An Evaluation of the Methodology of the 1972-73 U.S. Consumer Expenditure Survey

ABSTRACT - This paper evaluates the accuracy/quality of various types of expenditure data collected in the 1972-73 Consumer Expenditure Survey. Recommendation for changes in the methodology - some of which already have been implemented by the Census Bureau, are made.

Citation:

Robert B. Pearl (1981) ,"An Evaluation of the Methodology of the 1972-73 U.S. Consumer Expenditure Survey", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 254-261.

Advances in Consumer Research Volume 8, 1981      Pages 254-261

AN EVALUATION OF THE METHODOLOGY OF THE 1972-73 U.S. CONSUMER EXPENDITURE SURVEY

Robert B. Pearl, Survey Research Laboratory, University of Illinois

[This work was undertaken under a Joint Statistical Agreement between the Survey Research Laboratory and the Research Center for Measurement Methods of the Bureau of the Census.]

ABSTRACT -

This paper evaluates the accuracy/quality of various types of expenditure data collected in the 1972-73 Consumer Expenditure Survey. Recommendation for changes in the methodology - some of which already have been implemented by the Census Bureau, are made.

INTRODUCTION

Collection of data on consumer expenditures is probably the most widespread of all statistical endeavors in this country, engaging both the public and private sectors. Such information is, of course, virtually the lifeblood of the market research industry. The Government is also heavily involved in this field for a variety of purposes. Expenditures by consumers constitute one of the principal components of the Gross National Product accounts. Data on outlays for food, housing, and other living costs are an essential ingredient in establishing benefit levels for social programs. Perhaps most important, this kind of information provides the technical framework for construction of consumer price indexes.

Although the objectives and type of detail may differ a great deal among these various public and private undertakings, many of the technical problems are common to all. In particular, nearly every survey organization has experienced serious reporting errors and biases in attempting to collect information from consumers about their purchasing behavior. This is one reason the findings from the large-scale Government expenditure survey conducted in 1972-73 should be of general interest to practitioners in this field and to others who use or could use similar techniques.

The 1972-73 Survey comprised two major components:

1.  An interview panel consisting of about 10,000 households each year which was visited on a quarterly basis primarily to obtain the larger items of expenditure and certain repetitive items (rent, utilities, etc.) Particular categories were covered either quarterly or on a semi-annual or annual basis, depending primarily on expenditure size.

2.  A diary operation consisting of about 200-250 households per week asked to keep a diary or record of all expenditures for the subsequent, two-week period. Although the main focus of the diary was the smaller items of expenditure, the fact that all categories were covered provided various options in compiling estimates as well as many research opportunities. For further in-formation on the methodology used for this survey, see other papers presented in this session.

PURPOSE

Because of the critical uses which are being made or which could be made of the data and the marked changes in methodology, considerable attention has been devoted to evaluation of the survey results. The objectives have been to assess the adequacy of the survey procedures in general, the relative merits of using interviews as compared to diary keeping for particular categories of expenditures, and the implications of the findings with respect to improvements in methodology for future use in either a continuing survey program or other similar undertakings. Since there was little prospect that all of the methodological issues would be resolved, a determination of further research needs was also an important goal.

This paper presents a summary of the findings and conclusions deriving from this evaluation effort. Detailed results appear in publications in the Census Bureau's Technical Paper Series (Nos. 45 and 46).

Summary of Findings for Expenditure Categories

The following is an attempt to assess the adequacy of the expenditures data obtained in the 1972-73 survey for the various categories of goods and services. The general approach used in this appraisal has been to compare the estimates from the quarterly panel with those from the diary operation, where the same subject was covered in both, and to relate either or both to various independent sources of expenditure data. The principal objectives are to assess which of the survey procedures appeared to be more effective for particular categories of expenditures and to determine what types of improvements and modifications may be suggested by the results. The conclusions can only be tentative because of major uncertainties about the validity and comparability of the independent data used as a standard and because adequate detail was often unavailable to explore the subject in sufficient depth. Nevertheless, in a substantial number of cases, persistent patterns emerged across category lines which pointed in rather specific directions.

The most frequently used of the independent data sources are the Personal Consumption Expenditure (PCE) estimates prepared by the Department of Commerce in conjunction with the Gross National Product Accounts. The other independent data derive mainly from Government administrative, census, or survey sources although some private sources are also used. In some sectors, such as education and health, the conceptual differences between the PCE and the 1972-73 survey data were such that dependence had to be placed entirely on the other independent sources for comparative purposes.

It should be noted that the survey results used in this evaluation are derived from special tabulations of re-weighted original data tapes. They do not reflect editing charges which may have been made at later stages of processing by BLS. As a result, the figures may differ somewhat from those already published or to be published by that Agency or which may be compiled from the public-use data tapes recently issued. Certain differences in time reference and conceptual approach would also contribute to the disparities.

Table 1 presents a summary of the findings for the various expenditure categories. For purposes of summarization, a number of the detailed categories have been combined and averaged. The table designates the "best" survey source, that is, the one generally closest to the independent data, in cases where the two survey estimates are significantly different. The ratios of the "best" survey estimates to the independent estimates are indicated in terms of broad class intervals, allowing insofar as possible for conceptual differences between the sources, but the actual computed values are also provided. Although the magnitude of these ratios may provide some indication of the adequacy of the survey estimates, the margins can be regarded only as rough approximations, for reasons stated above, and have been used primarily for purposes of detecting any consistent and meaningful patterns which have a bearing on the methodology.

1. Food and beverage expenditures.  After allowance insofar as possible for various conceptual incomparabilities, there appeared to be a reasonably close correspondence between the diary estimates of food purchases for home use and the independent sources. The fact that the home-maker--the usual diary keeper for the family--is ordinarily responsible for most of the purchases was undoubtedly a positive factor. The allocation of maximum space on the diary record to this expenditure class probably contributed as well.

There were considerable disparities, however, in the precision with which various food categories were reported. The reporting was apparently most complete for relatively costly items, such as meat and poultry, and for those used promptly and on a daily basis, such as milk, eggs, fruit juices, and bread and fresh-baked items. The coverage seemed to be considerably less complete for food staples which are bought less frequently, with each purchase used over a considerable period of time. One of various possible explanations for these differences is that many respondents may not start keeping their diaries promptly--or do not make entries, as requested, on a daily basis--but later attempt to reconstruct the omitted periods by memory. In doing so, items which represent the main course in a meal or which are purchased and used relatively frequently might be more readily recalled.

A less anticipated finding was the close correspondence between the survey and independent estimates for meals in restaurants or other eating places, where a substantial proportion of the outlays would be made by individual family members other than the homemaker. The prominent positioning of the section for reporting purchased meals on the diary record and some emphasis to this subject at the time of the diary checking procedure might have contributed to this outcome. At the same time, the marked deficiency for alcoholic beverages confirms the continued failure of household surveys to measure a sector where there is considerable sensitivity about reporting. [Some special check questions, asked at the time diaries were collected in an effort to overcome this deficiency, provided data that would have increased the survey estimates by about 15 percent (no actual correction was made, however). Even on that basis, the survey aggregates would have mounted to only about half of the independent levels.]

2. Small expenditures other than food.  For various small expenditure items other than food, for which the diary was the principal if not only source, a predominant factor appeared to be the role of the various family members in making purchases. Where the responsibility was principally that of the homemaker, such as for laundry or cleaning products or household services, the reporting appeared to be considerably more complete then in cases where other members were substantially involved, as for toiletries or hair care. Even for those expenditures where the homemaker predominated, however, the reporting appeared to be generally less adequate than for food purchases, probably partly a reflection of the much smaller amount of space and attention given to non-food items on the diary record. Some limited efforts to measure small non-food expenditures through summary questions in the quarterly panel did not appear to be especially productive, although in one or two cases the figures appeared to be more complete than the weakest of the diary estimates.

3. Clothing expenditures.  As was anticipated to some extent, this expenditure category represented one of more troublesome sectors, with neither survey source exhibiting any clear cut overall advantage and neither corresponding very closely with the independent data. [Clothing expenditures were covered each quarter in the interview panel. Some 16 broad categories were specifically probed, using detailed item checklists for each. A small section was provided for clothing expenditures on the diary record and it was, inadvertently, located in an especially inconspicuous position on the form.] Following the pattern observed throughout the analysis--and expected from previous experience--the larger items (suits, coats, etc.) were apparently more adequately reported and the quarterly panel emerged as the superior source in this case. Also not surprisingly, the diary procedure represented the "best" source for a diversified category such as accessories, where it was probably difficult to communicate the full range of items in an interview procedure, For no apparent reason, the diary estimates also provided the closer correspondence with the independent data for footwear, although this subject was probed in much greater detail in the quarterly panel.

For the broad range of middle and lower priced clothing products, the advantage seemed to alternate between the two survey sources, without any consistent relationship to the importance of the item. One problem which complicated the appraisal--and which extended to most other expenditure classes as well--was the existence of a large residual clothing group in the diary estimates, consisting mainly of incomplete or inadequate entries which could not be assigned to specific categories.

4. Household appliances.  Reporting of expenditures for major appliances was apparently one of the more successful survey outcomes. The high cost associated with these items and the more probing nature of the "inventory" technique, described earlier, probably contributed to this favorable showing. [The inventory was updated only once, at the final quarterly visit, for large appliances. The updating occurred twice, at 6-month intervals, for small appliances.] There was some evidence in favor of pooling the two estimates obtainable for a given year under this approach, the one based directly on the survey for that year and the other derived indirectly from the initial inventory in the survey for the following year. Comparison between those two sets of estimates indicated no significant differences for several of the major appliance groups and no evident superiority of one over the other in relation to the independent data. Pooling of the estimates would have roughly the effect of doubling the sample size at virtually no additional cost, a considerable gain for items with especially large variances.

The picture was considerably less favorable for small applicances, which would be consistent with the findings for other product categories. The inventory approach is apparently less successful where a multiplicity of small products is involved (such as for small kitchen appliances) or where the family may possess several articles of the same kind (e.g., cameras or radios). An interesting byproduct of the inventory approach in these instances, however, was that estimates which included the value of gifts received from outside the household---which derived directly from the inventory on item on hand--invariably exceeded those which included the cost of item bought by the family as gifts to be given to others, which was based on their recollection of purchases. This finding suggests that the inventory method, based on items in the family's possession, has certain advantages over the more conventional approach which requires recall of previous actions.

5. Household furnishings.  This broad category provided a rather clear-cut demonstration of the relationship between the size of an expenditure and the likelihood of its being reported. [Most items in this class were covered at 6-month intervals in the quarterly panel. The exception was household linens which were covered each quarter. A smell section was provided on the diary form for "housewares, furnishings, hardware, and garden supplies."] The closest correspondence with the independent estimates was found for furniture, the most costly class, followed by the next most significant group--floor, window, and furniture coverings. In both cases, the quarterly panel appeared to be the superior source, partly on the basis of sampling variances. The survey estimates fell considerably short for household linens and especially for smaller products such as dinnerware and cookware, luggage, and decorative items. In these latter instances, the diary estimates, although themselves deficient, appeared to be at least equivalent in coverage to those from the interview panel. [In fact, for the dinnerware-cookware category, the diary estimates were appreciably higher and, thereby, closer to the independent levels.]

6. Automobile and vehicle expenses.  In most categories, the survey data corresponded closely with the independent sources. The main disparities were for tires and other accessories. Although the homemaker would normally have less responsibility for vehicle expenses than most others, the diary estimates held up surprisingly well for some of the smaller categories, especially gasoline. The quarterly panel, as expected, was clearly the superior source for vehicle purchases and also appeared to provide the more reliable data for vehicle registration, insurance, and similar items. [The "inventory" approach was used for vehicle purchases in the interview panel, with the inventory updated only once at the final quarterly visit. Questions were asked each quarter, however, for gasoline purchases and vehicle maintenance and repair. A section was provided for "gas, oil, tools, parking fees, and other vehicle expenses" on the diary form.]

7. Housing expenditures.  Once again, with the exception of one rather diffuse category (fuel purchases), the survey and independent estimates corresponded rather closely. [A rather complex set of procedures, including use of the "inventory" approach in some cases, was used to determine housing expenses in the quarterly panel. For home repairs and maintenance, detailed questions (adapted from the Census Bureau's special survey on this subject) were asked each quarter. Utilities were also covered each quarter using the "last payment" approach (asking for the amount of the last bill and the period it covered). A section for "rent, utilities, fuel, phone, insurance" was provided on the diary record.] Also, in the main, the diary-based data, again unexpectedly, matched those from the interview panel. The exception, in the case of mortgage payments, may have resulted from the fact that some such disbursements are made directly through bank accounts, which could easily escape the attention of the diary keeper.

8. Health expenditures.  One of the more pleasant surprises was the relatively close correspondence between the survey and independent estimates for most health expenditures, usually considered to be one of the more treacherous areas in this kind of undertaking. [Detailed inquires were made in the interview panel at 6-month intervals for health expenditures. A section for "personal care, drugs, and medical supplies" was provided on the diary form but other medical expenses would presumably be entered in the catch-all section.] A less optimistic reading might be that the survey results were at least as good as most previous endeavors of a similar nature, without attempting to categorize their accuracy in an absolute sense. In any event, the diary procedure once more provided the most unexpected outcome, in that not only for small items such as drugs and medicines, but also for most professional health services, the data seemed to hold up surprisingly well. An exception was for hospital services but even the quarterly panel data in that instance seemed somewhat deficient, possibly because of complications introduced by the pervasive role of health insurance and other third-party payors. [The survey objective was to measure "out of pocket" costs, net of insurance and other reimbursements.] Also, for health insurance premiums themselves, where the interview panel data appeared to be reasonably adequate, the diary procedure was understandably ineffective, in view of the fact that payments are often made through payroll deductions which would rarely come to mind in completing a household diary.

9. Education, travel, and miscellaneous expenses.  Because of the diversity of items, it is more difficult summarize for this remaining expenditure group. [Education expenses were covered at 6-month intervals in the interview panel. Detailed inquires were made each quarter on trips and vacations. Other items were covered at various intervals depending on expenditure size (but with a quarterly inquiry for watches and jewelry). Most of these items were not specifically mentioned on the diary record.] Although the evidence was limited, there appeared to be some deficiency in reporting of education expenditures, possibly because of the difficulty of adequately covering college students in a household survey, the quarterly panel still appeared to be the better source for the larger payments (tuition, etc.), but it seemed desirable to alter the survey procedures for students attending college away from home, whereby they would be interviewed directly at their college quarters instead of obtaining proxy information, in some instances, from their parents at home. No direct comparisons were possible for other school expenses, but it appeared likely, based on other findings, that the smaller expenses--books, school supplies, meals outside of school, boarding arrangements, etc.--as well as related items, such as recreational lessons, might better be approached through the diary procedure.

Comparisons with independent sources for travel and vacations could only be made in terms of the reported number of trips taken rather than directly for expenditures. On this basis, the quarterly panel data appeared to be consistent with other findings. Since the validity of expenditures data may depend more on whether or not an expenditure is reported than on the precise amount of the expenditure, even this limited finding with regard to trips taken could be meaningful.

For the remaining subjects--transportation costs and miscellaneous products and services--a few generalizations may suffice since the patterns appeared consistent with those already cited for items of a similar nature. Public transportation expenses, which were available in detail only from the diary, were drastically underreported in relation to the independent sources, probably because such outlays are usually made on an individual basis. Among the miscellaneous categories, the larger items, such as pianos, organs, and funeral expenses, were much more adequately reported in the quarterly panel and the survey estimates agreed, in large measure, with the independent data. A large disparity for watches and jewelry could have resulted, in part, from sensitivity in reporting such expenditures, whereas a similar deficiency for moving expenses could reflect some undercoverage of recent movers in the survey.

CONCLUSIONS AND RECOMMENDATIONS

The survey findings, although predictable in a number of respects, carry some implications for further efforts in this field, not only on behalf of Government programs but possibly for market research and similar private endeavors. A rather unexpected result was the relatively effective performance of the diary procedure in some sectors, such as housing and health, where the quarterly panel has been assumed to be the only realistic option. The diary results also appeared to be about as effective as those from the quarterly panel in many cases where both sets were evidently deficient, including some situations where a great deal of attention had been devoted to expenditure reporting in the interview procedure.

A general rule of thumb suggested by these findings might be that, unless a clear-cut reason existed for using an interview procedure such as the quarterly panel, dependence might better be placed on diary-keeping for purposes of expenditure data collection. The diary might be a dubious source for items with exceptionally large variances (vehicles, appliances, furniture, etc.) or where unusual payment arrangements might require special questioning, such as insurance paid through payroll deductions, mortgage payments made automatically through bank accounts, hospital bills paid largely but belatedly by insurance, etc. The diary might also be inferior to an interview process in cases where a composite set of questions may be necessary in obtaining complete responses as may be the case, for example, in cataloging trips and vacations. In other instances, the diary procedure appears to be at least as good a risk as the interview approach, and probably a less costly one as well.

A number of modifications and improvements in the diary procedure are clearly necessary, however, in order to overcome some rather evident deficiencies. Some possible steps in this direction are the following:

1. Limit the range of items any one family would be asked to report--The use of a diary covering all items of expenditure, as was done in the 1972-73 Survey, may have certain theoretical benefits, but considerably proscribes the ability to improve the overall process sufficiently to satisfy the expanded requirements just cited. Evidently, as stated earlier, one of the reasons for the more successful coverage of food purchases than other small items in the 1972-73 diary was the much greater amount of space and attention accorded the former. Moreover, for all categories including food, there was a considerable undifferentiated residual group, resulting mainly from incomplete or inadequate entries which could not be classified in detail, which detracted materially from the usefulness of the results. The general lack of space and the inability of interviewers to focus on so wide a range of items in reviewing the diaries probably largely accounted for this later deficiency.

Most market research endeavors and similar undertakings are already restricted in terms of item coverage. However, since exhaustive detail is often requested about specific items (type of packaging, package sizes and weights, brand names, ingredients, marketing features, etc.), it is possible that better results would be obtained if the samples were split and particular respondents were asked to report on only a subset of products. For more comprehensive projects, such as the 1972-73 Survey, where the amount of detail is more limited, a possibility would be to split the sample along category lines. One subsample, for example, might be asked to report only on food and the supermarket products, a second on clothing and household linens, a third on health expenditures, etc. There is obviously some practical limit to the number of subsamples that could be simultaneously operated and a good deal of thought and some experimentation would be necessary to derive a workable plan.

Probably even more important than space considerations, the use of this kind of specialized approach would make it feasible to provide for a more focused set of check questions and procedures at the time of diary pickup to overcome some of the disparities noted in the present survey (such as underreporting of certain food items relative to others). In fact, a modified procedure would likely entail much more of a combination of interviewing and record keeping than is now the case. A good deal of developmental work would be required in devising effective checking procedures.

2. Vary length of record-keeping periods-- As previously noted, the 1972-73 Survey provided for two weeks of record keeping for each sample family, covering all items of expenditure. In most private undertakings also, the practice is to use a standard reporting period for those products which are covered. If specialized subsamples are developed as proposed above, it is obvious that either a larger overall sample would be needed or much higher sampling variances would have to be accepted. One way out of this dilemma would be to vary the length of the record-keeping period depending on the variances of the subjects covered for a given subsample. For example, for a low variance category, such as frequently purchased food items, a period as short as a week might be adequate. For most categories besides food, however, such as clothing or health expenditures, record keeping periods of up to 3 months or longer might be considered. The fact that only a limited set of items is covered might reduce the reporting burden sufficiently to secure extended cooperation of this kind without an unacceptable attrition in the sample.

In such a system, the use of less costly collection methods already employed in many private endeavors, such as having respondents mail in completed diaries on a periodic basis (monthly, semi-monthly, etc.) would be more practical. Provision would, of course, have to be made for follow-up (by mail or telephone, where possible) for non-respondents or to carry out special checking for returns that did not meet prescribed standards.

3. Provide separate diaries, where indicated, for individual members--As noted, only one person, usually the homemaker, probably maintained the diary for the entire family in the 1972-73 Survey. Not surprisingly, the results were clearly more favorable for the kinds of expenditures for which the homemaker was mainly responsible than for those likely to be made by other members. One possible way of obtaining more consistent results, where the expenditures to be reported are of a more dispersed nature, would be to provide separate diaries for all family members above a certain age (perhaps 12 and over) on which to record their individual disbursements. For this purpose, the diaries could be briefer and less formal than the main record for the family. Experimentation with various versions would obviously be important in developing a procedure of this kind.

4. Providing monetary or other incentives for cooperation--Most market research endeavors provide various cash or non-cash incentives to promote greater cooperation. There is considerable reluctance to follow this course in Government surveys, although the approach is not unknown. In fact, an experiment with cash incentives was conducted in the early stages of the 1972-73 Survey to determine whether cooperation could thereby be improved. The results were inconclusive in this regard and the incentives were dropped from the procedure. As noted earlier, cooperation was high in spite of the absence of material rewards.

Since most studies have shown that the quality of reporting as well as the degree of cooperation appear to be benefited by some type of material inducement, the matter of providing incentives should be seriously considered in any operation where there is primary dependence on diary keeping. The rewards should, of course, be attuned as closely as possible to achieving the main objectives. If, for example, it is important for individual family members to keep separate records, the incentives should be offered only if all such members agree to cooperate, which would presumably promote intra-family pressure for individual compliance. Similarly, if cooperation over extended periods is requested, a useful approach is to offer a small reward for each sub-period with a sizeable bonus for completing the entire cycle. Special consideration for prompt and complete returns by mail (where used) might be still another element. Payment of incentives might also be made contingent on retention by respondents of cash register tapes, bills, and other evidences of expenditures which could be consulted in reviewing the diaries.

5. Continue exploration of timing biases--Although not mentioned up to this point, perhaps the most conclusive survey finding was affirmation of the traditional bias found in diary operations, whereby a higher level of expenditures is reported in the earlier as opposed to the latter stages of the record keeping period. In a 2-week diary procedure, for example, the estimates for the first week are almost invariably higher than those for the second week. Differences of this nature were found for virtually every expenditure category in the 1972-73 survey. The margins of difference varied a good deal, however, and seemed to be almost random in nature. There was no evident relationship of the weekly differences for a category to other survey measures, such as the completeness of reporting relative to the independent estimates.

Since separate pages were provided in the diary for each reporting day, it was possible to examine the day by day recordings in this context. It was found that the higher levels for the first week diaries were primarily the result of exceptional amounts of expenditures recorded for the first reporting day (about 40-50 percent above the daily average). Levels were also above average for the first day of the second diary week (8th reporting day) but not nearly to the extent found in the first week. There was a tendency for a gradual decline in expenditure levels in the course of each week. There were indications that this pattern persisted for most categories of expenditures.

Although various theories have offered for this phenomenon, the one that appears most likely on the basis of the evidence is that there is a certain amount of "telescoping'' of expenditures on the first day. According to this theory, many respondents may not begin their diaries immediately, but skip the first day or two and then attempt to reconstruct the information from memory. In doing so, they may inadvertently include some expenditures made prior to the reporting period. Another suggestion that the patterns can also be attributed to some element of fatigue as the period lengthens can also be supported by the findings. Still another hypothesis to the effect that respondents may alter their buying habits temporarily because a diary is being kept does not appear to be supportable as a major element because the differences are so widespread among categories.

Among the suggestions made to control for this tendency is to extend the period of reporting somewhat (such as to 17 days for two-week diaries) and then to exclude the entries for the early and final days from the estimation. Another is to conduct a brief retrospective interview covering a day or two prior to the reporting period and to use that information as a "bounding" or control device in examining the early diary entries. The problem with such procedures is that, by reducing the impact of the exceptional diary levels at the outset, they would accentuate the tendency for understatement in the overall diary levels in general. The better approach might be first to attempt improvements in diary reporting through various of the possibilities cited earlier in order to be able to afford the use of any devices for controlling telescoping.

6. Explore use of universal product codes--A promising technological development that could affect future survey work in this field is the inclusion of "universal product codes" on most canned and packaged supermarket and drug store items (and likely to extend to many others). A useful experiment would be to ask respondents to record these codes in their diaries where available, as well as brief product descriptions, to assess how accurately this information is reported. If the effort is sufficiently productive, respondents might be relieved of the necessity of describing products in any detail where code numbers are available. More importantly, this step could result in far more accurate and consistent classification of products reported in surveys and a major reduction in the coding effort required at the data processing stage.

A related effort would be to make more effective use of the much more descriptive cash register tapes now being provided in automated checkout systems. If retained by respondents, they could certainly augment and, in some cases, take the place of diary entries.

In a comprehensive expenditure data program, such as required for purposes of the Consumer Price Index, there would be need for an interview survey capability, especially for high variance items, even if greater dependence were placed on a diary procedure. An interview approach would also be the indicated procedure in market research surveys related to durable goods and similar products. The main lesson from the 1972-73 Survey was that exhaustive interview surveys are likely to be self-defeating in this sector, and that the inquires at any one time should be limited to those products and services not amenable to the diary technique. The use and possible extension of the so-called inventory approach for durable goods was also generally supported by the findings. Also confirmed was the notion of using variable time references in interviews depending on the size and importance of particular items.

One final matter might be mentioned which has a bearing primarily on comprehensive surveys such as chose conducted by the Government, but which would be of interest to analysts of consumer behavior in general. The major recommendation made--curtailing the scope of the interview phase and introducing specialized diaries--would mean that only very partial expenditure data would be available for individual families, either for annual or shorter periods. Unless some counter measures were taken, there would be some evident limitation in the ability to analyze individual consumer behavior.

Actually, some of the largest items, which distinguish one family's expenditures from another's, would still be obtained by interview under these proposals for periods up to a year, but most categories would not be covered. A first step to close the gap might be to add a few questions, perhaps on an annual basis, to identify exceptional expenditures in other sectors, such as large orthodontic bills, major house additions, purchase of expensive paintings, and the like.

Beyond that point, a further although admittedly experimental step would be to obtain certain limited information in the interview procedure on behavioral patterns which would help in imputing data from the diary operation or another source to the individual family record. For example, in the health sector, questions could be asked on the frequency of doctor or dental visits, consistent usage of costly drugs, etc., which, together with demographic and socioeconomic characteristics, could provide a basis for imputing a value for health expenditures to the family from another source. (The other source would, of course, also require the inclusion of such behavioral items.) Obviously, considerable research and testing would be needed before embarking on a large-scale endeavor of this kind.

TABLE 1

SUMMARY OF FINDINGS FOR EXPENDITURE CATEGORIES: 1972-73 EXPENDITURE SURVEY   -  ESTIMATES COMPARED TO INDEPENDENT SOURCES

*Signifies that comparisons are based on 2 years of data. For other categories, comparisons were based on only 1 year, usually 1972.

1. The "best" survey estimate is defined as the one--quarterly panel or diary estimate--which was closest to the independent figures. Where a double asterisk (**) is appended to the code, this indicates the specified source was the only one for which a comparison with the independent estimates could be made in the required detail. In most of these latter cases, the specified source was also the only realistic source for the data in question.

2. The ranges in this column are not always entirely consistent with the computed ratios in the next column, but attempt to make allowances for conceptual differences between the survey and independent sources, disparities among the independent sources themselves, and other factors for which numerical adjustments could not be made.

3. The ratios are based on weighted averages of the survey data (numerator of ratio) and of the independent data (denominator of ratio) for the various individual categories combined on a given line of the table. Where the two survey estimates were not significantly different (Code "N" in the "best" survey source column), the quarterly panel and the diary data were averaged. Where more than one independent source is specified, those data were averaged for this purpose. Revised ratios reflect PCE data adjusted to 1972 benchmarks. Preliminary ratios based on PCE data projected from 1967 benchmarks (and used in the preliminary evaluation report--Census Technical Report No. 45).

4. The independent sources, in alphabetical order, are as follows:

AHS--Annual Housing Survey, Bureau of the Census; A. M. Best--Insurance company premium data compiled by A. M. Best and Co.; Cen-Bus--1972 Census of Business Merchandise Line Data; Cen-Serv--1972 Census of Selected Service Industries; CHAS--Center for Health Administration Studies, University of Chicago; ERS--Economic Research Service, U.S. Dept. of Agriculture; HIS--Health Interview Survey, National Center for Health Statistics, U.S. Dept. of HEW; MRCA--Panel data from the Market Research Corp. of America; NCES--National Center for Education Statistics, U.S. Dept. of HEW; NTS--National Travel Survey, Bureau of the Census; PCE--Personal Consumption Expenditures, CNP Accounts, U.S. Dept. of Commerce; SSA--Social Security Administration, U.S. Dept. of HEW; SORAR--Survey of Residential Alterations and Repairs, Bureau of the Census.

5. Uncertainty about the magnitude of a conceptual difference precludes a more precise judgment for this borderline value. The ratio of the survey to the independent estimate is understated because the PCE figure used as a base relates to "off premise" consumption and includes (in addition to food purchased for home use) snacks purchased in scores (and other establishments which are not eating places) but consumed outside the home. The survey estimate, however, relates only to food purchased for home use. The reverse situation holds for the PCE and survey estimates for "purchased meals and snacks" where the ratio is overstated.

6. This is an average value for the eight major categories included in this bracket. Of these, three were significantly more completely reported then indicated by the coded mean value, but there was no clear-cut distinction among these by expenditure size.

7. Using optimum estimate (including value of gifts received and averaging the two alternative estimates for the year), see text. Under the standard estimation, the racing for major appliances would be "A, B" but that for minor appliances would be "D".

8. Survey estimates used in the computation were diary estimates for dinnerware-cookware and hand tools and average of the two estimates for decorative items and luggage.

9. Computed ratio is understated (magnitude uncertain) because the PCE base figure (but not the survey estimate) includes fuel used in rental quarters which is supplied and directly paid for by the landlord.

10. Comparison based on college tuition only.

11. Comparison based on number of trips taken rather than expenditure.

12. Uncertainty about allocation of a large unclassifiable residual (about 15 percent of transportation expenditures) precludes a more precise judgment for these categories.

----------------------------------------