Variance of each single residual

[Federico Nutarelli]

Hi Statlist,

I am struggling into constructing the residuals' specific variance for the following model:

xtreg tasso_crescita_sales_prod L.log_sales L.dummy_2 L2.dummy_2 L3.dummy_2 mean_gr_rate_atc2 recalls_sales ageprodcat1 ageprodcat2 ageprodcat3 ageprodcat4 newmolfirm newmolmarket i.Year, fe vce(cluster idpr)

What I would like to do is to generate a variable consisting of the variance of regression residuals so that each residual has its own variance (not gen var = e(sigma_e) which generates a unique value). So, if we call the residuals AV_it (resulting from AV_it = y - yhat where yhat = predict yhat, xbu) , I would like to obtain Var(AV_it ). Afterwords my aim is to compute the average over N_t , where N_t is the number of units with available data in t of such variances. Something like:

Sum for i ranging from 1 to N_t of Var(AV_it ) over (N_t )²


Is it possible in STATA to do compute the variance of such residuals?

[Carlo Lazzaro]

Federico:
do you mean something along the following lines?

use "http://www.stata-press.com/data/r15/nlswork.dta"
xtreg ln_wage age
predict e, e
quetly sum e
g var_e=(e-r(mean))^2/(r(N)-1)
list e var_e in 1/10
. list e var_e in 1/10

     +----------------------+
     |         e      var_e |
     |----------------------|
  1. | -.4115784   5.94e-06 |
  2. | -.8527393   .0000255 |
  3. | -.3099484   3.37e-06 |
  4. | -.1382197   6.70e-07 |
  5. | -.1786144   1.12e-06 |
     |----------------------|
  6. | -.2140789   1.61e-06 |
  7. |  .4826497   8.17e-06 |
  8. |  .5032557   8.88e-06 |
  9. |  .3161016   3.50e-06 |
 10. |  .4728792   7.84e-06 |
     +----------------------+

.

[Federico Nutarelli]

Thank you very much for the reply Carlo,

Ok then. I see that the only way is to construct it by hand. The question was also meant to open a quick discussion about residuals of fixed effects. If I had to construct confidence intervals, is it safer to take the prediction given by predict, e or the predictions given by predict, ue? As far as I have understood from the help, if I do something like:

predict yhat, xbu
gen av_t = tasso_crescita_sales - yhat

I am taking "e" since I am ruling out the individual heterogeneity. But am still quite undecided on which one are the "proper residuals" to construct confidence intervals.

Furthermore, I have individuals constituted by products and Years. when doing r(N) I am taking the bulk of all individuals in all years. Is it correct or something like by(Year) or by(individual) should be done?


Thank you again!

[Federico Nutarelli]

Better, to quote "Econometric Analysis of Cross Section and Panel Datahttps://jrvargas.files.wordpress.com › 2011/01 › wooldridge_j-_2002_eco..." (pag. 271-272) what I would like to obtain are the (u_hat_it)^2 and not the (sigma_u)