What I get is the following:

        Code:
        . mmregress DelayVar IncompetenceVar Fraud FundingCanc Freq NewbackersPerc NumberFAQ2 NumberComments2 NumberUpdates2 Daysoffunding Funded Raised Goal Averagepledgeperbacker2 if(Fraud<3 & DelayVar<6 & IncompetenceVar<3), dummies( g1 g2 g3 g4 g5 g6 g7 g8 g9) eff(.95)
        The total number of p-subsets to check for each iteration is 103
        
        DelayVar       Coef.   Std. Err.      t    P>t     [95% Conf. Interval]
        
        IncompetenceVar     .289974   .0728719     3.98   0.000     .1469402    .4330079
        Fraud    .2481416   .0768818     3.23   0.001      .097237    .3990462
        FundingCanc    .6379088   .3024372     2.11   0.035     .0442812    1.231536
        Freq    .0302155   .0390517     0.77   0.439    -.0464358    .1068667
        NewbackersPerc   -.0071803   .0022608    -3.18   0.002    -.0116178   -.0027427
        NumberFAQ2    .0084887   .0062017     1.37   0.171    -.0036841    .0206615
        NumberComments2     .000102   .0000552     1.85   0.065    -6.37e-06    .0002103
        NumberUpdates2    .0135703   .0039442     3.44   0.001     .0058286    .0213119
        Daysoffunding    .0086065   .0037836     2.27   0.023       .00118     .016033
        Funded    4.98e-06   1.21e-06     4.11   0.000     2.61e-06    7.36e-06
        Raised   -3.09e-08   2.01e-07    -0.15   0.878    -4.26e-07    3.64e-07
        Goal    9.94e-07   1.13e-06     0.88   0.381    -1.23e-06    3.22e-06
        Averagepledgeperbacker2    .0001908    .000146     1.31   0.192    -.0000957    .0004773
        g1   -1.026183    .539233    -1.90   0.057    -2.084596    .0322306
        g2           0  (omitted)
        g3    .2501159   .5503033     0.45   0.650    -.8300262    1.330258
        g4    .1508545   .5313165     0.28   0.777    -.8920201    1.193729
        g5    .0262855   .5289113     0.05   0.960    -1.011868    1.064439
        g6   -.0373713   .5262491    -0.07   0.943      -1.0703     .995557
        g7    .1890573   .5285254     0.36   0.721    -.8483389    1.226454
        g8    .2552847   .5321202     0.48   0.632    -.7891676    1.299737
        g9     .143642   .5313393     0.27   0.787    -.8992774    1.186561
        _cons    1.575538   .5639777     2.79   0.005     .4685559    2.682521
        
        Scale parameter=  .9928156
        No year dummy turns out significant, but what worries me is that g2 is omitted for collinearity issues. What can I do in this case?
        Also, is it enough for a reviewer asking to put time dummies in the regression? Do I need to implement some tests?

        Thanks for your help,
        Francesco

        Comment


        • #5
          Francesco:
          the community-contributed command -mmregress- omitted -g2- to shelter your results from the so called dummy trap (https://en.wikipedia.org/wiki/Dummy_...le_(statistics)).
          It is difficul to say whether you revised analysis is in live with rewier's expectations/requests (there's an Italian saying that goes "one head, one world").
          Please use CODE delimiters instead of screenshots in yiour future posts (the FAQ explain why). Thanks.
          Kind regards,
          Carlo
          (Stata 16.0 SE)

          Comment

          Working...
          X