Frequently Asked Questions about the Survey of Consumer Finances
Users of the SCFs are strongly encouraged to examine the information
provided in the technical codebooks and summary articles. These documents
contain essential information about the surveys and have the answers
to many of the most commonly asked questions about the SCFs.
Q: I tried to import the SAS transport file created using PROC
CPORT with StatTransfer and SPSS, but the error log says that NO
transport file is found or transport file is bad. Is there something
wrong with the SAS transport file?
A: StatTransfer and SPSS do not support SAS transport files
created using PROC CPORT. Only uncompressed SAS transport files
created using PROC COPY with an XPORT option are supported by
StatTransfer and SPSS. All SCF SAS data sets are available in both
forms. SAS users can download either form of SAS transport files.
However, StatTransfer and SPSS users should download the COPY/EXPORT
version of the SAS transport files.
Q: I downloaded a SCF SAS data set using a PC and I tried to
use it on the Unix machine. My SAS error log reported "SAS transport
file is bad". What is wrong with the data set?
A: SAS transport files are not text files. They are binary
files in ASCII format. If you downloaded the data sets using a PC, and
attempted to use the data sets on an Unix machine, you must be certain
that the binary format is preserved when transferring the data sets
from the PC to the Unix platform.
Q: I have tried to download files from the SCF Web site, but it
does not seem to be working. What could be wrong?
A: There are numerous reasons why you may have trouble
downloading files from the SCF Web site. Because of the size of some
of the files, it takes a substantial amount of time to download them. Internet
problems may break the connection before the file is completely
retrieved. It is best to compare the size of the downloaded file to
the size reported on the SCF site to ensure you have received the
complete file. If you do not receive the complete file, the best
option is to attempt the download again. Also, some files require
certain software for downloading. For example, you must have Acrobat
Reader software to download PDF files from the SCF site. This
software is available at no charge from
Adobe.
Q: I downloaded an SCF SAS data set on my PC, but the size of the
zipped and unzipped data sets is slightly different than the amount of
disk space given in the instructions. Is there something wrong?
A: No. The given disk space is for users who download the data
using Unix. Unix systems and PC systems store bytes differently. Thus,
depending on the system used, the space required for each data set will
be slightly different.
NOTE: To make sure you have downloaded the entire data set, you should always run a PROC CONTENTS on the data set.
Q: How are missing values (item nonresponse) dealt with in the SCF?
A: Missing values are replaced by use of a
multiple imputation technique.
Q: How do I compute statistics like means, medians, and frequencies
using the weights and all five implicates?
A: NOTE: The code below applies to the 1989-2004 SCFs only.There are two approaches: an exact method, and a method that is exact for means and frequencies but only a close approximation for medians.
For the exact method, compute the desired statistic separately for each implicate using the sample weight (X42001). The final point estimate is given by the average of the estimates for the five implicates.
The second method is given in the following steps using SAS (similar methods apply in other languages):
Begin by dividing the weight (X42001) by 5 so that the sum of the weights across all five implicates equals the correct population total.
LIBNAME IN 'scf data set library'; DATA SCF; SET IN.scf data set; WGT=X42001/5; RUN;Next, using PROC UNIVARIATE with the FREQ option compute the desired statistic. NOTE: You must use the FREQ or WEIGHT option to compute weighted medians with PROC UNIVARIATE.
PROC UNIVARIATE DATA=SCF; FREQ WGT; VAR selected variables; RUN;Or, using PROC MEANS with the FREQ option, compute the desired statistic. NOTE: The WEIGHT option can also be used with PROC MEANS, but the computed standard deviations will be unweighted. Consult the SAS Procedures Guide for more information.
PROC MEANS DATA=SCF; FREQ WGT; VAR selected variables; RUN;Or, using PROC FREQ with the WEIGHT option, compute the desired frequencies or crosstabulations.
PROC FREQ DATA=SCF; WEIGHT WGT; TABLES selected variables; RUN;
Q: How can I compute proper standard errors of estimates from
the SCF data that account for both imputation and sampling
variance?
A: A SAS program to compute standard errors using replicate
weights is provided in the 2004 Codebook. Similar code can
be used for the 1989-2001 SCFs.
Q: What adjustments should be made to regressions when using all
five implicates of an SCF data set?
A: Users who want to estimate regressions should be cautious in their
treatment of the implicates. Many regression packages will treat each
of the five implicates as an independent observation and
correspondingly inflate the reported significance of results. Users
who want to calculate regression estimates but who have no immediate
use for proper significance tests (perhaps for exploratory work)
could regress the average of the dependent and independent values across the
implicates or multiply the standard errors of the regression (on all
observations) by the square root of five.For details on the proper treatment of the implicates in regression, see Multiple Imputation for Nonresponse in Surveys, by Donald B. Rubin, Wiley, 1987. Here we provide two examples to compute standard errors that account for multiple imputations in regressions.
A SAS MACRO written by Catherine Montalto and Jaimie Sung will compute regressions and standard errors accounting for multiple imputations. Further documentation for this model are given in their paper "Multiple Imputation in the 1992 Survey of Consumer Finances," Financial Counseling and Planning, Volume 7, 1996, pages 133-146.
A second approach, which is applicable to other types of models (probits, etc.) is given in the SAS MACRO MISECOMP and the Stata program StataMICode.do in the 2004 Codebook
Q: The numbers I have computed do not exactly match the numbers
reported in the various Federal Reserve Bulletin articles on
the SCFs. Why?
A: Estimates made using the public versions of the SCF data may differ
from those reported in the Bulletin for two main reasons. First, some
of the articles used preliminary data, which have been superseded by
the final data provided on this site. Second, to protect the privacy of
respondents, the data made available to the public have been
systematically altered using statistical procedures that should not
significantly affect the estimates made with the final public data.
Where the differences are substantial, it is likely that either the
user has made an error or the differences are actually not
statistically significant. For more on disclosure review, see the
following papers:Multiple Imputation in the Survey of Consumer Finances
Multiple Imputation and Disclosure Protection: The Case of the 1995 Survey of Consumer Finances
Disclosure Review and the 1998 Survey of Consumer Finances
Analyzing the Disclosure Review Procedures for the 1995 Survey of Consumer Finances
The Challenges of Preparing Sensitive Data for Public Release
Top of page