The 2001 Residential Finance Survey was designed to provide data about the financing of nonfarm, privately owned, residential properties.
The program was conducted by mailing questionnaires to a sample of property owners and to lenders who held mortgages on the sample properties. An option was also provided to lenders who wanted to respond electronically. Telephone and personal visit followup was done for nonresponse cases.
A sample of about 50,000 addresses was drawn from the address file for Census 2000. These addresses were limited to counties and independent cities in the 394 sampling areas used for the Census Bureau's American Housing Survey National Sample.
The "pre-survey contact with lenders" letters were mailed in mid-June, 2000. The "owner seeker" letters were mailed to addresses or sent to the regional offices for field interviewing in early April, 2001. The homeowner questionnaires were mailed in early April, 2001, with a followup mailing mid-June. The rental and vacant property questionnaires were mailed in late May, 2001, with a followup mailing in late June. The lender questionnaires were mailed in August and September, 2001, to those lenders wishing to receive paper questionnaires. Those lenders that wished to respond electronically to data files made available by the Census Bureau were sent information in three phases: in mid-August and late September, 2001, and in late March, 2002.
The 2001 Residential Finance Survey was essentially a centralized mail-out/mail-back operation conducted by the National Processing Center in Jeffersonville, Indiana. The mail enumeration of property owners occurred mainly in the months of April 2001 through March 2002, while the lenders were polled in August 2001 through March 2002.
Field interviewing for property owners was limited to cases in which the owner was either not identified or failed to respond to the original mailed questionnaire and followup letters. Followup enumeration was conducted by field representatives from the 12 census regional offices.
Field representatives were trained for the followup enumeration of property owners through an extensive home study course and a 1-day classroom training session conducted by supervisory personnel at each of the 12 census regional offices.
Followup interviews of lenders (banks, savings and loans, insurance companies, etc.) were made by the National Processing Center in Jeffersonville, Indiana. If the lenders had not received the initial mailout of questionnaires, new documents were mailed to them. Those lenders who indicated they would respond electronically but had not done so were contacted by staff at the Census Bureau's headquarters office in the Washington, DC area.
The homeowner and rental and vacant property questionnaires were received in Jeffersonville, Indiana, National Processing Center. Each questionnaire was screened for completeness, mortgage status, tenure, consistency between expected and actual number of housing units, and determination of whether the property was within the scope of the survey.
After screening, a clerical edit operation was performed to ensure that the proper questionnaire (homeowner or rental) was completed by the respondent, that the answers referred to the sample address, and that all sampled addresses were associated with the correct property. Questionnaires which failed the clerical edit were referred to professional staff at headquarters in Washington for resolution. Upon completion of the clerical edit, questionnaires for nonmortgaged properties were sent to data capture, where the questionnaires were scanned. Images of the questionnaire pages were edited, based on guidelines and limits determined by the Washington staff. Those cases where there were questions which failed an edit were referred to the Washington staff for electronic review online.
Additional clerical edits were performed on questionnaires for mortgaged properties. One of the most important edits was to determine which lender held the mortgage(s) for the property. Prior to the 2001 Residential Finance Survey, a “pre-survey” contact of lenders was conducted. As a result of this survey, a database of approximately 6,000 lenders was established. If a mortgaged property questionnaire listed a lender not on the database, that lender was added to the database. After mortgaged property edits were completed, the questionnaires were forwarded to data capture, where the questionnaires were also scanned and the images of the questionnaire pages edited. The Washington staff also reviewed the mortgaged questionnaires online if the questionnaires failed preliminary edits. Once the lenders were identified for mortgaged properties, they were placed in a mortgage/lender database. This database was used to create and mail out lender questionnaires. If a lender indicated they did not want to receive a paper questionnaire, but wished to respond electronically, a separate database for electronic reporting was created.
Lender questionnaires were returned to Jeffersonville, Indiana. Each questionnaire was screened for completeness and consistency before being sent to data capture, where the questionnaire was scanned. Each questionnaire (homeowner, rental and vacant, and lender) was assigned a unique property address control number. Based on this number, a data file was created by linking the property (homeowner or rental and vacant) questionnaire to the corresponding lender questionnaire(s). After editing this file, professional staff in Washington was able to view the scanned images of the property questionnaire and corresponding lender questionnaire(s) and resolve any discrepancies. The mortgage edit was done to ensure that the property owner and mortgage lender were reporting for the same mortgage and that all mortgages on the property were accounted for.
Specifically, mortgage edit comprised the following activities:
1. The reduction of the number of “No reports.” In this volume no attempt was made to eliminate the “Not reported” category unless it could be done on the basis of other information provided by the property owner or mortgage lender. For example, if the owner failed to report the year the property was acquired, and also did not report that the mortgage was made at the time of acquisition, but the lender reported the year the mortgage was made, this was assumed also to be the year of acquisition. It was possible to eliminate the “Not reported” category for most mortgage items for two reasons. First, there were two sources of information (the property owner and the mortgage lender). Second, if a few facts are known about a mortgage, it is frequently possible to compute the missing loan information.
2. The interpretation of respondents’ notes, which were numerous and frequently complex. To cite a few examples:
3. Proration. This occurred when an owner reported his/her project as a single property, but the lender’s response indicated more than one property was involved with the holding. By definition, a property is what is covered by a single first mortgage. It was necessary to prorate the owner’s data on various property items; that is, number of units, value, expenses. Proration was usually done on the basis of the face amounts of the mortgages and/or the number of housing units on the property.
4. Consistency checks. The editors made a number of consistency checks. For example:
5. Allocation of mortgage information. Numerous steps were taken to ensure as complete a response to the 2001 Residential Finance Survey as possible. Despite these efforts, lender reports for a significant number of mortgages were not received. This occurred for several reasons:
The property owner and mortgage lender were asked a certain number of similar questions about the mortgage on the property. This was done to ensure that both were reporting on the same property and the same mortgage. Cases where the property owner made a complete report about the property, but where the mortgage lender did not report, became eligible for allocation of lender information.
In most cases the allocation of lender information was done by trained and experienced headquarters staff. In a small number of cases involving interest only loans, a computer program was written and the allocation done by computer. In all cases, the allocation of lender information was based on information provided by the property owner. Allocated lender records are identified as such on the computer file.
Three distinct computer edits were performed on the survey data. The first edit made a recheck of selected “key” items to ensure that they were answered. It also made consistency checks for each data record (for example, if the property had two mortgages, there must be two mortgage documents for that property). This edit also checked for duplicate records for the same property. All cases which “failed edit” were reviewed and corrected by professional staff.
The second edit was the allocation edit. In a small number of cases as described above, a lender record was allocated by computer based on information reported by the property owner.
The third computer edit made a final check for internal consistency between items within a data record. For a few individual items, response allocations were made based on this check. In addition, this edit performed a “clean-up” function propr to data tabulation (for example, blanking items which should not have been answered based on the questinnaire “skip” pattern).
The weighting of data (including nonresponse adjustment and ratio estimation) along with the calculation of standard errors, medians, means, and all ratios, and the final tabulation of the data were performed by computer.