How do I run a Ttest in Stata when I have multiple groups but I only want to compare specific groups? (More Information Below)

[Michael Hensley]

I am trying to do a basic Ttest. There 6 groups based on gender, race and ethnicity. I want to run a simple ttest between group 1 and 5, but I am not sure the best way to do it. From my understanding the group variable can only take two values, and I have seen people ignoring a group based on if commands, but I have 6 groups. I don't want to have to create a new pairwise variable based on each two combinations.

[Rich Goldstein]

you don't really supply enough information but, if your group is all in one variable, you want something like the following:

ttest var1 if inlist(group,1,5), by(group)

clearly you should replace "var1" with whatever quantitative variable you want to test; you may also need to rename what I have called "group"; please read the FAQ (note that the above code assumes both variables are numeric; a data example using -dataex- would have reduced the need for assumptions/guesses)

[Joseph Coveney]

I want to run a simple ttest between group 1 and 5, but I am not sure the best way to do it.

Two ways are shown below.

.ÿ
.ÿversionÿ16.1

.ÿ
.ÿclearÿ*

.ÿ
.ÿsetÿseedÿ`=strreverse("1578925")'

.ÿ
.ÿquietlyÿsetÿobsÿ30

.ÿgenerateÿbyteÿgrpÿ=ÿmod(_n,ÿ6)ÿ+ÿ1

.ÿ
.ÿgenerateÿdoubleÿoutÿ=ÿrnormal()

.ÿ
.ÿ*
.ÿ*ÿBeginÿhere
.ÿ*
.ÿ//ÿt-testÿ(useÿifÿknownÿheteroscedastic)
.ÿttestÿoutÿifÿinlist(grp,ÿ1,ÿ5),ÿby(grp)ÿwelch

Two-sampleÿtÿtestÿwithÿunequalÿvariances
------------------------------------------------------------------------------
ÿÿÿGroupÿ|ÿÿÿÿÿObsÿÿÿÿÿÿÿÿMeanÿÿÿÿStd.ÿErr.ÿÿÿStd.ÿDev.ÿÿÿ[95%ÿConf.ÿInterval]
---------+--------------------------------------------------------------------
ÿÿÿÿÿÿÿ1ÿ|ÿÿÿÿÿÿÿ5ÿÿÿ-.2536269ÿÿÿÿ.6935084ÿÿÿÿ1.550732ÿÿÿ-2.179115ÿÿÿÿ1.671861
ÿÿÿÿÿÿÿ5ÿ|ÿÿÿÿÿÿÿ5ÿÿÿ-.2595319ÿÿÿÿ.3805594ÿÿÿÿ.8509566ÿÿÿ-1.316134ÿÿÿÿ.7970703
---------+--------------------------------------------------------------------
combinedÿ|ÿÿÿÿÿÿ10ÿÿÿ-.2565794ÿÿÿÿ.3729116ÿÿÿÿÿ1.17925ÿÿÿ-1.100164ÿÿÿÿ.5870052
---------+--------------------------------------------------------------------
ÿÿÿÿdiffÿ|ÿÿÿÿÿÿÿÿÿÿÿÿÿ.005905ÿÿÿÿ.7910621ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-1.848549ÿÿÿÿ1.860359
------------------------------------------------------------------------------
ÿÿÿÿdiffÿ=ÿmean(1)ÿ-ÿmean(5)ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿtÿ=ÿÿÿ0.0075
Ho:ÿdiffÿ=ÿ0ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿWelch'sÿdegreesÿofÿfreedomÿ=ÿÿ7.31305

ÿÿÿÿHa:ÿdiffÿ<ÿ0ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿHa:ÿdiffÿ!=ÿ0ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿHa:ÿdiffÿ>ÿ0
ÿPr(Tÿ<ÿt)ÿ=ÿ0.5029ÿÿÿÿÿÿÿÿÿPr(|T|ÿ>ÿ|t|)ÿ=ÿ0.9942ÿÿÿÿÿÿÿÿÿÿPr(Tÿ>ÿt)ÿ=ÿ0.4971

.ÿ
.ÿ//ÿPreferredÿotherwise
.ÿregressÿoutÿi.grp

ÿÿÿÿÿÿSourceÿ|ÿÿÿÿÿÿÿSSÿÿÿÿÿÿÿÿÿÿÿdfÿÿÿÿÿÿÿMSÿÿÿÿÿÿNumberÿofÿobsÿÿÿ=ÿÿÿÿÿÿÿÿ30
-------------+----------------------------------ÿÿÿF(5,ÿ24)ÿÿÿÿÿÿÿÿ=ÿÿÿÿÿÿ2.36
ÿÿÿÿÿÿÿModelÿ|ÿÿ12.9967343ÿÿÿÿÿÿÿÿÿ5ÿÿ2.59934685ÿÿÿProbÿ>ÿFÿÿÿÿÿÿÿÿ=ÿÿÿÿ0.0709
ÿÿÿÿResidualÿ|ÿÿ26.4532529ÿÿÿÿÿÿÿÿ24ÿÿ1.10221887ÿÿÿR-squaredÿÿÿÿÿÿÿ=ÿÿÿÿ0.3294
-------------+----------------------------------ÿÿÿAdjÿR-squaredÿÿÿ=ÿÿÿÿ0.1898
ÿÿÿÿÿÿÿTotalÿ|ÿÿ39.4499871ÿÿÿÿÿÿÿÿ29ÿÿ1.36034438ÿÿÿRootÿMSEÿÿÿÿÿÿÿÿ=ÿÿÿÿ1.0499

------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿoutÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿtÿÿÿÿP>|t|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
-------------+----------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿgrpÿ|
ÿÿÿÿÿÿÿÿÿÿ2ÿÿ|ÿÿ-.6343209ÿÿÿ.6639936ÿÿÿÿ-0.96ÿÿÿ0.349ÿÿÿÿ-2.004736ÿÿÿÿ.7360946
ÿÿÿÿÿÿÿÿÿÿ3ÿÿ|ÿÿÿÿ1.08155ÿÿÿ.6639936ÿÿÿÿÿ1.63ÿÿÿ0.116ÿÿÿÿ-.2888657ÿÿÿÿ2.451965
ÿÿÿÿÿÿÿÿÿÿ4ÿÿ|ÿÿÿ1.231445ÿÿÿ.6639936ÿÿÿÿÿ1.85ÿÿÿ0.076ÿÿÿÿ-.1389705ÿÿÿÿÿ2.60186
ÿÿÿÿÿÿÿÿÿÿ5ÿÿ|ÿÿÿ-.005905ÿÿÿ.6639936ÿÿÿÿ-0.01ÿÿÿ0.993ÿÿÿÿÿ-1.37632ÿÿÿÿ1.364511
ÿÿÿÿÿÿÿÿÿÿ6ÿÿ|ÿÿÿ.0439541ÿÿÿ.6639936ÿÿÿÿÿ0.07ÿÿÿ0.948ÿÿÿÿ-1.326461ÿÿÿÿÿ1.41437
ÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿ_consÿ|ÿÿ-.2536269ÿÿÿ.4695144ÿÿÿÿ-0.54ÿÿÿ0.594ÿÿÿÿ-1.222657ÿÿÿÿ.7154032
------------------------------------------------------------------------------

.ÿtestÿ5.grp

ÿ(ÿ1)ÿÿ5.grpÿ=ÿ0

ÿÿÿÿÿÿÿF(ÿÿ1,ÿÿÿÿ24)ÿ=ÿÿÿÿ0.00
ÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿFÿ=ÿÿÿÿ0.9930

.ÿ
.ÿ//ÿor
.ÿ
.ÿregressÿoutÿibn.grp,ÿnoconstant

ÿÿÿÿÿÿSourceÿ|ÿÿÿÿÿÿÿSSÿÿÿÿÿÿÿÿÿÿÿdfÿÿÿÿÿÿÿMSÿÿÿÿÿÿNumberÿofÿobsÿÿÿ=ÿÿÿÿÿÿÿÿ30
-------------+----------------------------------ÿÿÿF(6,ÿ24)ÿÿÿÿÿÿÿÿ=ÿÿÿÿÿÿ1.97
ÿÿÿÿÿÿÿModelÿ|ÿÿ13.0284092ÿÿÿÿÿÿÿÿÿ6ÿÿ2.17140154ÿÿÿProbÿ>ÿFÿÿÿÿÿÿÿÿ=ÿÿÿÿ0.1101
ÿÿÿÿResidualÿ|ÿÿ26.4532529ÿÿÿÿÿÿÿÿ24ÿÿ1.10221887ÿÿÿR-squaredÿÿÿÿÿÿÿ=ÿÿÿÿ0.3300
-------------+----------------------------------ÿÿÿAdjÿR-squaredÿÿÿ=ÿÿÿÿ0.1625
ÿÿÿÿÿÿÿTotalÿ|ÿÿ39.4816621ÿÿÿÿÿÿÿÿ30ÿÿÿ1.3160554ÿÿÿRootÿMSEÿÿÿÿÿÿÿÿ=ÿÿÿÿ1.0499

------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿoutÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿtÿÿÿÿP>|t|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
-------------+----------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿgrpÿ|
ÿÿÿÿÿÿÿÿÿÿ1ÿÿ|ÿÿ-.2536269ÿÿÿ.4695144ÿÿÿÿ-0.54ÿÿÿ0.594ÿÿÿÿ-1.222657ÿÿÿÿ.7154032
ÿÿÿÿÿÿÿÿÿÿ2ÿÿ|ÿÿ-.8879478ÿÿÿ.4695144ÿÿÿÿ-1.89ÿÿÿ0.071ÿÿÿÿ-1.856978ÿÿÿÿ.0810823
ÿÿÿÿÿÿÿÿÿÿ3ÿÿ|ÿÿÿ.8279229ÿÿÿ.4695144ÿÿÿÿÿ1.76ÿÿÿ0.091ÿÿÿÿ-.1411072ÿÿÿÿ1.796953
ÿÿÿÿÿÿÿÿÿÿ4ÿÿ|ÿÿÿ.9778181ÿÿÿ.4695144ÿÿÿÿÿ2.08ÿÿÿ0.048ÿÿÿÿÿÿ.008788ÿÿÿÿ1.946848
ÿÿÿÿÿÿÿÿÿÿ5ÿÿ|ÿÿ-.2595319ÿÿÿ.4695144ÿÿÿÿ-0.55ÿÿÿ0.586ÿÿÿÿ-1.228562ÿÿÿÿ.7094982
ÿÿÿÿÿÿÿÿÿÿ6ÿÿ|ÿÿ-.2096728ÿÿÿ.4695144ÿÿÿÿ-0.45ÿÿÿ0.659ÿÿÿÿ-1.178703ÿÿÿÿ.7593573
------------------------------------------------------------------------------

.ÿtestÿ1.grpÿ=ÿ5.grp

ÿ(ÿ1)ÿÿ1bn.grpÿ-ÿ5.grpÿ=ÿ0

ÿÿÿÿÿÿÿF(ÿÿ1,ÿÿÿÿ24)ÿ=ÿÿÿÿ0.00
ÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿFÿ=ÿÿÿÿ0.9930

.ÿ
.ÿexit

endÿofÿdo-file


.

[Bruce Weaver]

Hello Michael Hensley. If it is sensible and defensible to assume homogeneity of variance, then I would carry out a planed contrast that uses the error term from an omnibus ANOVA. Here's an example using the dataset Joseph generated in #3.

clear *
set seed `=strreverse("1578925")'
quietly set obs 30
generate byte grp = mod(_n, 6) + 1
generate double out = rnormal()
tabstat out, statistics( count mean sd var ) by(grp)
quietly anova out grp
* Planned contrast of Group 1 vs Group 5
contrast {grp 1 0 0 0 -1 0}, effects nowald

Output from -contrast-:

. * Planned contrast of Group 1 vs Group 5
. contrast {grp 1 0 0 0 -1 0}, effects nowald

Contrasts of marginal linear predictions

Margins      : asbalanced

------------------------------------------------------------------------------
             |   Contrast   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         grp |
        (1)  |    .005905   .6639936     0.01   0.993    -1.364511     1.37632
------------------------------------------------------------------------------

[Michael Hensley]

you don't really supply enough information but, if your group is all in one variable, you want something like the following:

ttest var1 if inlist(group,1,5), by(group)

clearly you should replace "var1" with whatever quantitative variable you want to test; you may also need to rename what I have called "group"; please read the FAQ (note that the above code assumes both variables are numeric; a data example using -dataex- would have reduced the need for assumptions/guesses)

Thanks! This works perfectly for me.

[Bruce Weaver]

I failed to point out that the result from my -contrast- in #4 is equivalent to Joseph Coveney's result from -test 5.grp- following -regress- in #3. That approach will have greater degrees of freedom than an ordinary unpaired t-test using only the two groups of interest, so should have quite a bit more power.


PS- Does anyone else get ÿ characters instead of blank spaces when copying Joseph's code and pasting it into a DO (or text) editor?

[Mike Lacy]

As soon as I highlight his code on the screen in my browser, that character (ASCII 255) shows up for me, and displays when I paste the material into a text editor.