******************************************************************** ******************************************************************** **** Online supplement: Code for MAIHDA analysis **** Title manuscript: "Intersectional inequalities in **** mental health by education, income, gender, and age **** before and during the COVID-19 pandemic in the **** Netherlands: A longitudinal study" **** **** Author: S.E. Verra **** Adapted from: code developed by C. Evans ******************************************************************** ******************************************************************** ************************************* **** CODEBOOK FOR VARIABLES *** ************************************* /* 1157 observations in 53 strata (of 54 possible) Gender 1 = 'female' 2 = 'male' education 1 = 'low' 2 = 'intermediate' 3 = 'high' income 1 = 'low income' 2 = 'intermediate income' 3 = 'high income' age 1 = '25 to 39 yrs' 2 = '40 to 64 yrs' 3 = '65 yrs and older' Mentalhealth14 = score in 2014 mentalhealth21 = score in 2021 ChangeMH = score 2021 - score 2014 */ ************************************************************** ****** ****** Prep Log and Other Setup Stuff ****** ************************************************************** clear cd "Z:\Stata" capture log close log using "Z:\MAIHDA", replace set more off set cformat %9.1f sysdir set PLUS "Z:\plus" global MLwiN_path "C:\Program Files\MLwiN v3.06\mlwin.exe" global MLwiNScript_path "C:\Program Files\MLwiN v3.06\mlnscript.exe" ************************************************************** ****** ****** Examine Data and Set Key Values ****** ************************************************************** use "Z:\Data aangepast\GLOBE n=1157.dta", clear *** Generate variable which records stratum size bysort stratum: generate n = _N order stratum n *** Construct individual ID variable generate individual = _n order individual, after(stratum) *** Construct Constant and Denomincator Vectors for use in MLwiN *** generate cons = 1 generate denomintator = 1 recast float mentalhealth14 mentalhealth21 ChangeMH, force *** Construct Dummy Variables for Main Effects for Use Later *** tabulate Gender14, gen(dum) rename dum1 female rename dum2 male tabulate age3, gen(dum) rename dum1 young_age rename dum2 mid_age rename dum3 old_age tabulate Edu3_14, gen(dum) rename dum1 low_edu rename dum2 mid_edu rename dum3 high_edu tabulate Inc3_14, gen(dum) rename dum1 low_inc rename dum2 mid_inc rename dum3 high_inc save, replace ****************************************************************** ****** ****** Create Temp Data Set ****** ****************************************************************** use "Z:\Data aangepast\GLOBE n=1157.dta", clear by stratum, sort: generate selectfromlist = (_n==1) drop if selectfromlist == 0 keep stratum low_edu mid_edu high_edu low_inc mid_inc high_inc male female young_age mid_age old_age ChangeMH mentalhealth14 n mentalhealth21 save "strata.dta", replace * count number of strata tab stratum, nofreq display r(r) * 53 strata ****************************************************************** ****** ****** DESCRIPTIVE STATISTICS ****** ****************************************************************** use "Z:\Data aangepast\GLOBE n=1157.dta", clear sort stratum individual *** Examine distribution of outcome variables histogram mentalhealth14, bin(25) percent ytitle(Percentage of participants) xtitle(Mental health score 2014) title(Distribution of mental health score 2014) histogram mentalhealth21, bin(25) percent ytitle(Percentage of participants) xtitle(Mental health score 2021) title(Distribution of mental health score 2021) histogram ChangeMH, bin(25) percent ytitle(Percentage of participants) xtitle(Change in mental health status 2014-2021) title(Distribution of change in mental health score (2014-2021)) *** Examine categories tabulate Gender14 tabulate age3 tabulate Edu3_14 tabulate Inc3_14 *** Distribution of sample across intersectional strata tabulate stratum ****************************************************************** ****** ****** TABLE 1 ****** ****************************************************************** *** Summarize the outcomes across total sample tabstat mentalhealth14, statistics(n mean p50 sd median min max ) format(%3.1f) tabstat mentalhealth21, statistics(n mean p50 sd median min max ) format(%3.1f) tabstat ChangeMH, statistics(n mean p50 sd median min max ) format(%3.1f) *** Summarize the outcomes across each variable of interest sort Gender14 by Gender14: tabstat mentalhealth14, statistics(mean sd) by Gender14: tabstat mentalhealth21, statistics(mean sd) by Gender14: tabstat ChangeMH, statistics(mean sd) sort age3 by age3: tabstat mentalhealth14, statistics(mean sd) by age3: tabstat mentalhealth21, statistics(mean sd) by age3: tabstat ChangeMH, statistics(mean sd) sort Edu3_14 by Edu3_14: tabstat mentalhealth14, statistics(mean sd) by Edu3_14: tabstat mentalhealth21, statistics(mean sd) by Edu3_14: tabstat ChangeMH, statistics(mean sd) sort Inc3_14 by Inc3_14: tabstat mentalhealth14, statistics(mean sd) by Inc3_14: tabstat mentalhealth21, statistics(mean sd) by Inc3_14: tabstat ChangeMH, statistics(mean sd) *** Compare to stratum means collapse (sum) n = cons /// (mean) CMH_stratum_mean = ChangeMH /// (mean) MH14_stratum_mean = mentalhealth14 /// (mean) MH21_stratum_mean = mentalhealth21 , /// by(stratum Gender14 age3 Edu3_14 Inc3_14) tabstat CMH_stratum_mean, statistics(n mean sd median min max) tabstat MH14_stratum_mean, statistics(n mean sd median min max) tabstat MH21_stratum_mean, statistics(n mean sd median min max) ****************************************************************** ****** ****** TABLE 2: stratum-level descriptive statistics ****** ****************************************************************** *** Load the stratum-level data use "strata.dta", clear *** Generate binary indicators for whether each stratum has more than X individuals generate nlessthan5 = (n < 5) generate nmorethan5 = (n >= 5) generate n5to9 = (n >= 5 & n<10) generate n10to14 = (n >= 10 & n<15) generate n15to19 = (n >= 15 & n<20) generate n20to24 = (n >= 20 & n<25) generate n25to29 = (n >= 25 & n<30) generate n30plus = (n >= 30) *** Tabulate the binary indicators tabulate nlessthan5 tabulate n5to9 tabulate n10to14 tabulate n15to19 tabulate n20to24 tabulate n25to29 tabulate n30plus tabulate nmorethan5 sort n list stratum n if n < 10 list stratum n if n < 5 ****************************************************************** ****** ****** Fit models using Bayesian MCMC in MLwiN ****** ****************************************************************** ****************************************************************** ****** ****** OUTCOME 1: MENTAL HEALTH SCORE 2014 ****** ****************************************************************** use "Z:\Data aangepast\GLOBE n=1157.dta", clear recast float mentalhealth14 sort stratum individual *** Fit model using PQL2/quasilikelohood to determine mcmc initialization runmlwin mentalhealth14 cons , /// level2(stratum: cons) /// level1(individual: cons) /// rigls /// nopause *** Fit model using MCMC runmlwin mentalhealth14 cons , /// level2(stratum: cons, residuals(u, savechains("m14a_u.dta",replace))) /// saves residuals for each stratum level1(individual: cons) /// mcmc(burnin(5000) chain(50000) thinning(50) /// savechains("m14a_beta.dta", replace)) initsprevious /// saves beta&variance parameters for the models nopause drop u0se *** Calculate the VPC / ICC from the chains use "m14a_beta.dta", clear rename RP2_var_cons_ sigma2u14A rename RP1_var_cons_ sigma2e14A generate icc = sigma2u14A/(sigma2u14A + sigma2e14A) mcmcsum icc, variables *** store for PCV calculation summarize sigma2u14A scalar sigma2u14A =r(mean) di scalar(sigma2u14A) *------------------------------------------------------------------* * PREPARE FIXED-PART PAREMETER CHAINS 2014A *------------------------------------------------------------------* use "m14a_beta.dta", clear drop deviance RP2_var_cons_ RP1_var_cons_ rename FP1_* b_* format %9.2f b_* compress save "m14a_beta_prepped.dta", replace isid iteration codebook iteration, compact *------------------------------------------------------------------* * PREPARE STRATUM RANDOM EFFECTS CHAINS 2014A *------------------------------------------------------------------* use "m14a_u.dta", clear drop residual idnum rename value u format %9.2f u sort stratum iteration order stratum iteration compress save "m14a_u_prepped.dta", replace isid stratum iteration codebook iteration, compact *------------------------------------------------------------------* * MERGE DATA, FIXED-PART PARAMETER AND RANDOM EFFECT CHAINS TOGETHER 2014A *------------------------------------------------------------------* use "strata.dta", clear isid stratum cross using "m14a_beta_prepped.dta" isid stratum iteration sort stratum iteration merge 1:1 stratum iteration using "m14a_u_prepped.dta", nogenerate assert(match) isid stratum iteration compress save "m14a_data_prepped.dta", replace *------------------------------------------------------------------* * CALCULATE VALUES OF INTEREST 2014A *------------------------------------------------------------------* *** Expected value based on fixed and random part use "m14a_data_prepped.dta", clear gen cons = 1 generate expectedvalue = b_cons*cons + u label var expectedvalue "Expected value based on main effects and interactions" format %9.3f expectedvalue *** Expected value based only on the fixed-part generate fixedeffect = b_cons*cons label var fixedeffect "Expected value based only on the intercept" format %9.3f fixedeffect *** Expected value based only on the random-part generate randomeffect = u label var randomeffect "Random Effect" format %9.3f randomeffect *** Calculate the mean, 2.5th and 97.5th percentiles of the MCMC chains bysort stratum (iteration): egen expmn = mean(expectedvalue) bysort stratum (iteration): egen explo = pctile(expectedvalue), p(2.5) bysort stratum (iteration): egen exphi = pctile(expectedvalue), p(97.5) format %9.3f expmn explo exphi bysort stratum (iteration): egen FEmn = mean(fixedeffect) bysort stratum (iteration): egen FElo = pctile(fixedeffect), p(2.5) bysort stratum (iteration): egen FEhi = pctile(fixedeffect), p(97.5) format %9.3f FEmn FElo FEhi bysort stratum (iteration): egen REmn = mean(randomeffect) bysort stratum (iteration): egen RElo = pctile(randomeffect), p(2.5) bysort stratum (iteration): egen REhi = pctile(randomeffect), p(97.5) format %9.3f REmn RElo REhi *** Drop chains and just keep their summaries (mean, 2.5th and 97.5th) drop iteration b* u* expectedvalue fixedeffect randomeffect duplicates drop isid stratum *** Ranks sort expmn generate exprank = _n order exprank, after(exphi) sort FEmn generate FErank = _n order FErank, after(FEhi) sort REmn generate RErank = _n order RErank, after(REhi) *** Sort the data sort stratum isid stratum *** Compress and save the data compress save "m14a_results.dta", replace ******************************************************************** * MODEL 2 - MAIN EFFECTS MODEL 2014B ******************************************************************** *** Load the data use "Z:\Data aangepast\GLOBE n=1157.dta", clear recast float mentalhealth14 sort stratum individual *** Set output format of coefficients, SEs, and confidence limits to 1dp set cformat %9.1f *** Fit model using PQL2 runmlwin mentalhealth14 cons female young_age mid_age low_edu mid_edu low_inc mid_inc , /// level2(stratum: cons) /// level1(individual: cons) /// rigls /// nopause *** Fit model using MCMC runmlwin mentalhealth14 cons female young_age mid_age low_edu mid_edu low_inc mid_inc , /// level2(stratum: cons, residuals(u, savechains("m14b_u.dta", replace))) /// level1(individual: cons) /// mcmc(burnin(5000) chain(50000) thinning(50) /// savechains("m14b_beta.dta", replace)) initsprevious /// nopause drop u0se *** Calculate the VPC / ICC from the chains use "m14b_beta.dta", clear rename RP2_var_cons_ sigma2u14B rename RP1_var_cons_ sigma2e14B generate icc = 100*sigma2u14B/(sigma2u14B + sigma2e14B) mcmcsum icc, variables *** Calculate the Proportional Change in Variance (PCV) (as a percentage) generate pcv = 100*((sigma2u14A - sigma2u14B) / sigma2u14A) summarize pcv *------------------------------------------------------------------* * PREPARE FIXED-PART PAREMETER CHAINS 2014B *------------------------------------------------------------------* use "m14b_beta.dta", clear drop deviance RP2_var_cons_ RP1_var_cons_ rename FP1_* b_* format %9.2f b_* compress save "m14b_beta_prepped.dta", replace isid iteration codebook iteration, compact *------------------------------------------------------------------* * PREPARE STRATUM RANDOM EFFECTS CHAINS 2014B *------------------------------------------------------------------* use "m14b_u.dta", clear drop residual idnum rename value u format %9.2f u sort stratum iteration order stratum iteration compress save "m14b_u_prepped.dta", replace isid stratum iteration codebook iteration, compact *------------------------------------------------------------------* * MERGE DATA, FIXED-PART PARAMETER AND RANDOM EFFECT CHAINS TOGETHER 2014B *------------------------------------------------------------------* use "strata.dta", clear isid stratum cross using "m14b_beta_prepped.dta" isid stratum iteration sort stratum iteration merge 1:1 stratum iteration using "m14b_u_prepped.dta", nogenerate assert(match) isid stratum iteration compress save "m14b_data_prepped.dta", replace *------------------------------------------------------------------* * CALCULATE VALUES OF INTEREST 2014B *------------------------------------------------------------------* *** Expected value based on fixed and random part use "m14b_data_prepped.dta", clear gen cons = 1 generate expectedvalue = (b_cons*cons /// + b_female*female /// + b_young_age*young_age /// + b_mid_age*mid_age /// + b_low_edu*low_edu /// + b_mid_edu*mid_edu /// + b_low_inc*low_inc /// + b_mid_inc*mid_inc /// + u ) label var expectedvalue "Expected value based on main effects and interactions" format %9.3f expectedvalue *** Expected value based only on the fixed-part generate fixedeffect = (b_cons*cons /// + b_female*female /// + b_young_age*young_age /// + b_mid_age*mid_age /// + b_low_edu*low_edu /// + b_mid_edu*mid_edu /// + b_low_inc*low_inc /// + b_mid_inc*mid_inc /// ) label var fixedeffect "Expected value based only on main effects" format %9.3f fixedeffect *** Expected value based only on the random-part generate randomeffect = u label var randomeffect "Random Effect" format %9.3f randomeffect *** Calculate the mean, 2.5th and 97.5th percentiles of the MCMC chains bysort stratum (iteration): egen expmn = mean(expectedvalue) bysort stratum (iteration): egen explo = pctile(expectedvalue), p(2.5) bysort stratum (iteration): egen exphi = pctile(expectedvalue), p(97.5) format %9.3f expmn explo exphi bysort stratum (iteration): egen FEmn = mean(fixedeffect) bysort stratum (iteration): egen FElo = pctile(fixedeffect), p(2.5) bysort stratum (iteration): egen FEhi = pctile(fixedeffect), p(97.5) format %9.3f FEmn FElo FEhi bysort stratum (iteration): egen REmn = mean(randomeffect) bysort stratum (iteration): egen RElo = pctile(randomeffect), p(2.5) bysort stratum (iteration): egen REhi = pctile(randomeffect), p(97.5) format %9.3f REmn RElo REhi *** Drop chains and just keep their summaries (mean, 2.5th and 97.5th) drop iteration b* u* expectedvalue fixedeffect randomeffect duplicates drop isid stratum *** Ranks sort expmn generate exprank = _n order exprank, after(exphi) sort FEmn generate FErank = _n order FErank, after(FEhi) sort REmn generate RErank = _n order RErank, after(REhi) *** Sort the data sort stratum isid stratum *** Compress and save the data compress save "m14b_results.dta", replace *** List strata with statistically significant interaction effects use "m14b_results.dta", clear list stratum REmn RElo REhi if RElo<0 list stratum REmn RElo REhi if RElo>0 ****************************************************************** ****** ****** OUTCOME 2: MENTAL HEALTH SCORE 2021 ****** ****************************************************************** ****************************************************************** *** MODEL 1, Null MODEL for CHANGE IN MENTAL HEALTH 2021A ****************************************************************** *** Load the data use "Z:\Data aangepast\GLOBE n=1157.dta", clear sort stratum individual *** Set output format of coefficients, SEs, and confidence limits to 1dp set cformat %9.1f *** Fit model using PQL2/quasilikelohood to determine mcmc initialization runmlwin mentalhealth21 cons , /// level2(stratum: cons) /// level1(individual: cons) /// rigls /// nopause *** Fit model using MCMC runmlwin mentalhealth21 cons , /// level2(stratum: cons, residuals(u, savechains("m21a_u.dta",replace))) /// saves residuals for each stratum level1(individual: cons) /// mcmc(burnin(5000) chain(50000) thinning(50) /// savechains("m21a_beta.dta", replace)) initsprevious /// saves beta&variance parameters for the models nopause drop u0se *** Calculate the VPC / ICC from the chains use "m21a_beta.dta", clear rename RP2_var_cons_ sigma2u21A rename RP1_var_cons_ sigma2e21A generate icc = sigma2u21A/(sigma2u21A + sigma2e21A) mcmcsum icc, variables *** store for PCV calculation summarize sigma2u21A scalar sigma2u21A =r(mean) di scalar(sigma2u21A) *------------------------------------------------------------------* * PREPARE FIXED-PART PAREMETER CHAINS 2021A *------------------------------------------------------------------* use "m21a_beta.dta", clear drop deviance RP2_var_cons_ RP1_var_cons_ rename FP1_* b_* format %9.2f b_* compress save "m21a_beta_prepped.dta", replace isid iteration codebook iteration, compact *------------------------------------------------------------------* * PREPARE STRATUM RANDOM EFFECTS CHAINS 2021A *------------------------------------------------------------------* use "m21a_u.dta", clear drop residual idnum rename value u format %9.2f u sort stratum iteration order stratum iteration compress save "m21a_u_prepped.dta", replace isid stratum iteration codebook iteration, compact *------------------------------------------------------------------* * MERGE DATA, FIXED-PART PARAMETER AND RANDOM EFFECT CHAINS TOGETHER 2021A *------------------------------------------------------------------* use "strata.dta", clear isid stratum cross using "m21a_beta_prepped.dta" isid stratum iteration sort stratum iteration merge 1:1 stratum iteration using "m21a_u_prepped.dta", nogenerate assert(match) isid stratum iteration compress save "m21a_data_prepped.dta", replace *------------------------------------------------------------------* * CALCULATE VALUES OF INTEREST 2021A *------------------------------------------------------------------* *** Expected value based on fixed and random part use "m21a_data_prepped.dta", clear gen cons = 1 generate expectedvalue = b_cons*cons + u label var expectedvalue "Expected value based on main effects and interactions" format %9.3f expectedvalue *** Expected value based only on the fixed-part generate fixedeffect = b_cons*cons label var fixedeffect "Expected value based only on the intercept" format %9.3f fixedeffect *** Expected value based only on the random-part generate randomeffect = u label var randomeffect "Random Effect" format %9.3f randomeffect *** Calculate the mean, 2.5th and 97.5th percentiles of the MCMC chains bysort stratum (iteration): egen expmn = mean(expectedvalue) bysort stratum (iteration): egen explo = pctile(expectedvalue), p(2.5) bysort stratum (iteration): egen exphi = pctile(expectedvalue), p(97.5) format %9.3f expmn explo exphi bysort stratum (iteration): egen FEmn = mean(fixedeffect) bysort stratum (iteration): egen FElo = pctile(fixedeffect), p(2.5) bysort stratum (iteration): egen FEhi = pctile(fixedeffect), p(97.5) format %9.3f FEmn FElo FEhi bysort stratum (iteration): egen REmn = mean(randomeffect) bysort stratum (iteration): egen RElo = pctile(randomeffect), p(2.5) bysort stratum (iteration): egen REhi = pctile(randomeffect), p(97.5) format %9.3f REmn RElo REhi *** Drop chains and just keep their summaries (mean, 2.5th and 97.5th) drop iteration b* u* expectedvalue fixedeffect randomeffect duplicates drop isid stratum *** Ranks sort expmn generate exprank = _n order exprank, after(exphi) sort FEmn generate FErank = _n order FErank, after(FEhi) sort REmn generate RErank = _n order RErank, after(REhi) *** Sort the data sort stratum isid stratum *** Compress and save the data compress save "m21a_results.dta", replace ******************************************************************** * MODEL 2 - MAIN EFFECTS MODEL 2021B ******************************************************************** *** Load the data use "Z:\Data aangepast\GLOBE n=1157.dta", clear sort stratum individual *** Set output format of coefficients, SEs, and confidence limits to 1dp set cformat %9.1f *** Fit model using PQL2 runmlwin mentalhealth21 cons female young_age mid_age low_edu mid_edu low_inc mid_inc , /// level2(stratum: cons) /// level1(individual: cons) /// rigls /// nopause *** Fit model using MCMC runmlwin mentalhealth21 cons female young_age mid_age low_edu mid_edu low_inc mid_inc , /// level2(stratum: cons, residuals(u, savechains("m21b_u.dta", replace))) /// level1(individual: cons) /// mcmc(burnin(5000) chain(50000) thinning(50) /// savechains("m21b_beta.dta", replace)) initsprevious /// nopause drop u0se *** Calculate the VPC / ICC from the chains use "m21b_beta.dta", clear rename RP2_var_cons_ sigma2u21B rename RP1_var_cons_ sigma2e21B generate icc = sigma2u21B/(sigma2u21B + sigma2e21B) mcmcsum icc, variables *** Calculate the Proportional Change in Variance (PCV) (as a percentage) generate pcv = 100*((sigma2u21A - sigma2u21B) / sigma2u21A) summarize pcv *------------------------------------------------------------------* * PREPARE FIXED-PART PAREMETER CHAINS 2021B *------------------------------------------------------------------* use "m21b_beta.dta", clear drop deviance RP2_var_cons_ RP1_var_cons_ rename FP1_* b_* format %9.2f b_* compress save "m21b_beta_prepped.dta", replace isid iteration codebook iteration, compact *------------------------------------------------------------------* * PREPARE STRATUM RANDOM EFFECTS CHAINS 2021B *------------------------------------------------------------------* use "m21b_u.dta", clear drop residual idnum rename value u format %9.2f u sort stratum iteration order stratum iteration compress save "m21b_u_prepped.dta", replace isid stratum iteration codebook iteration, compact *------------------------------------------------------------------* * MERGE DATA, FIXED-PART PARAMETER AND RANDOM EFFECT CHAINS TOGETHER 2021B *------------------------------------------------------------------* use "strata.dta", clear isid stratum cross using "m21b_beta_prepped.dta" isid stratum iteration sort stratum iteration merge 1:1 stratum iteration using "m21b_u_prepped.dta", nogenerate assert(match) isid stratum iteration compress save "m21b_data_prepped.dta", replace *------------------------------------------------------------------* * CALCULATE VALUES OF INTEREST 2021B *------------------------------------------------------------------* *** Expected value based on fixed and random part use "m21b_data_prepped.dta", clear gen cons = 1 generate expectedvalue = (b_cons*cons /// + b_female*female /// + b_young_age*young_age /// + b_mid_age*mid_age /// + b_low_edu*low_edu /// + b_mid_edu*mid_edu /// + b_low_inc*low_inc /// + b_mid_inc*mid_inc /// + u ) label var expectedvalue "Expected value based on main effects and interactions" format %9.3f expectedvalue *** Expected value based only on the fixed-part generate fixedeffect = (b_cons*cons /// + b_female*female /// + b_young_age*young_age /// + b_mid_age*mid_age /// + b_low_edu*low_edu /// + b_mid_edu*mid_edu /// + b_low_inc*low_inc /// + b_mid_inc*mid_inc /// ) label var fixedeffect "Expected value based only on main effects" format %9.3f fixedeffect *** Expected value based only on the random-part generate randomeffect = u label var randomeffect "Random Effect" format %9.3f randomeffect *** Calculate the mean, 2.5th and 97.5th percentiles of the MCMC chains bysort stratum (iteration): egen expmn = mean(expectedvalue) bysort stratum (iteration): egen explo = pctile(expectedvalue), p(2.5) bysort stratum (iteration): egen exphi = pctile(expectedvalue), p(97.5) format %9.3f expmn explo exphi bysort stratum (iteration): egen FEmn = mean(fixedeffect) bysort stratum (iteration): egen FElo = pctile(fixedeffect), p(2.5) bysort stratum (iteration): egen FEhi = pctile(fixedeffect), p(97.5) format %9.3f FEmn FElo FEhi bysort stratum (iteration): egen REmn = mean(randomeffect) bysort stratum (iteration): egen RElo = pctile(randomeffect), p(2.5) bysort stratum (iteration): egen REhi = pctile(randomeffect), p(97.5) format %9.3f REmn RElo REhi *** Drop chains and just keep their summaries (mean, 2.5th and 97.5th) drop iteration b* u* expectedvalue fixedeffect randomeffect duplicates drop isid stratum *** Ranks sort expmn generate exprank = _n order exprank, after(exphi) sort FEmn generate FErank = _n order FErank, after(FEhi) sort REmn generate RErank = _n order RErank, after(REhi) *** Sort the data sort stratum isid stratum *** Compress and save the data compress save "m21b_results.dta", replace *** List strata with statistically significant interaction effects use "m21b_results.dta", clear list stratum REmn RElo REhi if REhi<0 list stratum REmn RElo REhi if RElo>0 *** Creating data files for a figure of MH change per intersection use "m21b_results.dta", clear keep stratum n expmn explo exphi FEmn FElo FEhi REmn RElo REhi sort stratum compress save "scores2021.dta", replace use "m14b_results.dta", clear keep stratum n expmn explo exphi FEmn FElo FEhi REmn RElo REhi sort stratum compress save "scores2014.dta", replace ****************************************************************** ****** ****** OUTCOME 3: CHANGE IN MENTAL HEALTH SCORE ****** 2014 - 2021 ****** ****************************************************************** ****************************************************************** * MODEL 1, Null MODEL for CHANGE IN MENTAL HEALTH ****************************************************************** use "Z:\Data aangepast\GLOBE n=1157.dta", clear sort stratum individual *** Set output format of coefficients, SEs, and confidence limits to 1dp set cformat %9.1f *** Fit model using PQL2/quasilikelohood to determine mcmc initialization runmlwin ChangeMH cons , /// level2(stratum: cons) /// level1(individual: cons) /// rigls /// nopause *** Fit model using MCMC runmlwin ChangeMH cons , /// level2(stratum: cons, residuals(u, savechains("m1_u.dta",replace))) /// saves residuals for each stratum level1(individual: cons) /// mcmc(burnin(5000) chain(50000) thinning(50) /// savechains("m1_beta.dta", replace)) initsprevious /// saves beta&variance parameters for the models nopause drop u0se *** Calculate the VPC / ICC from the chains use "m1_beta.dta", clear rename RP2_var_cons_ sigma2uCA rename RP1_var_cons_ sigma2eCA generate icc = sigma2uCA/(sigma2uCA + sigma2eCA) mcmcsum icc, variables *** store for PCV calculation summarize sigma2uCA scalar sigma2uCA =r(mean) di scalar(sigma2uCA) *------------------------------------------------------------------* * PREPARE FIXED-PART PAREMETER CHAINS CMHA *------------------------------------------------------------------* use "m1_beta.dta", clear drop deviance RP2_var_cons_ RP1_var_cons_ rename FP1_* b_* format %9.2f b_* compress save "m1_beta_prepped.dta", replace isid iteration codebook iteration, compact *------------------------------------------------------------------* * PREPARE STRATUM RANDOM EFFECTS CHAINS CMHA *------------------------------------------------------------------* use "m1_u.dta", clear drop residual idnum rename value u format %9.2f u sort stratum iteration order stratum iteration compress save "m1_u_prepped.dta", replace isid stratum iteration codebook iteration, compact *------------------------------------------------------------------* * MERGE DATA, FIXED-PART PARAMETER AND RANDOM EFFECT CHAINS TOGETHER CMHA *------------------------------------------------------------------* use "strata.dta", clear isid stratum cross using "m1_beta_prepped.dta" isid stratum iteration sort stratum iteration merge 1:1 stratum iteration using "m1_u_prepped.dta", nogenerate assert(match) isid stratum iteration compress save "m1_data_prepped.dta", replace *------------------------------------------------------------------* * CALCULATE VALUES OF INTEREST CMHA *------------------------------------------------------------------* *** Expected value based on fixed and random part use "m1_data_prepped.dta", clear gen cons = 1 generate expectedvalue = b_cons*cons + u label var expectedvalue "Expected value based on main effects and interactions" format %9.3f expectedvalue *** Expected value based only on the fixed-part generate fixedeffect = b_cons*cons label var fixedeffect "Expected value based only on the intercept" format %9.3f fixedeffect *** Expected value based only on the random-part generate randomeffect = u label var randomeffect "Random Effect" format %9.3f randomeffect *** Calculate the mean, 2.5th and 97.5th percentiles of the MCMC chains bysort stratum (iteration): egen expmn = mean(expectedvalue) bysort stratum (iteration): egen explo = pctile(expectedvalue), p(2.5) bysort stratum (iteration): egen exphi = pctile(expectedvalue), p(97.5) format %9.3f expmn explo exphi bysort stratum (iteration): egen FEmn = mean(fixedeffect) bysort stratum (iteration): egen FElo = pctile(fixedeffect), p(2.5) bysort stratum (iteration): egen FEhi = pctile(fixedeffect), p(97.5) format %9.3f FEmn FElo FEhi bysort stratum (iteration): egen REmn = mean(randomeffect) bysort stratum (iteration): egen RElo = pctile(randomeffect), p(2.5) bysort stratum (iteration): egen REhi = pctile(randomeffect), p(97.5) format %9.3f REmn RElo REhi *** Drop chains and just keep their summaries (mean, 2.5th and 97.5th) drop iteration b* u* expectedvalue fixedeffect randomeffect duplicates drop isid stratum *** Ranks sort expmn generate exprank = _n order exprank, after(exphi) sort FEmn generate FErank = _n order FErank, after(FEhi) sort REmn generate RErank = _n order RErank, after(REhi) *** Sort the data sort stratum isid stratum *** Compress and save the data compress save "m1_results.dta", replace ******************************************************************** * MODEL 2 - MAIN EFFECTS MODEL CMHB ******************************************************************** *** Load the data use "Z:\Data aangepast\GLOBE n=1157.dta", clear sort stratum individual *** Set output format of coefficients, SEs, and confidence limits to 1dp set cformat %9.1f *** Fit model using PQL2 runmlwin ChangeMH cons female young_age mid_age low_edu mid_edu low_inc mid_inc , /// level2(stratum: cons) /// level1(individual: cons) /// rigls /// nopause *** Fit model using MCMC runmlwin ChangeMH cons female young_age mid_age low_edu mid_edu low_inc mid_inc , /// level2(stratum: cons, residuals(u, savechains("m2_u.dta", replace))) /// level1(individual: cons) /// mcmc(burnin(5000) chain(50000) thinning(50) /// savechains("m2_beta.dta", replace)) initsprevious /// nopause drop u0se *** Calculate the VPC / ICC from the chains use "m2_beta.dta", clear rename RP2_var_cons_ sigma2uCB rename RP1_var_cons_ sigma2eCB generate icc = sigma2uCB /(sigma2uCB + sigma2eCB) mcmcsum icc, variables *** Calculate the Proportional Change in Variance (PCV) (as a percentage) generate pcv = 100*((sigma2uCA - sigma2uCB) / sigma2uCA) summarize pcv *------------------------------------------------------------------* * PREPARE FIXED-PART PAREMETER CHAINS CMHB *------------------------------------------------------------------* use "m2_beta.dta", clear drop deviance RP2_var_cons_ RP1_var_cons_ rename FP1_* b_* format %9.2f b_* compress save "m2_beta_prepped.dta", replace isid iteration codebook iteration, compact *------------------------------------------------------------------* * PREPARE STRATUM RANDOM EFFECTS CHAINS CMHB *------------------------------------------------------------------* use "m2_u.dta", clear drop residual idnum rename value u format %9.2f u sort stratum iteration order stratum iteration compress save "m2_u_prepped.dta", replace isid stratum iteration codebook iteration, compact *------------------------------------------------------------------* * MERGE DATA, FIXED-PART PARAMETER AND RANDOM EFFECT CHAINS TOGETHER CMHB *------------------------------------------------------------------* use "strata.dta", clear isid stratum cross using "m2_beta_prepped.dta" isid stratum iteration sort stratum iteration merge 1:1 stratum iteration using "m2_u_prepped.dta", nogenerate assert(match) isid stratum iteration compress save "m2_data_prepped.dta", replace *------------------------------------------------------------------* * CALCULATE VALUES OF INTEREST CMHB *------------------------------------------------------------------* *** Expected value based on fixed and random part use "m2_data_prepped.dta", clear gen cons = 1 generate expectedvalue = (b_cons*cons /// + b_female*female /// + b_young_age*young_age /// + b_mid_age*mid_age /// + b_low_edu*low_edu /// + b_mid_edu*mid_edu /// + b_low_inc*low_inc /// + b_mid_inc*mid_inc /// + u ) label var expectedvalue "Expected value based on main effects and interactions" format %9.3f expectedvalue *** Expected value based only on the fixed-part generate fixedeffect = (b_cons*cons /// + b_female*female /// + b_young_age*young_age /// + b_mid_age*mid_age /// + b_low_edu*low_edu /// + b_mid_edu*mid_edu /// + b_low_inc*low_inc /// + b_mid_inc*mid_inc /// ) label var fixedeffect "Expected value based only on main effects" format %9.3f fixedeffect *** Expected value based only on the random-part generate randomeffect = u label var randomeffect "Random Effect" format %9.3f randomeffect *** Calculate the mean, 2.5th and 97.5th percentiles of the MCMC chains bysort stratum (iteration): egen expmn = mean(expectedvalue) bysort stratum (iteration): egen explo = pctile(expectedvalue), p(2.5) bysort stratum (iteration): egen exphi = pctile(expectedvalue), p(97.5) format %9.3f expmn explo exphi bysort stratum (iteration): egen FEmn = mean(fixedeffect) bysort stratum (iteration): egen FElo = pctile(fixedeffect), p(2.5) bysort stratum (iteration): egen FEhi = pctile(fixedeffect), p(97.5) format %9.3f FEmn FElo FEhi bysort stratum (iteration): egen REmn = mean(randomeffect) bysort stratum (iteration): egen RElo = pctile(randomeffect), p(2.5) bysort stratum (iteration): egen REhi = pctile(randomeffect), p(97.5) format %9.3f REmn RElo REhi *** Drop chains and just keep their summaries (mean, 2.5th and 97.5th) drop iteration b* u* expectedvalue fixedeffect randomeffect duplicates drop isid stratum *** Ranks sort expmn generate exprank = _n order exprank, after(exphi) sort FEmn generate FErank = _n order FErank, after(FEhi) sort REmn generate RErank = _n order RErank, after(REhi) *** Sort the data sort stratum isid stratum *** Compress and save the data compress save "m2_results.dta", replace *** List strata with statistically significant interaction effects use "m2_results.dta", clear list stratum REmn RElo REhi if REhi<0 list stratum REmn RElo REhi if RElo>0 *** Close log file capture log close ******************************************************************************** exit