Running regression including fixed effects and clustered standard errors

[Samuel Pfaff]

Hello everyone,

i have panel data on a monthly basis over a time span of 14 years for 26 cities. Simply speaking, each monthly observation contains information about a respective city, the date of observation (e.g. "2014m1"), the temperature (ab_temp), as well as the Google Search Volume (DSVI_city) of that city in the given month.

I want to regress DSVI_city on ab_temp. I am told to:

1) Include year-month fixed effects in the regression
2) Cluster standard errors by city and year-month

I tried the following code:

. xtset city_id date

Panel variable: city_id (unbalanced)
 Time variable: date, 2004m1 to 2017m12, but with gaps
         Delta: 1 month

. xtreg DSVI_city ab_temp, fe vce(cluster city_id)

Fixed-effects (within) regression               Number of obs     =      3,588
Group variable: city_id                         Number of groups  =         26

R-squared:                                      Obs per group:
     Within  = 0.0004                                         min =         42
     Between = 0.0061                                         avg =      138.0
     Overall = 0.0004                                         max =        168

                                                F(1,25)           =       3.54
corr(u_i, Xb) = 0.0010                          Prob > F          =     0.0716

                               (Std. err. adjusted for 26 clusters in city_id)
------------------------------------------------------------------------------
             |               Robust
   DSVI_city | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
     ab_temp |   .0038789   .0020617     1.88   0.072    -.0003673    .0081252
       _cons |  -.0020442   .0005277    -3.87   0.001    -.0031311   -.0009573
-------------+----------------------------------------------------------------
     sigma_u |   .0168631
     sigma_e |  .59288721
         rho |  .00080831   (fraction of variance due to u_i)
------------------------------------------------------------------------------

So far, I think that I only clustered the standard errors by city and not by city and year-month.
My line of reasoning was that I thought it wouldn't make sense to cluster the standard errors by city and year-month, as this would give me as many clusters as observations (as for each city, there is only one unique year-month observation)...


Given on the information provided, could anyone be so kind and provide me with some feedback whether I implemented the code correctly?
(I am still quite new to Stata and already went through many different posts and explanations, but I am not sure if I understood everything properly...)

Thanks a lot in advance for your time and support-it is highly appreciated!

Samuel

[Carlo Lazzaro]

Samuel:
some comments about your post:
1) your code is correct, but your regression model needs more predictors (the withnin R-sq is alarming low; the F test tells that your model is not more informative the the mean of the regresssand).
More substantively, you have a T>N panel dataset: hence -xtreg- is outperformed by -xtregar- or -xtgls-;
2) the advice you received about clustering your standard errors (SEs) also on -i.month_year- sounds obscure and I would skip it altogether.

[Rhys Williams]

Hi Sam,

To include year-fixed effects you would need to include "i.date":

 
 xtreg DSVI_city ab_temp i.date, fe vce(cluster city_id)

I agree with you that I don't think it would make sense to cluster by city and year-month but I could be wrong on this.

Best,
Rhys

[Joro Kolev]

You have not included year/month fixed effects. You have included city fixed effects.

There is also two way clustering, see this thread here
https://www.statalist.org/forums/for...ols-regression

[Joro Kolev]

I think the easiest thing for you to do, is what posters #11 and #12 suggest on https://www.statalist.org/forums/for...ols-regression

Something like

 
 reghdfe DSVI_city ab_temp, absorb(city_id date)  cluster(city_id date)

[Samuel Pfaff]

Dear Joro,

thanks a lot for very helpful post and sorry for my late response!

Please allow me one short follow-up question:

As I only have to include year-month fixed effects, I would use:

reghdfe DSVI_city ab_temp, absorb(date) cluster(city_id date)

Am I correct?

Best regards,

Samuel

[Joro Kolev]

Yes Samuel, this sounds correct.

This puts date fixed effects, and does two way clustering by date and city.

Dear Joro,

thanks a lot for very helpful post and sorry for my late response!

Please allow me one short follow-up question:

As I only have to include year-month fixed effects, I would use:

reghdfe DSVI_city ab_temp, absorb(date) cluster(city_id date)

Am I correct?

Best regards,

Samuel

[Samuel Pfaff]

perfect, thank you Joro!