Double-cluster pooled OLS

[Canny Lee]

Good day to all. Is there any researcher know how to interpret the double-clustering pooled OLS? My thesis is about the firm performance and political connection. I'm using roa roe as firm performance and I have adopted the cluster2 by Petersen, M. A. (2009). Here is my attached file for double clustering pooled OLS.

[Joro Kolev]

You interpret OLS coefficients the same way, regardless of how you cluster your standard errors.

[Canny Lee]

Good day to sir, would you mind to explain more on this? I'm a new STATA user. I know how to run the data, but i'm not good in interpret the data...

[Carlo Lazzaro]

Canny:
as you can see from the following toy-example, the sample estimates for -mpg- and -_cons- do not vary, regardless clustering or not the standard errors:

. sysuse auto.dta
(1978 Automobile Data)

. regress price mpg

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(1, 72)        =     20.26
       Model |   139449474         1   139449474   Prob > F        =    0.0000
    Residual |   495615923        72  6883554.48   R-squared       =    0.2196
-------------+----------------------------------   Adj R-squared   =    0.2087
       Total |   635065396        73  8699525.97   Root MSE        =    2623.7

------------------------------------------------------------------------------
       price |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         mpg |  -238.8943   53.07669    -4.50   0.000    -344.7008   -133.0879
       _cons |   11253.06   1170.813     9.61   0.000     8919.088    13587.03
------------------------------------------------------------------------------

. regress price mpg, vce(cluster foreign)

Linear regression                               Number of obs     =         74
                                                F(0, 1)           =          .
                                                Prob > F          =          .
                                                R-squared         =     0.2196
                                                Root MSE          =     2623.7

                                (Std. Err. adjusted for 2 clusters in foreign)
------------------------------------------------------------------------------
             |               Robust
       price |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         mpg |  -238.8943   57.46835    -4.16   0.150     -969.099    491.3103
       _cons |   11253.06   595.4638    18.90   0.034     3686.976    18819.15
------------------------------------------------------------------------------

.

[Canny Lee]

Reply to Joro Kolev, sir may I ask, I just interpret with the OLS coefficients and standard error in between the dependent and independent variables?

[Canny Lee]

Is it that i need to explain about the t-statics? Do you have any reference or note in interpreting the double-cluster pooled ols?

[Joro Kolev]

As Carlo showed, what I said was a bit weak: It is not only that the interpretation of OLS estimates is the same no matter how you cluster (which is anyways true), but your parameter estimates are numerically the same (which is a bit stronger statement).

Your standard errors and t-statistics will be different depending on how you cluster. By the way, you did not show us your actual Stata estimation command and what Stata returned, so we need to assume that you know what "double-clustering pooled OLS" means, and that you have done this correctly. (What did you use to do double clustering, user written -reghdfe- or -ivreg2-? It is best for you just to show what you typed at Stata, and what Stata returned to you, rather than showing your formatted table of output. )

So in short you interpret your estimates as you would do for any OLS, but you just say that your standard errors and (t-statistics) are taking into account two way clustering.

And looking at your table, I think that you did not do OLS. In your table you are saying that you control for Year and Industry fixed effects--this is two way fixed effects model, you could say OLS, but saying two way fixed effects is better. So I am guessing what you did was something like this

reghdfe y x, absorb(industry year) cluster(industry year)

Reply to Joro Kolev, sir may I ask, I just interpret with the OLS coefficients and standard error in between the dependent and independent variables?

[Canny Lee]

Reply to Joro Kolev,
I'm doing the double cluster command in this way which is cluster2 roa1_w polcon1_w lnsize_w lnage_w mtb_w leverage_w turnover_w lnbsize_w bindep_w big4 dual y2015-y2019 d2-d14, fcluster(firm) tcluster(year)
I'm using the cluster2 command by Petersen, M. A. (2009).
From my lecturer, she explained that the double cluster have adjusted the standard error for both firm and year. The double cluster POLS is better than Fixed effect model and random effect model for firm-level data. Is it the statement is correct?

[Hong Il Yoo]

It is not correct as a general statement. Note that you can estimate fixed effects and random effects models with two-way clustered standard errors. You may be interested in checking out -ssc install vcemway- which is intended to provide a one-stop solution for multiway clustering in Stata. You can find the background paper for the command [here].

[Sally Ahmed]

Hello I am trying to run a clustered pooled regression by year and industry but I have an error in the command I don't know what is the right command:

reg AQ ESGCombinedScore SIZE_W GROWTH_W RETURNONASSETS_W MTB_W RDINT_W LEV_W CYCLE_W WGI  if year>2001 & year<2019, cluster (year IndustryX)

[Carlo Lazzaro]

Sally:
-regress- doe not support double-clustering.
That said, if you have both N and T dimensions, why not considering an -xt- suite commands as your first choice?
However, assuming that your regressand is continuous and that -fe- is the way to go with your dataset, please note that -xtreg,fe- does not support double-clustering either. Hence, you should switch to the community-contributed module -reghdfe- (that, in turn, has no cousin for the -re- specification).

[Hong Il Yoo]

Sally:
-regress- doe not support double-clustering.
That said, if you have both N and T dimensions, why not considering an -xt- suite commands as your first choice?
However, assuming that your regressand is continuous and that -fe- is the way to go with your dataset, please note that -xtreg,fe- does not support double-clustering either. Hence, you should switch to the community-contributed module -reghdfe- (that, in turn, has no cousin for the -re- specification).

My package -vcemway- allows you to apply -regress-, -xtreg, fe- and -xtreg, re- with multiway clustering. You can download it by typing -ssc install vcemway-. It has an intuitive syntax diagram:

vcemway regress y x, cluster(id1 id2)
vcemway xtreg y x, fe cluster(id1 id2)
vcemway xtreg y x, re cluster(id1 id2) nonest

and the like. The background paper is available at this link to the Stata Journal: [Link].