Announcement

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

  • Should RESET Test determine which model is more appropriate?

    Hello everybody,

    I am trying to estimate the relationship between vote preferences and COVID-19 vaccination rate among 3107 counties (fips) over 480 days. I have built two models and I am uncertain which should be chosen and whether this decision should be led by the RESET tests results. Two critical points:
    1. RESET Test results are quite different (H0 is rejected in the first and not rejected in the second);
    2. In both models, I try to account for cross-sectional dependence (by using Driscoll-Kraay SE in the first model and clustering by US state and week in the second model)
    Concerning the variables: qdate denotes quarters, mdate denotes months and rooted_partisanship is a categorical variable based on vote preferences.

    Code:
    * First model results
    ivreghdfe series_complete_pop_pct i.rooted_partisanship##i.qdate ///
    L.(total_cases_rate total_deaths_rate sqtotal_cases_rate) , a(mdate fips i.state_ID#i.mdate) ///
    dkraay(2) partial(i.rooted_partisanship i.qdate)
    
    Estimates efficient for homoskedasticity only
    Statistics robust to heteroskedasticity and clustering on date
    and kernel-robust to common correlated disturbances (Driscoll-Kraay)
      kernel=Bartlett; bandwidth=2
      time variable (t):  date
      group variable (i): num_fips
    
    Number of clusters (date) =        484                Number of obs =  1500912
                                                          F( 13,   483) = 42791.46
                                                          Prob > F      =   0.0000
    Total (centered) SS     =  32377794.18                Centered R2   =   0.1147
    Total (uncentered) SS   =  32378046.71                Uncentered R2 =   0.1147
    Residual SS             =  28662941.98                Root MSE      =    4.376
    
    -------------------------------------------------------------------------------------------
                              |               Robust
      series_complete_pop_pct | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
    --------------------------+----------------------------------------------------------------
    rooted_partisanship#qdate |
                   swing#245  |  -2.139201   .2081888   -10.28   0.000    -2.548268   -1.730133
                   swing#246  |  -4.726556    .061988   -76.25   0.000    -4.848355   -4.604756
                   swing#247  |  -6.212122   .0574571  -108.12   0.000    -6.325019   -6.099225
                   swing#248  |  -6.843741   .0372809  -183.57   0.000    -6.916993   -6.770488
                   swing#249  |  -6.958978   .0450317  -154.54   0.000     -7.04746   -6.870495
                safe_rep#245  |  -4.589379   .3073205   -14.93   0.000    -5.193229   -3.985529
                safe_rep#246  |  -8.476939   .1058513   -80.08   0.000    -8.684925   -8.268953
                safe_rep#247  |  -10.87777    .105965  -102.65   0.000    -11.08598   -10.66956
                safe_rep#248  |   -12.0281   .0762365  -157.77   0.000     -12.1779    -11.8783
                safe_rep#249  |  -12.40196   .0656108  -189.02   0.000    -12.53088   -12.27304
                              |
             total_cases_rate |
                          L1. |   .0636378   .0028855    22.05   0.000      .057968    .0693075
                              |
            total_deaths_rate |
                          L1. |  -1.561511   .0700126   -22.30   0.000    -1.699078   -1.423945
                              |
           sqtotal_cases_rate |
                          L1. |  -.0000606   4.16e-06   -14.58   0.000    -.0000687   -.0000524
    -------------------------------------------------------------------------------------------
    
    * Results from RESET Test
    
    . test y_h_2
    
     ( 1)  y_h_2 = 0
    
           F(  1,   483) =  589.01
                Prob > F =    0.0000
    
    *  Second model results
    . asdoc reghdfe series_complete_pop_pct i.rooted_partisanship##i.qdate ///
    L.(total_cases_rate total_deaths_rate sqtotal_cases_rate), ///
    a(mdate fips i.state_ID#i.mdate) cluster(i.state_ID i.wdate) replace
    
    HDFE Linear regression                            Number of obs   =  1,500,912
    Absorbing 3 HDFE groups                           F(  13,     49) =          .
    Statistics robust to heteroskedasticity           Prob > F        =          .
                                                      R-squared       =     0.9545
                                                      Adj R-squared   =     0.9544
    Number of clusters (state_ID) =         50        Within R-sq.    =     0.1148
    Number of clusters (wdate)   =         70         Root MSE        =     4.3764
    
                                         (Std. err. adjusted for 50 clusters in state_ID wdate)
    -------------------------------------------------------------------------------------------
                              |               Robust
      series_complete_pop_pct | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
    --------------------------+----------------------------------------------------------------
          rooted_partisanship |
                       swing  |          0  (omitted)
                    safe_rep  |          0  (omitted)
                              |
                        qdate |
                         245  |          0  (omitted)
                         246  |          0  (omitted)
                         247  |          0  (omitted)
                         248  |          0  (omitted)
                         249  |          0  (omitted)
                              |
    rooted_partisanship#qdate |
                   swing#245  |  -2.167068   .7982988    -2.71   0.009     -3.77131   -.5628264
                   swing#246  |  -4.732485   1.176687    -4.02   0.000    -7.097127   -2.367844
                   swing#247  |  -6.227344   1.106175    -5.63   0.000    -8.450285   -4.004403
                   swing#248  |  -6.828386   1.205607    -5.66   0.000    -9.251145   -4.405628
                   swing#249  |  -6.977102   1.141394    -6.11   0.000    -9.270819   -4.683385
                safe_rep#245  |  -4.589702    .974014    -4.71   0.000    -6.547056   -2.632347
                safe_rep#246  |   -8.47663   1.351906    -6.27   0.000    -11.19339   -5.759874
                safe_rep#247  |  -10.88409   1.265858    -8.60   0.000    -13.42793   -8.340253
                safe_rep#248  |  -12.02892    1.54244    -7.80   0.000    -15.12857   -8.929275
                safe_rep#249  |  -12.40258   1.540502    -8.05   0.000    -15.49834   -9.306829
                              |
             total_cases_rate |
                          L1. |   .0636085   .0108333     5.87   0.000     .0418382    .0853789
                              |
            total_deaths_rate |
                          L1. |  -1.561054   .3374245    -4.63   0.000    -2.239133   -.8829736
                              |
           sqtotal_cases_rate |
                          L1. |  -.0000605   .0000135    -4.49   0.000    -.0000877   -.0000334
                              |
                        _cons |   34.84693    1.41406    24.64   0.000     32.00527    37.68859
    -------------------------------------------------------------------------------------------
    
    * RESET Test results
    . test y_h_2 
     ( 1)  y_h_2 = 0
    
           F(  1,    49) =    2.26
                Prob > F =    0.1391
    Thank you in advance for your time and your help!
Working...
X