Calculating the total of multiple regression estimates and standard errors, that are obtained from the same model

[Jeroen Sanderink]

Dear People,

Is there a command in Stata to calculate the sum/average of several coefficients, and standard errors?
The model I am analyzing is (sort of) the following:

The output of this regression model contains three values for T (d = 3). Is there a way to take the average of the coefficients and the standard deviations?
And is it possible to replace these averages with the regression estimates after that (for outreg2 purposes) ?

I hope I made myself clear enough, if there are any furthers questions I am happy to elucidate.
Thanks in advance,

Jeroen

[Steve Samuels]

Please read FAQ 12 (http://www.statalist.org/forums/help) and follow its instructions. It will be possible to answer your questions only if you show us the Stata code you use to fit the model and the results, copied and pasted from the output window or log, between CODE delimiters.

I don't understand what you mean by "replace these averages with the regression estimates after that", so illustrate by hand, if necessary (I don't use outreg2).

[Jeroen Sanderink]

Dear Steve,

To be more clear, I analyse the effect of coffeeshops on house prices.

set matsize 800
set more off
local coffeeshops coffeeshop1 coffeeshop2 coffeeshop3 //these represent observations (houses) that are located within 600 meter of an active coffeeshop (so the treatment group)
//the distance of the three nearest coffeeshops are determined, if a coffeeshop is outside the range of 600 meter or not active, this dummy (coffeeshop1, coffeeshop2 or coffeeshop3) is 0.

local Ai withinrange1 withinrange2 withinrange3 //indicate if the house is situated within a range of 600 meters of a (future) coffeeshop
local housecharacteristics LNHouseSize nRooms nFloors Pool i.HouseSubType OpenPorch i.ParkFacility i.MaintenanceIn i.MaintenanceEx Monument ///
CostsForSeller ConstrPer_1906_1930 ConstrPer_1931_1944 ConstrPer_1945_1959 ConstrPer_1960_1970 ConstrPer_1971_1980 ConstrPer_1981_1990 ///
ConstrPer_1991_2000 ConstrPer_after2000 //represents Xi in the formula above

regress LNPrice `coffeeshops' `housecharacteristics' i.Year i.Month

//i.Year and i.Month represents period fixed effects

The output of this regression is as follows:

				Number of obs =	646131
Source	SS	df	MS	F(106,646024) =	19218.03
				Prob > F =	0.0000
Model	169306.847	106	1597.23441	R-squared =	0.7592
Residual	53691	.864646024	.083111253	Adj R-squared =	0.7592
Total	222998	.711646130	.34512979	Root MSE = .	28829
LNPrice	Coef.	Std.	Err.	t	P>t	[95% Conf. Interval]
coffeeshop1	-.0004755	.0048719	-0.10	0.922	-.0100242	.0090733
coffeeshop2	.0297427	.0056136	5.30	0.000	.0187403	.0407452
coffeeshop3	.0424199	.007703	5.51	0.000	.0273223	.0575176
withinrange1	-.0650802	.0049458	-13.16	0.000	-.0747739	-.0553866
withinrange2	.0049671	.0060811	0.82	0.414	-.0069517	.016886
withinrange3	-.0784532	.007826	-10.02	0.000	-.0937919	-.0631145
LNHouseSize	.6006385	.0015604	384.94	0.000	.5975803	.6036968
nRooms	.0268163	.0003644	73.60	0.000	.0261021	.0275304
nFloors	.0002785	.0007278	0.38	0.702	-.0011479	.0017049
Pool	.1728571	.009129	18.93	0.000	.1549645	.1907498
Other Housing Characteristics	+60 variables
Period Fixed Effects	37 variables

What I like to now is what the average coefficient is of the dummies coffeeshop1 - coffeeshop3, and its average standard error. And also for the withinrange1 - withinrange3 dummies.

After that I use the following outreg2 code to make a nice word table. As you can see I used only coffeeshop1 and withinrange1. My second question is if I can replace the coffeeshop1 output variables (coefficient and standard error) with the average of the three variables, so that the table show the average, and not just one coffeeshop or all three coffeeshops.

outreg2 using Table3.doc, append ///
    ctitle(Housing, characteristics) dec(4) nocons noni nor2 noobs label ///
    keep(coffeeshop1 withinrange1 LNHouseSize nRooms nFloors Pool i.HouseSubType OpenPorch i.ParkFacility i.MaintenanceIn i.MaintenanceEx Monument ///
    CostsForSeller ConstrPer_1906_1930 ConstrPer_1931_1944 ConstrPer_1945_1959 ConstrPer_1960_1970 ConstrPer_1971_1980 ConstrPer_1981_1990 ///
    ConstrPer_1991_2000 ConstrPer_after2000) ///
    addtext(Housing Characteristics, Yes, Year fixed effects, Yes, PC4 fixed effects, No) ///
    addstat(Number of Observations, e(N), Adjusted R2, e(r2_a))

[Jeroen Sanderink]

The correct formula for the above regression output is the following:

Thanks in advance!

[Steve Samuels]

Thanks for the code, but the FAQ asks that you show results between CODE delimiters too; notice that the table you gave doesn't have monospace font so that entries don't line up. Here's how to get the averages. I don't know outreg2 so I can't help with that part of your question.

scalar av_b  =  (_b[coffeeshop1] +   _b[coffeeshop2] + _b[coffeeshop3])/3
scalar av_se = (_se[coffeeshop1] +_se[coffeeshop2] +_se[coffeeshop3])/3

[Jeroen Sanderink]

Thanks for your feedback! I will try something with outreg2, if I succeed I will post it here!

[Jeroen Sanderink]

I found an answer to my second question, how to change the coefficient and standard error that are obtained from the regression. This can be used to include the calculated averages in the outreg2 output table.

see below for the solution

//the below program change the coefficient of coffeeshop1 to the average of the three coffeeshops coefficients
program changeb, eclass
    tempname bmat
    matrix `bmat' = e(b)
    matrix `bmat'[1,1] = (_b[coffeeshop1] +   _b[coffeeshop2] + _b[coffeeshop3])/3 //_b[coffeeshop1] etc. are coefficient that are obtained from the regression (to show them type "matrix list e(b)")
    ereturn repost b = `bmat'
end

//the below program change the standard error of coffeeshop1 to the average of the three coffeeshop standard errors
program changese, eclass
    tempname semat
    matrix `semat' = e(V)
    matrix `semat'[1,1] = ((_se[coffeeshop1] +_se[coffeeshop2] +_se[coffeeshop3])/3)*((_se[coffeeshop1] +_se[coffeeshop2] +_se[coffeeshop3])/3) (to show them type "matrix list e(V)")
    ereturn repost V = `semat'
end

local coffeeshops coffeeshop1 coffeeshop2 coffeeshop3
local Ai withinrange1 withinrange2 withinrange3
regress LNPrice `coffeeshops' `Ai' i.Year i.Month

//after the regression you just have to recall the programs
changeb
changese

//
outreg2 using Table3.doc, replace title("Table 3 - Name of table") ///
    ctitle(Standard, DID) tdec(4) bdec(5) nocons noni nor2 noobs label ///
    keep(coffeeshop1 withinrange1) //coffeeshop1 is now the average of the three coffeeshops

You can write a similar program for other variables such as withinrange1 in this case

[Steve Samuels]

Very nice! You can simplify a bit by ereturning b and V in one program