1
$\begingroup$

Let us say this is an output of a model I ran in Stata, where int_retis a continuous variable and time1-time17 are time dummy variables. Reference is time0. I notice that the coefficients are significant at alpha=0.05 up til time 5 after which it stops being significant. I also notice that the coefficients for time decrease in magnitude from time1 to time5.

I am wondering if it is allowable for me to test if there was a "decreasing trend" between time1 and time 5 using the test command in Stata.

Example: test time1=time2=time3=time4=time5=0 or

  test time6=time7=time8=time9=time10=time11=time2=time13=0

I know that you have to enter all the dummy variables (except reference) in a model, but can you test the significance just a subset of the dummy variables (meaning, NOT all dummy variables) to determine "trend?"

Thanks

         |             Linearized
 int_ret |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
   time1 |   3.803034   1.569459     3.24   0.001     1.690201    8.557009
   time2 |   2.983938   1.145295     2.85   0.005     1.403566    6.343762
   time3 |   2.963121   1.412266     2.28   0.023      1.16145    7.559592
   time4 |   2.709894   1.137635     2.37   0.018     1.187634    6.183323
   time5 |   2.319515    1.00564     1.94   0.053     .9894511    5.437508
   time6 |   3.240125   2.107322     1.81   0.071     .9026434    11.63074
   time7 |   1.474064   .6102438     0.94   0.349     .6534528    3.325207
   time8 |   1.348993   .6327848     0.64   0.524     .5366498    3.391004
   time9 |   1.211208   .4933164     0.47   0.638     .5440385    2.696546
  time10 |   .7559254   .4048442    -0.52   0.602     .2638907    2.165379
  time11 |   2.224805   1.034081     1.72   0.086     .8925419    5.545686
  time12 |   1.676299   .6908775     1.25   0.211     .7457992    3.767742
  time13 |   1.233598   .6585384     0.39   0.694     .4321069    3.521729
| cite | improve this question | | | | |
$\endgroup$
0
$\begingroup$

Why not just include a time trend term in the model corresponding to the time dummies and segment the dataset to only include the five time steps you are concerned with? Applying a t-test to the coefficient on the trend term will tell you whether there is a statistically significant negative trend.

| cite | improve this answer | | | | |
$\endgroup$
  • $\begingroup$ OK, I think I get what you are saying, so pls correct me if I am wrong... First, generate a variable TIMETREND where TIMETREND=1 for time0, TIMETREND=2 for time1, ...,TIMETREND=14 for time13 (TIMETREND goes from 1 to 14 b/c there are 14 time points) then, regress int_ret TIMETREND if (time>=0 &time<=5) The coefficient and sign on TIMETREND will tell me if there was significant decreasing trends b/w time 0 & time 5. Did I read your response correctly? Thanks again $\endgroup$ – Marquis de Carabas Feb 15 '14 at 3:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.