APPENDIX 2 Examples of SAS and SUDAAN Programs Combining Respondent and Interval File Data Using SAS

Transcription

1 APPENDIX 2 Examples of SAS and SUDAAN Programs Combining Respondent and Interval File Data Using SAS As mentioned in the section called "Organization and Use of the Data File," selected interval variables have been placed on the respondent file, and selected respondent variables have been placed on the interval file, but users may need to merge in additional variables for their analyses. Below are 2 examples of SAS programs that will yield either an interval-based file or a respondent-based file. The user must tailor these examples to their own computing environments (e.g., adding their own file definition statements). Example 1: Adding Respondent Variables to a Pregnancy (Interval) Based File This template program will yield a sasfile with 21,332 records, assuming that the user does not subset observations from the interval file. The respondent-based variables that are not already on the interval file will be added to EACH interval record with the same CASEID (case identification number). DATA RESPOND; INFILE IN1; INPUT CASEID 1-8 : (insert other variables desired from respondent file) : ; DATA INTERVAL; INFILE IN2; INPUT CASEID 1-8 : (insert other variables desired from interval file) : ; DATA PREGFILE; MERGE RESPOND INTERVAL (IN=A); BY CASEID; IF A; - 1 -

2 Example 2: Adding Interval Variables to a Respondent Based File This template program will extract the most recent live birth for each respondent from the interval file and merge with selected variables from the respondent file. The resulting sasfile will yield a respondent-based sasfile with less than 10,847 records, because only those respondents who have ever had a live birth will be included. This program may be helpful if the user wishes to examine, for example, breastfeeding and maternity leave for the most recent birth. For this program, the following interval variables are needed: CASEID = Case identification number OUTCOME = Outcome of pregnancy (=1 if live birth) PREGORDR = Pregnancy order or number DATA RESPOND; INFILE IN1; INPUT CASEID 1-8 : (insert other variables desired from respondent file) : ; DATA INTERVAL; INFILE IN2; INPUT CASEID 1-8 OUTCOME 288 : (insert other variables desired from interval file) : ; IF OUTCOME=1; /* keep only live births */ PROC SORT; BY CASEID PREGORDR; /* sort PREGORDR within CASEID */ DATA LASTPREG; SET INTERVAL; BY CASEID; IF LAST.CASEID THEN OUTPUT; /* keep only the last birth for each R */ DATA LASTBRTH; MERGE RESPOND LASTPREG (IN=A); BY CASEID; IF A; /* yields 1 record for each woman who had a live birth */ Example of SUDAAN Run to Obtain Standard Errors The following example consists of 3 parts: 1) Guidelines for setting up the SUDAAN procedures: corresponds to the two procedures below - 2 -

3 2) Example of SUDAAN program and output for PROC DESCRIPT (Among its functions are: producing basic statistics such as means and percentages and their standard errors) 3) Example of SUDAAN program and output for PROC LOGISTIC (fits logistic regression models to binary data) Documentation and support for SUDAAN can be obtained through Research Triangle Institute: Phone: (919) Fax: (919)

4 1) General guidelines for setting up SUDAAN program code -- Procedure Statements: Include DESIGN=WR -- SUDAAN Design Parameters Statement NEST Statement: a. COL_STR (NHIS collapsed strata) b. PANEL (NHIS national panel identifier (values=1-4)) TOTCNT Statement: No statement required WEIGHT Statement: POST_WT (Final 1995 NSFG analysis weight) -- Categorical variables: - SUBGROUP statement: identifies categorical classing (explanatory) variables. Variables must have values ranging from 1-n. In the two examples, they are AGECAT, WRELIG, and RACE2. - LEVEL statement: identifies the number of valid levels (AGECAT values: 1-6, WRELIG values: 1-3, RACE2 values: 1-3) -- Non-categorical variables: (In the two examples, CONPILL is the "outcome" variable. PARITY is an explanatory variable in the PROC LOGISTIC example. PARITY: Number of live births (continuous) CONPILL: 1 if currently using the Pill, 0 otherwise (binary) -- PROC DESCRIPT: VAR = computation statement that names the response/outcome variable (In the example, "CONPILL" CATLEVEL = indicates that the VAR variable(s) is/are categorical and selects the level of the variable(s) to be analyzed. (In the example: "1") -- PROC LOGISTIC: MODEL statement specifies the dichotomous dependent variable (CONPILL) as a function of one or more independent variables (AGECAT, WRELIG, PARITY, RACE2) 2) Program lines 1-11 (out of 18): PROC DESCRIPT program statements - 4 -

5 S U D A A N Software for the Statistical Analysis of Correlated Data Copyright Research Triangle Institute June 1996 Release PROC DESCRIPT DATA="Y:\\DATA\\PILLGVAR" FILETYPE=SAS DESIGN=WR; 2 NEST COL_STR PANEL; 3 WEIGHT POST_WT; 4 VAR CONPILL; 5 CATLEVEL 1; 6 SUBGROUP AGECAT WRELIG RACE2; 7 LEVELS 6 3 3; 8 TABLES AGECAT WRELIG RACE2; 9 SETENV LINESIZE=100 PAGESIZE=60; 10 PRINT NSUM TOTAL PERCENT SEPERCENT / STYLE=NCHS TOTALFMT=F8.0 PERCENTFMT=F6.2 SEPERCENTFMT=F6.3; 11 TITLE "NSFG CYCLE V - Standard Error Table for Pill Use (DESIGN=WR)"; Number of observations read : Weighted count : Number of observations skipped : 0 (WEIGHT variable nonpositive) Denominator degrees of freedom : 186 Output from PROC DESCRIPT: - 5 -

6 Date: Research Triangle Institute Page : 1 Time: 14:52:07 The DESCRIPT Procedure Table : 1 NSFG CYCLE V - Standard Error Table for Pill Use (DESIGN=WR) by: Variable, AGECAT. Variable Sample SE AGECAT Size Total Percent Percent CONPILL: 1 Total NSFG CYCLE V - Standard Error Table for Pill Use (DESIGN=WR) by: Variable, WRELIG. Variable Sample SE WRELIG Size Total Percent Percent CONPILL: 1 Total White Protestants White Catholics Others NSFG CYCLE V - Standard Error Table for Pill Use (DESIGN=WR) by: Variable, RACE2. Variable Sample SE RACE2 Size Total Percent Percent CONPILL: 1 Total Hispanics NonHispanic Blacks Others ) Program lines (out of 18): PROC LOGISTIC program statements - 6 -

7 12 PROC LOGISTIC DATA="Y:\\DATA\\PILLGVAR" FILETYPE=SAS DESIGN=WR; 13 NEST COL_STR PANEL; 14 WEIGHT POST_WT; 15 SUBGROUP AGECAT WRELIG RACE2; 16 LEVELS 6 3 3; 17 MODEL CONPILL = AGECAT WRELIG PARITY RACE2; 18 TITLE "NSFG CYCLE V - Logistic Regression Model for Pill Use (DESIGN=WR)"; Number of observations read : Weighted count: Number of observations skipped : 0 (WEIGHT variable nonpositive) Observations used in the analysis : Weighted count: Observations with missing values : 0 Weighted count: 0 Denominator degrees of freedom : 186 Number of non-zero responses: 1806 Number of zero responses : 9041 LOGISTIC has converged in 4 iterations Multiple R-Square for the dependent variable CONPILL: * Normalized Log-Likelihood with Intercepts Only : * Normalized Log-Likelihood Full Model : Approximate Chi-Square (-2 * Log-L Ratio) : Degrees of Freedom : 10 Approximate P-Value : 0.00 Note: The approximate Chi-Square is not adjusted for clustering. Refer to hypothesis test table for adjusted test

8 Output from PROC LOGISTIC: Date: Research Triangle Institute Page : 1 Time: 14:52:07 The LOGISTIC Procedure Table : 1 For response variable CONPILL NSFG CYCLE V - Logistic Regression Model for Pill Use (DESIGN=WR) Independent P-value Variables and Beta T-Test Effects Coeff. SE Beta T-Test B=0 B= Intercept AGECAT WRELIG White Catholics White Protestants Others PARITY RACE2 Hispanics NonHispanic Blacks Others For response variable CONPILL NSFG CYCLE V - Logistic Regression Model for Pill Use (DESIGN=WR) Contrast Degrees of P-value Freedom Wald F Wald F OVERALL MODEL MODEL MINUS INTERCEPT INTERCEPT... AGECAT WRELIG PARITY RACE

9 Output from PROC LOGISTIC (continued): Date: Research Triangle Institute Page : 3 Time: 14:52:07 The LOGISTIC Procedure Table : 1 For response variable CONPILL NSFG CYCLE V - Logistic Regression Model for Pill Use (DESIGN=WR) Independent Variables and Lower 95% Upper 95% Effects Odds Ratio Limit Limit Intercept AGECAT WRELIG White Catholics White Protestants Others PARITY RACE2 Hispanics NonHispanic Blacks Others