Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | "Nick Cox" <n.j.cox@durham.ac.uk> |
To | <statalist@hsphsun2.harvard.edu> |
Subject | RE: st: RE: RE: estimation with a time trend. |
Date | Mon, 5 Jul 2010 17:12:27 +0100 |
I agree with Dr Buis. Nick n.j.cox@durham.ac.uk natasha agarwal On Mon, Jul 5, 2010 at 4:49 PM, Maarten buis <maartenbuis@yahoo.co.uk> wrote: > --- On Mon, 5/7/10, natasha agarwal wrote: >> I wanted to introduce a time trend in my estimation where >> my years go from 2001-2005. >> >> With time trend I mean: >> >> panel identifier year time trend >> 1 2001 1 >> 1 2002 2 >> 1 2003 3 >> 1 2004 4 >> 1 2005 5 >> 2 2001 1 >> 2 2002 2 >> >> The only way I knew how to do it was to use the egen >> -group- command.That is why I wanted to do the egen >> -group- command. > > I am still confused: What is "time trend" supposed to > measure that cannot be measured with "year"? The results with the time dummies showed that the estimated coefficients on the time dummies was increasing over time. I am pasting the results of the same. Fixed-effects (within) regression Number of obs = 46959 Group variable: number Number of groups = 12911 R-sq: within = 0.2300 Obs per group: min = 1 between = 0.4727 avg = 3.6 overall = 0.4656 max = 5 F(7,12910) = 626.65 corr(u_i, Xb) = 0.1507 Prob > F = 0.0000 (Std. Err. adjusted for 12911 clusters in number) Robust lnrval Coef. Std. Err. t P>t [95% Conf. Interval] lnk .2036338 .0107336 18.97 0.000 .1825942 .2246733 lnw .5677307 .0168696 33.65 0.000 .5346637 .6007977 lnvfdi .0217376 .0198568 1.09 0.274 -.0171848 .0606599 y14 -.3255905 .0171191 -19.02 0.000 -.3591466 -.2920345 y15 -.2313728 .0130272 -17.76 0.000 -.256908 -.2058375 y16 -.1464979 .0097877 -14.97 0.000 -.1656832 -.1273126 y17 -.1293191 .0081628 -15.84 0.000 -.1453195 -.1133188 y18 (dropped) _cons .3633778 .2559695 1.42 0.156 -.1383602 .8651158 sigma_u .95102062 sigma_e .55962159 rho .74279561 (fraction of variance due to u_i) So I decided to introduce one a time variable like a trend which is shown as before. Given your > earlier remarks you did not think that "the 0 outside the > range of your data" was a problem in your fixed effects > model. So what problem did you want to solve by creating > that "time trend" variable? > > Mind you, I don't think it is a bad idea to ensure that > 0 happens at a meaningful point in time, but it does not > seem to be the problem that you wanted to solve. We need > to make sure we understand what the problem is you want > to solve before we can help. > > Again, there is no need to use -egen group()-, and it is > dangerous as your year variable may be unequally spaced > (even if you think that it is not, real data has the > nasty property of always deviating from what you think > should be true in your data...). You could easily check > it before hand, but there is an easier and much saver > solution: just subtract a constant: > > *--------- begin example -------- > drop _all > input id year > 1 2001 > 1 2002 > 1 2003 > 1 2004 > 1 2005 > 2 2001 > 2 2002 > end > gen trend = year - 2000 > list > *--------- end example ---------- > (For more on examples I sent to the Statalist see: > http://www.maartenbuis.nl/example_faq ) > > To repeat: do not use -egen group()- for this purpose, > it is dangerous. It will work in this example, but it > can just too easily backfire in real data. > * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/