Stratified t-test in Stata

[Iche Anya]

1. Please can any one explain the procedure for conducting stratified t-test in Stata (assuming intervention arm and strata to be the two factors in a cluster-level analysis of a stratified cluster randomised trial)?

[Joseph Coveney]

You might want to post your question in the General forum. This Sandbox forum is for testing the post's display.

More to your question: couldn't you use mixed with its degrees-of-freedom corrections to get the stratified t statistic and its p-value in the context of a cluster-randomized trial? Something like that below?

.ÿversionÿ14.2

.ÿ
.ÿclearÿ*

.ÿsetÿmoreÿoff

.ÿsetÿseedÿ1374242

.ÿ
.ÿ//ÿClusters
.ÿquietlyÿsetÿobsÿ50

.ÿgenerateÿbyteÿcidÿ=ÿ_n

.ÿlabelÿvariableÿcidÿ"ClusterÿID"

.ÿgenerateÿdoubleÿcid_uÿ=ÿrnormal()

.ÿ
.ÿ//ÿCluster-randomizedÿinterventionÿarm
.ÿgenerateÿbyteÿtrtÿ=ÿmod(_n,ÿ2)

.ÿlabelÿvariableÿtrtÿTreatment

.ÿlabelÿdefineÿTrtÿ0ÿControlÿ1ÿExperimental

.ÿlabelÿvaluesÿtrtÿTrt

.ÿ
.ÿ//ÿIndividuals
.ÿquietlyÿexpandÿ50

.ÿbysortÿcid:ÿgenerateÿdoubleÿpid_uÿ=ÿrnormal()

.ÿgenerateÿintÿpidÿ=ÿ_n

.ÿlabelÿvariableÿpidÿ"ParticipantÿID"

.ÿ
.ÿ//ÿStrata
.ÿgenerateÿbyteÿsidÿ=ÿmod(_n+4,ÿ5)

.ÿlabelÿvariableÿsidÿ"Stratum"

.ÿlabelÿdefineÿStrataÿ0ÿ"Veryÿlow"ÿ1ÿLowÿ2ÿMediumÿ3ÿHighÿ4ÿ"Veryÿhigh"

.ÿlabelÿvaluesÿsidÿStrata

.ÿ
.ÿgenerateÿdoubleÿoutÿ=ÿcid_uÿ+ÿpid_uÿ+ÿsidÿ/ÿ25ÿ+ÿtrtÿ/ÿ2.5ÿ+ÿ0ÿ*ÿsidÿ*ÿtrt

.ÿlabelÿvariableÿoutÿOutcome

.ÿformatÿoutÿ%3.2f

.ÿ
.ÿsortÿcidÿpidÿsid

.ÿforeachÿvarÿofÿvarlistÿ*tÿ*dÿ{
ÿÿ2.ÿÿÿÿÿÿÿÿÿlocalÿvariable_labelÿ:ÿvariableÿlabelÿ`var'
ÿÿ3.ÿÿÿÿÿÿÿÿÿcharÿdefineÿ`var'[varname]ÿ`variable_label'
ÿÿ4.ÿ
.ÿÿÿÿÿÿÿÿÿlocalÿlabel_lengthÿ=ÿstrlen("`variable_label'")
ÿÿ5.ÿÿÿÿÿÿÿÿÿifÿ"`var'"ÿ==ÿ"out"ÿformatÿ`var'ÿ%`label_length'.2f
ÿÿ6.ÿÿÿÿÿÿÿÿÿelseÿifÿ!inlist("`var'",ÿ"trt",ÿ"sid")ÿformatÿ`var'ÿ%`label_length'.0f
ÿÿ7.ÿÿÿÿÿÿÿÿÿelseÿ{
ÿÿ8.ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ//ÿnoÿop
.ÿÿÿÿÿÿÿÿÿ}
ÿÿ9.ÿ}

.ÿlistÿcidÿtrtÿpidÿsidÿoutÿifÿinrange(_n,ÿ1,ÿ5)ÿ|ÿinrange(_n,ÿ`=_N-4',ÿ`=_N'),ÿ///
>ÿÿÿÿÿÿÿÿÿnoobsÿsepby(trt)ÿsubvarnameÿnocompress

ÿÿ+------------------------------------------------------------------+
ÿÿ|ÿClusterÿIDÿÿÿÿÿÿTreatmentÿÿÿParticipantÿIDÿÿÿÿÿStratumÿÿÿOutcomeÿ|
ÿÿ|------------------------------------------------------------------|
ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿExperimentalÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿVeryÿlowÿÿÿÿÿ-2.28ÿ|
ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿExperimentalÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿÿÿÿLowÿÿÿÿÿ-0.12ÿ|
ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿExperimentalÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ3ÿÿÿÿÿÿMediumÿÿÿÿÿÿ1.04ÿ|
ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿExperimentalÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ4ÿÿÿÿÿÿÿÿHighÿÿÿÿÿ-1.64ÿ|
ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿExperimentalÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ5ÿÿÿVeryÿhighÿÿÿÿÿ-0.81ÿ|
ÿÿ|------------------------------------------------------------------|
ÿÿ|ÿÿÿÿÿÿÿÿÿ50ÿÿÿÿÿÿÿÿControlÿÿÿÿÿÿÿÿÿÿÿÿÿ2496ÿÿÿÿVeryÿlowÿÿÿÿÿÿ2.47ÿ|
ÿÿ|ÿÿÿÿÿÿÿÿÿ50ÿÿÿÿÿÿÿÿControlÿÿÿÿÿÿÿÿÿÿÿÿÿ2497ÿÿÿÿÿÿÿÿÿLowÿÿÿÿÿÿ0.41ÿ|
ÿÿ|ÿÿÿÿÿÿÿÿÿ50ÿÿÿÿÿÿÿÿControlÿÿÿÿÿÿÿÿÿÿÿÿÿ2498ÿÿÿÿÿÿMediumÿÿÿÿÿÿ0.22ÿ|
ÿÿ|ÿÿÿÿÿÿÿÿÿ50ÿÿÿÿÿÿÿÿControlÿÿÿÿÿÿÿÿÿÿÿÿÿ2499ÿÿÿÿÿÿÿÿHighÿÿÿÿÿÿ0.24ÿ|
ÿÿ|ÿÿÿÿÿÿÿÿÿ50ÿÿÿÿÿÿÿÿControlÿÿÿÿÿÿÿÿÿÿÿÿÿ2500ÿÿÿVeryÿhighÿÿÿÿÿ-1.06ÿ|
ÿÿ+------------------------------------------------------------------+

.ÿ
.ÿ*
.ÿ*ÿBeginÿhere
.ÿ*
.ÿmixedÿoutÿi.trtÿi.sidÿ||ÿcid:ÿ,ÿremlÿdfmethod(kroger)ÿnolrtestÿnolog

Mixed-effectsÿREMLÿregressionÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿÿ=ÿÿÿÿÿÿ2,500
Groupÿvariable:ÿcidÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿgroupsÿÿ=ÿÿÿÿÿÿÿÿÿ50

ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿObsÿperÿgroup:
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿminÿ=ÿÿÿÿÿÿÿÿÿ50
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿavgÿ=ÿÿÿÿÿÿÿ50.0
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿmaxÿ=ÿÿÿÿÿÿÿÿÿ50
DFÿmethod:ÿKenward-RogerÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿDF:ÿÿÿÿÿÿÿÿÿÿÿminÿ=ÿÿÿÿÿÿ48.00
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿavgÿ=ÿÿÿ1,647.45
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿmaxÿ=ÿÿÿ2,446.00

ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿF(5,ÿÿÿ481.04)ÿÿÿÿ=ÿÿÿÿÿÿÿ5.05
Logÿrestricted-likelihoodÿ=ÿ-3651.2343ÿÿÿÿÿÿÿÿÿÿProbÿ>ÿFÿÿÿÿÿÿÿÿÿÿ=ÿÿÿÿÿ0.0002

-------------------------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿoutÿ|ÿÿÿÿÿÿCoef.ÿÿÿStd.ÿErr.ÿÿÿÿÿÿtÿÿÿÿP>|t|ÿÿÿÿÿ[95%ÿConf.ÿInterval]
--------------+----------------------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿtrtÿ|
Experimentalÿÿ|ÿÿÿ.5816661ÿÿÿ.2588292ÿÿÿÿÿ2.25ÿÿÿ0.029ÿÿÿÿÿ.0612551ÿÿÿÿ1.102077
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿsidÿ|
ÿÿÿÿÿÿÿÿÿLowÿÿ|ÿÿÿ.1032732ÿÿÿ.0633495ÿÿÿÿÿ1.63ÿÿÿ0.103ÿÿÿÿÿ-.020951ÿÿÿÿ.2274974
ÿÿÿÿÿÿMediumÿÿ|ÿÿÿ.1541544ÿÿÿ.0633495ÿÿÿÿÿ2.43ÿÿÿ0.015ÿÿÿÿÿ.0299302ÿÿÿÿ.2783786
ÿÿÿÿÿÿÿÿHighÿÿ|ÿÿÿ.2309975ÿÿÿ.0633495ÿÿÿÿÿ3.65ÿÿÿ0.000ÿÿÿÿÿ.1067733ÿÿÿÿ.3552217
ÿÿÿVeryÿhighÿÿ|ÿÿÿÿ.247529ÿÿÿ.0633495ÿÿÿÿÿ3.91ÿÿÿ0.000ÿÿÿÿÿ.1233049ÿÿÿÿ.3717532
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿ_consÿ|ÿÿÿ-.359458ÿÿÿÿ.187354ÿÿÿÿ-1.92ÿÿÿ0.060ÿÿÿÿ-.7352911ÿÿÿÿ.0163751
-------------------------------------------------------------------------------

------------------------------------------------------------------------------
ÿÿRandom-effectsÿParametersÿÿ|ÿÿÿEstimateÿÿÿStd.ÿErr.ÿÿÿÿÿ[95%ÿConf.ÿInterval]
-----------------------------+------------------------------------------------
cid:ÿIdentityÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(_cons)ÿ|ÿÿÿÿ.817341ÿÿÿ.1709359ÿÿÿÿÿÿ.5424827ÿÿÿÿ1.231461
-----------------------------+------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(Residual)ÿ|ÿÿÿ1.003289ÿÿÿ.0286888ÿÿÿÿÿÿ.9486069ÿÿÿÿ1.061124
------------------------------------------------------------------------------

.ÿ
.ÿexit

endÿofÿdo-file


.

[Iche Anya]

Thanks very much. I thought i was posting the question in the general forum. I will do that. Notwithstanding, `The text am consulting ("cluster randomized trials" by Hayes and Moulton 2009) recommended cluster-level analysis for trials with small number of clusters per arm (less than 15). With proportion as the primary outcome measure, this involves (1). a one-stage process of estimating the mean of cluster summaries/proportions in each arm and using the mean difference for significance testing with stratified t-test in Stata (2). Adjustment for covariates employs individual-level analysis with random effect logistic regression (for proportion). There is ". xi: regress outcome i. arm* i. stratum" command in the Stata output of an example in the text. Please, can such analysis be done with the point-and-click GUI In Stata?