Estimation of the Variance Using Bootstrap Weights User's Guide for
Contents
1. E 2 The Final Weight variable included in the bootstrap weight file a RY 3 The prefix of the boostraps weight variables kx bi 4 The number of bootstrap weights to use note For testing B must be gt 2 km IT IS NECESSARY TO USE ALL THE BOOTSTRAP WEIGHTS WHEN PERFORMING THE FINAL a ANALYSIS THE COMPLETE BOOTVARE_V30 SAS PROGRAM MUST THEN BE RUN Fx Refer to AppendixC XYZ to obtain this information bi seve S VR abe aes ERN UR A Pd ERU DR eee ad ERE ERRER 309 et ident er personi d et fwgt fwgt Wet bsw bsw et B 500 XCKOkCk KK kk EE OKOE OK OK C ECKE KK KKK KAKA KK EC EO OK EEEE EEEE GEEK Ge GG GG EGG KG e I E GGG AK KK AK KKK SPECIFY THE DIRECTORY AND THE NAME OF THE FILE THAT CONTAINS THE MACROS THE PROGRAM MACROE V30 5AS IF NO MODIFICATIONS HAVE BEEN MADE BY THE USER a EHE ES SREA ES ELELL VENT AER OAS TEMERE PEE EER CHEER ERE RDADAREARARE E i nclude C VBOOTVAR MACROE V30 SAS 11 EEELE SEER LAL ROC RO EORR OE ODER OO EORODRE OPER SE TOR ROO ER Se OR Ne ee eR ee ee a RUE Dek SECTION 2 e RERLERERLERERA ERE AER EROR ORG GR OE RE UE PEER EERE EER ARLE RA ERE OR EROR OE EROR OR ERU S EU e e f x KX xk This section lets the user specify the different analyses of interest x KX xk RERREREERE ER OE REE ERA AE REAE EER ER EERE EERE ERASE ERE S ERASERS ep e ERE RAE RE REEF TO OBTAIN VARIANCE ESTIMATES OF A TOTAL RUN St otal variable name t ot al dia2. The prefix of the boostraps weight variables R vx 4 The number of bootstrap weights to use note For testing B must be gt 2 tt IT 1S NECESSARY TO USE ALL THE BOOTSTRAP WEIGHTS WHEN PERFORMING THE FINAL k2 ANALYSIS THE COMPLETE BOOTVARE V30 SAS PROGRAM MUST THEN BE RUN ue x x ae Refer to AppendixC_XYZ to obtain this information ae Ss ese Va AVES ERREUR ETT EREDHERCEEER EDEN i ee ee et ident unique identifier variable s et fwgt final weighf l et bsw prefix_of_bootstrap_weight_variables et B number of weights to use COE OK GO OE GE GO GE EE OG OG GEO E COE GO GG E EEEE GG GG EG GO EEEE EEEE e X GG EEEE EEEE E EEEE I e KK KKK SPECIFY THE DIRECTORY AND THE NAME OF THE FILE THAT CONTAINS THE MACROS xx THE PROGRAM MACROE V30 SAS IF NO MODIFICATIONS HAVE BEEN MADE BY THE USER pe Idee iu MUR Hi include directory name of macroe V30 sas MACROE V30 SAS 5 CECI Roo REET RRR ERE RAE EERE SE JOE E ERASER ERE ER RAE E RRR EER EKG GIO eee eee eer nae SECTION 2 ex REARAA ERE REOR ERAES LEAR RARER ERLE ELE RELL A EARL Oeo do e E oe ERLE AREER A ERE FL x X x x This section lets the user specify the different analyses of interest j RS RR eR RUE ESRB QEGEESRUR AERE ee ene he ee mam ee GEGESRORUB ean eR ee RR eee RR ee eR ae RI De eR FORE ON E TO OBTAIN VARIANCE ESTIMATES OF A TOTAL RUN Wtotal variable name TO OBTAIN VARIANCE ESTI MATES OF A RATIO RUN
3. Watio numerator variable denominator variable TO OBTAIN VARIANCE ESTI MATES OF A DIFFERENCE BETWEEN RATIOS RUN NOTE see the comment at the beginning of section 2 Wliff rat VARI VAR2 VAR3 VAR4 where varl the numerator variable of the first ratio var2 the denominator variable of the first ratio var3 the numerator variable of the second ratio var4 the denominator variable of the second ratio TO OBTAIN VARIANCE ESTIMATES OF REGRESSION PARAMETERS RUN Wregress dependent variable independent variables no comma TO OBTAIN VARIANCE ESTIMATES OF LOGISTIC REGRESSION PARAMETERS RUN l ogreg dependent variable independent variables no comma woutput Displays the results on the screen Do not modify TO SAVE THE RESULTS IN A FILE RUN remove the data out Results filename set amp result fon end of BOOTVARE V30 SAS program i i Appendix A Appendix B APPENDIX B This is a complete example showing how to use the program BOOTVARE_V30 SAS First the analysis data file is created step 1 Then BOOTVARE_V30 SAS is adapted to obtain the desired analysis The results that are produced follow the programs Example This example uses the cycle 3 1998 cross sectional file of the National Population Health Survey general component This example 1 Computes the total number and the proportion of diabetics in the population
4. else diab 0 sex if dhc8 sexzl then males l else males 0 if dhc8 sexz2 then femaleszl else females 0 diabetes sex mdiab diab males male diabetics fdiab diab females female diabetics keep ist of variables to keep s recommended that only the necessary variables be kept rder to reduce the runtime of BOOTVARE_V30 SAS RTANT the identification variables and if necessary breakdown variable ex province sex must be kept The ht variable also must be kept if point estimates are ulated at this step 4 Ke 3 3 3X OH run KKAK AKA KAKA RR KK KAR ke KARA KAKA KK e ek RK e x e e e e Ge KK Calculation of point estimates Suggested but not required x RRERRAR ER AES ERA ERE EER EKER EER AR ERK ERK S proc freq datazinl Name of analysis file table variables of interest weight weight variable run proc logistic datazinl Name of analysis file model dependent variable independent variables weight weight variable run Appendix A BOOTVARE V30 SAS Program The parts in bold need to be changed RRR IO XO XO OO OE RR OE OO OE IO OO OO E OI OO E OI OO E OE IO E EGO RK KR RRR KR RR KK IG EGG J dupl SECTION 1 dia PERE ERLE ERE EGEEOR EAE SE hee ee Wee eee A aoe Roe eR ce Re ee eR oie Ce eRe ee eR SI aR RL obe ROR eR e dee x Yo ER This section lets the user specify the different parameters of ids pes interest variable names directory names
5. file names etc Ber fees Xx RRR EO OO E OE OO OE IO OE OO E OI OO E GIO E OE IG RK RR RK RK RRR RR RR KK GEN J KAKA KAKA KKK ok k ok ke ok ok ok kk ke k ek ek e e x e e e oe ee A KK e e e KA KR KK KER KKH KKK SPECIFY THE NAME OF THE FOLLOWING 2 DIRECTORIES directories ox o AERA SARE OO ODE OE REE AER ERE RS AERE ERED ERE ERLE EEA ERK ERA EA ERR li bname inl name of the directory containing analysis file step 1 s ex c data libname out name of the directory to save results in E ex ce output CK Ck Kk e eG OG ko e o e e e e OG Sk Ge e eG e GG KG e eG KKK e eG S S GS X e x x X B Xx X Xx x x X x X X X X Xx X x KK xv x x KK SPECIFY THE NAME OF THE ANALYSIS FILE CREATED IN STEP 1 without extension CK CK Kk e eG KKK eG e e ok e ke e e Ok S e ke eG e E OG e eG GG S e e e KR S KKK X e x x X x Xx X X x x Xx X X X x e X X KK KKK KK Wet Mfile inl Name of analysis file XC kk KA X KG X X GG GO E GE KR Ge eG e RK GG GE GO EG E RK KA RK KA RR KR e KR RK oe oe oe oe Xe e x x SPECIFY THE NAME OF THE FILE CONTAINING THE BOOTSTRAP WEI GHTS 4 du Only run one of the two following series of commands a F comment the other one out or erase it EEXOOODOOGODDOEODOEOEODOOEODOOERODOEEGOOEEGDOEERGOOEEEGOOEEGOOEEREGOOGERGOOEREG GE EXECUTE THIS PART IF THE BOOTSTRAP WEIGHTS ARE IN SAS FORMAT remove the libname in2 directory name containing bootstrap weights file ex c bootstrp Wet
6. Variance Estimation for a Total using 500 bootstrap replicates Obs prc8 cur type var n Estimate bs sd bs cv CIL95 CIU95 1 All Total diab 785 698848 46 27250 05 3 90 645438 37 752258 55 2 All Total hdiab 392 378528 18 20925 33 5 53 337514 54 419541 83 3 10 Total diab 99 20741 31 1845 88 8 90 17123 38 24359 23 4 10 Total hdiab 35 7029 11 1380 61 19 64 4323 11 9735 10 5 24 Total diab 199 205292 21 15960 30 7 77 174010 03 236574 40 6 24 Total hdiab 104 110452 77 10944 25 9 91 89002 04 131903 51 7 35 Total diab 374 362439 56 19721 07 5 44 323786 27 401092 86 8 35 Total hdiab 190 198237 67 15854 33 8 00 167163 17 229312 16 9 59 Total diab 113 110375 38 10847 03 9 83 89115 19 131635 56 10 59 Total hdiab 63 62808 64 8568 12 13 64 46015 12 79602 15 Variance Estimation for a Ratio using 500 bootstrap replicates Obs prc8 cur type vari var2 ni Estimate bs sd bs cv CIL95 CIU95 1 All Ratio diab total 785 0 0306 0 0012 3 90 0 0282 0 0329 2 All Ratio hdiab males 392 0 0335 0 0019 5 53 0 0299 0 0371 3 10 Ratio diab total 99 0 0385 0 0034 8 90 0 0318 0 0453 4 10 Ratio hdiab males 35 0 0263 0 0052 19 64 0 0162 0 0365 5 24 Ratio diab total 199 0 0287 0 0022 7 77 0 0243 0 0331 6 24 Ratio hdiab males 104 0 0312 0 0031 9 91 0 0252 0 0373 7 35 Ratio diab total 374 0 0322 0 0018 5 44 0 0288 0 0356 9 59 Ratio diab total 113 0 0283 0 0028 9 83 0 0228 0 0337 10 59 Ratio hdiab males 63 0 0324 0 0044 13 64 0 0237 0 0411 Variance Estimation for a Logistic Regr
7. bsampzi n2 SAS file name containing the weights without extension EXECUTE THIS PART IF THE BOOTSTRAP WEIGHTS ARE IN ASCII TXT FORMAT remove the data bootwt l et datafide directory location a and bootstrap weights file with extension include directory location and file name of layout with extension run s yl et bsamp boot wt XC KK KARR KR KR KR AK KR KKK KR KKK RK RK RK RRR RK RK E RK KA e KA e KR ARK e e KR KK KR KKK RR KR KR KR RR AR KK KKK KK e e x x SPECIFY IF DESIRED THE BREAKDOWN VARI ABLE S EG PROVINCE SEX ETC t2 rx Write the name of the breakdown variable s below ve x ps f the analysis includes all of the data in the file created in step 1 put ul a dot let classes ik If more than one variable leave a blank between each variable du udi et classes varl var2 si DO NOT ERASE OR COMMENT OUT THIS COMMAND ue oorr ore ere oer bro or oer oerte eroe do or o obrero rore ebore oorr beoe brereebrereerbreeeerere et classes breakdown variable s or a dot KAKA KK oko kk ok ek ok e ok ok AKA KKK ek ek ek eG GG GE GEO EG Ek e e Ek e ke ok ok ke ok ok ok ook ke ke ko ke ke EEEE EEEE ee ke eee e e e KK KK SPECIFY THE FOLLOWING INFORMATION SPECIFIC TO THE SURVEY YOU ARE USING r i You must specify x le The unique identifier variable s separated by a space x niu 2 The Final Weight variable included in the bootstrap weight file du ili 3
8. more details The analysis file must contain e The necessary variables for the analysis derived variables including dichotomous variables and input variables that do not need to be modified To reduce the runtime of the program DO NOT keep unnecessary variables e The identification variable s of the respondents e If needed the breakdown variable s identifying the groups for which a separate analysis is desired ex province sex etc e Ifthe analysis is only of interest for a certain subgroup for example a province or an age group keep only the records that are part of this subgroup REMARKS e tis recommended that point estimates be calculated at this step to be sure that the desired estimate is being calculated correctly and that the program BOOTVARE V30 SAS correctly calculates the same estimate In this case it is necessary to keep the weight variable when creating the analysis file e Means are estimated using the ratio macro in Step 2 Dichotomous variables identifying the records that are part of the group of interest must be created for the denominator The users must create their own program to prepare the analysis file containing the necessary variables for the analysis An example of a program that creates this file is included in Appendix A the program STEPI SAS Step 2 Variance Calculation Using the BOOTVARE V30 SAS Program Once the new SAS data file is created in step 1 the next step consists
9. of running the BOOTVARE V30 SAS program Before running it the desired parameters and analyses must be specified This program calls the MACROE V30 SAS program MACROE V30 SAS contains the program code of the various macros For standard use of the variance estimation program no modification of the MACROE_V30 SAS program by the user is necessary Changes may be required in certain cases as explained later The BOOTVARE V30 SAS program is included in Appendix A The parts that are to be changed by the user are given in bold type The rest of the program does not need to be changed The program is divided into fwo sections The first section is for specifying the required parameters and the second section is for listing the desired analyses Section 1 In this section the user must specify e The name of the directory of the analysis file created in step 1 and of the output file containing the results The name of the data file analysis file created in step 1 e The name and directory of the bootstrap weights file e The breakdown variable s to specify that the analysis is to be performed separately for specific sub groups ex provinces sex e The identification variable s of the respondents the weight variables and the number of bootstrap weights e The name of the directory where the program MACROE V30 SAS is located N B AppendixC XYZ contains survey specific information file names certain variable names number of bootstr
10. 30 SAS program calculates estimates of the variance of totals ratios differences between ratios and linear or logistic regression parameters Variance estimation is performed in two steps and involves the use of three SAS programs The first step consists of creating a data file containing the variables required for the analysis first program The second step involves using BOOTVARE V30 SAS and MACROE V30 SAS to estimate the variances Step 1 Creation of the Analysis File The user needs to create a SAS data file which will be used as the input file for the program estimating the variance in step 2 The following tasks must be done in this step 1 Reading of the input file 2 Creation of the variables required for the analysis 1 Reading of the input file The analysis file is created from the survey data file The file layout must be provided in order to read in the variables contained in the file See AppendixC XYZ for the file and variable names 2 Creation of the variables required for the analysis Variables derived from the input variables should be created in this step It may be necessary to create dichotomous variables 1 or 0 which identify records that have a characteristic of interest such variables will take a value of 1 for records that have the characteristic and a value of O otherwise The total of a dichotomous variable will sum the weights of the records with the characteristic See the example in Appendix B for
11. Estimation of the Variance Using Bootstrap Weights User s Guide for the BOOTVARE V30 SAS Program VERSION 3 0 1 Introduction This guide is for users of the SAS program BOOTVARE V30 SAS which was created to estimate the variance using the bootstrap method Section 2 of this guide briefly explains the bootstrap resampling method used to estimate the variance Section 3 gives detailed instructions for using the BOOTVARE V30 SAS program as well as a description of the preliminary steps that are required The programs are given in Appendix A Appendix B contains a complete example programs and results Finally the survey specific parameters required for executing the programs file names identification variables etc are provided in the document AppendixC XYZ where XYZ identifies the survey Changes from the Previous Version The biggest change made to the program is that a single version of the program supports all the Statistics Canada surveys that use the BOOTVAR program Users only need to specify a few parameters see the document AppendixC_XYZ in section 1 of the program Please note that the program was tested and works with SAS versions 6 12 and 8 2 2 Bootstrap Method The sampling designs for Statistics Canada s surveys are generally complex Since the variance for such designs cannot be estimated with simple formulas resampling methods are often used to estimate the variance The bootstrap method consists of subsamp
12. S program XXX OE OO IOIOIO E OO OE OO E OE OE OE OE IO EE IGI E IIO OOo Ib RR RRR RRR RRR KK KH KK ER SECTION 1 Xo OK kk Gk E OEOEOKCOE C EE ECKE OG OK AKA COE EEEE E EEEE C EGG GE KG X GEEK GO KG EEEE EEEE EEEE E EEEE EEEE ER This section lets the user specify the different parameters of dai pees interest variable names directory names file names etc TEX x x xk FOX OI OO OIX E OO EO OO E OE OE E OE OO EE IO E IIO OO Ib RRR RRR RRR RRR KK RR KH KK IEE EEEE Gk GE EEEE E EEEE GO GG GE EG E GO X GG EG G0 X G EEEE EEEE E EEEE KK KK KKK KK KK KK KK KKK SPECIFY THE NAME OF THE FOLLOWING 2 DIRECTORIES directories only TOS RE Pee he ee ee Rk ee ee eR Peo eee eee eee Re ee he ee Re Re he eee eee ee RR ee oe RRM libname inl C BOOTVAR e ex c data li bname out C BOOTVAR f ex czVoutput e kx Ed oec odora coe LEER ERATED doe oo SLEEK REDE ER LES ER EARL ED LAE ES ASHER p pee pee o cp er SPECIFY THE NAME OF THE ANALYSIS FILE CREATED IN STEP 1 without extension FACOG CCGG CCG CCGG CCGG CCGG CCGG CGR GRIGG ERK REE et Mfile inl diabetes LEE EEEE EEE KAKA KR KKK RAK AR AKA KAKA EEE EEEE EEEE EEEE EEEE EEE EEEE EEEE EEEE EEE e e Ge x x SPECIFY THE NAME OF THE FILE CONTAINING THE BOOTSTRAP WEIGHTS bii i NB Only run one of the two following series of commands kx comment the other one out or erase it JOGO GGG GCG GGG GG RRR REG EXECUTE THIS PART IF THE BOOTSTRAP WEIGH
13. TS ARE IN SAS FORMAT remove the libname in2 D boot let bsamp in2 b5h EXECUTE THIS PART IF THE BOOTSTRAP WEIGHTS ARE IN ASCII TXT FORMAT remove the oH wn E Rn data bootwt et datafidz directory location and bootstrap weights fi wit with extension i nclude directory location and file name of Tayout t Ie th extension de 36 GE de 3 run et bsamp bootwt XO kck kk kk Ek Gk kk E Eo e o AKA KAKA KKK ok ok ok ko ke ok kk e e E GG Gk X GE GEO GE E e EEEE e x e x e e e oe e EEEE EEEE SPECIFY IF DESIRED THE BREAKDOWN VARIABLE S EG PROVINCE SEX ETC i Write the name of the breakdown variable s below x x If the analysis includes all of the data in the file created in step 1 put M a dot Wet classes ubi f more than one variable leave a blank between each variable ubi et classes varl var2 kx DO NOT ERASE OR COMMENT OUT THIS COMMAND ae ORK RR ROR RR ROR OK ROR OR KOR KOR AOR KR KOR ROK KOK ROK KKK KOR RR KOR RR KK ROK RK RK OK KR RK RR RK RR RAR KR KKK KR KR KR KR KR KK oe KK KK KKK RK et classes prc8 cur XO Kok OE OR RRR ORK RR ROKR ROR KOR RR KOR ROK ROK 0 e 0 KOR ROR KOR ROR KOK ROK RK ROK OR KR RK ORR RK RRR KOR KR e e KOR KOR KOR KR KOR KK KOR KKK KR X e x x SPECIFY THE FOLLOWING INFORMATION SPECIFIC TO THE SURVEY YOU ARE USING ii ibi You must specify ibi 1 The unique identifier variable s separated by a Space
14. and for the men for each province only four provinces will be kept 2 Studies the relationship between diabetes sex and type of interview proxy or not for each province The different parameters needed in the program specified in AppendixC_Health are NPHS Household Component Name of bootstrap A Weight Weigh Prefixe of weights file eut Identification variable be the of on the bootstrap weights bootstrap H weights file weights Name of data file variables on the data ASCII format txt SAS format sd2 or sas7bdat REALUKEY WT58 M file Step 1 LERE EEE kk EE KA KK E EEEE EEEE GEEK GE KG GG E KG e EG x E EG GI ERG GO EE EEEE EEEE STEP ISAS This program creates the SAS datafile containing the necessary variables for the BOOTVARE V30 SAS program XCk KK KK KR KKK KARR X X GE X KO Kok ke RK KK KEK RA X ook o KR KR KK KR GGG GG X G LI BNAME inl C BOOTVAR Appendix B Creation of the SAS data file containing the variables and cases required for the analysis Note that this file should be as small as possible containing only necessary variables and cases in order to reduce ti me and memory requirements especially if regression type analysis are to be done data inl diabetes file to be used with BOOTVARE V30 5AS Wet datafi s VDatalh35 txt jnclude D Layout h35_i sas keep only 4 provinces if prc8 cur in 10 24 35 59 Cr
15. ap weights Section 2 In this section the user lists the analyses for which estimates of the variance are desired The following types of analyses are supported Totals Ratios including means Differences between ratios Regression models linear or logistic For means To estimate the variance of a mean the macro for ratios can be used The numerator is the variable of interest and the denominator is a dichotomious variable that identifies the population of interest For totals ratios and differences between ratios The variable of interest must be positive or null For differences between ratios To calculate the difference between ratios it may be necessary for the user to modify the macro diff rat in the MACROE V30 SAS program to suit their needs See the notes included in the BOOTVARE V30 SAS program for more details For regressions Categorical variables will be treated as continuous variables Dichotomous variables must be created in step 1 for each possible value except one of the categorical variable in order to treat this variable properly Modification to the program for testing purposes Running the program could take long especially for complex model analyses It is possible to reduce the number of bootstrap weights used in order to test the program However to obtain the final estimates of the variance it is important to use all of the bootstrap weights provided To test the program all that is require
16. b t ot al mdi ab TO OBTAIN VARIANCE ESTIMATES OF A RATIO RUN ratio numerator variable denominator variable Yratio diab total Wratio mdi ab males TO OBTAIN VARI ANCE ESTI MATES OF A DIFFERENCE BETWEEN RATIOS RUN NOTE see the comment at the beginning of section 2 Wiiff rat VARI VAR2 VAR3 VAR4 where varl the numerator variable of the first ratio ii var2 the denominator variable of the first ratio var3 the numerator variable of the second ratio vara the denominator variable of the second ratio TO OBTAIN VARI ANCE ESTIMATES OF REGRESSION PARAMETERS RUN Wregress dependent variable independent variables no comma TO OBTAIN VARI ANCE ESTI MATES OF LOGISTIC REGRESSION PARAMETERS RUN ogreg dependent variable independent variables no comma Wlogreg diab nonproxy females Woutput Displays the results on the screen Do not modify TO SAVE THE RESULTS IN A FILE RUN remove the data out results set Gresult run end of BOOTVARE V30 SAS program 12 Appendix B Pan Appendix B Results and interpretation The tables on the next page present the results of the analyses done using the programs from the example Results for the totals and ratios are presented in the first and second tables For totals and ratios the first lines in the output give the results for the entire file ALL Results for each category of
17. d is to modify the parameter that specifies the number of bootstrap weights in the first part of the BOOTVARE_V30 SAS program Results Obtained with BOOTVARE V30 SAS The following results are differences between ratios TYPE VARI to VAR4 n nl n3 YHAT BS_SD BS_CV CIL95 CIU95 The following results are obtained after running BOOTVARE V30 SAS for totals ratios and See Appendix B for an interpretation of the results Type of estimate total ratio diff ratio Variables used to calculate the estimates Tailles d chantillons pour les totaux n et les ratios n1 et n3 Parameter estimate Standard deviation Coefficient of variation Lower limit of the 9596 confidence interval Upper limit of the 95 confidence interval obtained after running BOOTVARE V30 SAS for linear and logistic regressions See Appendix B for an interpretation of the results PARAM Parameter to estimate BETA Parameter estimate ODDS Odds ratio logistic regression only WALD Wald statistic logistic regression only PVALUE P value of the Wald statistic logistic regression only BSVAR Variance of the parameter estimate BS SD Standard deviation of the parameter BS CV Coefficient of variation for the parameter estimate CIL95 Lower limit of the 9596 confidence interval for the odds ratio if logistic regression CIU9S Upper limit of the 95 confidence interval for the odds ratio if logistic regression Appendix A conta
18. eation of Dichotomous Variables qu examples are presented below using NPHS cycle 3 variables diabetes if ccc8 1jzl then diab 1 else diab 0 Dichotomous variable 0 1 for type of interview nonproxy 0 if am58 pxy 2 then nonproxys if am58 pxyz2 then nonproxy l sex total zl if dhc8 sexzl then maleszl else males 0 if dhc8 sex 2 then femal es 1 else femal es 0 diabetes sex mdiab diab males male diabetics f fdiab diab females female diabetics keep diab total males females mdiab fdiab nonproxy wt58 realukey personid prc8 cur t is recommended that only the necessary eru be kept t jn order to reduce the runtime of BOOTVARE V30 5 IMPORTANT the identification variables and TAa ME dg the breakdown variable ex province sex must be kept The weight variable also must be kept if point estimates are calculated at this step xi run EEEE ok ke ke ok o e ok e ke ok e k e k e xk e ke e koe ee e e e e e e x e x Calculation of point estimates Suggested but not required BGS ICG ICE RIE I PROC SORT DATA inl diabetes BY prc8 cur RUN di ia at nl e di ab md i by ore tur weight wt 58 run proc f ia t b proc logi E ata in1 diabetes model d ab nonproxy females by prc8 cur ght wt58 LE Relationship between diabetes sex and type of interview ux di wei Th run 10 Appendix B Step 2 BOOTVARE V30 SA
19. ession Dependent variable diab using 500 bootstrap replicates Obs PRC8_CUR param beta odds wald pvalue bs_var bs_sd bs_cv CIL95 CIU95 1 10 Intercept 4 00372 0 01825 831205 16 0 00000 0 00002 0 00439 0 11 0 00964 0 02685 2 10 nonproxy 0 85783 2 35803 2 09 0 14824 0 35204 0 59333 69 17 1 19511 3 52096 3 10 females 0 46625 1 59401 0 87 0 35123 0 25016 0 50016 107 27 0 61370 2 57432 4 24 Intercept 3 89892 0 02026 2226070 39 0 00000 0 00001 0 00261 0 07 0 01514 0 02539 5 24 nonproxy 0 89121 2 43807 4 94 0 02618 0 16065 0 40082 44 97 1 65247 3 22367 6 24 females 0 36991 0 69080 10 75 0 00104 0 01273 0 11283 30 50 0 46965 0 91194 7 35 Intercept 3 57575 0 02799 1325488 73 0 00000 0 00001 0 00311 0 09 0 02191 0 03408 8 35 nonproxy 0 60946 1 83943 6 35 0 01171 0 05845 0 24177 39 67 1 36557 2 31329 10 59 Intercept 3 99955 0 01832 971323 00 0 00000 0 00002 0 00406 0 10 0 01037 0 02628 11 59 nonproxy 1 05756 2 87934 1 67 0 19693 0 67172 0 81959 77 50 1 27295 4 48574 12 59 females 0 46337 0 62916 10 16 0 00143 0 02113 0 14535 31 37 0 34428 0 91404 14
20. ins the BOOTVARE_V30 SAS program preceded by an example of a program which prepares the analysis file STEPI SAS Appendix B contains a complete example programs and results Finally AppendixC XYZ contains survey specific information file names names of certain variables number of bootstrap weights APPENDIX A Programs to Run STEP 1 SAS Program Used as an example the users can use their own program The parts in bold need to be changed XC EEE EEEE EEEE KOR KOR KOR KR KK KOK RK ROKR KR KOR KR e e e KR KOR KR KOR KKK e e KOK e e KOK STEP1 SAS ii This program creates the SAS datafile containing the necessary variables for the BOOTVARE V30 SAS program T KAKA AKA ko K K K K K K K K K K ke ok ke o ek ek ex X E Xo oe E oe e KAKA KKK E eG x k LIBNAME inl directory to save file in Creation of the SAS data file containing the variables and cases required for the analysis Note that this file should be as small as possible containing only necessary variables and cases in order to reduce ti me and memory requirements especially if regression type analysis are to be done data inl Name of analysis file file to be used with BOOTVARE_ V30 SAS let datafidz name and location of source file i nclude name and location of layout Creation of Dichotomous Variables ee examples are presented below using NPHS cycle 3 variables diabetes if ccc8 ljzl then diab l
21. ling the initial sample Within each stratum a simple random sample SRS is selected with replacement from n 1 clusters within the n clusters of the stratum The process is repeated B times creating B new samples or replicates Weights are recalculated for each of the B samples the B weights are called the bootstrap weights The bootstrap weights are used to calculate B estimates which are then used to estimate the variance The bootstrap weights have already been generated and are available with the data The BOOTVARE V30 SAS program uses these bootstrap weights to estimate the variance for simple statistics such as totals and ratios as well as for more complex analyses like regressions These estimates of the variance should be used to derive quality indicators and to apply the survey s rules for releasing the estimates Here are the main steps for estimating the variance of a particular estimate using the bootstrap method A Calculate an estimate total ratio etc using the final weight included in the data file This estimate is the point estimate B Calculate the same estimate this time using each of the B bootstrap weights contained in the bootstrap file B estimates total ratio etc are then obtained C Finally calculate the variance of the B estimates This variance is the estimate of the variance of the point estimate calculated in A 3 Variance Estimation with the BOOTVARE V30 SAS Program The BOOTVARE V
22. the breakdown variable s follow For example if we want the ratio of the number of diabetic males to the total number of males in Ontario we look at observation 8 in the second table The region 35 corresponds to the province of Ontario see the data dictionary document included on the CD ROM for the codes associated with each province and the variable Type indicates the type of analysis in this case a ratio We find the variables mdiab VARI as the numerator of the ratio and males V AR2 as the denominator The column n1 indicates that the data file contains 190 diabetics males in Ontario sample size for the numerator The estimate of the ratio is 3 57 YHAT with a standard deviation of 0 29 BS SD and a coefficient of variation of 8 00 BS CV The 95 confidence interval for this estimate is 3 01 4 14 CIL95 CIU95 Results from the logistic regression are shown in the third table For example the estimate of the parameter for the variable females in Ontario observation 9 is 0 34507 BHAT and the odds ratio is 0 70817 ODDS The Wald s statistic for this parameter and its associated p value are 14 28 WALD and p 0 00016 PVALUE respectively The estimates of the variance and the standard deviation for the parameter estimate are 0 00834 BS VAR and 0 09132 BS SD and the coefficient of variation is 26 47 BS CV Finally the confidence interval for the odds ratio is 0 52918 0 88717 CIL95 CIUO95 13 Appendix B
Download Pdf Manuals
Related Search