Analysis panel data which includes country, industry and year fixed effects.

[Alex Potter]

Hi,

I am a fresher of STATA, and expect to analysis the panel data by simple regression, which includes the country, industry and year fixed effects.

by checking online, I think I can use "xtreg Y X1 X2 X2, fe" but it seems that it dose not include the country, industry and year fixed effects.

Could you please tell me how to do it? or give me some suggestions about this/

Thanks for your attention to this matter.

[Carlo Lazzaro]

Alex:
welcome to this forum.
As per Jeff Wooldridge' s memento (that you can find in many of his most recent threads), please note that the only fixed effect that you impose is the one included in your -xtset- code.
Assuming that you're interested in industry:

xtset industry year

Then you can add all the categorical variables you need for -country- and -year-. (but they are not fixed effect. If that were the case, our regressions will be full of fixed effects floating around).
Last but not least, you chose an -fe- estimator: as you're surely aware of, -the -fe- machinery wipes out time-invariant predictors.

[Phil Bromiley]

To add to Carlo's useful response, you might consider user-written reghdfe.

[Alex Potter]

Hi Carlo :
Many thanks for your advice.

But I still confuse about it. the following is part of my data,

Company_ID	year	Y	X1	X2	X3	X4	X5	X6	Industry	Country
130057	1994	0	0.008482	0	0.001069	1.752031361	0.693147181	Medium	MANUFACTURING	US
130057	1995	0	0.028821	0	0.015976	2.886118307	1.098612289	Medium	MANUFACTURING	US
130057	1996	0.693147181	0.031338	0	0.0079	2.002018961	1.386294361	Medium	MANUFACTURING	US
130057	1997	0	0.090801	0	0.00618	2.731637747	1.609437912	Medium	MANUFACTURING	US
130057	1998	0	0.140647	0	0	0.983591186	1.791759469	Small	MANUFACTURING	US
130057	1999	0	1.172691	0.803213	0	9.305220884	1.945910149	Small	MANUFACTURING	US
130057	2000	0	0.037084	0	0.003455	0.231832316	2.079441542	Small	MANUFACTURING	US
130057	2001	0	0.029063	0	0.002095	0.104206668	2.197224577	Small	MANUFACTURING	US
130057	2002	0	0.043314	0	0	0.247358121	2.302585093	Small	MANUFACTURING	US
130057	2003	0	0.071889	0	0.002701	0.577602327	2.397895273	Small	MANUFACTURING	US
130057	2004	0	0.016134	0	0.000234	0.094743734	2.48490665	Small	MANUFACTURING	US
130057	2005	0	0.038466	0	0.03692	0.313566113	2.564949357	Medium	MANUFACTURING	US
130057	2006	0	0.045024	0	0.015556		2.63905733	Medium	MANUFACTURING	US
130057	2007	0	0.0988	0	0.006043		2.708050201	Medium	MANUFACTURING	US
130057	2008	0	0.025239	0	0.000255		2.772588722	Medium	MANUFACTURING	US
130057	2009	0	0.018938	0	8.11E-05		2.833213344	Small	MANUFACTURING	US
130057	2010	0		0			2.890371758		MANUFACTURING	US
130062	1994	0	1.309941	0.185963	0.079193	0.001003551	0.693147181	Large	MINING	US
130062	1995	0	1.16531	0.293465	0.090752	0.001100877	1.098612289	Large	MINING	US
130062	1996	1.386294361	1.276097	0.170919	0.103396	0.000957189	1.386294361	Large	MINING	US
130062	1997	0.693147181	1.123869	0.282267	0.078176	0.000598709	1.609437912	Large	MINING	US
130062	1998	1.098612289	0.945815	0.38888	0.07801	0.000560875	1.791759469	Large	MINING	US
130062	1999	0	0.948716	0.368267	0.07909	0.000351205	1.945910149	Large	MINING	US
130062	2000	0.693147181	0.984882	0.353806	0.092756	0.000200387	2.079441542	Large	MINING	US
130062	2001	0	0.932501	0.360461	0.087381	0.000249935	2.197224577	Large	MINING	US
130062	2002	0	0.950388	0.329933	0.067601	0.001029873	2.302585093	Large	MINING	US
130062	2003	0.693147181	0.946514	0.313037	0.051774	0.001196948	2.397895273	Large	MINING	US
130062	2004	1.791759469	0.98034	0.307387	0.069378	0.001295922	2.48490665	Large	MINING	US
130062	2005	1.609437912	1.028134	0.295059	0.090987	0.001399736	2.564949357	Large	MINING	US
130062	2006	1.098612289	1.092924	0.285045	0.106118	0.000756856	2.63905733	Large	MINING	US
130062	2007	1.098612289	1.109753	0.421522	0.098712	0.000693418	2.708050201	Large	MINING	US
130062	2008	0.693147181	1.095104	0.448489	0.085159	0.000650948	2.772588722	Large	MINING	US
130062	2009	0	1.069983	0.386304	0.042982	0.000700464	2.833213344	Large	MINING	US
130062	2010	0	1.160512	0.335925	0.044204		2.890371758	Large	MINING	US
130088	1994	0	0.362183	0.09485	0.046347	0	0.693147181	Large	RETAIL TRADE	US
130088	1995	0	0.352445	0.158885	0.058568	0	1.098612289	Large	RETAIL TRADE	US

(each Company_ID represents a company, Y is the dependent variable, X1-X6 are the explanatory variables, specifically, X6 is a dummy variable.
it covers 11 industries and 20 countries, not the just US)


How could I control the industry and country fixed effects of this panel data and then analysis this by a linear regression?

I have read some post or log of panel data and fixed effects, but all of them describe a part of the process, and it makes me even more confused. For example, a video said I should use "xtset Company_ID year" at the beginning, it strange.

Could you please give me some suggestions about this specific question?

Many thanks in advance.

[Carlo Lazzaro]

Alex:
1) the -xt-commands suite requires that you -xtset- your data beforehand: this is not strange, as it allows Stata to learn that your data are panel;
2) for what you have in mind, I think you could follow two approaches:

xtset Company_ID year
xtreg  xtreg <depvar>  <other predictors> i.year i.industry <controls>, fe

As expected, all time-invariant predictors will be wiped out by the -fe- machinery.

The second approach, as per Phil's wise advice, implies the community-contributed command -reghdfe-:

reghdfe <depvar> <other predictors> <controls>, abs(Company_ID year industry)

[Alex Potter]

Hi Carlo,

Many thanks for your reply and kind help. May I ask one more questions please?

1)In the Code:
xtset Company_ID year xtreg xtreg <depvar> <other predictors> i.year i.industry <controls>, fe Why you write "i.year" rather than "i.country". it seems the "i.country" is meaning of country-fixed effect?

so as I expect to use the industry and country-fixed effect, should I write the code like below,
(apology that I write the year-fixed effect in the earliest post, I should not do that)

xtset Company_ID year
xtreg xtreg <depvar> <other predictors> i.country i.industry <controls>, fe

Many thanks in advance.

[Carlo Lazzaro]

Alex:
I would recommend you to add -i.year- to your -xtreg,fe- code, as this predictor will inform you about how time contributes to explain within panel variations when adjusted for the other predictors.
Again please note the the only fixed effect y
that you will get with your code is the one related to compact (as per -xtset- coding).

[Alex Potter]

Hi Carlo,

Many thanks for your advice.

I use the following code,

xtreg LN_APPLN PPETA LEV CAPEXTA RDTA Q SHORT_LIQ LONG_LIQ LN_TA LN_AGE LN_MV Employees_1 i.Country i.industry i.year , fe

However, I get
note: LONG_LIQ omitted because of collinearity
note: 2.Country omitted because of collinearity
note: 3.Country omitted because of collinearity
note: 4.Country omitted because of collinearity
note: 5.Country omitted because of collinearity
note: 6.Country omitted because of collinearity
note: 7.Country omitted because of collinearity
note: 8.Country omitted because of collinearity
note: 9.Country omitted because of collinearity
note: 10.Country omitted because of collinearity
note: 11.Country omitted because of collinearity
note: 13.Country omitted because of collinearity
note: 14.Country omitted because of collinearity
note: 15.Country omitted because of collinearity
note: 16.Country omitted because of collinearity
note: 17.Country omitted because of collinearity
note: 18.Country omitted because of collinearity
note: 19.Country omitted because of collinearity
note: 22.Country omitted because of collinearity
note: 23.Country omitted because of collinearity
note: 24.Country omitted because of collinearity
note: 25.Country omitted because of collinearity
note: 27.Country omitted because of collinearity
note: 28.Country omitted because of collinearity
note: 29.Country omitted because of collinearity
note: 30.Country omitted because of collinearity
note: 32.Country omitted because of collinearity
note: 33.Country omitted because of collinearity
note: 35.Country omitted because of collinearity
note: 36.Country omitted because of collinearity
note: 37.Country omitted because of collinearity
note: 38.Country omitted because of collinearity
note: 39.Country omitted because of collinearity
note: 40.Country omitted because of collinearity
note: 41.Country omitted because of collinearity
note: 42.Country omitted because of collinearity
note: 2.industry omitted because of collinearity
note: 3.industry omitted because of collinearity
note: 4.industry omitted because of collinearity
note: 5.industry omitted because of collinearity
note: 6.industry omitted because of collinearity
note: 7.industry omitted because of collinearity
note: 8.industry omitted because of collinearity
note: 9.industry omitted because of collinearity
note: 10.industry omitted because of collinearity

Do you know how to fix it?
Many thanks for your help.

[Carlo Lazzaro]

Alex:
there's nothing to fix here: as expected, the -fe- estimator got ride of time-invariant predictors.
If you're intetested in estimating the coefficients of the omitted variable, you should switch to a differenti estimator: see -xtreg,re- or -xthybrid-.

[Alex Potter]

Hi Carlo,

Many thanks for your advice. But I did not understand it. What should I do if I want to control the country, industry and year fixed effect?

Besides, is the code

xtset company_ID year,yearly
xtreg LN_APPLN PPETA LEV CAPEXTA RDTA Q SHORT_LIQ LONG_LIQ LN_TA LN_AGE LN_MV i.Country i.Industry i.year, fe

do the same thing as the Code:

reg LN_APPLN PPETA LEV CAPEXTA RDTA Q SHORT_LIQ LONG_LIQ LN_TA LN_AGE LN_MV i.Country i.Industry i.year

I can get a result without collinearity by the second code（ reg ）, however, I am not sure whether the second approach can control the country, industry and year fixed effect.

Many thanks for your help, and sorry for my poor understanding ability

[Carlo Lazzaro]

Alex:
if you want to obtain fixed effects for country, industry and year, you should switch to the community-contributed command -reghdfe-.
Please note that investigating the fixed effect (ie, the heterogeneity) in a panel data set is not the same thing as controlling for other predictors.
With your -regress- code you could obtain (for some regressors)the same results as per -xtreg,fe- if you include -i.company_ID- among the predictors.
I think you should consult any decent textbook on panel data econometrics in order to get yourself familiar with the building blocks of this theoretically demanding research field.

[Shiwani Varal]

Hi everyone,
I found the above discussion very helpful for my analysis, and am working towards similar issue. I am currently using Stata 16.1. I am also running a Panel data analysis for a sample period of 2011-2018 for American firms. In my research, if I want to control for election years as my dependent variables may be influenced by political factors. So, could someone advise me on what would be a good way to go about this? So far I am thinking of creating a dummy variable to represent the election years, but I am not sure.

Best,
Shiwani

[Carlo Lazzaro]

Shiwany:
-i.election_year- is the way to go.

[Shiwani Varal]

Hi Carlo,

Thank you so much for helping me! Just to be a little more clear. I will also be controlling for industry and year. So, I will first need to construct a variable called "election_year", and then run the following:

regress D_var Indep_var control_var i.election_year i.year i.industry

is that the way?

Best,
Shiwani

[Carlo Lazzaro]

Shiwany:
if you want to estimate so many fixed effects, I would recommend you to switch to the community-contributed command -reghdfe-.