NEWSLETTER vol.1, no. 4 U.S. Department of Health, Education and Welfare. Standardized Micro-Data Tape Transcripts. (DHEW Pub. No. 76-1213). Washington: GPO 1976. 30 pp. Describes data bases for health statistics available from the Federal government. Prices, technical configurations. Available from GPO. U.S. Department of Labor. Bureau of Labor Statistics. BLS Data Bank Files and Statistical Routines , (draft brochure). Describes 30 data files available from the Bureau of Labor Statistics. Describes publications. Write: Commissioner, BLS, Dept. of Labor, Washington, DC 20212. U.S. General Accounting Office. 1976 Congressional Sourcebook Series (OPA 76-23) Federal Information Sources and Systems; A Directory for the Congress . Washing- ton, DC: GPO 1976. 456 pp. paperback. An indexed reference guide to over 1,000 Federal sources and information systems in 63 Federal agencies, which con- tain budgetary, fiscal and program-related data. Available from GPO. U.S. National Archives and Records Service. Catalog of Machine-Readable Records in the National Archives of the United States . Washington: GPO, 1977. 37 pp. Describes 99 files created by Federal agencies and retained in the National Ar- chives. Prices, available technical configurations. Write: National Archives (NNR), Washington, DC 20408. University of Waterloo. Leisure Studies Data Bank. April, 1977. 23 pp. Lists 24 files created by private and Federal Canadian agencies. In French and Eng- lish. Write: Dept. of Recreation, University of Waterloo, Waterloo, Ontario, Canada. N2L 3G1 . DISCUSSION PAPER/gary m. grandon Using SPSS MULT RESPONSE to Generate Filtered Marginals Gary M. Grandon Social Science Data Center The University of Connecticut, Storrs With the release of SPSS's Version 7.0 this year, several additions have been added to its already extensive battery of analysis programs. MULT RESPONSE is one of these programs. It generates frequency counts and bivariate tables for "dummy" coded multiple response questions. These multiple response variables are a real nuisance to the analyst and the program at face value provides a means for their interpretation. Considerations for the analysis of multiple response data are not the subject of this paper though, but rather the use of this program for quite another purpose; the generation of "filtered marginals." Filtered marginals are typically basic frequency counts and percentages for specified variables for each of a number of subpopulations within a study. SPSS has in the past provided *SELECT IF statements to facilitate the processing of subpopulations. Each such statement followed by FREQUENCIES procedure statements will generate filtered marginals. The inherent problem with this type of coding 30 sist NEWSLETTER vol.1, no. 4 is that each *SELECT IF statement requires a complete reading of the data set. For large studies this procedure can be quite inefficient! Another alternative to the *SELECT IF procedure would be to create a new variable with codes indica- ting each filter group (subpopulation) . Then the very fast CROSSTABS procedure can be used with the new filter group variable as one variable in the bivariate table and the variable upon which frequencies are to be generated as the other. A large number of frequencies can be generated in this manner with only one read of the data. The problem with this CROSSTABS procedure is that it requires that each filter group or subpopulation be discrete. That is, any one subject can only be in one filter group. This restriction is removed by using the MULT RESPONSE technique suggest in this paper. The following setup using the SPSS - OSIRIS interface to shorten the presentation illustrates the use of the MULT RESPONSE procedures to perform filtered marginals: 1 RUN NAME OSIRIS VARS INPUT MEDIUM N OF CASES WEIGHT IF IF MULT RESPONSE STATISTICS READ INPUT DATA FINISH YANKELOVICH—COMBINED STUDIES-1976 V7,V8,V11,V13,V18,V28,V64,V75 TAPE UNKNOWN V75 ((V7 EQ 1 OR 2 OR 3) AND (V8 EQ 6) AND (Vll EQ 1) AND (V18 EQ 1)F1=1 ((V7 EQ 5 or 6) AND (V8 EQ 1 OR 2) AND (Vll EQ )3) AND {V18 EQ 1)F2=1 GROUPS=FILTERS (Fl F2(l))/ VARIABLES=V13(1,4),V28(1,3),V64{1,3)/ TABLES=FILTERS BY VI 3 TO V64/ 1 Not only does the technique illustrated above produce the same kind of results as the *SELECT IF process but it formats the output in such a way as to allow the user to scan differences across subpopulations within a single tabular display. A single procedure statement like the one above can process 20 filter groups and up to 100 analysis variables. Each frequency variable will have its own table. A resource expenditure comparison between the *SELECT IF procedure and the MULT RESPONSE procedure was performed on the 1976 Yankelovich Combined Study ar- chived at The Social Science Data Center of the University of Connecticut. The study has a N of 7,977 subjects. Nineteen filter groups were identified and frequencies for 45 analysis variables were obtained using both techniques. Table 1 provides relevant resource comparisons for the techniques. The CROSSTABS tech- nique was not included due to its unique group membership constraint. Clearly costs are much lower for the MULT RESPONSE procedure. sist NEWSLETTER vol.1, no. 4 '^SELECT IF and MULT RESPONSE techniques for filtered marginals. RESOURCE *SELECT IF MULT RESPONSE NUMBER OF JOBS 2 1 KBYT-SEC 21,016 24,957 ELAP-KBS 180,493 73,752 READER EXCP** 93 31 PRINTER EXCP 14,630 3,844 PUBLIC DISK EXCP 10,476 360 PRIVATE DISK EXCP 10 5 TAPE EXCP 72 36 TAPE MOUNTS 2 1 COST $31.70 $13.39 ELAPSED TIME 13.48 min. 6.23 min. *Comparisons were performed on the University of Connecticut's Research Computer Center IBM 360/65 - IBM 370/155 systems running under OS and shared spool HASP. Local costs may vary from installation to installa- tion. **Each EXCP represents the "Execution of a channel program" and indicates the movement of one block of data. The MULT RESPONSE procedure can further be extended to filtered bivariate tables by including a second BY_ statement on the TABLES card. This technique is not illustrated due to space limitations. Readers are referred to their local installations for further documentation of SPSS Version 7.0. DISCUSSION PAPER/riohard c. roistacher The following article describes the work of Richard C. Roistacher and Barbara Noble at the Center for Advanced Computation at the University of Illinois. They are involved in the development of guidelines for the descriptive materials which accompany a data file. A Source Documentation Style Manual by Richard Roistacher Center for Advanced Computation University of Illinois Urbana, Illinois Barbara Noble and Richard Roistacher of the University of Illinois' Center for Advanced Computation are currently developing a style manual for the documentation of machine readable data. The manual, which is being developed as part of a pro- ject funded by the U.S. Department of Justice's Law Enforcement Assistance Admini- stration, is presently available in draft form. The manual, conforming to the