Comparing means across groups using svy commands and esttab

[Sofie Johansen]

Dear STATA users

I hope someone can help me. I am having a really hard time making a - what I thought would be a quite simple table.

I want to make a table over background factors related to placement breakdown looking similar to this one below:

	Probability of placement breakdown	Significance
Gender
Boys (n=545)	0.21	0.720
Girls (n=478)	0.23	0.720
Age
0-5 (n=56)	0.07	0.000
6-12 (n=318)	0.22	0.958
13-17 (n=587)	0.23	0.320
18-22 (n=52)	0.23	0.741

Placement breakdown is a dummyvariable called placementbreakdown 1 = placement breakdown 0 = no placement breakdown
Gender is also a dummy variable called gender 1 = boy 0 = girl.
And age consists of multiple dummy variables called age_0_5, age_6_12, age_13_17 and age_18_22.

If we start with the gender variable, I might be able to figure out the rest on my own.

My data is survey data collected in clusteres, so we use the svy-command to compensate for that.

What I have tried to do is:

. svy: mean breakdown, over(gender)
(running mean on estimation sample)

Survey: Mean estimation

Number of strata = 1 Number of obs = 1,007
Number of PSUs = 156 Population size = 1,007
Design df = 155

--------------------------------------------------------------------
| Linearized
| Mean std. err. [95% conf. interval]
-------------------+------------------------------------------------
c.breakdown@gender |
0 | .2252632 .024941 .175995 .2745313
1 | .2142857 .0237364 .1673971 .2611744
--------------------------------------------------------------------

. test [email protected][email protected]

Adjusted Wald test

( 1) [email protected] - [email protected] = 0

F( 1, 155) = 0.13
Prob > F = 0.7200

. estadd scalar p_diff = r(p)

added scalar:
e(p_diff) = .71999581

. esttab ., cells("b") stats(p_diff) nostar noabbrev nonumber eqlabels(none) collabels(Probability of placement breakdown) nomtitle

-------------------------
Probability of placement breakdown
-------------------------
[email protected] .2252632
[email protected] .2142857
-------------------------
p_diff .7199958
-------------------------

I have tried several ways but this is the closest i get to the table above. It is not very interpretable especially since there is no labels on the variables.

Are there any suggestions on how I could do this better/more interpretable?

Thank you in advance,

Sofie.

[Andrew Musau]

See #2 here https://www.statalist.org/forums/for...tab-using-over

[Sofie Johansen]

Thank you Andrew! That was very helpfull. Still not exactly the result I was looking for, yet though.

This is my preliminary result:

. svy: regress breakdown ibn.gender, nocons
(running regress on estimation sample)

Survey: Linear regression

Number of strata = 1 Number of obs = 1,007
Number of PSUs = 156 Population size = 1,007
Design df = 155
F(2, 154) = 66.84
Prob > F = 0.0000
R-squared = 0.2196

------------------------------------------------------------------------------
| Linearized
breakdown | Coefficient std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
gender |
Girl | .2252632 .024941 9.03 0.000 .175995 .2745313
Boy | .2142857 .0237364 9.03 0.000 .1673971 .2611744
------------------------------------------------------------------------------

. mat list e(b)

e(b)[1,2]
0. 1.
gender gender
y1 .22526316 .21428571

. test 0.gender=1.gender

Adjusted Wald test

( 1) 0bn.gender - 1.gender = 0

F( 1, 155) = 0.13
Prob > F = 0.7200

. estadd scalar p_diff = r(p)

added scalar:
e(p_diff) = .71999581

. esttab, cells("b") wide se nostar label stats(p_diff) collabels(Probability of placement breakdown) noabbrev nomtitle nonumber

---------------------------------
Probability of placement breakdown
---------------------------------
Girl .2252632
Boy .2142857
---------------------------------
p_diff .7199958
---------------------------------

Do you know if there is somehow I can get the variable label (in this example: "Gender of the child") automatically included in the tabel (in a collumn above the variable values or so)?

Something like this:

	Probability of placement breakdown
Gender of the child
Girl	.2252632
Boy	.2142857
p_diff	.7199958

In this example it is not really necessary (since the label values are quite informative). But I have several variables that need some kind of explanation to be interpreted e.g. yes/no-questions. You need some informations about the question to understand what "yes" and "no" stands for. I found codes where you can type a label manually, but I am looking for a way to do this automatically since I have to do it with more than 120 + variables.

[Andrew Musau]

Add -label- as an option in esttab. Otherwise show an example where the labels fail to appear. But note that your variables need to be labeled in the first place.

[Sofie Johansen]

Thank you and sorry. Thought I made an example with my last post, but maybe it more clear with anouther example. When I use the command (including the label option), I get this table:

. esttab, cells("b") wide se nostarlabelstats(p_diff) collabels(Probability of placement breakdown) noabbrev nomtitle nonumber

---------------------------------
Probability of placement breakdown
---------------------------------
No .2252632
Yes .2142857
---------------------------------
p_diff .7199958
---------------------------------

What I would like is a table, that also includes the variable label. Something like this (so it is clear what yes and no stands for) :

	Probability of placement breakdown
School age
No	.2252632
Yes	.2142857
p_diff	.7199958

I know I can redefine the labels values and give them a more informative name, but again since I work with over 120+ variables, I would prefer it there is an easier way.

Hope someone can help me.

[Andrew Musau]

Sorry, my mind was on value labels when I read your question in #3. By example, I mean a reproducible example in the sense of FAQ Advice #12. As long as the dependent variable in svy:regress is labeled, here is an automatic way (highlighted)

webuse nhanes2f
svyset psuid [pweight=finalwgt], strata(stratid)
svy: regress zinc ibn.race, nocons
esttab, not nostar noobs label nonumbers mlab("Mean", lhs("`:var lab `e(depvar)''"))

Res.:

. esttab, not nostar noobs label nonumbers mlab("Mean", lhs("`:var lab `e(depvar)''"))

---------------------------------
serum zinc (mcg/dL)          Mean
---------------------------------
White                       87.50
Black                       85.09
Other                       83.57
---------------------------------

[Sofie Johansen]

Thank you so much for your quick respond. Now I get this:

. esttab, cells("b") wide se nostar label stats(p_diff) collabels(Probability of placement breakdown) noabbrev nomtitle nonumber mlab("Mean", lhs("`:var lab `e(depvar)''"))

---------------------------------
breakdown Mean
Probability of placement breakdown
---------------------------------
No .219888
Yes .21843
---------------------------------
p_diff .9683106
---------------------------------

But, actually it is the label of the independent variable I want included in the table (in your example race).

I tried to just change 'depvar' to 'indepvar', but that was not possible:

esttab, cells("b") wide se nostar label stats(p_diff) collabels(Probability of placement breakdown) noabbrev nomtitle nonumber mlab("Mean", lhs("`:var lab `e(indepvars)''"))
nothing found where name expected

---------------------------------
Mean
Probability of placement breakdown
---------------------------------
No .219888
Yes .21843
---------------------------------
p_diff .9683106
---------------------------------

Any suggestions?

[Andrew Musau]

In the regression, there are two variables. It seems that you want the label of the categorical variable, e.g., gender in your case. From the stored results, the only way I see is to parse the command line to extract the variable name. From my example:

webuse nhanes2f
svyset psuid [pweight=finalwgt], strata(stratid)
svy: regress zinc ibn.race, nocons
esttab, not nostar noobs label nonumbers mlab("Mean", lhs("`:var lab `=ustrregexra("`e(command)'",  "(.*ibn\.)(.*)\,\s+\w+", "$2")''"))

Res.:

. esttab, not nostar noobs label nonumbers mlab("Mean", lhs("`:var lab `=ustrregexra("`e(command)'",  "(.*ibn\.)(.*)\,\s+\w+"
> , "$2")''"))

---------------------------------
1=white, 2=black, ~r         Mean
---------------------------------
White                       87.50
Black                       85.09
Other                       83.57
---------------------------------