capture program drop gmm_style_Ltrunc
program define gmm_style_Ltrunc
version 9.0
syntax varlist [if] [in] [, nt(varlist numeric max=2) GMMLiv(numlist integer >=0) *]
*capture drop _GMML_*
local nindeces : list sizeof nt
*di as result "Number of indeces for observations in your data-set: `nindeces'"
*di as result "General description of the dataset"
if `nindeces'==0 {
di as error "You cannot create GMM-style IVs with cross-section data. Use nt() option to set time series or panel data."
exit 198
}
if `nindeces'==1 {
di as text "You are creating GMM-style IVs in levels for a time series"
tokenize `nt'
local time="`1'"
qui tsset `1'
*return list
local nperiods=r(tmax)-r(tmin)+1
local time_min=r(tmin)
local time_max=r(tmax)
*di as text "time variable: `time'"
*di as text "first year: `time_min'"
*di as text "last year: `time_max'"
}
if `nindeces'==2 {
di as text "You are creating GMM-style IVs in levels for a panel"
tokenize `nt'
local time="`2'"
qui tsset `1' `2'
*return list
local nperiods=r(tmax)-r(tmin)+1
local time_min=r(tmin)
local time_max=r(tmax)
*di as text "time variable: `time'"
*di as text "first year: `time_min'"
*di as text "last year: `time_max'"
}
tokenize `varlist'
local i 1
while "``i''"!="" {
di in blue "_____ variable: ``i'' _____"
if "`gmmliv'" != "" {
local nlag : list sizeof gmmliv
*di as result "Elements in GMML(): `nlag'"
if `nlag'==1 {
*local optname gmmliv()
local primo_lag =`gmmliv'
di as result "Lag selection in GMML(): from t-`primo_lag' to the last available lag"
if `primo_lag'==0 {
*di as text "Are you sure that the first IV is t (exogenous variables)?"
}
if `primo_lag'==1 {
*di as text "Are you sure that the first IV is t-1 (predetermined variables)?"
}
if `primo_lag'==2 {
*di as text "Are you sure that the first IV is t-2 (endogenous variables)?"
}
forvalues tau=`=`time_min'+`primo_lag'' / `time_max' {
forvalues lag = `primo_lag' / `=`tau'-`time_min'' {
qui g _GMML_``i''_`tau'L`lag' = L`lag'.``i'' if `time' == `tau'
qui summ _GMML_``i''_`tau'L`lag'
if r(N)==0 {
drop _GMML_``i''_`tau'L`lag'
}
else if r(N)!=0 {
qui recode _GMML_``i''_`tau'L`lag' (.=0)
}
} /* IV generation */
} /* for each year */
} /* one lag */
if `nlag'==2 {
*local optname gmmliv()
tokenize `gmmliv'
local primo_lag `1'
local ultimo_lag `2'
di as result "Lag selection in GMML(): from t-`primo_lag' to t-`ultimo_lag'"
if `primo_lag'==0 {
*di as text "Are you sure that the first IV is t (exogenous variables)?"
}
if `primo_lag'==1 {
*di as text "Are you sure that the first IV is t-1 (predetermined variables)?"
}
if `primo_lag'==2 {
*di as text "Are you sure that the first IV is t-2 (endogenous variables)?"
}
if `primo_lag'>`ultimo_lag' {
di as error "This procedure uses lags as IV, not forewards (use GMM_style_Ftrunc procedure)"
exit 198
}
tokenize `varlist'
forvalues tau=`=`time_min'+`primo_lag'' / `time_max' {
*di "`tau' `time_min' `primo_lag' `time_max'"
forvalues lag = `primo_lag' / `ultimo_lag' {
*di "`primo_lag' `ultimo_lag' `time'"
qui g _GMML_``i''_`tau'L`lag' = L`lag'.``i'' if `time' == `tau'
qui summ _GMML_``i''_`tau'L`lag'
if r(N)==0 {
drop _GMML_``i''_`tau'L`lag'
}
else if r(N)!=0 {
qui recode _GMML_``i''_`tau'L`lag' (.=0)
}
} /* IV generation */
} /* for each year */
} /* two lags */
} /* gmmliv != missing */
if "`gmmliv'" == "" {
di as error "You must select lag(s) for GMM-style IVs in levels; use the gmml() option"
exit 198
}
local i=`i'+1
} /* close varlist */
end
**********************************************
* `1' individuals
* `2' years
* `3' variables
* gmml() lags selection
* COMMAND: gmm_style_Ltrunc list_of_variables, nt(id year) gmml()
* 1 index --> from t-index to the last available lag
* 2 index --> from t-index1 to t-index2
*(example from abdata and panel_dynamic_application)
* it creates the matrix of GMM-style IVs so that also ivreg2 or other similar commands could operate like xtabond2 gmm
* IVs in levels, for eq(diff) *
*******************************
*******************************
capture program drop gmm_style_Dtrunc
program define gmm_style_Dtrunc
version 9.0
syntax varlist [if] [in] [, nt(varlist numeric max=2) GMMDiv(numlist >=0) *]
*capture drop _GMMD_*
local nindeces : list sizeof nt
di as result "Number of indeces for observations in your data-set: `nindeces'"
if `nindeces'==0 {
di as error "You cannot create GMM-style IVs with cross-section data. Use nt() option to set time series or panel data."
exit 198
}
if `nindeces'==1 {
di as text "You are creating GMM-style IVs in first-differences for a time series"
tokenize `nt'
local time="`1'"
qui tsset `1'
*return list
local nperiods=r(tmax)-r(tmin)+1
local time_min=r(tmin)
local time_max=r(tmax)
di as text "time variable: `time'"
di as text "first year: `time_min'"
di as text "last year: `time_max'"
}
if `nindeces'==2 {
di as text "You are creating GMM-style IVs in first-differences for a panel"
tokenize `nt'
local time="`2'"
qui tsset `1' `2'
*return list
local nperiods=r(tmax)-r(tmin)+1
local time_min=r(tmin)
local time_max=r(tmax)
di as text "time variable: `time'"
di as text "first year: `time_min'"
di as text "last year: `time_max'"
}
tokenize `varlist'
local i 1
while "``i''"!="" {
di in blue "_____ variable: ``i'' _____ "
if "`gmmdiv'" != "" {
local nlag : list sizeof gmmdiv
*di as result "Elements in GMMD(): `nlag'"
if `nlag'==1 {
*local optname gmmdiv()
local primo_lag =`gmmdiv'
di as result "Lag selection in GMMD(): from t-`primo_lag' to the last available lag"
if `primo_lag'==0 {
*di as text "Are you sure that the first IV is t (exogenous variables)?"
}
if `primo_lag'==1 {
*di as text "Are you sure that the first IV is t-1 (predetermined variables)?"
}
if `primo_lag'==2 {
*di as text "Are you sure that the first IV is t-2 (endogenous variables)?"
}
forvalues tau=`=`time_min'+`primo_lag'' / `time_max' {
forvalues lag = `primo_lag' / `=`tau'-`time_min'' {
qui g _GMMD_``i''_`tau'L`lag' = D.L`lag'.``i'' if `time' == `tau'
qui summ _GMMD_``i''_`tau'L`lag'
if r(N)==0 {
drop _GMMD_``i''_`tau'L`lag'
}
else if r(N)!=0 {
qui recode _GMMD_``i''_`tau'L`lag' (.=0)
}
} /* IV generation */
} /* for each year */
} /* one lag */
if `nlag'==2 {
*local optname gmmdiv()
tokenize `gmmdiv'
local primo_lag `1'
local ultimo_lag `2'
di as result "Lag selection in GMMD(): from t-`primo_lag' to t-`ultimo_lag'"
if `primo_lag'==0 {
*di as text "Are you sure that the first IV is t (exogenous variables)?"
}
if `primo_lag'==1 {
*di as text "Are you sure that the first IV is t-1 (predetermined variables)?"
}
if `primo_lag'==2 {
*di as text "Are you sure that the first IV is t-2 (endogenous variables)?"
}
if `primo_lag'>`ultimo_lag' {
di as error "This procedure uses lags as IV, not forewards (use GMM_style_Ftrunc procedure)"
exit 198
}
tokenize `varlist'
forvalues tau=`=`time_min'+`primo_lag'' / `time_max' {
*di "`tau' `time_min' `primo_lag' `time_max'"
forvalues lag = `primo_lag' / `ultimo_lag' {
*di "`primo_lag' `ultimo_lag' `time'"
qui g _GMMD_``i''_`tau'L`lag' = D.L`lag'.``i'' if `time' == `tau'
qui summ _GMMD_``i''_`tau'L`lag'
if r(N)==0 {
drop _GMMD_``i''_`tau'L`lag'
}
else if r(N)!=0 {
qui recode _GMMD_``i''_`tau'L`lag' (.=0)
}
} /* IV generation */
} /* for each year */
} /* two lags */
} /* gmmdiv != missing */
if "`gmmdiv'" == "" {
di as error "You must select lag(s) for GMM-style IVs in first-differences; use the gmmd() option"
exit 198
}
local i=`i'+1
} /* close varlist */
end
**********************************************
* `1' individuals
* `2' years
* `3' variables
* gmmd() lags selection
* COMMAND: gmm_style_Dtrunc id year list_of_variables, gmmd()
* 1 index --> from t-index to the last available lag
* 2 index --> from t-index1 to t-index2
*(example from abdata and panel_dynamic_application)
* it creates the matrix of GMM-style IVs so that also ivreg2 or other similar commands could operate like xtabond2 gmm
* IVs in first-differences, for eq(level) *
*******************************
***************************** MAIN PROGRAM *****************************************************************
*****************************************************************************************************************
*********************************************************************************************************************
capture program drop pca2
program define pca2, rclass
version 9.0
syntax varlist [if] [in] [, nt(varlist numeric max=2) VARiance(numlist integer >0 <=100) AVG COVariance PREfix(string) SEE /*
*/ TOGVar TOGLD GMMLiv(numlist integer >=0 missingokay) GMMDiv(numlist integer >=0 missingokay) RETain *]
set more off
/*
local m 1
local l = length(`"`nt'"')
while `m' <= `l' {
local char`m' = substr(`"`nt'"', `m', 3)
if substr(`"`nt'"', `m', 3) == " " {
local char`m' " "
}
local m = `m' + 1*
}
di `char1'
di `char2'
*/
local nvars : list sizeof varlist
local varlistpca `varlist'
tokenize `varlist'
local firstvar `1'
local lastvar ``nvars''
*di "`firstvar' `lastvar' `nvars'" "
/*
tempname s_trace s_ExplainedVarVar s_RetainedEigenCountVar s_trace_PV_perc s_ExplainedVarVar_var /*
*/ l_SecondIndex /*
*/ s_RetainedEigenCountAvg s_ExplainedVarAvg s_ExplainedVarAvg_var s_RetainedPC /*
*/ primo_lag1D ultimo_lag1D
*/
tempname s_trace l_SecondIndex primo_lag1D ultimo_lag1D
**** TYPE OF DATA *******
local nindeces : list sizeof nt
*di as result "Number of indeces for observations of your data-set: `nindeces'"
di as result "General description of the dataset"
if `nindeces'==0 {
di as text "Is your data-set a cross-section?"
}
if `nindeces'==1 {
di as text "Is your data-set a time-series?"
tokenize `nt'
local time_var ="`1'"
*di as result "time var: `time_var'"
tsset `time_var'
local nperiods=r(tmax)-r(tmin)+1
local time_min=r(tmin)
local time_max=r(tmax)
}
if `nindeces'==2 {
*di as text "Is your data-set a panel?"
tokenize `nt'
local ind_var = "`1'"
local time_var = "`2'"
*di as result "panel var: `ind_var'" " " "time var: `time_var'"
tsset `ind_var' `time_var'
*return list
local nperiods=r(tmax)-r(tmin)+1
local time_min=r(tmin)
local time_max=r(tmax)
}
**** NAME FOR SCORES *******
* by default, if the name of the prefix is not esplicitly indicated by option pre(),
* the _BM_ prefix is used
if `"`prefix'"' == "" {
local prefix1 = "_BM_"
di as result "The prefix is: `prefix1'"
* Change the default prefix if variables called by the same name already exist:
capture descr `prefix1'*
*di _rc
*di "`prefix1'"
if _rc==0&r(N)!=. {
di as error "Variables starting with the default `prefix1' already exist in your data-set: please use the pre() option"
*capture drop `prefix'*
exit 198
}
} /* end prefix */
* Verify the validity of the selected prefix:
if `"`prefix'"' != "" {
local prefix1 = "_`prefix'_"
di as result "The prefix is: `prefix1'"
* Change the selected prefix if variables called by the same name already exist:
capture descr `prefix1'*
di _rc
*di "`prefix1'"
if _rc==0&r(N)!=. {
di as error "Variables starting with the selected `prefix1' already exist in your data-set: please use the pre() option to select another name"
exit 198
}
} /* end prefix */
marksample touse, novarlist
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
********** GMM LEV DIFFERENT LAGS VAR BY VAR & TOGVAR ***************************
*********************************************************************************
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
local lagslgroups 0
*di "START LOOP LAGSL: lagslgroups: `lagslgroups'"
local 0, `options'
local _optlagsl `options'
*local 0, `_optlagsl'
*di " opzioni: `0'"
*di "START LOOP LAGSL: _optlagsl: `_optlagsl'"
syntax [, LAGSL(string) *]
while "`lagsl'" != "" {
if "`gmmliv'"!="" {
di as error "Either use gmml or lagsl"
exit 198
}
local optionsarg `options'
*di "optionsarg: `optionsarg'"
local 0 `lagsl'
*di "lagsl : `lagsl'"
local _optlagsl `lagsl'
*di "_optlagsl: `_optlagsl'"
capture syntax varlist, [Ll(string)]
/*
if _rc {
di as err _n "lags(`0') invalid."
exit 198
}
*/
local nbasevarsl : list sizeof varlist
local basevarsl `varlist'
*di "varlist originally in lagsl:`varlist'"
*di "varlist derived from lagsl:`basevarsl'"
*di "varlist in pca2 syntax :`varlistpca'"
*di "nbasevarsl: `nbasevarsl'"
foreach lag of numlist `lagslgroups' {
if "`ll'" == "" {
di as err _n `"ll(`ll') must have at least one argument."'
exit 198
}
if `:word count `ll'' == 1 {
di "ONE LAG in ll()"
local lagliml1`lag' = `: word 1 of `ll'' + 0
*di "lag in ll(): `lagliml1`lag''"
if `lagliml1`lag'' == . {
di as error "Lags can not be missing"
exit 198
}
local lagslgroups = `lagslgroups' + 1
local lagslvars "`lagslvars' `basevarsl'"
*di "lagslgroups: `lagslgroups'"
*di "lagslvars: `lagslvars'"
local nlagslvars : list sizeof lagslvars
*di "nlagslvars: `nlagslvars'"
gmm_style_Ltrunc `basevarsl', nt(`ind_var' `time_var') gmml(`lagliml1`lag'')
} /* ONE LAG */
else {
if `:word count `ll'' == 2 {
di "TWO LAGS in rit()"
forvalues a = 1/2 {
*di "how many lags?`a'"
capture local lagliml`a'`lag' = `: word `a' of `ll'' + 0
if _rc {
di as err _n `"ll(`ll') invalid."'
exit 198
}
} /* chiude il loop*/
} /* word count */
if `lagliml1`lag'' == . {
di as error "Lags can not be missing"
exit 198
}
if `lagliml2`lag'' == . {
di as error "Lags can not be missing"
exit 198
}
*di "First lag in rit(): `lagliml1`lag''"
*di "Second lag in rit(): `lagliml2`lag''"
local lagslgroups = `lagslgroups' + 1
local lagslvars "`lagslvars' `basevarsl'"
*di "lagslgroups: `lagslgroups'"
*di "lagslvars: `lagslvars'"
local nlagslvars : list sizeof lagslvars
*di "nlagslvars: `nlagslvars'"
gmm_style_Ltrunc `basevarsl', nt(`ind_var' `time_var') gmml(`lagliml1`lag'' `lagliml2`lag'')
} /* TWO LAGS */
} /* for each lagslgroup */
local 0, `optionsarg'
syntax [, LAGSL(string) *]
} /* chiude while lags */
local lagslgroups 0
local _optlagsl 0
if "`nlagslvars'"!=""&"`gmmdiv'"=="" {
if `nlagslvars'<`nvars' {
di as error "The number of variables in pca2 command is different from the number of variables in the options"
exit 198
}
}
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
********** GMM DIF DIFFERENT LAGS VAR BY VAR & TOGVAR ***************************
*********************************************************************************
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
local lagsdgroups 0
*di "START LOOP LAGSD: lagsdgroups: `lagsdgroups'"
local 0, `options'
local _optlagsd `options'
*di "START LOOP LAGSD: _optlagsd: `_optlagsd'"
syntax [, LAGSD(string) *]
while "`lagsd'" != "" {
if "`gmmdiv'"!="" {
di as error "Either use gmmd or lagsd"
exit 198
}
local optionsarg `options'
*di "optionsarg: `optionsarg'"
local 0 `lagsd'
*di "lagsd : `lagsd'"
local _optlagsd `lagsd'
*di "_optlagsd: `_optlagsd'"
capture syntax varlist, [Ll(string)]
local nbasevarsd : list sizeof varlist
local basevarsd `varlist'
*di "varlist originally in lagsd: `varlist'"
*di "varlist derived from lagsd: `basevarsd'"
*di "varlist in pca2 syntax :`varlistpca'"
*di "nbasevarsd: `nbasevarsd'"
foreach lag of numlist `lagsdgroups' {
if "`ll'" == "" {
di as err _n `"ll(`ll') must have at least one argument."'
exit 198
}
if `:word count `ll'' == 1 {
di "ONE LAG in rit()"
local laglimd1`lag' = `: word 1 of `ll'' + 0
*di "lag in rit(): `laglimd1`lag''"
if `laglimd1`lag'' == . {
di as error "Lags can not be missing"
exit 198
}
local lagsdgroups = `lagsdgroups' + 1
local lagsdvars "`lagsdvars' `basevarsd'"
di "lagsdgroups: `lagsdgroups'"
di "lagsdvars: `lagsdvars'"
local nlagsdvars : list sizeof lagsdvars
*di "nlagsdvars: `nlagsdvars'"
gmm_style_Dtrunc `basevarsd', nt(`ind_var' `time_var') gmmd(`laglimd1`lag'')
} /* ONE LAG */
else {
if `:word count `ll'' == 2 {
di "TWO LAGS in rit()"
forvalues a = 1/2 {
*di "how many lags?`a'"
capture local laglimd`a'`lag' = `: word `a' of `ll'' + 0
if _rc {
di as err _n `"ll(`ll') invalid."'
exit 198
}
} /* chiude il loop*/
} /* word count */
if `laglimd1'`lag' == . {
di as error "Lags can not be missing"
exit 198
}
if `laglimd2'`lag' == . {
di as error "Lags can not be missing"
exit 198
}
*di "First lag in rit(): `laglimd1`lag''"
*di "Second lag in rit(): `laglimd2`lag''"
local lagsdgroups = `lagsdgroups' + 1
local lagsdvars "`lagsdvars' `basevarsd'"
di "lagsdgroups: `lagsdgroups'"
di "lagsdvars: `lagsdvars'"
local nlagsdvars : list sizeof lagsdvars
*di "nlagsdvars: `nlagsdvars'"
gmm_style_Dtrunc `basevarsd', nt(`ind_var' `time_var') gmmd(`laglimd1`lag'' `laglimd2`lag'')
} /* TWO LAGS */
} /* for each lagsdgroups */
local 0, `optionsarg'
syntax [, LAGSD(string) *]
} /* close while lags */
if "`nlagsdvars'"!=""&"`gmmliv'"=="" {
if `nlagsdvars'<`nvars' {
di as error "The number of variables in pca2 command is different from the number of variables in the options"
exit 198
}
}
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
*** START PCA ***
************** PCA VAR BY VAR LEV ******************
if "`togvar'"=="" & "`nlagslvars'"!="" {
foreach var of varlist `lagslvars' {
di "_____ PCA LEV VAR BY VAR: `var'"
local newvarlistl_`var' _GMML_`var'_*
local newvarlistldl_`var' _GMML_`var'_*
local newvarlistldllags_`var' _GMML_`var'_*
if "`togld'"=="" {
di as text "You are applying PCA to GMM-style LEV lags of more than one variable,"
di as text "keeping the variables separated with different lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistl_`var'' if `touse'
}
if "`see'"!="" {
pca `newvarlistl_`var'' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistl_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistl_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels for `var' __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLEV`var'N1-`prefix1'varscoreLEV`var'N`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLEV`var'N1, score
}
ret scalar var_byvar_LEV_`var' = s_ExplainedVarVar_var
ret scalar nscores_byvar_LEV_`var' = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLEV`var'N1-`prefix1'avgscoreLEV`var'N`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLEV`var'N1, score
}
ret scalar var_byavg_LEV_`var' = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LEV_`var' = s_RetainedEigenCountAvg
} /*close option mean */
} /* close togld=="" */
} /* close for each var */
*if "`retain'"=="" {
* capture drop _GMML_*_*
* }
} /* close if togvar=="" */
************ PCA VARTOG LEV *********************
if "`togvar'"!="" &"`nlagslvars'"!="" {
di "_____ PCA LEV VAR TOGETHER: `lagslvars'"
local newvarlistl_tog _GMML_*_*
local newvarlistldl_tog _GMML_*_*
local newvarlistldllags_tog _GMML_*_*
if "`togld'"=="" {
di as text "You are applying PCA to GMM-style LEV lags of more than one variable,"
di as text "keeping the variables together with different lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistl_tog' if `touse'
}
if "`see'"!="" {
pca `newvarlistl_tog' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistl_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistl_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels for the variables `varlistpca' together"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLEVtog1-`prefix1'varscoreLEVtog`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLEVtog1, score
}
ret scalar var_byvar_LEV_tog = s_ExplainedVarVar_var
ret scalar nscores_byvar_LEV_tog = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLEVtog1-`prefix1'avgscoreLEVtog`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLEVtog1, score
}
ret scalar var_byavg_LEV_tog = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LEV_tog = s_RetainedEigenCountAvg
} /*close option mean */
} /*close togld=="" */
*if "`retain'"=="" {
* capture drop _GMML_*_*
* }
} /*chiude if togvar!="" */
****************** PCA VAR BY VAR DIF *************************
if "`togvar'"=="" & "`nlagsdvars'"!="" {
foreach var of varlist `lagsdvars' {
di "_____ PCA DIF VAR BY VAR: `var'"
local newvarlistd_`var' _GMMD_`var'_*
local newvarlistldd_`var' _GMMD_`var'_*
local newvarlistlddlags_`var' _GMMD_`var'_*
if "`togld'"=="" {
di as text "You are applying PCA to GMM-style DIF lags of more than one variable,"
di as text "keeping the variables separated with different lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistd_`var'' if `touse'
}
if "`see'"!="" {
pca `newvarlistd_`var'' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistd_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistd_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in first_differences for `var' __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreDIF`var'N1-`prefix1'varscoreDIF`var'N`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreDIF`var'N1, score
}
ret scalar var_byvar_DIF_`var' = s_ExplainedVarVar_var
ret scalar nscores_byvar_DIF_`var' = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreDIF`var'N1-`prefix1'avgscoreDIF`var'N`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreDIF`var'N1, score
}
ret scalar var_byavg_DIF_`var' = s_ExplainedVarAvg_var
ret scalar nscores_byavg_DIF_`var' = s_RetainedEigenCountAvg
} /*close option mean */
} /* close togld=="" */
} /* close for each var */
* if "`retain'"=="" {
* capture drop _GMMD_*_*
* }
} /* close if togvar=="" */
******************** PCA VARTOG DIF ****************************
if "`togvar'"!="" & "`nlagsdvars'"!=""{
di "_____ PCA DIF VAR TOGETHER: `lagsdvars'"
local newvarlistd_tog _GMMD_*_*
local newvarlistldd_tog _GMMD_*_*
local newvarlistlddlags_tog _GMMD_*_*
if "`togld'"=="" {
di as text "You are applying PCA to GMM-style DIF lags of more than one variable,"
di as text "keeping the variables together with different lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistd_tog' if `touse'
}
if "`see'"!="" {
pca `newvarlistd_tog' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistd_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistd_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in first-differences for the variables `varlistpca' together"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreDIFtog1-`prefix1'varscoreDIFtog`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreDIFtog1, score
}
ret scalar var_byvar_DIF_tog = s_ExplainedVarVar_var
ret scalar nscores_byvar_DIF_tog = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreDIFtog1-`prefix1'avgscoreDIFtog`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreDIFtog1, score
}
ret scalar var_byavg_DIF_tog = s_ExplainedVarAvg_var
ret scalar nscores_byavg_DIF_tog = s_RetainedEigenCountAvg
} /*close option mean */
} /*close togld=="" */
*if "`retain'"=="" {
* capture drop _GMMD_*_*
* }
} /* chiude il togvar!="" */
************ PCA VAR BY VAR LEV & DIF TOGETHER ********************
if "`togld'"!="" & "`nlagslvars'"!="" & "`nlagsdvars'"!="" {
if "`togvar'"=="" {
foreach var of varlist `varlistpca' {
di "_____ PCA LEV&DIF VAR BY VAR: `var'"
local newvarlistld_`var' "`newvarlistldl_`var'' `newvarlistldd_`var''"
di "`newvarlistld_`var''"
di as text "You are applying PCA to GMM-style LEV & DIF lags of more than one variable,"
di as text "keeping the variables separated with different lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistld_`var'' if `touse'
}
if "`see'"!="" {
pca `newvarlistld_`var'' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistld_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistld_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels & first_differences together for `var' _________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLD`var'N1-`prefix1'varscoreLD`var'N`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLD`var'N1, score
}
ret scalar var_byvar_LD_`var' = s_ExplainedVarVar_var
ret scalar nscores_byvar_LD_`var' = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLD`var'N1-`prefix1'avgscoreLD`var'N`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLD`var'N1, score
}
ret scalar var_byavg_LD_`var' = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LD_`var' = s_RetainedEigenCountAvg
} /*close option mean */
} /* close for each var */
} /* close if togvar=="" */
****************** PCA VARTOG LEV & DIF TOGETHER ****************
if "`togvar'"!="" {
di "_____ PCA LEV&DIF VAR TOGETHER: `varlistpca'"
local newvarlistld_tog "`newvarlistldl_tog' `newvarlistldd_tog'"
di "`newvarlistld_tog'"
di as text "You are applying PCA to GMM-style LEV & DIF lags of more than one variable,"
di as text "keeping the variables together with different lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistld_tog' if `touse'
}
if "`see'"!="" {
pca `newvarlistld_tog' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistld_tog' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistld_tog' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels & first_differences together for the variables `varlistpca' together _________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
*di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLDtog1-`prefix1'varscoreLDtog`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLDtog1, score
}
ret scalar var_byvar_LD_tog = s_ExplainedVarVar_var
ret scalar nscores_byvar_LD_tog = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLDtog1-`prefix1'avgscoreLDtog`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLDtog1, score
}
ret scalar var_byavg_LD_tog = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LD_tog = s_RetainedEigenCountAvg
} /*close option mean */
} /* close togvar!="" */
} /* close togld!="" */
* POSSO TOGLIERLO?
* if "`retain'"=="" {
* capture drop _GMM*_*_*
* }
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
********** GMM LEV SAME LAGS VAR BY VAR & TOGVAR ********************************
*********************************************************************************
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
if "`gmmliv'"!="" {
/*
if `nvars'==1 {
di as text "You are applying PCA to GMM-style lags of one variable."
}
*/
*else if `nvars'>1 { }
if "`nlagslvars'"!="" {
di as error "Either use gmml or lagsl"
exit 198
}
local nlag1 : list sizeof gmmliv
*di as result "Elements in GMML(): `nlag1'"
if `nlag1'==1 {
local primo_lag1l =`gmmliv'
gmm_style_Ltrunc `varlistpca', nt(`ind_var' `time_var') gmml(`primo_lag1l')
}
if `nlag1'==2 {
tokenize `gmmliv'
local primo_lag1l `1'
local ultimo_lag1l `2'
*tokenize `varlist'
gmm_style_Ltrunc `varlistpca', nt(`ind_var' `time_var') gmml(`primo_lag1l' `ultimo_lag1l')
}
} /* close gmmliv !"== */
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
********** GMM DIF SAME LAGS VAR BY VAR & TOGVAR ********************************
*********************************************************************************
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
if "`gmmdiv'"!="" {
if "`nlagsdvars'"!="" {
di as error "Either use gmmd or lagsd"
exit 198
}
local nlag1 : list sizeof gmmdiv
*di as result "Elements in GMMD(): `nlag1'"
if `nlag1'==1 {
local primo_lag1d =`gmmdiv'
gmm_style_Dtrunc `varlistpca', nt(`ind_var' `time_var') gmmd(`primo_lag1d')
}
if `nlag1'==2 {
tokenize `gmmdiv'
local primo_lag1d `1'
local ultimo_lag1d `2'
*tokenize `varlist'
gmm_style_Dtrunc `varlistpca', nt(`ind_var' `time_var') gmmd(`primo_lag1d' `ultimo_lag1d')
}
} /* close gmmdiv !"== */
************** PCA VAR BY VAR LEV SAME LAGS *******************
if "`togvar'"=="" & "`gmmliv'"!="" {
foreach var of varlist `varlistpca' {
di "_____ PCA LEV VAR BY VAR: `var'"
local newvarlistl_`var' _GMML_`var'_*
local newvarlistldl_`var' _GMML_`var'_*
local newvarlistldlgmm_`var' _GMML_`var'_*
if "`togld'"=="" {
di as text "You are applying PCA to GMM-style LEV lags of one or more than one variable,"
di as text "keeping the variables separated with the same lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistl_`var'' if `touse'
}
if "`see'"!="" {
pca `newvarlistl_`var'' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistl_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistl_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels for `var' __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLEV`var'N1-`prefix1'varscoreLEV`var'N`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLEV`var'N1, score
}
ret scalar var_byvar_LEV_`var' = s_ExplainedVarVar_var
ret scalar nscores_byvar_LEV_`var' = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLEV`var'N1-`prefix1'avgscoreLEV`var'N`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLEV`var'N1, score
}
ret scalar var_byavg_LEV_`var' = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LEV_`var' = s_RetainedEigenCountAvg
} /*close option mean */
} /* close togld=="" */
} /* close for each var =="" */
*if "`retain'"=="" {
* capture drop _GMML_`var'_*
* }
} /* close if togvar=="" */
************** PCA TOGVAR LEV SAME LAGS *******************
if "`togvar'"!="" & "`gmmliv'"!="" {
di "_____ PCA LEV VAR TOGETHER: `varlistpca'"
local newvarlistl_tog _GMML_*_*
local newvarlistldl_tog _GMML_*_*
local newvarlistldlgmm_tog _GMML_*_*
if "`togld'"=="" {
di as text "You are applying PCA to GMM-style LEV lags of one or more than one variable,"
di as text "keeping the variables together with the same lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistl_tog' if `touse'
}
if "`see'"!="" {
pca `newvarlistl_tog' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistl_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistl_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels for the variables `varlistpca' together"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLEVtog1-`prefix1'varscoreLEVtog`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLEVtog1, score
}
ret scalar var_byvar_LEV_tog = s_ExplainedVarVar_var
ret scalar nscores_byvar_LEV_tog = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLEVtog1-`prefix1'avgscoreLEVtog`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLEVtog1, score
}
ret scalar var_byavg_LEV_tog = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LEV_tog = s_RetainedEigenCountAvg
} /*close option mean */
} /*close togld=="" */
*if "`retain'"=="" {
* capture drop _GMML_*_*
* }
} /* close if togvar!="" */
************** PCA VAR BY VAR DIF SAME LAGS *******************
if "`togvar'"=="" & "`gmmdiv'"!="" {
foreach var of varlist `varlistpca' {
di "_____ PCA DIF VAR BY VAR: `var'"
local newvarlistd_`var' _GMMD_`var'_*
local newvarlistldd_`var' _GMMD_`var'_*
local newvarlistlddgmm_`var' _GMMD_`var'_*
if "`togld'"=="" {
di as text "You are applying PCA to GMM-style DIF lags of one or more than one variable,"
di as text "keeping the variables separated with the same lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistd_`var'' if `touse'
}
if "`see'"!="" {
pca `newvarlistd_`var'' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistd_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistd_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in first-differences for `var' __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreDIF`var'N1-`prefix1'varscoreDIF`var'N`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreDIF`var'N1, score
}
ret scalar var_byvar_DIF_`var' = s_ExplainedVarVar_var
ret scalar nscores_byvar_DIF_`var' = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreDIF`var'N1-`prefix1'avgscoreDIF`var'N`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreDIF`var'N1, score
}
ret scalar var_byavg_DIF_`var' = s_ExplainedVarAvg_var
ret scalar nscores_byavg_DIF_`var' = s_RetainedEigenCountAvg
} /*close option mean */
} /* close togld=="" */
} /* close for each var =="" */
*if "`retain'"=="" {
* capture drop _GMMD_`var'_*
* }
} /* close if togvar=="" */
************** PCA TOGVAR DIF SAME LAGS *******************
if "`togvar'"!="" & "`gmmdiv'"!="" {
di "_____ PCA DIF VAR TOGETHER: `varlistpca'"
local newvarlistd_tog _GMMD_*_*
local newvarlistldd_tog _GMMD_*_*
local newvarlistlddgmm_tog _GMMD_*_*
if "`togld'"=="" {
di as text "You are applying PCA to GMM-style DIF lags of one or more than one variable,"
di as text "keeping the variables together with the same lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistd_tog' if `touse'
}
if "`see'"!="" {
pca `newvarlistd_tog' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistd_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistd_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in first-differences for the variables `varlistpca' together"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreDIFtog1-`prefix1'varscoreDIFtog`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreDIFtog1, score
}
ret scalar var_byvar_DIF_tog = s_ExplainedVarVar_var
ret scalar nscores_byvar_DIF_tog = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreDIFtog1-`prefix1'avgscoreDIFtog`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreDIFtog1, score
}
ret scalar var_byavg_DIF_tog = s_ExplainedVarAvg_var
ret scalar nscores_byavg_DIF_tog = s_RetainedEigenCountAvg
} /*close option mean */
} /*close togld=="" */
*if "`retain'"=="" {
* capture drop _GMMD_*_*
* }
} /* close if togvar!="" */
************************** PCA LEV & DIF TOGETHER SAME LAGS VAR BY VAR & TOGVAR *****************
if "`togld'"!="" & "`gmmliv'"!="" & "`gmmdiv'"!="" {
if "`togvar'"=="" {
foreach var of varlist `varlistpca' {
di "_____ PCA LEV&DIF VAR BY VAR: `var'"
local newvarlistld_`var' "`newvarlistldl_`var'' `newvarlistldd_`var''"
di "`newvarlistld_`var''"
di as text "You are applying PCA to GMM-style LEV & DIF lags of more than one variable,"
di as text "keeping the variables separated with the same lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistld_`var'' if `touse'
}
if "`see'"!="" {
pca `newvarlistld_`var'' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistld_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistld_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels & first_differences together for `var' __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLD`var'N1-`prefix1'varscoreLD`var'N`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLD`var'N1, score
}
ret scalar var_byvar_LD_`var' = s_ExplainedVarVar_var
ret scalar nscores_byvar_LD_`var' = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLD`var'N1-`prefix1'avgscoreLD`var'N`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLD`var'N1, score
}
ret scalar var_byavg_LD_`var' = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LD_`var' = s_RetainedEigenCountAvg
} /*close option mean */
} /* close for each var */
} /*close if togvar=="" */
******* PCA LEV & DIF SAME LAGS TOGVAR ***************************
if "`togvar'"!="" {
di "_____ PCA LEV&DIF VAR TOGETHER: `varlistpca'"
local newvarlistld_tog "`newvarlistldl_tog' `newvarlistldd_tog'"
di "`newvarlistld_tog'"
di as text "You are applying PCA to GMM-style LEV & DIF lags of more than one variable,"
di as text "keeping the variables together with the same lags structure"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistld_tog' if `touse'
}
if "`see'"!="" {
pca `newvarlistld_tog' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistld_tog' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistld_tog' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels & first_differences together for the variables `varlistpca' together __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLDtog1-`prefix1'varscoreLDtog`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLDtog1, score
}
ret scalar var_byvar_LD_tog = s_ExplainedVarVar_var
ret scalar nscores_byvar_LD_tog = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLDtog1-`prefix1'avgscoreLDtog`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLDtog1, score
}
ret scalar var_byavg_LD_tog = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LD_tog = s_RetainedEigenCountAvg
} /*close option mean */
} /*close if togvar!="" */
} /*close if togld!="" */
* POSSO TOGLIERLO?
*if "`retain'"=="" {
* capture drop _GMM*_*_*
* }
************************** PCA LEV & DIF TOGETHER DIFFERENT LAGS VAR BY VAR & TOGVAR *****************
if "`togld'"!="" & "`gmmliv'"!="" & "`nlagsdvars'"!="" {
if "`togvar'"=="" {
foreach var of varlist `varlistpca' {
di "_____ PCA LEV&DIF VAR BY VAR: `var'"
local newvarlistld_`var' "`newvarlistldlgmm_`var'' `newvarlistlddlags_`var''"
di "`newvarlistld_`var''"
di as text "You are applying PCA to GMM-style LEV & DIF lags of more than one variable,"
di as text "keeping the variables separated"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistld_`var'' if `touse'
}
if "`see'"!="" {
pca `newvarlistld_`var'' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistld_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistld_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels & first_differences together for `var' __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLD`var'N1-`prefix1'varscoreLD`var'N`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLD`var'N1, score
}
ret scalar var_byvar_LD_`var' = s_ExplainedVarVar_var
ret scalar nscores_byvar_LD_`var' = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLD`var'N1-`prefix1'avgscoreLD`var'N`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLD`var'N1, score
}
ret scalar var_byavg_LD_`var' = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LD_`var' = s_RetainedEigenCountAvg
} /*close option mean */
} /* close for each var */
} /*close if togvar=="" */
******* PCA LEV & DIF SAME LAGS TOGVAR ***************************
if "`togvar'"!="" {
di "_____ PCA LEV&DIF VAR TOGETHER: `varlistpca'"
local newvarlistld_tog "`newvarlistldlgmm_tog' `newvarlistlddlags_tog'"
di "`newvarlistld_tog'"
di as text "You are applying PCA to GMM-style LEV & DIF lags of more than one variable,"
di as text "keeping the variables together"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistld_tog' if `touse'
}
if "`see'"!="" {
pca `newvarlistld_tog' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistld_tog' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistld_tog' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels & first_differences together for the variables `varlistpca' together __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLDtog1-`prefix1'varscoreLDtog`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLDtog1, score
}
ret scalar var_byvar_LD_tog = s_ExplainedVarVar_var
ret scalar nscores_byvar_LD_tog = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLDtog1-`prefix1'avgscoreLDtog`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLDtog1, score
}
ret scalar var_byavg_LD_tog = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LD_tog = s_RetainedEigenCountAvg
} /*close option mean */
} /*close if togvar!="" */
} /*close if togld!="" */
* POSSO TOGLIERLO?
*if "`retain'"=="" {
* capture drop _GMM*_*_*
* }
************************** PCA LEV & DIF TOGETHER DIFFERENT LAGS VAR BY VAR & TOGVAR *****************
if "`togld'"!="" & "`gmmdiv'"!="" & "`nlagslvars'"!="" {
if "`togvar'"=="" {
foreach var of varlist `varlistpca' {
di "_____ PCA LEV&DIF VAR BY VAR: `var'"
local newvarlistld_`var' "`newvarlistldllags_`var'' `newvarlistlddgmm_`var''"
di "`newvarlistld_`var''"
di as text "You are applying PCA to GMM-style LEV & DIF lags of more than one variable,"
di as text "keeping the variables separated"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistld_`var'' if `touse'
}
if "`see'"!="" {
pca `newvarlistld_`var'' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistld_`var'' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistld_`var'' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels & first_differences together for `var' __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLD`var'N1-`prefix1'varscoreLD`var'N`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLD`var'N1, score
}
ret scalar var_byvar_LD_`var' = s_ExplainedVarVar_var
ret scalar nscores_byvar_LD_`var' = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLD`var'N1-`prefix1'avgscoreLD`var'N`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLD`var'N1, score
}
ret scalar var_byavg_LD_`var' = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LD_`var' = s_RetainedEigenCountAvg
} /*close option mean */
} /* close for each var */
} /*close if togvar=="" */
******* PCA LEV & DIF SAME LAGS TOGVAR ***************************
if "`togvar'"!="" {
di "_____ PCA LEV&DIF VAR TOGETHER: `varlistpca'"
local newvarlistld_tog "`newvarlistldllags_tog' `newvarlistlddgmm_tog'"
di "`newvarlistld_tog'"
di as text "You are applying PCA to GMM-style LEV & DIF lags of more than one variable,"
di as text "keeping the variables together"
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `newvarlistld_tog' if `touse'
}
if "`see'"!="" {
pca `newvarlistld_tog' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `newvarlistld_tog' if `touse', covariance
}
if "`see'"!="" {
pca `newvarlistld_tog' if `touse', covariance
}
}
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about PCA of IV in levels & first_differences together for the variables `varlistpca' together __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscoreLDtog1-`prefix1'varscoreLDtog`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscoreLDtog1, score
}
ret scalar var_byvar_LD_tog = s_ExplainedVarVar_var
ret scalar nscores_byvar_LD_tog = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscoreLDtog1-`prefix1'avgscoreLDtog`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscoreLDtog1, score
}
ret scalar var_byavg_LD_tog = s_ExplainedVarAvg_var
ret scalar nscores_byavg_LD_tog = s_RetainedEigenCountAvg
} /*close option mean */
} /*close if togvar!="" */
} /*close if togld!="" */
if "`retain'"=="" {
capture drop _GMM*_*_*
}
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
*********** FINALLY ******************************************************
**** Classical IVs or different variables taken together *****************
*&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
if "`gmmdiv'"=="" & "`gmmliv'"=="" & "`nlagslvars'"=="" & "`nlagsdvars'"=="" {
if `nvars'==1 {
di as error "You have only one variable: you cannot apply PCA on it"
di as error "Use gmm() and/or lags() options if you want GMM-style lags of a variable taken alone"
exit 198
}
else if `nvars'>1 {
di as text "You are applying PCA to different variables or to standard lags of the same variable(s),"
di as text "taken together."
if "`covariance'"=="" {
if "`see'"=="" {
qui pca `varlistpca' if `touse'
}
if "`see'"!="" {
pca `varlistpca' if `touse'
}
}
if "`covariance'"!="" {
if "`see'"=="" {
qui pca `varlistpca' if `touse', covariance
}
if "`see'"!="" {
pca `varlistpca' if `touse', covariance
}
}
} /* close nvars>1 */
***** COMPUTATION OF SCORES AFTER PCA
di as text " _________ Some information about standard PCA for the variables `varlistpca' __________"
scalar `s_trace'=e(trace)
di as result "Trace of the matrix: " _col(80) `s_trace'
if "`variance'"!="" {
local optname variance()
local PV=`variance'
di as result "Percentage of selected variability to be explained: " _col(80) `PV' "%"
}
if `"`PV'"' == "" {
local PV = 90
di as result "By default percentage of selected variability to be explained: " _col(80) `PV' "%"
}
* variability criterion *
scalar s_ExplainedVarVar=0
scalar s_RetainedEigenCountVar=0
local s_trace_PV_perc = (`PV'/100)*`s_trace'
*di as result "Percentage of trace due to the selected variability:" _col(80) `s_trace_PV_perc'
while ( s_ExplainedVarVar < `s_trace_PV_perc' ) {
scalar s_RetainedEigenCountVar = s_RetainedEigenCountVar+1
scalar s_ExplainedVarVar = s_ExplainedVarVar + ( el( e(Ev), 1, s_RetainedEigenCountVar) )
}
*di as result "Sum of eigenvalues corresponding to the retained eigenvectors: " _col(80) s_ExplainedVarVar
scalar s_ExplainedVarVar_var = (s_ExplainedVarVar / `s_trace')*100
di as result "Percentage of variance explained by the variability criterion: " _col(80) s_ExplainedVarVar_var "%"
di as result "Number of retained scores according to the variability criterion: " _col(80) s_RetainedEigenCountVar
if s_RetainedEigenCountVar >1 {
local l_SecondIndex = scalar(s_RetainedEigenCountVar)
qui predict double `prefix1'varscorepca1-`prefix1'varscorepca`l_SecondIndex', score
}
else {
qui predict double `prefix1'varscorepca1, score
}
ret scalar var_byvar_togvar = s_ExplainedVarVar_var
ret scalar nscores_byvar_togvar = s_RetainedEigenCountVar
if "`avg'"!="" {
* average criterion *
scalar s_RetainedEigenCountAvg = 0
scalar s_ExplainedVarAvg = 0
forvalues p=1/`e(f)' {
* this is a trick to find if the eigenvalue is greater that avg(eigenvalues)
scalar delta = ( el(e(Ev),1,`p'))-(`s_trace' / `e(f)')
if delta > 0 {
* autovalore oltre la media
scalar s_RetainedEigenCountAvg = s_RetainedEigenCountAvg+1
scalar s_ExplainedVarAvg = s_ExplainedVarAvg + el(e(Ev),1,`p')
}
else {
* ho esaurito gli autovalori > media quindi esco dal loop forvalues
continue, break
}
}
scalar s_RetainedPC = s_RetainedEigenCountAvg
scalar s_ExplainedVarAvg_var = (s_ExplainedVarAvg / `s_trace')*100
di as result "Percentage of variance explained by the average criterion: " _col(80) s_ExplainedVarAvg_var "%"
di as result "Number of retained scores according to the average criterion: " _col(80) s_RetainedEigenCountAvg
if s_RetainedEigenCountAvg>1 {
local l_ThirdIndex = scalar(s_RetainedEigenCountAvg)
quietly predict double `prefix1'avgscorepca1-`prefix1'avgscorepca`l_ThirdIndex', score
}
else {
quietly predict double `prefix1'avgscorepca1, score
}
ret scalar var_byavg_togvar = s_ExplainedVarAvg_var
ret scalar nscores_byavg_togvar = s_RetainedEigenCountAvg
} /*close option mean */
} /* pca together without gmm/lags */
end
**************************