[FRB logo]
[SCF logo]
 1986 Survey of Consumer Finances
Here: Introduction
Next: Question Text, Variable Names, & Responses


CODEBOOK FOR 1986 SURVEY OF CONSUMER FINANCES

Robert B. Avery and Arthur B. Kennickell
Board of Governors of the Federal Reserve System



Table of Contents
1. Introduction
2. Question Text, Variable Names, & Responses



INTRODUCTION


The Federal Reserve Board has sponsored a series of Surveys of Consumer Finances (SCF) dating back to the 1940's. After a several-year hiatus in the 1970s, the survey effort was renewed in 1983 and now continues, in expanded form, on a triennial basis. The 1986 survey is the second of this new series. The earlier surveys, conducted annually by the Survey Research Center (SRC) at the University of Michigan between 1946 and 1970, concentrated on collecting information on household expectations and purchase intentions. Only very sketchy information was gathered on household finances, although some data on debts and income were collected in the later years. Although the surveys were discontinued in 1970, in 1977 the Federal Reserve and other federal bank regulatory agencies sponsored an ad hoc survey on consumer credit usage. Also conducted by SRC, this survey gathered detailed data on consumer debts and attitudes toward debt. Information was also collected on assets, thereby permitting the calculation of household wealth.

In 1983 the Board decided to initiate a new series of Surveys of Consumer Finances with an emphasis on the measurement of household wealth. The recognition of a number of federal agencies, particularly the Federal Reserve and Department of Health and Human Services, that they had a joint interest in collecting high-quality household financial data precipitated this decision. Joint survey sponsorship from several agencies produced a much richer questionnaire than would have been likely under narrower guidance. The 1983 survey gathered detailed data on household balance sheets, income and employment history, expectations of retirement and future pension and Social Security benefits, and attitudes toward and use of consumer credit. In addition, SRC performed a linked survey of the employers of survey respondents designed to gather detailed information on the structure of firms' pension plans. In order to ensure that the entire wealth distribution was adequately sampled, the 1983 SCF supplemented its cross-section sample with an additional sample of wealthy respondents drawn from tax files.

The 1986 wave of the SCF reinterviewed 1983 survey respondents. While this questionnaire obtained less detailed balance sheet data, sufficient information was collected to compute saving for these households over 1983-86. Employment information was also obtained to update the work history of the panel. In addition, the 1986 survey opened new lines of inquiry to explore the broad role of the family in economic behavior. Planning is now underway for a third wave of the new SCF, scheduled to take place in 1989.

This manual describes the set of recoded, edited, and imputed variables developed at the Federal Reserve Board for the 1986 wave of the SCF. It documents the procedures used for editing the raw survey responses, the statistical methods used for imputing missing data, the construction of new variables from the original variables, and the addition of new variables which have been created by matching survey data with information from other sources. The manual also presents technical material on the survey's design and weights. It should be used in conjunction with a number of other documents, including the comparable manual for the 1983 wave, cited below; the SRC release titled the "1986 Survey of Consumer Finances," which describes those variables representing the raw responses of the survey respondents; and a comparable release from SRC for the 1983 SCF. Additional information on the SCFs, particularly the 1986 wave, can be obtained from:

Robert B. Avery, Gregory E. Elliehausen, and Arthur
B. Kennickell, "Changes in Consumer Installment Debt: Evidence from
the 1983 and 1986 Surveys of Consumer Finances," ,Federal Reserve
Bulletin, Vol. 73, No. 10 (October 1987), pp. 761-778.

Robert B. Avery and Arthur B. Kennickell, "Savings and Wealth:
Evidence from the 1986 Survey of Consumer Finances," presented at
the May 1988 NBER Conference on Research in Income and Wealth.

Robert B. Avery, Gregory E. Elliehausen, Glenn B. Canner, and Thomas
A. Gustafson, "1983 Survey of Consumer Finances," Federal Reserve
Bulletin, 70 (September 1984): pp. 679-92.

______________________ , "1983 Survey of Consumer Finances: A Second
Report," Federal Reserve Bulletin, 70 (December 1984): pp. 857-868.

Robert B. Avery and Gregory E. Elliehausen, 1983 Survey of Consumer
Finances: Technical Manual and Codebook, Board of Governors of the
Federal Reserve System, 1985.

______________________ , "Financial Characteristics of High-Income
Families," Federal Reserve Bulletin, 72 (March 1986): pp. 163-77.

Robert B. Avery, Gregory E. Elliehausen, and Arthur B. Kennickell,
"Measuring Wealth with Survey Data: An Evaluation of the 1983 Survey
of Consumer Finances," Review of Income and Wealth, December 1988.

The data from the 1983 and 1986 SCF waves and the supplemental pension survey are available on request from the National Technical Information Service, 5283 Port Royal Road, Springfield, Virginia 22161 (telephone 703-487-4600).

THE 1983 SURVEY OF CONSUMER FINANCES
The 1983 SCF was conducted through in-person household interviews between February and August 1983. Complete details on the survey can be found in the 1983 Survey of Consumer Finances: Technical Manual and Codebook. Briefly, the survey -- sponsored jointly by The Federal Reserve Board; the Department of Health and Human Services; the Department of the Treasury, Office of Tax Analysis; the Federal Trade Commission; the Department of Labor; the Office of the Comptroller of the Currency; and the Federal Deposit Insurance Corporation -- was conducted by the Survey Research Center[1]. The sample group was drawn from a dual frame: 5,396 households were selected for interviewing from a standard area-probability frame and 5,000 returns of high-income taxpayers were selected by the Statistics of Income Division (SOI) of the Internal Revenue Service. Of the households selected for the area-probability sample, 3,824 (71 percent) ultimately participated; this figure is somewhat lower than expected due to a higher than expected rate of vacant dwellings[2]. At first, the SOI sample members were approached indirectly. The Comptroller of the Currency sent introductory letters requesting participation in the study to a sample of high-income 1980 taxpayers selected by SOI and residing in the same sampling areas used for the area probability sample[3]. Of the 459 households that subsequently agreed to participate, 438 ultimately completed interviews. While the completion rate in the SOI sample is not high, it appears to be not radically out of line with response rates to previous SOI mail surveys.

Within each survey household, the "economically dominant" (primary) family (or individual) was interviewed[4]. The same questionnaire was used to interview respondents in both the area- probability and high-income samples, and field interviewers were not told which households were part of the high-income sample. The length of the interview averaged 74 minutes for the area- probability sample and 87 minutes for the high-income sample. Information was collected (at the household level) on the following items:

Financial Assets. The amount, type, and source of each household checking, savings, and money market account. Household holdings of publicly traded stock, bonds, certificates of deposit, mutual funds, trusts, and notes or land contracts. Face and cash value of life insurance.

Tangible Assets. The value and purchase terms of the household's principal residence, other real estate, and automobiles. Limited data on the value and characteristics of businesses in which a member of the household had both a management and equity interest.

Pension Assets. I.R.A., Keogh, thrift, profit sharing, 401K, and other tax-deferred account holdings. Estimates of the present value and other information on each current or expected pension, including Social Security, of the household head and spouse.

Household Debts. The purpose, amount outstanding, source, and terms of each household debt, including home mortgages and home equity loans, lines of credit, credit cards, and other consumer loans.

Demographic Data. Education, employment and marital history, age, race, and earnings for the head and spouse of each household. A detailed breakdown of income by source.

Attitudes and Financial Decision Making. Attitudes toward saving and credit use. Data on the use of financial services, methods of choosing among competing sources, and the number and proximity of the respondent's financial institutions.

A unique and powerful feature of the 1983 SCF is the range of information on pensions and retirement. In addition to data gathered from respondents (and their spouses) about their expected dates of retirement and their expected benefits from Social Security and pensions, the names of their employers were also obtained. Using this information, the pension plans were identified, and detailed information on the plans was drawn from the Summary Plan Descriptions filed with the Department of Labor. Publicly available software has been developed to translate this complex information into more digestible figures, such as present values of benefits and effects of various decisions (such as early retirement) on pension benefits. This offers the opportunity to compare respondents' perceptions of their pension benefits and coverage with the "reality" as provided by the firms[5].

SURVEY DESIGN
The 1986 SCF -- sponsored by the Federal Reserve Board, the Department of Health and Human Services, the Office of the Comptroller of the Currency, and the General Accounting Office -- was a more limited 27-minute re-interview survey with respondents to the 1983 SCF[6]. When 1983 couples divorced or separated, the survey followed both parties. However, other members who left the household, such as young adults were not followed. A total of 2,822 interviews were completed, by telephone, between June and September 1986. As in 1983, the unit of observation was the family.

The 1986 SCF was designed primarily to update essential information in the 1983 SCF -- the household balance sheet and employment data. Sufficient information was collected that household net worth could be estimated, although more aggregated asset and debt categories were used (roughly 25 categories versus 85 in 1983). Limited information was also solicited on the disposition of assets in divorce or upon death of a spouse, purchases and sales of houses, and major expenditures for health, durables, charity, and education. Substantial new information was collected on gifts made or received by the household and the recipient or source of those gifts. In a related vein, data were gathered on the sharing of living quarters over the three-year period. In addition, income, marital, and employment history over the intervening three-year period was gathered.

1986 SAMPLE WEIGHTS
Because observations for the 1983 SCF were drawn from two different sampling frames, and because the 1986 survey is a re-interview with only a portion of the 1983 sample, construction of appropriate sample weights is a particularly important issue. Weights for the 1983 survey were constructed in two phases. Relative weights for the area-probability sample were constructed to compensate for differential non-response rates across the survey's 75 primary sampling units. Those weights were further post-stratified by region and degree of urbanization to reflect population estimates from the March 1983 Current Population Survey (CPS).

Construction of weights for the full 1983 sample, including both the area-probability and high-income sub-samples, posed a more difficult problem. Full information on the high-income sampling procedure is not available, nor is the information collected from survey respondents sufficient to construct a fully accurate measure of the income concepts that the IRS likely used in drawing the sample. Additional complications stem from the fact that the high-income observations were selected from a 1980 sampling frame, but reported data as of 1983, and the fact that the reporting basis for tax files (individuals or married couples) is not always the same as the survey (families).

These problems led to the design of 1983 sampling weights for the high-income sample (and area-probability observations with income above a certain level) using a post-stratification scheme based on control totals for an "extended" income measure (roughly, equal to IRS-defined adjusted gross income plus excluded realized capital gains) constructed from the 1982 Tax Model File (TMF) of the IRS[7]. Weights were determined so that the weighted number of survey observations for six cells with extended income above $80,000 matched TMF control totals. The original weights of the area-probability observations with income below $80,000 were adjusted so that the weighted number of SCF households equaled the population estimated from the CPS. High-income sample observations with income below $80,000 were arbitrarily assigned the same weight as observations in the $80,000 to $90,000 group.

Construction of the 1986 sample weights proved even more problematic. As we argue later, there are at least two different interpretations of the 1986 sample relevant for analysis. Thus, at least two different 1986 weights are necessary. For many purposes, it is useful to view the 1986 sample as a sub-sample of the 1983 survey. However, it is also possible to view the 1986 sample as a new cross-section, representative of the structure of the 1986 household population. Accordingly, we have constructed two sampling weights for the 1986 sample. The first weight -- which we term the "1983/86 weight" -- was designed to allow the sample of households reached in 1986 to represent the entire 1983 sample. The second 1986 sample weight -- deemed the "1986 weight" -- may be used to represent the 1986 household population.

As the basis for both weights, the original 1983 weights were adjusted for attrition using a multi-stage probability model estimated with a variety of 1983 financial, demographic and other characteristics; the 1983 weight was multiplied by the inverse of the estimated probability of being in the 1986 sample. As can be seen in table 1, the 1986 sample is clearly not a random sample from the 1983 survey.

Each weight was further adjusted for changes in marital status and for deviations from externally observed measures of the population structure. At the second stage of calculation, for the 1983/86 weight, the base weight was halved for all households that separated or divorced; for the 1986 weight, households married since the 1983 survey received half of the base weight[8]. Although the sample follows separations and divorces of primary family heads and spouses, these second-stage adjustments do not fully reflect external measures of the structure of the population. Finally, to compensate for this difference, the weights were post-stratified to age, marital status (including the sex of single respondents), and homeownership cells[9]. As discussed below, this adjustment was performed in either of two ways, by using information from 1983 and 1986 independently, or by exploiting the information in the change in the population structure between the two years.

Table 1
                                      Sample Attrition

                	Weighted	Percent of	Weighted
	                percent of	1983 group	percent of
	                1983 sample	in 1986	        1986 sample

Age (head)
  under 25	            8.0	          56.6	           8.6
  25 to 34	           22.6	          62.8	          23.7
  35 to 44	           19.5	          68.9	          20.8
  45 to 54	           15.5	          67.7	          14.4
  55 to 64 	           15.0	          69.3	          14.5
  65 or more	           19.4	          56.6	          18.0

Marital Status
  married	           60.6	          67.8	          64.3
  sep/divorced male	    4.9	          56.9	           4.4
  sep/divorced female	   10.7	          66.5	          11.2
  widowed male	            1.9	          46.2	           1.3
  widowed female	    9.5	          58.0	           8.1
  never married male	    6.3	          57.4	           5.7
  never married female	    6.1	          49.4	           5.0

Race
  Caucasian	           82.3	          67.5	          82.3
  Nonwhite or Hispanic	   17.7	          47.7	          17.7

Family Income (1982)
  less than $10,000	   24.0	          46.5	          21.2
  $10,000 to $19,999	   26.8	          62.1	          28.7
  $20,000 to $29,999	   19.3	          69.8	          19.7
  $30,000 to $49,999	   19.7	          75.2	          20.1
  $50,000 to $99,999	    8.2	          77.0	           8.3
  $100,000 or more	    2.0	          80.0	           2.1

Family Net Worth
  less than $100,000	   76.6	          61.2	          77.4
  $100,000 to $249,999	   14.7	          72.0	          13.5
  $250,000 to $999,999	    7.1	          75.0	           7.7
  $1,000,000 to $2,499,999  1.2	          77.2	           1.1
  $2,500,000 or more	    0.5           71.4	           0.4

Homeownership
  homeowners	           63.4	          70.9	          64.4
  other	                   36.6	          52.1	          35.6

Education of the head
  0 to 8 grades	           14.5	          55.1	          14.2
  9 to 12 grades	   44.9	          61.6	          45.6
  some college	           17.7	          66.4	          17.4
  college graduate	   22.9	          72.7	          22.8

Labor force participation
  Married
    only head working	   19.0	          70.6	          20.6
    only spouse working	    4.0	          56.9	           3.9
    head & spouse working  27.8	          73.1	          31.1
    neither working	    9.8	          52.0	           8.7
  Single
    working	           22.4	          65.1	          21.5
    not working	           17.0	          49.1	          14.2

Sample
  area-probability	   98.2	          63.7	          98.1
  high-income	            1.8	          82.9	           1.9

Total	                  100.0	          64.0	         100.0

*Groups are defined by their 1983 status.
For the 1986 weight, there are two qualifications to these post-stratification schemes. First, the 1983 SCF interviewed only independent households. However, because at any given time a significant proportion of younger adults are in school or the military, or live with their parents, those younger people living independently at the time of the 1983 SCF are unlikely to represent the population of households of their cohort three years later. Therefore, the sample has not been weighted to represent the population of households with heads aged 24 and under in 1986. Households that fell into that group in 1986 were assigned weights adjusted for attrition and marital status change, but were not post-stratified to 1986 control totals. For this reason, we strongly recommend that only those households aged 25 or more in 1986 be used for analysis of changes[10]. Second, some individuals who were heads of households (or spouses of heads) in 1983 moved into living arrangements where in 1986 they would no longer have been defined as a head or spouse using SRC definitions (such as young adults moving back to their parents' home). In these circumstances, the post-stratification scheme used the age, sex, and marital status of the 1986 household member who would have been identified as the head, instead of the characteristics of the 1983 respondent.

What follows is a more formal description of the derivation of the 1986 weights. Let W83 , original weight in the 1983 survey[11]. Let W8386 be the weight that weights the 1986 sample to represent the 1983 population. Let W86 be the weight that weights the 1986 sample to represent the 1986 population. These weights are closely related. Suppose for the moment that there had been no attrition from the sample between 1983 and 1986 and that the household universe in each year was equivalent (absent the effects of aging), but there had been some number of divorces and new marriages. In this case,
	W8386 = W83 /2  if a couple has been divorced or
	      = W83 otherwise;
or summarize the relationship between W83 and W8386 by
	W8386 = W83 * phi(1) ; and

	W86 = W8386 /2 if a single person has married outside the sample,
	W86 = W8386 * 2 if a couple has been divorced, or
	W86 = W8386 otherwise;
or summarize the relationship of W8386 to W86 for each individual by
	W86 = W8386 * phi(2) .
First we relax the assumption of no sample attrition between 1983 and 1986[12]. Let the inverse of the true probability of each household's remaining in the sample be given by b. A part of b can be estimated using the information on the distribution of characteristics observed in the 1983 sample, and a part, say G, is related to unobservable characteristics. Let this relationship be summarized as
	b = a * G
In this case, the relationships among the weights are summarized by
	W8386 = W83 * phi(1) * a * G   and
	W86 = W8386 * phi(2) * a * G  .
The quantities W83, phi(1), phi(2), and a are all directly estimable. One way to estimate the missing parameter G is to use information available about the structure of the population in 1983 and 1986 to estimate G iteratively in the following way. Let the population in 1983 and in 1986 be divided into s = 1,...,n cells and let p83(s) and p86(s) be the true population counts for cell s. Define W8386(i), W86(i) and G(i) as the ith-round estimates of W8386, W83 and G, respectively. Let sigma8386(i)(s) be a scaling factor such that
	sigma8386(i)(s) * 
             sum(over W8386 in cell s) W8386(i) = p83(s) ;
similarly, let sigma86(i) be defined by
        sigma86(i)(s) *
             sum(over W86 in cell s) W86(i) = p86(s) ;
Dropping the cell-specific notation for simplicity, the iterations are defined as follows[13]:
         
        W8386(0) = W83*phi(1)*a*sigma8386(0)

        W86(0) = W8386(0)*phi(2)*a*sigma86(0)


        W8386(1) = W8386(0)*phi(1)*a*sigma8386(1)

        W86(1) = W8386(1)*phi(2)*a*sigma86(1)

        .         .
        .         .
        .         .


        W8386(T) = W8386(T)*phi(1)*a*sigma8386(T)

        W86(T) = W8386(T)*phi(2)*a*sigma86(T)
Implicitly, the final estimate of G for each case is given by
        G(T) = {product(i=0 to T) sigma8386(i)} *
               {product(i=0 to T) sigma86(i)}
In terms of this notation, the difference between the sets of weights given in this codebook that use 1983 and 1986 data independently for post-stratification (C1013, C1015 and C1016) and the weights that use both years of data jointly (C1014, C1017 and C1018) is that the former uses only the first round of this iterative process while the latter is based on the 30th iteration (change is minimal after the 10th iteration). As discussed further below, however, this difference can also be interpreted in terms of the comparability of the household universes in each year.

As we have proceeded, the estimation of the attrition adjustment model that underlies the calculation of a is a multi-stage process. A set of the 1983 final area-probability sample were excluded in 1986 when neither telephone nor address information was obtainable; all of the original high-income sample were approached. For the area-probability cases, a probit model is estimated for this first-stage exclusion (that adjustment is identically unity for the high-income cases).

At the second stage, attrition is modeled separately for different 1983 populations: single males, single females and married couples. The modeling for the single people is straightforward; either the individuals were found or not found. The 1983 married couples are somewhat more complicated: if the couple remain married to each other, they can be found or not found; if they divorce, either can be found separately, or either or both can be not found. For this group, the probabilities of these events are computed using two models sequentially: the first model calculates the probability that both halves of a 1983 married couple are found (still married to each other, married to someone else, or as single people), and the second model calculates the probability that one person was found (either alone or married to someone else) given that both halves were not found. In terms of the notation above, the final estimate of a is given by 
			  ( Pr2(found), for 1983 single males, found;
			  |
			  | Pr3(found), for 1983 single female, found;
			  |
   a = Pr1(in 1st         |
          stage sample) * { Pr4(both found), for 1983 married couples,
                   	  |   both found; or
			  |
			  | (1-Pr4(both found))*Pr5(one found), for
			  (  1983 married couples, one found.
All of these probability models were estimated as weighted probits using nearly the same set of independent variables drawn entirely from the 1983 data. These variables include: a quadratic spline on the age of the 1983 head of household, dummies for years of education, indicators of income (logarithm of total family income, dummy for capital income, logarithm of capital income, interactions of age of the head of the household with the logarithms of total family income and capital income), indicators of wealth (logarithm of net wealth, dummy for home ownership, dummy for mortgage on the principal residence, dummy for consumer debt and logarithm of consumer debt, dummy for credit card debt, logarithm of paper assets), household composition variables (number of people in the household, number of children in the household, age of the youngest and oldest children, dummy for children outside the household), an interaction term of the age of the head of the household with the dummy for homeownership, number of years the respondent had lived in the county of residence in 1983, a dummy reflecting whether the respondent had given "saving to buy a house" as a reason for saving in 1983, a dummy for membership in the high-income sample, and variables reflecting the level of interest and suspicion that interviewers perceived in the respondents in 1983.

The post-stratification cells were designed to reflect CPS population characteristics while preserving differences due to different styles of measurement in the CPS and the SCF[14]. For W8386, the post-stratification cells were a cross of 1983 family type (single female, single male, or married couple) by age class (1983 ages 22 to 41, 42 to 61, and 62 and over) by 1983 housing tenure (homeowner or not); for W86, 1986 characteristics were used to define identical cells except in the case of the age groups (age cells were 25 to 44, 45 to 64, and 65 and over). The 1983 control totals were derived from the full sample of the 1983 SCF; to preserve subtle differences between Census and SRC sampling procedures, the 1986 control totals were calculated as the 1986 CPS cell counts less the difference between the counts for the same cells in the 1983 CPS and the 1983 SCF.

The post-stratification procedures followed in creating these weights make implicit assumptions about the dynamics of the underlying household universe. For the versions of W8386 that reweight the sample to align with 1983 controls independently of 1986 information (C1015 and C1016), the assumption is simply that the adjusted 1986 sample is still representative of the population from which it was originally drawn. However, the question is somewhat more complex for the 1986 weights (C1013 and C1014) and the versions of W8386 that use both 1983 and 1986 information for post-stratification (C1017 and C1018); all of these weights implicitly assume that the 1986 household universe is simply an aging of the 1983 universe with additions at the younger end for newly-formed households[15]. While it is obvious that this assumption is not strictly valid (e.g., there is immigration), there is little information available to assess how significantly it deviates from reality.

As one check of this difference, we have taken the population of heads and spouses represented in the 1983 CPS, aged that population three years with adjustments for mortality, and compared it with the group of heads and spouses actually observed in the 1986 CPS (table 2). As expected, the largest deviations appear at the younger end of the age distribution where people are moving from living in institutional arrangements (dormitories, the military, jail, etc.) to living in households. While this difference is particularly great below the age 25 cut-off we have chosen in computing our weights, the pattern is pronounced into the 30s for both sexes. For older people, the differences reflect movements into institutional quarters, particularly nursing homes. However, it is surprising that this over-prediction begins in the 56 to 60 age group. This may reflect either substantial population shifts that we have not identified or errors in the mortality factors applied.

Table 2
     Difference in Actual and Predicted 1986 CPS Household Population,
                              By Age Groups
                          (Thousands of Persons)

Age group	Males		    Females	        Males and Females
	   Actual  Actual	Actual  Actual	        Actual  Actual
        -Predicted           -Predicted              -Predicted
 
21 to 25   2,719   4,417	3,105	 6,219	         5,824	10,636
26 to 30   1,555   7,753	1,053	 8,825	         2,608	16,578
31 to 35     558   8,206	  530	 9,139	         1,088	17,345
36 to 40     272   7,682	  129	 8,358	           401	16,041
41 to 45    -255   6,110	  -99	 6,642	          -354	12,752
46 to 50      48   5,200	  -66	 5,528	           -18	10,728
51 to 55     155   4,984	   -4	 5,367	           152	10,351
56 to 60     -49   5,011	  -88	 5,569	          -137	10,581
61 to 65    -260   4,572	    2	 5,479	          -258	10,050
66 to 70    -157   3,787	 -107	 4,592	          -264	 8,379
71 to 75    -267   2,807	 -248	 3,781	          -515	 6,589
76 to 80    -156   1,736	 -292	 2,558	          -447	 4,294
Over 80	    -490   1,245	 -583	 2,288	        -1,073	 3,529
One final complication for W8386 is the treatment of households that passed out of the sampled universe of households either by moving to an institution, or by dying (these cannot be distinguished from the data coded). For some purposes turning on questions of differential mortality, it may be useful to have estimates of the 1983 population including this group of households. For this reason, versions of W8386 including such households have been provided (C1016 and C1018). Note that W86 is identically zero for these cases.

NON-RESPONSE AND IMPUTATIONS OF MISSING DATA
In any household survey, some responses to survey questions will be missing due to respondents' lack of knowledge or unwillingness to answer. In keeping with most comparable surveys, extensive steps were undertaken to impute missing data for the 1983 and 1986 SCFs. The size and complexity of the SCF made imputation difficult. While the sample was too small to impute missing values with "hot deck" or matching techniques used by the Census Bureau, the comprehensiveness of the questionnaire offered opportunities for inference not found with shorter surveys, allowing the use of a number of different methods for imputation.

For the 1983 survey, which was cleaned first, missing values were imputed using: (1) formulas based on respondent information that was closely related to the missing items; (2) randomized draws from conditional frequency distributions (used primarily for discrete variables); and (3) a variety of regression models (missing values were assigned the regression prediction plus a random disturbance designed to preserve the second moments of the sample)[16]. Income and asset regression imputations were done simultaneously, using an iterative technique in order to preserve their full covariance structures[17].

The area-probability and high-income samples were handled separately. Missing values for all observations in the high-income sample were imputed. In the area-probability sample, however, 159 of the original 3824 observations in that sample were discarded because virtually all dollar amounts for income and assets were missing. For the remaining observations, all missing values were imputed.

Similar methods were used to impute missing values for the 1986 survey. Specific account, however, was taken of the panel nature of the data. Thus, regression-based imputations used 1983 values (whether imputed or not) in the prediction equations. These procedures were designed to preserve the inter-temporal correlations among the data. Special problems were created, however, when data were originally missing in 1983, but not 1986. In these cases, it was necessary to re-impute 1983 values using 1986 data. As with the 1983 data, asset and income data were imputed simultaneously and the area-probability and high-income samples were handled separately.

Several important qualifications should be made about the relative quality of the 1983 and 1986 data. The questionnaire for the 1983 survey was far more extensive than that for the 1986 survey; and the 1983 survey took place in person, while the 1986 survey was conducted by telephone. Some information, such as the value of principal residences and home mortgages, was solicited in similar ways in both years. Other data, such as financial assets, were collected in much more aggregated form in 1986. In making comparisons across the two years, three areas may be particularly subject to problems of measurement error. First, only the size of payments on consumer installment debts and some investment property mortgages were collected in 1986; thus, amounts outstanding on these debts had to be estimated using independent information on average terms. Second, for automobiles, only purchase data were collected in 1986; thus, assumptions had to be made about the treatment of existing automobiles to value the 1986 stock. Finally, it appears that problems in the ordering of questions in the business and employment sections of the 1986 survey caused many businesses to go unreported. While imputations were made in some of these instances based on 1983 data, we suspect that aggregate 1986 businesses in the survey are still underestimated.

The significance of the imputation method is illustrated by the figures in table 3. Column 4 shows the percentage of owners in each of 12 asset and debt items who did not disclose the full dollar value of their holdings in 1986. Column 3 shows the percentage of the sample total value for each item that was created by imputation. As described, business items were particularly troublesome. Virtually every asset category, however, had at least 10 percent of its value imputed. Overall, 17.4 percent of the sample wealth was imputed.

Table 3
    
                                        Imputations

                	   --  FINAL SAMPLE --	--  IMPUTATIONS --

                             Percent   Means of	  Percent  Percent
	                     owning    owners  	  missing  imputed
	----------------------------------------------------------

 ASSETS	                       97.3%	173663	   38.5%    16.2%
   Principal residence	       66.1%	 80335	    8.1%     5.7%
   Other real estate (gross)   22.1%	120124	   22.5%    18.5%
   Public stock	               19.8%	 75594	   20.3%    25.8%
   Bonds	               20.6%	 26359	   13.3%    12.3%
   Check/savings accounts      88.7%	  7709	   22.6%    31.1%
   IRAs	                       27.1%	 19580	    8.2%    11.1%
   CDs/Money market	       27.6%	 33877	   15.0%    17.9%
   Business assets (net)       12.8%	206413	   56.5%    38.5%
   Automobiles	               88.5%	  7306	     ---      ---
   Profit sharing & thrifts    15.0%	 27033	   27.0%    16.9%
   Miscellaneous	       44.6%	 23512	    3.2%    10.5%

 DEBT	                       82.8%	 28745	   33.0%     8.7%
   Principal residence debt    38.4%	 34152	   10.6%     9.4%
   Installment debt	       76.1%	  4476	   30.5%     8.7%
   Other debt	               22.2%	  7782	    3.4%     1.8%
   Other real estate debt	9.4%	 59141	   14.8%     9.4%

 NET WORTH	              100.0%	145226	   51.3%    17.4%

 INCOME (GROSS)	              100.0%	 31731	    6.8%     7.9%

*Figures for gross assets, debts and networth are the percentage or
asset or debt owners or for net worth, the entire sample of households
aged 25 or older, who are missing any asset, debt, or wealth item. 

**This is the percentage of the aggregate weighted sample total of
each item which was imputed.
COMPARABILITY WITH OTHER DATA
Wealth information, and particularly data on changes in wealth at the household level, are not available from many sources. Perhaps the finest sources of household wealth data are the Federal Reserve Board's 1963 Survey of Financial Characteristics of Consumers (SFCC), and the follow-up re-interview survey in 1964[18]. Methodological work for this survey was conducted by the SRC, with interviewing performed by the Bureau of the Census. Like the 1983 SCF, the 1963 SFCC collected a more detailed inventory of assets and liabilities than is customary in other consumer surveys. The 1963 survey also used Federal tax information to oversample high-income households. A cross-section sample of housing units stratified by income reported in the 1960 Decennial Census was chosen to represent households with incomes below $50,000. Households with incomes of $50,000 or more were selected from a sample of 1960 Federal income tax returns. Although this sample selection procedure is not exactly the same as that used for the 1983 survey, it produced a heavy over-sampling of households in the upper end of the income distribution, making the 1963 sample the only household survey sample comparable to the full 1983 SCF sample.

Current data on wealth are available from the Internal Revenue Service, which uses federal estate tax returns to estimate total household wealth and its percentage distribution. Unfortunately, data from this source are available only in aggregate form, with very limited demographic breakdowns. Another source is the 1979 Income Survey Development Program of the Department of Health and Human Services, which provides information for a sample of households larger than that of most other surveys of wealth. The New York Stock Exchange has also periodically conducted surveys of household stockholders, doing a survey at roughly the same time as the 1983 SCF and, more recently, in 1985. Wealth data was also collected on respondents to the Panel Study of Income Dynamics (PSID) in 1984.

The most comprehensive recent survey of household wealth was conducted in 1984 (and repeated annually since) by the Bureau of the Census on participants in the Survey of Income and Program Participation (SIPP). This survey solicited information similar to the SCF for a very large sample of households. Its initial panel was a random cross-section of about 21,000 households selected by procedures similar to those used to select the area-probability sample for the 1983 SCF. Net worth information was collected between September and December 1984[19]. Aggregate wealth estimates from the earlier Surveys of Consumer Finances and SIPP are generally comparable to those from the area-probability sample of the 1983 SCF in their understatement of aggregate wealth relative estimates from independent sources. Using comparably defined categories, we estimate an aggregate net worth for the 1983 SCF area-probability sample of $8,277 billion versus a $7,740 billion total for the SIPP sample. The difference derives primarily from a smaller estimate of small business assets in the SIPP. The full-sample SCF estimate of the same net wealth concept is $9,610 billion. Thus, it appears that the major difference between the two surveys arises from the inclusion of the high-income sample in the SCF.

The annual March Current Population Survey is perhaps the most comprehensive U.S. household economic survey, soliciting economic information from approximately 59,000 households. The representativeness of the 1986 SCF is demonstrated by a comparison of the sample distribution of various demographic variables for the SCF and comparable March 1986 CPS survey in table 4. The CPS data are given for "primary families" defined comparably to families in the SCF. As can be seen, the 1986 SCF has a very similar distribution for most items.

Table 4
                          A Comparison of the 1986 SCF and CPS

                                 	SCF	                CPS
                        	Number      Weighted	Number	     Weighted
                        	of cases    share	of cases     share
                                --------------------    ---------------------
Age (head)
34 or less
 married                        661         16.9        9922         15.7
 unmarried male                 223          6.2        3435          6.2
 unmarried female               273          7.6        4105          7.2
35 to 44		       					  
 married                        555         13.6        7830         13.6
 unmarried male                  71          1.8        1388          2.7
 unmarried female               151          4.1        2165          4.2
45 to 54		       					  
 married                        492         10.5        6253         10.1
 unmarried male                  70          1.7         912          1.7
 unmarried female               118          3.2        1735          3.0
55 to 64		       					  
 married                        475          9.7        5967          9.5
 unmarried male                  62          1.5         897          1.5
 unmarried female               136          3.7        2130          3.6
			       					  
65 or more		       					  
 married                        452          9.9        5532          9.7
 unmarried male                  74          1.8        1437          2.3
 unmarried female               290          7.7        5293          9.2
			       					  
Race			       					  
 Caucasian                     3468         82.3       47515         86.6
 Nonwhite or Hispanic           635         17.7       11486         12.4
			       					  
Family Income		       					  
 less than $10,000              912         24.0       15053         21.3
 $10,000 to $19,999             982         26.8       15580         23.1
 $20,000 to $29,999             711         19.3       12072         18.9
 $30,000 to $49,999             717         19.7       11533         22.7
 $50,000 to $99,999             309          8.2        4480         12.3
 $100,000 or more               472          2.0         283          1.7
			       					  
Homeownership                  2766         63.4       38320         63.8
			       					  
Education of the Head	       					  
 0 to 8 grades                  560         14.5        9155         12.8
 9 to 12 grades                1713         44.9       27269         46.5
 some college                   678         17.7       10355         18.6
 college graduate              1152         22.9       12222         22.1
Labor Force Participation      					  
 single not working             635         17.0       11130         19.3
 single, working                833         22.4       12367         22.2
 married, neither working       389          9.8        7088          6.2
 married, one working          1077         23.0       14023         22.8
 married, both working         1169         27.8       14393         29.5
			       					  
Totals                         4103        100.0       59001        100.0
The representative quality of the 1986 SCF data can also be evaluated by a comparison with aggregate data sources. The most comparable of these is the Flow-of-Funds (FOF) accounts. A comparison of the levels of various wealth items from the 1983 SCF and FOF accounts shows remarkable consistency, heretofore not observed with survey data (see Avery, Elliehausen, and Kennickell (1988)). Survey-based estimates for many asset and debt categories were within 5 to 15 percent of aggregate estimates drawn from the Flow-of-Funds (FOF) accounts for the same time period[20]. Moreover, the sign of the discrepancy was not consistent -- survey-based estimates were not systematically higher or lower than FOF estimates. Estimates did differ substantially for checking and savings accounts, business, and some real estate categories. However, since these are areas where there are significant problems in the FOF, it is not clear whether discrepancies stem from measurement problems in the survey-based estimates or from the FOF or both.

A similar comparison can be made to see if changes in aggregate household wealth as measured by survey data correspond to changes as measured by the FOF. Table 5 shows estimates of levels and changes in levels of household wealth from FOF and the 1983 and 1986 SCFs. The survey estimates given are weighted sums of the various asset types using the appropriate statistical sampling weights. For all categories except stocks and bonds and other debt, the match of survey and FOF growth rates are roughly in the same ranges[21]. The problems with debts may stem from the methods used to estimate outstanding amounts. The behavior of stocks is more puzzling. One explanation may be that because the 1986 survey was a less detailed survey than the 1983 survey, respondents may have been less rigorous in recalling exact market values of shares. In a time of rapidly rising share prices, this may have induced substantial bias.

Table 5
    A Comparison of Survey-based Wealth AggregatesWith Flow-of-Funds Estimates
                                   ($ Billions)

			      Survey	           Flow of Funds
Item 	              1983    1986    %Growth	1983	1986    %Growth
---------------------------------------------------------------------------
Financial Institution
  accounts/CDs	      1032.9  1417.0  37.2%	1832.5	2485.3	35.6%
Stocks/bonds	      1543.8  2088.6  35.3%	1438.3	2456.9	70.8%
Principal Residences  4276.4  5112.6  19.6%	2703.4	3388.3	25.3%
Non-corp. business    1852.8  1951.6   5.3%	2347.1	2415.6	 2.9%

Home mortgages	       995.4  1293.8  30.0%	1064.6	1480.6	39.1%
Other Debt	       224.9   468.6 108.4%	 332.8	 519.7	56.2%
---------------------------------------------------------------------------

	Classifications are not exhaustive and differ from those used
in previous tables.  The 1986 survey figures were extrapolated to
include households under 25.  The Flow-of-Funds estimates were
adjusted to take out non-profit holdings.
MANUAL INSTRUCTIONS
In the remainder of the manual, information is given on all the variables included in the final data set. A brief description is given for each variable along with information on imputation and a listing of the values that the variable takes on. The question number corresponding to the actual 1986 survey questionnaire (e.g. H6) is also given for all variables except recodes. Variables are listed by number, with the numeric code used as the basis of the variable's internal name in the survey's SAS data set. All variables listed here have a "C" prefix followed by a four digit number ranging from 1001 to 1976. The original uncleaned survey responses are contained in variables with a "W" prefix. These variables range from W1 to W1359. The codes for the "W" variables are described in the original survey codebook released by the SRC. We note, though, that the "C" variables contain all the same types of information as those contained in the "W" variables, as well as additional, constructed variables. Thus, for most analyses, it is possible to use the "C" variables without reference to the "W" variables. Much of the time, the 1986 data will be used in conjunction with data from the 1983 SCF. The 1983 variables are denoted by "B" prefixes for the recodes and "V" variables for the original uncleaned responses. The 1983 variables are described fully in the 1983 Survey of Consumer Finances: Technical Manual and Codebook.

The range of allowable values for the variables is also given. The symbol xxxx is used for continuous variables with a statement of the units used and the sample range. For discrete variables with a small number of allowable codes, all possible values and their meanings are listed. The number of sample cases (out of the 2,822 observations for the 1986 survey) taking on each value of discrete variables is also given. If the listing is for several variables (such as the 1st, 2nd, and 3rd reasons for saving), then the case totals are given for the listed variables, in order, separated by slashes (e.g. 123/45/87 cases). If a variable reflects 1983 data, case totals will be for the 1986 sample, not the 1983 sample. Although useful in giving a flavor of the distribution of responses to questions, these case listings should not be used for statistical purposes, as they are unweighted distributions.

Most of the information collected for the 1986 SCF applies to the entire family unit. Some information, however, such as employment, marital history, and pension income, was collected individually for the survey respondent and their spouse (if he or she had one). For married couples, the respondent could have been either the husband or the wife. Unlike the 1983 datafile, the 1986 survey file is arranged by "respondent" and "spouse." This was done because the 1986 respondent had to be in the 1983 survey. Thus, there are always 1983 variables for the 1986 respondent, but not necessarily for the respondent's spouse (who could be a new spouse). To facilitate inter-year comparisons, the 1983 data have been rearranged on the 1986 tape to 1986 respondent and the 1983 spouse of the 1986 respondent. If another order is desired, such as 1986 "head" and "spouse," (where head is always the husband for married couples), the data can be easily reordered. Several variables, -- C1004, C1005 and C1006 -- can be used to do this.

Several different codes are used in the data set, including:

(1) The code "1" is almost always used for the answer "yes" to a question;

(2) The code "3" is generally used for the answer "sometimes" or "maybe";

(3) The code "5" is almost always used for the answer "no" to a question;

(4) The code "-4" is used to denote a "small negative number";

(5) The code "-6" is used to denote the answer "none," which is sometimes differentiated from zero;

(6) The code "-7" is used to denote a special "other" response which does not fit into existing codes. SRC has cards indicating what the actual response is. "-7" is also sometimes used to denote answers like "forever" or "never" when used for continuous variables;

(7) The code "-8" is used to denote the answer "don't know" (DK). Most DKs have been imputed, but some still exist for selected variables where imputation is not appropriate (e.g. attitudinal variables);

(8) The code "-9" is used to denote "not answered" (NA). This indicates either that the interviewer inadvertently did not ask a question or that a respondent refused to answer. Most NAs have been imputed, but a few remain;

(9) The code "0" is generally used to denote cases where a variable is inappropriate for a particular observation because the question which underlies the variable was not asked. For example, questions on spouses would be inappropriate for single households. Note that sometimes a question is asked, but the answer given is none or zero (such as "my business is worth nothing"). These answers are generally coded as -6 not 0. There are some instances, particularly with recoded variables, where 0 does denote none or nothing.

All variables on the tape are integers. All dollar amounts are given in whole dollars (although in answering the questions respondents may have rounded). Some variables had to be rescaled so that information would not be lost (such as percentage answers which are generally multiplied by 100).

ENDNOTES
1. The interview questionnaire for the household survey was prepared by Robert B. Avery and Gregory E. Elliehausen, of the Federal Reserve Board, and Thomas A. Gustafson, of the Department of Health and Human Services, with assistance from staffs of the sponsoring agencies. Field work and editing and coding of survey responses was performed under the direction of Richard T. Curtin of the SRC. Mr. Curtin and Timothy Carr and Miles Maxfield, of Mathematica Policy Research, administered the supplemental pension survey. The Statistics of Income Division of the Internal Revenue Service, provided the high-income sample.

2. Observations selected for the 1983 SCF were drawn from 75 primary sampling units in 37 states and the District of Columbia. For a further discussion of the SRC sampling procedures, see Irene Hess, Sampling for Social Research Surveys: 1947-1980, Ann Arbor: Institute for Social Research, 1985.

3. For a general description of the sample from which the survey sample was drawn, see Internal Revenue Service, Statistics of Income-1982, Individual Income Tax Returns, Washington D.C.: U.S. Government Printing Office, 1984.

4. This definition of family differs from that of the Census Bureau, which excludes single individuals. Because some persons within a household -- those not related to the primary family -- were not interviewed by the SRC, wealth figures will understate the U.S. household total. We estimate, however, that the understatement is only about 0.4% for 1983. Because the number of primary families and households is the same, we use the terms interchangeably in this manual.

5. The supplemental "Pension Provider Survey" is described in detail in "Survey of Consumer Finances: Employer-Sponsored Pension Benefit Plans," The Survey Research Center, 1986, by Richard T. Curtin.

6. The interview questionnaire for the 1986 SCF was prepared primarily by Robert B. Avery, Gregory E. Elliehausen, and Arthur B. Kennickell, with substantial assistance from Sharon Ward, Robert Schmidt, and Ron Wienk. Field work and editing and coding or survey responses was performed under the direction of Richard T. Curtin of the SRC.

7. The TMF is a stratified sample of 88,218 individual tax returns with a significant over-sampling of high incomes (see Michael Strudler, General Descriptive Booklet for the 1982 Individual Tax Model File, Statistics of Income Division, Internal Revenue Service, 1983.)

8. One way of understanding this division is to think of the 1983 sample couples as being composed of people with exactly the same characteristics and everyone outside the sample having a weight of zero; weights within families in both 1983 and 1986 are determined as the average of the weights of the head and spouse.

9. The decision to post-stratify by homeownership was made very reluctantly because of the feeling that the SCF should represent an independent assessment of wealth, of which homeownership is an important part. Unfortunately, homeownership in the 1986 SCF appears to be related to attrition in ways that could not be forecasted from 1983 information. Without post-stratification for homeownership (actually the 1986 CPS homeownership rate adjusted for differences between the 1983 CPS and SCF homeownership rate), the 1986 SCF sample over-predicted homeownership by up to 3 percentage points.

10. A total of 31 of the 2,822 households in the 1986 survey were headed by persons age 25 (head here is restricted to either the 1986 respondent or his/her spouse). Data from the 1983 survey, which can more reasonably be used to represent the entire age distribution of households with heads 18 or older, indicates that little wealth is missed by ignoring the population under 25. Asset, debt and net worth totals would be estimated to be only 0.9%, 3%, and 0.6% higher respectively, if individuals under 25 were included in 1983. The excluded households, however, add 4.2% to total income and increase the weighted number of households by 8.7%.

11. Unless otherwise noted, all weights should be taken to have an implicit index for each household.

12. In computing W8386, families that were dead or institutionalized at the time of the 1986 SCF were treated as having been found; those families are given a value of zero for W86.

13. To avoid extreme distortions in the weights, the product of a and s was constrained to the interval [.1, 10] for each case at each iteration.

14. As noted in the text, weights for cases with heads aged less than 25 in 1986 were treated differently. For these cases, an attrition adjustment was computed and W8386 was post-stratified to the number of cases with heads aged less than 22 in 1983 with an adjustment for changes in household composition for this group between 1983 and 1986. W86 is related to W8386 by the simple scale adjustment discussed above.

15. The assumption is most clear in the case of the weights that iterate on 1983 and 1986 information (C1014, C1017 and C1018); there the change in population structure is forced on the sample directly. 16. For a complete description of imputation methods used in 1983, see Avery and Elliehausen (1987).

17. The validity of this procedure rests on the assumption that the item nonresponse bias is "ignorable," that is, that there is no systematic selection bias in item nonresponse.

18. See Dorothy S. Projector and Gertrude S. Weiss, Survey of Financial Characteristics of Consumers, Washington D.C.: Board of Governors of the Federal Reserve System, 1966.

19. Detailed discussion of the survey findings can be found in "Household Wealth and Asset Ownership: 1984," Household Economic Studies Series P-70, No. 7 (July 1986), Bureau of the Census; and John M. McNeil and Enrique J. Lamas, "Year-Apart Estimates of Household Net Worth from the Survey of Income and Program Participation," NBER Conference on Research in Income and Wealth, Baltimore, 1987. Richard T. Curtin, F. Thomas Juster, and James N. Morgan, "Survey Estimates of Wealth: An Assessment of Quality," NBER Conference on Research in Income and Wealth, Baltimore, 1987, provide a detailed comparison of the PSID, SIPP, and 1983 SCF wealth data.

20. The FOF accounts are widely regarded as the most reliable source of aggregate data on the composition of national wealth.

21. Both the survey-based and FOF estimates of gross household wealth change overstate National Income Account (NIA) figures. NIA data show $1,954 billion in gross private savings for 1983 through 1985 while FOF-defined total household wealth grew $3,127 billion over roughly the same period and the survey-based estimates show a growth of $2,096 billion. See John F. Wilson, James L. Freund, Frederick O. Yohn and Walter Lederer, "Household Saving Measurement: Recent Experience from the Flow-of-Funds Perspective," NBER Conference on Research on Income and Wealth, Baltimore, March 1987 for a discussion of the differences between NIA and FOF measures of savings.


Top of page | Next: Question Text, Variable Names, & Responses
Home | Surveys | OSS | SCF index| 1986 SCF index

To comment on this site, please fill out our feedback form.
Last update: January 29, 1998, 5:00 PM