2

I never did something with panel data before, and could use some help.

I have data of 173 crowdfunding projects, measured at four different time points, with %funded, if the project contains a video, and the amount of comments and updates.

I want to know, in particular, the effect of the increase in updates on %funded, and the project having a video or not.

OLS, random, fixed everything is kind of new for me, could someone point me in the right direction?

Thank you!

edit:

I have decided to use change in %funded as my DV, and change in number of updates, the goal amount (constant), whether it contains a video (0/1) or whether it is featured (0/1) as my IVs. I've done the hausman test, which recommends the random effects model. 

    Random-effects GLS regression                   Number of obs      =       519
    Group variable: project                         Number of groups   =       173

    R-sq:  within  = 0.1407                         Obs per group: min =         3
           between = 0.2628                                        avg =       3.0
           overall = 0.2149                                        max =         3

                                                    Wald chi2(6)       =    116.26
    corr(u_i, X)   = 0 (assumed)                    Prob > chi2        =    0.0000

    ------------------------------------------------------------------------------
       cfunded_w |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
    -------------+----------------------------------------------------------------
          goal_w |  -.0001171   .0000382    -3.06   0.002    -.0001921   -.0000422
      cupdates_w |   7.712477   1.019072     7.57   0.000     5.715134    9.709821
           video |    9.30539   2.987579     3.11   0.002     3.449842    15.16094
     staffpicked |   6.385104   6.595511     0.97   0.333     -6.54186    19.31207
                 |
                 |
           _cons |   9.231501   2.746575     3.36   0.001     3.848312    14.61469
    -------------+----------------------------------------------------------------
         sigma_u |  14.286375
         sigma_e |  18.899747
             rho |  .36362048   (fraction of variance due to u_i)
    ------------------------------------------------------------------------------

I want to know more about the effects for each different time period (out of three)? adding if time<2 for example gives an error: insufficient observations

could someone enlighten me or point me in the right direction?

Should I just make three different Time Dummies and put T2, T3 in the regression when estimating for T1?

Thank you!

tkdgh35
  • 21
  • 2

1 Answers1

1

I assume you have a set of 300 identical observations with attributes taken at different points in time. I recommend using panel data models to make use of the larger sample size.

The first question you need to ask yourself is whether the variable video is substantially varying over time? If yes, create a dummy variable for video=1 and you can employ a 'fixed effect' model that makes your results more robust. In effect, unobserved factors that were not considered in your regression and that seen as fixed over time are technically eliminated. Your estimate of the coefficient on video will be closer to the true (unobserved) coefficient.

However, I guess this may not apply in your case, where I assume that video fairly constant throughout the entire time. In such a case, a random effect model makes use of the larger sample and, although the coefficient on video will be possibly biased, the model is slightly more robust with regards to the variances of the estimates. Using dynamic panel data, GMM and lagged instruments (e.g. lagged comments as it may be correlated with video) is a step up for you to consider if you want to tackle the bias from unobserved factors on your estimates.

Moreover, if %funded variable is binary, it may complicate things. You are looking here for probit with random effects and possibly dynamic panel data methods that use lagged variables as instruments.

That's just some prime thoughts, but you can see it can get fairly complicated depending on how serious you are. If you are interested you definitely should start reading a standard textbook for econometrics and continue with graduate readings. Link provides a discussion on stackexchange with some recommendation.

Community
  • 1
Majte
  • 1,842
  • 13
  • 22