diff --git a/Scripts/Analysis/DIF-ROSALI/dif_rosali.do b/Scripts/Analysis/DIF-ROSALI/dif_rosali.do deleted file mode 100644 index 80d267b..0000000 --- a/Scripts/Analysis/DIF-ROSALI/dif_rosali.do +++ /dev/null @@ -1,198 +0,0 @@ -*================================================================================================================================================= -* Date : 2024-01-23 -* Stata version : Stata 18 SE -* -* This program analyses simulated data accounting for DIF through a partial credit model -* -* ado-files needed : - pcm, rosali (version 5.5 October 25, 2023, available on gitea) -* -* -*================================================================================================================================================ -adopath+"/home/corentin/Documents/These/Recherche/ROSALI-SIM/Modules/rosali_custom" - - -local N = "100 200 300" - local ss = "1 2 3 4 5 6 7 8 9 10 11 1 13 14 15 16 17 18 19 20" - foreach s in `ss' { - foreach Nnn in `N' { - local Nn = `Nnn' - local path_data = "/home/corentin/Documents/These/Recherche/Simulations/Data/DIF/N`Nn'" - if (`s'<=4) { - local path_data = "/home/corentin/Documents/These/Recherche/Simulations/Data/NoDIF/N`Nn'" - } - local path_res = "/home/corentin/Documents/These/Recherche/Simulations/Analysis/ROSALI-DIF/N`Nn'" - local scenarios = "A B C D E F G" - foreach scen in `scenarios' { - clear - import delim "`path_data'/scenario_`s'`scen'_`Nn'.csv", encoding(ISO-8859-2) case(preserve) clear - rename TT tt - - if (`s'<=2) { - local nbitems=4 - } - else if (`s'<=4) { - local nbitems=7 - } - else if (`s'<=12) { - local nbitems=4 - } - else { - local nbitems=7 - } - -if (mod(`s',2)==0) { - local nbmoda=3 -} -else { - local nbmoda=1 -} - - if (`s'<=4) { - local nbdif=0 - } - else if (`s'<=8) { - local nbdif=1 - } - else if (`s'<=16) { - local nbdif=2 - } - else { - local nbdif=3 - } - * taillemat = Maximum J*M cases pour les items par et J*M cases pour les dif par + J cases pour les DIF detect + nbdif cases pour dif réel - local taillemat=2*`nbitems'*`nbmoda'+`nbitems'+`nbdif'+2 - local colna="" - forvalues i=1/`nbitems' { - forvalues z=1/`nbmoda' { - local colna = "`colna'"+"item`i'_`z' " - local colna = "`colna'"+"dif_`i'_`z' " - } - } - forvalues i=1/`nbitems' { - local colna = "`colna'"+"dif_detect_`i' " - } - - forvalues i=1/`nbdif' { - local colna = "`colna'"+"real_dif_`i' " - } -local colna = "`colna'" + "beta " + "se_beta" - - - mat outmat = J(1000,`taillemat',.) - mat colnames outmat= `colna' - di "Scenario `s'`scen' / N=`Nnn'" - forvalues k=1/1000 { - if (mod(`k',100)==0) { - di "`k'/1000" - } - preserve - qui keep if replication==`k' - - * ROSALI - qui rosali_original item1-item`nbitems' item1-item`nbitems', group(tt) - qui mat resmat=r(difitems) - local nbitems2 = 2*`nbitems' - - * Calculer le nbre d'items détectés - local nbdetect = 0 - local stop = 0 - forvalues jj=1/`nbitems' { - if (`stop'==0) { - mat testm=J(1,1,.) - if (resmat[1,`jj']==testm[1,1]) { - local stop = 1 - local nbdetect = `jj'-1 - } - } - } - - * Stocker les items détectés + - * Définition des contraintes - local csrt=0 - mat testm=J(1,1,0) - forvalues u=1/`nbdetect' { - local difitems`u'=resmat[1,`u'] - local i=`difitems`u'' - if (`nbmoda'==3 & resmat[1,`nbitems'+`i']!=testm[1,1]){ - local constrnt`u' = "constraint `u' 2*([1.item`i']_cons-([1.item`i']_cons+[1.item`i'] tt))=([2.item`i']_cons-([2.item`i']_cons+[2.item`i'] tt))" - qui `constrnt`u'' - local v=`u'+100 - local constrnt`u'_2 = "constraint `v' 3*([1.item`i']_cons-([1.item`i']_cons+[1.item`i'] tt))=([3.item`i']_cons-([3.item`i']_cons+[3.item`i'] tt))" - qui `constrnt`u'_2' - } - } - - * Définition du modèle - local mod "gsem " - local conformula = "" - forvalues i=1/`nbitems' { - local mod = "`mod'"+"(1.item`i'<-THETA@1)" - if (`nbmoda'==3) { - local mod = "`mod'"+"(2.item`i'<-THETA@2)(3.item`i'<-THETA@3)" - } - } - forvalues u=1/`nbdetect' { - local v=`difitems`u'' - local mod = "`mod'"+"(1.item`v'<-THETA@1 tt)" - if (`nbmoda'==3) { - local mod = "`mod'"+"(2.item`v'<-THETA@2 tt)(3.item`v'<-THETA@3 tt)" - } - local w= 100+`u' - if (resmat[1,`nbitems'+`v']!=testm[1,1]) { - local conformula = "`conformula'" + "`u' " + "`w' " - } - } - if ("`conformula'" != "") { - local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(`conformula')" - } - else { - local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent" - } - *calcul du modèle - qui `mod' - mat V=r(table) - mat W=V[1..2,1...] - - * compilation - forvalues j=1/`nbitems' { - forvalues z=1/`nbmoda' { - mat outmat[`k',colnumb(outmat,"item`j'_`z'")] = W[1,colnumb(W,"`z'.item`j':_cons")] - } - } - * compilation DIF - forvalues u=1/`nbdetect' { - local j=`difitems`u'' - forvalues z=1/`nbmoda' { - mat outmat[`k',colnumb(outmat,"dif_`u'_`z'")] = W[1,colnumb(W,"`z'.item`j':tt")] - mat outmat[`k',colnumb(outmat,"dif_detect_`u'")] = `j' - } - } - - * Stocker les items de DIF originaux - if (`nbdif' > 0) { - qui levelsof dif1 - local ldif1 = r(levels) - local diff1: word 1 of `ldif1' - qui mat outmat[`k',colnumb(outmat,"real_dif_1")]=`diff1' - if (`nbdif' > 1) { - qui levelsof dif2 - local ldif2 = r(levels) - local diff2: word 1 of `ldif2' - qui mat outmat[`k',colnumb(outmat,"real_dif_2")]=`diff2' - if (`nbdif' > 2) { - qui levelsof dif3 - local ldif3 = r(levels) - local diff3: word 1 of `ldif3' - qui mat outmat[`k',colnumb(outmat,"real_dif_3")]=`diff3' - } - } - } - qui mat outmat[`k',colnumb(outmat,"beta")]=W[1,colnumb(W,"THETA:tt")] - qui mat outmat[`k',colnumb(outmat,"se_beta")]=W[2,colnumb(W,"THETA:tt")] - restore - } - putexcel set "`path_res'/`s'`scen'_`Nn'_original.xls", sheet("outmat") replace - putexcel A1=matrix(outmat), colnames -} -} -} diff --git a/Scripts/Analysis/DIF-ROSALI/pcm_dif_rosali.do b/Scripts/Analysis/DIF-ROSALI/pcm_dif_rosali.do new file mode 100644 index 0000000..53e0e26 --- /dev/null +++ b/Scripts/Analysis/DIF-ROSALI/pcm_dif_rosali.do @@ -0,0 +1,364 @@ +*================================================================================================================================================= +* Date : 2024-01-23 +* Stata version : Stata 18 SE +* +* This program analyses simulated data accounting for DIF through a partial credit model +* +* ado-files needed : - pcm, rosali (version 5.5 October 25, 2023, available on gitea) +* +* +*================================================================================================================================================ +adopath+"/home/corentin/Documents/These/Recherche/ROSALI-SIM/Modules/rosali_custom" + + +local N = "50 100 200 300" + local ss = "1 2 3 4 5 6 7 8 9 10 11 1 13 14 15 16 17 18 19 20" + foreach s in `ss' { + foreach Nnn in `N' { + local Nn = `Nnn' + local path_data = "/home/corentin/Documents/These/Recherche/Simulations/Data/DIF/N`Nn'" + if (`s'<=4) { + local path_data = "/home/corentin/Documents/These/Recherche/Simulations/Data/NoDIF/N`Nn'" + } + local path_res = "/home/corentin/Documents/These/Recherche/Simulations/Analysis/ROSALI-DIF/N`Nn'" + local scenarios = "A B C D E F G" + foreach scen in `scenarios' { + clear + import delim "`path_data'/scenario_`s'`scen'_`Nn'.csv", encoding(ISO-8859-2) case(preserve) clear + rename TT tt + + if (`s'<=2) { + local nbitems=4 + } + else if (`s'<=4) { + local nbitems=7 + } + else if (`s'<=12) { + local nbitems=4 + } + else { + local nbitems=7 + } + +if (mod(`s',2)==0) { + local nbmoda=3 +} +else { + local nbmoda=1 +} + + if (`s'<=4) { + local nbdif=0 + } + else if (`s'<=8) { + local nbdif=1 + } + else if (`s'<=16) { + local nbdif=2 + } + else { + local nbdif=3 + } + * taillemat = Maximum J*M cases pour les items par et J*M cases pour les dif par + J cases pour les DIF detect + nbdif cases pour dif réel + local taillemat=2*`nbitems'*`nbmoda'+`nbitems'+`nbdif'+2 + local colna="" + forvalues i=1/`nbitems' { + forvalues z=1/`nbmoda' { + local colna = "`colna'"+"item`i'_`z' " + local colna = "`colna'"+"dif_`i'_`z' " + } + } + forvalues i=1/`nbitems' { + local colna = "`colna'"+"dif_detect_`i' " + } + + forvalues i=1/`nbdif' { + local colna = "`colna'"+"real_dif_`i' " + } +local colna = "`colna'" + "beta " + "se_beta" + + + mat outmat = J(1000,`taillemat',.) + mat colnames outmat= `colna' + di "Scenario `s'`scen' / N=`Nnn'" + forvalues k=1/1000 { + if (mod(`k',100)==0) { + di "`k'/1000" + } + preserve + qui keep if replication==`k' + + + * MERGE des modalités si non représentées + + if (`nbmoda'>1 & `Nn'==50) { + local com_z = 0 + qui gen comz = 0 + forvalues j = 1 / `nbitems' { + local recoda_`j' = 0 + qui tab item`j' if tt == 0, matrow(rect1_g0_`j') matcell(nbrt1_g0_`j') + local maxm`j'_t1_g0 = rect1_g0_`j'[r(r),1] + local minm`j'_t1_g0 = rect1_g0_`j'[1,1] + + qui tab item`j' if tt == 1, matrow(rect1_g1_`j') matcell(nbrt1_g1_`j') + local minm`j'_t1_g1 = rect1_g1_`j'[1,1] + local maxm`j'_t1_g1 = rect1_g1_`j'[r(r),1] + + local minm_`j' = min(`minm`j'_t1_g0',`minm`j'_t1_g1') + local maxm_`j' = max(`maxm`j'_t1_g0',`maxm`j'_t1_g1') + local nbm_`j' = `=`maxm_`j''-`minm_`j''' + + if `minm_`j'' != 0 & `com_z' == 0 { + local com_z = 1 + } + + qui count if item`j' == 3 & tt == 0 + local mod3plac = r(N) + qui count if item`j' == 3 & tt == 1 + local mod3tt = r(N) + local nb_rn3 = min(`mod3plac',`mod3tt') + if `nb_rn3'==0 { + qui replace comz = 1 + } + + forvalues m = 0/`=`nbm_`j''-1' { + qui count if item`j' == `m' & tt == 0 + local nb_rn1_g0 = r(N) + qui count if item`j' == `m' & tt == 1 + local nb_rn1_g1 = r(N) + local nb_rn = min(`nb_rn1_g0',`nb_rn1_g1') + if `nb_rn' == 0 { + qui replace comz = 1 + local recoda_`j' = 1 + if `m' == 0 | `m' < `minm`j'_t1_g0' | `m' < `minm`j'_t1_g1' { + local stop = 1 + forvalues kk = 1/`=`nbm_`j''-`m'' { + qui count if item`j' == `=`m' + `kk'' & tt == 0 + local v`kk'1_0 = r(N) + qui count if item`j' == `=`m' + `kk'' & tt == 1 + local v`kk'1_1 = r(N) + if (`v`kk'1_0' != 0 | `v`kk'1_1' != 0) & `stop' != 0 { + qui replace item`j'= `=`m'+`kk'' if item`j'==`m' + local zzz=`j'+`nbitems' + *qui replace item`zzz'=`=`m'+`kk'' if item``=`j'+`nbitems'''==`m' + *di "WARNING SCENARIO `k': items `j': answers `m' and `=`m'+`kk'' merged" + local stop = 0 + } + } + } + else if `m' == `=`nbm_`j''-1' | `m' >= `maxm`j'_t1_g1' { + local stop = 1 + forvalues kk = 1/`=`m'' { + qui count if item`j' == `=`m' - `kk'' & tt == 0 + local v`kk'1_0 = r(N) + qui count if item`j' == `=`m' - `kk'' & tt == 1 + local v`kk'1_1 = r(N) + if (`v`kk'1_0' != 0 | `v`kk'1_1' != 0) & `stop' != 0 { + qui replace item`j'= `=`m' - `kk'' if item`j'==`m' + local zzz=`j'+`nbitems' + *qui replace item`zzz'= `=`m' - `kk'' if item`zzz'==`m' + *di "WARNING SCENARIO `k': items `j': answers `m' and `=`m'+`kk'' merged" + local stop = 0 + } + } + } + else { + if runiform()>0.5{ + local stop = 1 + forvalues kk = 1/`m' { + qui count if item`j' == `=`m' - `kk'' & tt == 0 + local v`kk'1_0 = r(N) + qui count if item`j' == `=`m' - `kk'' & tt == 1 + local v`kk'1_1 = r(N) + if (`v`kk'1_0' != 0 | `v`kk'1_1' != 0) & `stop' != 0 { + qui replace item`j'= `=`m'-`kk'' if item`j'==`m' + local zzz=`j'+`nbitems' + *qui replace item`zzz'=`=`m'-`kk'' if item``=`j'+`nbitems'''==`m' + *di "WARNING SCENARIO `k': items `j': answers `m' and `=`m'+`kk'' merged" + local stop = 0 + } + } + } + else { + local stop = 1 + forvalues kk = 1/`=`nbm_`j''-`m'' { + qui count if item`j' == `=`m' + `kk'' & tt == 0 + local v`kk'1_0 = r(N) + qui count if item`j' == `=`m' + `kk'' & tt == 1 + local v`kk'1_1 = r(N) + if (`v`kk'1_0' != 0 | `v`kk'1_1' != 0) & `stop' != 0{ + qui replace item`j'=`=`m' + `kk'' if item`j'==`m' + local zzz=`j'+`nbitems' + *qui replace item`zzz'=`=`m' + `kk'' if item``=`j'+`nbitems'''==`m' + *di "WARNING SCENARIO `k': items `j': answers `m' and `=`m'+`kk'' merged" + local stop = 0 + } + else { + if `stop' != 0 { + qui replace item`j'= `nbm_`j'' if item`j'==`m' + local zzz=`j'+`nbitems' + *qui replace item`zzz'= `nbm_`j'' if item``=`j'+`nbitems'''==`m' + *di "WARNING SCENARIO `k': items `j': answers `m' and `=`m'+`kk'' merged" + local stop = 0 + } + } + } + } + } + } + } + qui levelsof item`j' + local val = r(levels) + local checker: word 1 of `val' + local checker2: word 2 of `val' + local checker3: word 3 of `val' + local nummoda=r(r) + local nbmoda_`j'=`nummoda' + if (`nummoda'==2) { + qui recode item`j' (`checker'=0) (`checker2'=1) + } + if (`nummoda'==3) { + if (`checker'!=0) { + qui recode item`j' (`checker'=0) (`checker2'=1) (`checker3'=2) + } + else if (`checker2'!=1) { + qui recode item`j' (`checker2'=1) (`checker3'=2) + } + else if (`checker3'!=2) { + qui recode item`j' (`checker3'=2) + } + } + } + + qui valuesof comz + local val = r(values) + local checker: word 1 of `val' + } + else { + forvalues jj=1/`nbitems' { + local nbmoda_`jj'=`nbmoda' + } + } + + + * ROSALI + qui rosali_original item1-item`nbitems' item1-item`nbitems', group(tt) + qui mat resmat=r(difitems) + local nbitems2 = 2*`nbitems' + + * Calculer le nbre d'items détectés + local nbdetect = 0 + local stop = 0 + forvalues jj=1/`nbitems' { + if (`stop'==0) { + mat testm=J(1,1,.) + if (resmat[1,`jj']==testm[1,1]) { + local stop = 1 + local nbdetect = `jj'-1 + } + } + } + + * Stocker les items détectés + + * Définition des contraintes + local csrt=0 + mat testm=J(1,1,0) + forvalues u=1/`nbdetect' { + local difitems`u'=resmat[1,`u'] + local i=`difitems`u'' + if (`nbmoda_`i''==3 & resmat[1,`nbitems'+`i']!=testm[1,1]){ + local constrnt`u' = "constraint `u' 2*([1.item`i']_cons-([1.item`i']_cons+[1.item`i'] tt))=([2.item`i']_cons-([2.item`i']_cons+[2.item`i'] tt))" + qui `constrnt`u'' + local v=`u'+100 + local constrnt`u'_2 = "constraint `v' 3*([1.item`i']_cons-([1.item`i']_cons+[1.item`i'] tt))=([3.item`i']_cons-([3.item`i']_cons+[3.item`i'] tt))" + qui `constrnt`u'_2' + } + if (`nbmoda_`i''==2 & resmat[1,`nbitems'+`i']!=testm[1,1]){ + local constrnt`u' = "constraint `u' 2*([1.item`i']_cons-([1.item`i']_cons+[1.item`i'] tt))=([2.item`i']_cons-([2.item`i']_cons+[2.item`i'] tt))" + qui `constrnt`u'' + } + } + + * Définition du modèle + local mod "gsem " + local conformula = "" + forvalues i=1/`nbitems' { + local mod = "`mod'"+"(1.item`i'<-THETA@1)" + if (`nbmoda_`i''==3) { + local mod = "`mod'"+"(2.item`i'<-THETA@2)(3.item`i'<-THETA@3)" + } + else if (`nbmoda_`i''==2) { + local mod = "`mod'"+"(2.item`i'<-THETA@2)" + } + } + forvalues u=1/`nbdetect' { + local v=`difitems`u'' + local mod = "`mod'"+"(1.item`v'<-THETA@1 tt)" + if (`nbmoda_`v''==3) { + local mod = "`mod'"+"(2.item`v'<-THETA@2 tt)(3.item`v'<-THETA@3 tt)" + } + else if (`nbmoda_`v''==2) { + local mod = "`mod'"+"(2.item`v'<-THETA@2 tt)" + } + local w= 100+`u' + if (resmat[1,`nbitems'+`v']!=testm[1,1] & `nbmoda_`v''==3) { + local conformula = "`conformula'" + "`u' " + "`w' " + } + else if (resmat[1,`nbitems'+`v']!=testm[1,1] & `nbmoda_`v''==2) { + local conformula = "`conformula'" + "`u' " + } + } + if ("`conformula'" != "") { + local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent constraint(`conformula')" + } + else { + local mod = "`mod'" + "(THETA<-tt), mlogit tol(0.01) iterate(500) latent(THETA) nocapslatent" + } + *calcul du modèle + qui `mod' + mat V=r(table) + mat W=V[1..2,1...] + + * compilation + forvalues j=1/`nbitems' { + forvalues z=1/`nbmoda_`j'' { + mat outmat[`k',colnumb(outmat,"item`j'_`z'")] = W[1,colnumb(W,"`z'.item`j':_cons")] + } + } + * compilation DIF + forvalues u=1/`nbdetect' { + local j=`difitems`u'' + forvalues z=1/`nbmoda_`j'' { + mat outmat[`k',colnumb(outmat,"dif_`u'_`z'")] = W[1,colnumb(W,"`z'.item`j':tt")] + mat outmat[`k',colnumb(outmat,"dif_detect_`u'")] = `j' + } + } + + * Stocker les items de DIF originaux + if (`nbdif' > 0) { + qui levelsof dif1 + local ldif1 = r(levels) + local diff1: word 1 of `ldif1' + qui mat outmat[`k',colnumb(outmat,"real_dif_1")]=`diff1' + if (`nbdif' > 1) { + qui levelsof dif2 + local ldif2 = r(levels) + local diff2: word 1 of `ldif2' + qui mat outmat[`k',colnumb(outmat,"real_dif_2")]=`diff2' + if (`nbdif' > 2) { + qui levelsof dif3 + local ldif3 = r(levels) + local diff3: word 1 of `ldif3' + qui mat outmat[`k',colnumb(outmat,"real_dif_3")]=`diff3' + } + } + } + qui mat outmat[`k',colnumb(outmat,"beta")]=W[1,colnumb(W,"THETA:tt")] + qui mat outmat[`k',colnumb(outmat,"se_beta")]=W[2,colnumb(W,"THETA:tt")] + restore + } + putexcel set "`path_res'/`s'`scen'_`Nn'_original.xls", sheet("outmat") replace + putexcel A1=matrix(outmat), colnames +} +} +}