Tobacco Documents Online

•.. for the assistance you have given the Census Bureau interviewer who just visited you. It is only through the cooperation of you and others who are being visited that a health survey such as this one can be carried on, and we thought you would like to know how the information you have given will be used. It will, of course, be held in confidence• When combined with information given by other persons in this and other com- munities, it will reflect health conditions throughout the United States and provide new knowledge to improve the health of the American people• It isbecause such knowledge is now lacking that Congress recently authorized the National Health Survey--of which the interviewing in this area is a part. The National Health Survey will be collecting information on other aspects of health, and it is possible that we may wish to ask for your further cooperation at some time in the future. Meanwhile, thank you for your help today. Surgeon General, Public Health Service Figure 2. Letter o£ ~pprec;ation to respondent. 8 T106653822

d,-~23dn3. This statement a~plie~ vdth ~ many ways ~ass~fion dete~es ~e d ~= pro~ect, decIde~ ~'k=~=r ~m ~ a of chosen b1~:~ ~ i~o~affon, ~uenc~ ~e pre~ence or absence of bias in ~a meas~emen~ ~zess, and genera~y con~fio~ ~ ufiHW s~,ey re~ts. For ~ese r~sons, anti adequate ~ea~ent of ~ds one matter is lengthy, a separate report on defi~fio~ and class~icati~ns is ~g pr~ar~ for issue in ~s s~ies. 0~y a few remarks are includ~ here. " V~erever ~ssibl~ 9renard-~efi~dons and classifications have ~en employS. ~us, "dwel- lin~ ', u~t,'~ "household," "Standard Metropolitan Area," "family," and many other terms are de- fined as in ~e ~cennial Pop~ation Census or other widely accepted operati~s. Similarly, the Inte~ational Statistical Class~icafion of Diseases, Injuries, and Causes ~ Dea~ is ~e basis of clas- sify~g health conditions; dem~raphic, s~ia], and economic measures have been ~ouped into classes which co~orm, it is ~lieved, to most common practice. Classifications have ~en pr~esi~ated and are fixed, and the questionnaire is ~eared in most areas to such a system. Ve~ little latitude is given the respondent to create new ~asses through replies, but translating replies ~toa specific c~e is still a relatively ~fficult pr~ess, especially for ~e mescal c~in~. Accordingly, m~ic~ c~ing for each condi- tion ini~ally has ~en done independently by 2 c~ers, with differences ~ing umpir~ by a c~ing experL Information conceding the naive of c~- iug ~fficulties is ~g assembled to permit par- tial verification and pr~ess-control techniques for the medical c~ing. The use of p~ch-card equipment for early phases of editing was dictated at first by consider- ation of workloads on available eq~pment, and of start-stop requirements for some of the steps. Some of these operations have ~en shifted to ~ computer, and others will ~ later. When reportsarereceived in Washin~on, ~ey go througha control procedure ate "con~ol desk." ~is pr~edure, in addition to routine housekeep- ing checks, includes 3 operations with statistical si~ificance. i. la the prelisting step, an "expected" n~- ~r ~ households in each se~ent was dete~ined in the field and r~orted to Washington. Incoming reports must account for ~e same num~r of . households or explain ~screpa,cies. Any segment for which an unexplained discrepancy is found reconciled through reconmct wi~ ~e Census Bu- r~u Regional ~fice if time ~rmits. In instances in which tab~ation cutoff ~me prevents this, the case is later call~ to ~e atten~on of the Regional ~ice so that it will ~ used in initiating neces- sa~ ti~tening of su~rvisory controls on listing, inte~ie%~ing, and cleric~ operations. 2. In s~me s_~grn~nts it will ,have keen fo-.md, eid-.er earlier in Washington or in ~e field, a chosen se~n-nent clearly centa~-~d more than 20 households. In these cases the Ee~'nent was ~ampled so tl--~t the final ~l~sample canLained roughly 6 h~useholds. Tile s~tbzampling fraction is noted at the Washington centrol desk and an ad- justing order is g~ansmitted to the computer. 3. A third D~e of review at the control desk adjusts the sampling fractJoll for households and persons from special dwelling places, such as re- formatories, homes for the aged, or hotels for transients. The general purpose of all operations at the Wasbdngton control desk is to assure that dam moving into the editing and tabulating stream are, with respect to coverage and weighting, in agree- ment with the survey design, within narrow toler- ances of error, Evaluation and Control of Data A substantial proportion of total resources of the household-interview survey is devoted to con- trol, evaluation, and improvement of quality of data input. There are 4 very broad areas of activity which have impact on quality of data, which are not discussed here in this connection, but which are pal~s of the Uo So National Health Survey, and which are listed by tide in order to place in perspective those items which are displayed in the following paragraphs as devices for control and improve- ment of data. The 4 areas are (I) the over-all survey design, including concepts, definitions, and general plan of operation; (2) operating control, in the sense of maintaining general adherence to design, including proper use of trainin~ and super- vision; (3) utilization of comparative analysis of data, including external checks against other sources of health information, and especially against medical and health records; and (4)pro- vision for organized outside review and criticism of both methods and products through the creation of both governmental and nongovernmental advi- sory committees, and the use of consultants. Aside, then, from these broad areas just named, there are 3 vlpes of operation which are integral par~s of survey procedure, and which are principally devices for control and improvement of quality of data. The Keinterview Procedure.--Already men- cloned, in connection with training and supervision of interviewers, is the reinterview program, The supervisor regularly recontacts about one sixth of the households in his part of the sample, and thus, about one sixth of the households assigned to each interviewer. The supervisor audits the household information previously secured by the original interviewer, and reinterviews I pre- designated member of fine household. Three main 9 T106653S23

ef ~imng an~ q~i~' con~olfo~ bili~'. ~ ~d is de~ecfi~ of ~t~ewer m ~ extent ~t the more e~en s~-~or c~ ~scover it. ~s m~ns e~ra~g ~e as~ption • at ~a su~e~i~or, us~g ~e s~e ~estio~e ~d ~ a~e p~ed~e ~a~ were ~ by ~e ~i- ~al ~te~ewe~, hut~more~orou~y ~ainad, will s~ure ~ta w~ch may ~ consid~ ~mn~rd w~ ~e ~te~iewer ~ho~d and pass~ly ~d. ~ ~e reinte~ew, ~ ad~ mus~ ~ inte~iewed as se~-res~ndents ~an proxy-respondents. ~us ~ere is a com~on~nt of variance from se~-~, pro~-~e~d~t as source. This componem has some ~ufing on measurement of inte~ew~ con~butions bias and variance, but its e~ate~ce ~y make possible ~o de~erm~e the exten~ d bias (ff any) caused by the proxy-respondent. Pr~esa~g checks and con~ds.~t ~ch c/pal p~easing step, c~trols are established either by verification ~ough duplicate or by sample verificar~on te~niquez, bas~l pn- ~y en prc~ezs con~oL ~ far, ~or rates ~yanl prelimina~, s~n~rds ~e ~ed to ~ at- tendc~ of ~e res~ibl~ ~;~radn~ su~sor. ~ a reco~on ~at steps ~e token ~o re- du~e th~ e~or ~te. F~er s~y of e~or ~[es and ~e~ pro~le impact en es~mates ~e ex- pect~ to I~d to a he~er-~anced se~ of stand- ards. ~ney will l~d also ~o ~t~r use ~ s~ple ver~on and reduc~ons in I~ percenl dupli- ca~ of e~fing steps. ~te~al ~fin~ and consistency checks.~efer- en~e was made ~r~er, and will ~ made aga~ when the derails of es~afion in the su~ey are ~scussed, to e~ng routines desired to make questionnaires in~e~a~y consistent, and to e~m- i~e "impossible" res~nses. ~s is an ar~ ~ w~ch the m~ber and ty~ of poss~le checks are ~mited. E~erience must ~ ~e ~ide in de- ciding how much e~ting is profit~le. As ~plied earlier, the firs~ objec~ves are to ~s~e ~ha~ data are consistent and not obviously incorrect. More p~ne~rat~ e~ts are to ~ tes~. 5. SAMPLE DESIGN :Th'e Multistage Design. As noted in the summax~] on page 4 of this report, the Health Household-Interview Survey rests on a highly stratified, constructively 2-stage probability design. Acmalselectton of sample units takes place in a multistage process, which is mod- ified further by the use of 3 selection zones and 41 subuniverses. The design is termed "construc- tively 2-stage" because the first sampling step is the selection of 372 primary sampling units from among some 1,900 areas into which the country has been divided, while the remaining steps lead effectively to a second or ultimate sampling stage in %vhich small segments or clusters of an expected 6 households are chosen for inclusion in the sam- ple from within the PSU's selected in the first step. The following paragraphs describe principal features of the design, and the manner in which the sample was drawn. Additional technical notes on selected aspects of the design are included in Ap- pendices [I through VII. In particular, algebraic statements of the estimating and variance equa- tions are given in Appendices II and IlL Still fur- ther insight on the topic can be gained from con- sulting Chapters 7, 8, 9, and 12 and AppendLx B of reference 2, since much of the theory underlying the sample design of the health survey is set forth in this b~ok. I0 Primary sampling units. The PSU is a county, a group of contiguous counties, or a Standard Met- tropolitan Area. A total of 1,900 PSU's exhaust the land area of the continental United States, Forma- tion of such PSU's is an art rather than a science, although several clear-cut principles and rules were helpful. Prominent among these are the fol- lowing 4: 1. PSU's should be units for which a wide va- riety of descriptive statistics is I available, since this permits the PSU's to be stratified or classi- fled in an efficient manner. 2. When the PSU is used by a large surveying organization, there are distinct economies in using the same set of PSU's for more than 1 survey. Consequently there are advantages in having the PSU conform to administrative structure in the field, and in having the unit adaptable to many so- cial andeconomic objectives. 3. For technical sampling reasons, the great- er the internal heterogeneity of the PSU, the more efficient it is. This principle tends to produce physically large units. 4. Contrastingly, costs per ultimate sample unit (i. e., cluster of sample households) tend to increase with transportation distances between ul- timate units within a PSU, and thus to increase with the size of the PSU. This factor has limited the size of a PSU to not more than a few neigh- boring counties, _J T106653824

Th~ aho;-e pzinaiples led to formation of I,g69 PSU's, w~ch are ~ss ~d ~ oi~r E~u ~'eyz, and ~'~ch, ~i~ a fe'~' ~c~p~o~, have ~e f~e~: ~ h~n~ bl~k or ez~ s~c~al com~n~nt of ~ PSU ~ a co.W; ~ch PSU ~ ~ W~e~ U~ed Smt~ contains a ;~p~adon ~f at least 7.5~ (19~0 Ce~us). and o~r parts ~ ~ co~ a pop.salon of at l~sz 10,~O; ~ch wes~e~ PSU conta~s not more ~ 2,0~ sq~re ~les and o~her PSU's no~ more ~an 1,500 square mfles--u~ess ~e single county is larger, which in ~e West resuh~ in many PSU's hav~g less ~n 7,5~ persons; and, ~ith ~he q~- ~cadon ~ each Stan~rd Metropolitan Ar~ a PSU. ~e PSU is kep~ as in~e~aHy contrasting as possible in s~i~conomic ~ems. S~dficad~ of PSU's~ampling ~eo~ makes it cl~r ~at if units to ~ ~mpled ~n ~ classi- fied into ~ego~es or strata whose mem~rs tend m ~ relatively ~ke wi~Nn strata and r~advely u~ike ~ween s~ata, and drawings ~de ~om • ose s~am. ~hen resulting sampling variances are z~uced over those of samples ~awn from an ~stratifi~ universe. ~e PSU's were s~at~ied accordingly, ~e principal m~es of stradficadon ~ing ge~rap~ l~a~ion, densiw of popula~on, rate of pop~a~ion ~ow~h ~ween 1940 and 1950, pr~o~on d n~whi~e, type of ~dus~ in pre- domi~nfly ur~n areas, and ~ of farming in ~ral areas. The general sampling desi~ con- ~emplat~ drawing f~st-s~age uni~ wi~h pro~bil- i~y pro~nio~te ~o size, wi~ 1 PSU to ~ ~a~ from ~ch s~ra~. F~zher, iz was desir~ ~hat separate estimates ~ obtainable r~dily for ~ch of 41 subu~verses--to ~ f~ther descried later, but characmriz~ often in [he survey as Tab Ar~s. ~ese s~cffi~ons, au~en~ed by an e~sGng s~raGficadon of ~e PSU's. set up by ~e Census Bur~u for o~er pu~oses, res~t~ in classifica- tion of ~he appro~mamly 1,9~ PSU's into 372 strata. Fur~er descripdon of ~e precise manner ia which ~s was done is Wen in Appen~x IV. ~aw~ first-stage u~ts~From ~cb of the $72 strata I PSU was selec~ for inclusion in ~he sample wi~ pro~bi~W propor~ioaam ~o its pop.salon. ~is m~nt. for ex~ple, that a s~ll PSU with 50,~ i~bimnts in 1950 had o~y 1/20 as much chance of inclusion in [he sample as did ~e 15rger PSU w{~ 1 million in~bi~an~s. ~&se ~eren~ial sampling ~ms were of course taken ~to consideration in subsequent sampling and es- ti~ting s~eps. As indicamd, the selection procure and s~c~icatioa ~a~ serrate worksheet est~ms ~ compu~ for ~ch of ~he Tab ~eas had i~u- enced s~a~ffi~tion. ~e Tab Ar~s ini~lly s~c- ffied were ~e 8 larges~ Standard Me~opolima Ar~s. and wi~ ~ch of 11 ge~raphic sections. ~e S s~sections composed of (1)smaller Stand- ard Metr~olimn Ar~s. (2) o~er ur~n ar~s. and (3) o~er ~ ar~s. ~e s~tions and 8 la~e S~'s are sho~m on ~e map in fig~e 3. It shoed he understc.--4 ~.at s~parate s~adsdcs ~hed for e~ of ~ ~ T~ #~s, hut ~er ~ ~ for Tab ~s ~n ~ more ~n one ~my in[o breeder ~e$o~es for which reH~le ~es ~ ~e pr~uced. In ~ome ~s~nces, effici~ s~ca~on re- s~t~ in a s~atum he~g compos~ of 1 sin~e large PSU. From such a stra~ ~e sin~e PSU enters ~e s~ple ~ cerm~D' and is ~lled a se~-representing PSU, Each of ~e 8 largest S~'s and 102 o~er PSU's ~me self-representing PSU's. Each PSU ~a~ into ~e sample from a nonme~opolitan strat~ was utilized later as the ~ame for both "o~er urban" and "other ~ral" mb~ation areas. Table 12 ~ Appen~x VIII shows ~e geo~ap~e dis~ibudon of ~ se~-represent- ing ann sonnet-representing PSU's. Selection z~es~For s~pling p~poses and in order to reduce over-all variance, ~e civilian pop~aOon in ~e United States is ~vid~ into 3 mutu~ly exclusive classes or selection zones: Zone A. ~ose ~rsons living in co~on ,dwelling places. ~ne B. ~ose ~rsons living in ar~s of "new h~sing." Zone C. ~ose persons living in large spec- ial dwe~ng places. Common dweH~g places include whatwo~d be or- ~narHy regarded as such--for example, private homes, a~rtment houses, and duplexes. Areas of new hous~g are simply ~ose in wMch considerable new housing has ~en built since ~e last population e~s~ (Apr~ 1950) and wMch have ~en recorded and mapped by ~e ~nsus Bureau. ~ese may elude ar~s which wo~d ~ class~ as ~lon~ng to eider ~ne A or Zone C except ~t they are positively identified as ~ing in ~ne B, Special" dwelling places include such places as ~tentia- ties, refomatories, homes for ~e aged, mental hospimls, and hotels for ~ansients, ~e 372 first-stage ~ts are identic~ for aH 3 zones, but later-stage samplMg is han~ed sepa- rately for each zone. For the large s~ial dwdling places, Zone C. lists of in~vidu~l institutions and organi~tions • e ~mple PSU's w~e assembled from a varie~ of sources. These list~ places are exelud~ from further ar~ sampling. S~cial inst~ctions for ~aw~g s~ples of ~rs~s from ~ne ~ are pre- par~ for ~e ~fferent ~s of s~etal dwelling places. Such ~rsons, constituting about 2 percent of ~e universe, have not ~en included ~n i~dal tab~ati~ of dam and are not ~scuss~ f~er in ~is accost. ~ne relatio~hip ~tween selecd~ ~nes A and B and ~een ~nes B and C is slightly more complex and makes ~e of t~e p~eiple ofstrat~i- ~fion aft~ ~p~ng3 and ~age ~8 ~ reference 2. ~e of ~e ris~ of ~ sampl~g, when using ~m on nmMr of households for a p~or y~r as the ~sis for selecOon, lies ~ ~e e~stence of large u~ts ~new cons~ucfion built since the prior II TI06653825

TABULATION AREAS: NATIONAL HEALTH SURVEY Standard Metropolitan 1950 Pop.. Are--~s in I,O00'S b in

;~u~ unless corrective action is ~aken will in:tease varia=ze. From the National. Hous[n~ ~e C~n~ B~u had av~a~le a r~cord of large new con~c~cn activities ~ many of ~ 572 sam- pl~ PSU's. Consider ~s~ PSU's as he~n~ strad- fi~ ~nto 2 classes: (I) ~hos= PSU's which contain ar~s of new hous~n~ so and (2)all o~er PSU's. SeEmen~s are from wi~ ~I sample PSU's ~o represent Zones A and C. ~nes A and C are mutually exclusive; ~at ~s, they do no~ overlap. ~e semen,s se]ec~ ~om ~e class i PSU's are ~hen e~mined to see ~ ~ey fa~ into areas classified as new construc- tion ar~s accor~g to ~e Housing Inventory. ~ose segments from class I which do not f~l into new cons~ction ar~s and all accents select~ from class 2 PSU's are retained and ~come the independen~ samples for Zones A and C. ~e seg- ments or parts of segments w~ch are contained in • e areas of new construction, and which were ini- tially drawn into ~e sample, are at t~s point de- leted from ~e original ~mple. An independent sample is then taken ~om among the new con- struction areas of Zone B at the same sampling rate as Zone A. ~er-all sampling ratios for all 3 se- lec~on Zones A, B, and C are identical within each Tab ~, Appro~mately 8 percent of the popula- tion and of the sample are accounted for by Zone B. Selection of so.eats in Zone A.--~us by far the major part (90 ~rcent) of ~e sample is found in Zone A. For many pu~oses, it is convenient to think of ~e sample as co, sisting only of Zone A. An outline is given here of the way in w~ch sam- pling within PSU's is carried out for selection Zone A. An example of the pr~ess is given in Appendix VI. ~e ~timate sampling ~it within ~e PSU is called a accent. It is a ge~raphically defined ~rcel w~ch contains an expected 6 households. Se~ents to ~ included in the sample are chosen separately for each Tab Ar~ in a series of steps or stages.. S~vey specifications resulted in a require- ment ~at over a peri~ of a y~r 144 segments are to ~ su~eyed in each Tab Ar~. Wi~in cho- sen segments, all households are interviewed. (As noted in Section &, if it develops that a selected segment con~ins obviously more than 20 house- holds, it is subsampl~ and appro~mately 6 house- holds in it are inte~iewed.) ~e selecOon pr~edure all.ares the num~r of segments to ~ interview~ to first-stage units in the Tab Ar~ in proportion to the size of the strat~ they represent. Se~ents are drawn with- in PSU's through a sequence of selection of suc- cessively smaller ~ts of area un~l finally a unit containing ~e e~ected 6 households is secure. ~is becomes the dtimate sampling unit. An illus- traticn of d;.~ prc.:ed,lre is ~iven in Ap;en~ VL S~ples for ~ y~r, qua~er, and ~veek.--~- fial ~amp~n~ is ~ ~n in a %ray ~~ch nhe ~e~ents repD~ed for ~ch ~en~r q~ner an ~i~nden~ ~mple of the land ar~ of d~ U~ Sm~es. ~ qua~erly samples are ad~ve and ~us ~e annual sample is 4 dines ~e size of quarterly samples. ~e samples are also random- ~ed by necks %%~n each quarter, so ~ each week's in~e~,iews become a random sample of ~he population and the wee~y samples are ad~tive within the qua~er. ~e dezail by which ~is is complished is iHuszrated in Append~ VII. ~e f~l s~ey desi~ is off.tire over each quarter. wee~y samples are unbiased bu~ n~es~riIy fol- low a more res~ic~ed design, on ~he average de- pendin~ upon s f~s~-s~a~e selection of 60 rather than 372 PSU's. Mapping of Segments and Listing of Households For each segment in the" sample, the inter- viewer is furnished 2 maps: a ~ and a Seg.___- merit Map. The Key Map shows the general location of the segment and may be a county highway map or a city street or block map. The se~nent number and approximate location of the segment (shown by the large dot beneath "Hillcrest Avenue" in figure 4 are entered on the Key Map. SCAt.E . MT. ALTO, RD. Fi~ure 4. Key Nap showin~ Segment 0534. 13 TI06653827

seme cas~s n~) s~rucn~res are shs"a,'n on uhe map. Two illus~rations cf S-~grnent Maps are aho',vn in figures 5 and 6. T'n~ s-~gment b~,mndaries, in any case, are o'itlined on d~._~ map. E___S_T.~ 0534 MT. ALTO ROAD N Figure 5. One type of Se~men~ Map. 14 N. CUSHING DRIVE 2100 2102 2104 2106 D A A A A [] APTS [] [] ~ S ~ D 2 2101 2103 2105 2107 2109. 2111 FOURTH STREET Figure 6. A second type of Segment Map. D - D~elllng F - Flat S - Store Apts. - Apartments A - Garage The numbers inside indicate the number of floors ~n the structure and the numbers a/on~ TI06653828

Intervie-,~rs are mstrncted to list all ~s~s or c~ de~c~cns of ~ places ~o#le live or mi~h[ live, incl~n5 su~ places as OrPhan' house d~veHin~s, a#~ents, duplexes, ~a~ers, tents, bonehead, conve~ hexes, and rented r~oms, ~¢lu~g evening ~'~ch Hes in- side • e de~n~ se~ent." ~e ins~cGon is sup- po~ed and ampl~ied by ~e maps and a 93-page ~dexed lisGng m~u~. Th~ Hs~g operaNon is con- ducted at a ~e prior ¢o ~terviewing, ~hus pro- viding 2 checks on coverage: one at ~e ~e of list- ing and a second at the time of interviewing. Summa~ of ~its~veral differen~ ~nds of u~s and ca~ories are menNuned in ~is section and in the App~n~ces. It may ~ helpful to reca- pillage in caps~e form ~e principal elements of teminol~,. Tab Ar~ - ~e of 41 s~universes, defined by geographic boundaries and by size and density of population. PSU - Primary sampling units consist- ing of I or a group of contig- uous counties: about 1,900 of them in the United States; 372 in the NHS sample, Strata -372 socioeconomic classes into which the' PSU's are grouped. E__D - Enumeration. District, .a geo- graphic subdivision of a PSU, usually containing between 50 and 1,000 households. ~- A subdivision of an ED containing an expected 6 households. This is normally the uhimate sam- piing unit in the survey. Selection Zones - Strata in a different dimension, based upon type of dwelling unit, and utilized in reducing vari- ance. .Dwelling Unit - Place where persons live or might live. This is the unit list- ed for subsequent interviewing purposes. Elementary Unit- There are 4 elementary units or channels for processing in- formation which are utilized in the survey: (1) the household; (2) t_he person; (3) the health con- dition--illness, injury, chronic condition, or impairment; and (4) the episede of hospitaliza- tion. The Estimating Process Some aspects of the estimating process were treated in Section 4 under Editing and Processing, and other aspects are influenced, of course, by the ~-~ple design ~ch has just ]~e~a discussed. In ",',hat follows in t/:_~ pre=~.~nt ~ecdcn, ks f~_--us of a~tention is on the estimating problem as such. Th.~ exfimattcn pro:ess in the h~dth su~:ey is b-nsically simple, although actual prccedure in- clu!es a ccnsid~able numl:~r of steps. Leading reasons for the ap#arent comptexiW are 4 in num- ter, growing largely eut of the fact that the survey pre-~uces a variety of estimates in several dimen- sions: Geographic scope.--The survey yields work- sheet figures for the Nation as a whole, and also for constituent Tab Areas. The Tab Areas can be combined into geographic divisions of the country, or into classes ~hich reflect size and density of the population in the community. Type of statistic.--Three variations may be distinguished under this heading. (I)The number or proportion of persons in the population with a specified characteristic, such as having I or more chronic conditions, or not having visited a physi- cian within the year immediately previous to the week of interview. (2)Estimated volumes of events arising from, tabulating answers to such direct questions as, "How many days were you in the hos- pital, not counting the day you left?", with editing converting the reply to "Number of hospital days in past year." (3) The incidence of a particular disease or health condition, bulk up from cumulat- ing occurrences over 2-week periods as reported by persons interviewed in successive weeks. The first type of statistic named above will be recognized as an instance of binomial estimation (modified of course by the structure of the sample design), since each individual respondent will either have or not have the specified characteristic. The second type of statistic is like the first except that the population measures involved are quantitative rather than qualitative variables, and consequently estimation is not binomial. The distinction between the second and third types of statistic is sharpened perhaps with an example. Approximately 115,000 persons are interviewed each year in the health survey, about 2,200 each week. Each of the 115,000 persons, in effect, gives the interviewer the num- ber of days he spent in the hospital in the previous year, and thus provides data which per~nit an esti- mate of the number of days of hospitalization ex- perienced by living persons in the year previous to the week of the interview. This is a type (2) esti- mate. Similarly, each week approximately 2,200 persons report their days of hospitalization in the previous 2 weeks. Summing these reports over 52 weeks of interviewing and taking account of the l- week overlap in reference periods for adjacent weeks of interviewing would provide the basis for a second estimate of a year's hospitalization, this time the resulting statistic being of type (3). More is said later on the procedure whereby estimates of type (3) are produced. Some might also wish to distinguish, under this tide, between estimates of an aggregate, such as total number of physician visits for a specified ~5 T[08653829

~e~ew$ ~ ~o~adon ~;er aweek, a q~rter, ~ ye~, or o~er ~e ~e~-~s. ~s~ r~erence per~s for ~c~ence or vol~ of even~ ~n ~e v~ed widely, mn~g for some items from a week to any m~dpl~ d wee~ ~ ~e histoD, of ~e s~ey. Fo~ of es~ate.--Tne sm~s~cs prc4uc~ ~om ~e s~a~i~ desi~ ~ough 2 stages of ra- ~o es~afion are ~e pr~ucts ofa desi~ which is mu~h more efficiem ~an a simpl~ ~ndom sample would have ~en. hu~ which necessary require sdmewhat more elaborate computation. St~ps in estimationrln ~e interest of brining out main ~r~ds of the es~ma~g s~o~. obscured as litde as possible by ~e crosscurrents jus~ not~. ~e reminder of this section is written mostly around ~e pr~ucdon of est~ates of an average n~ber of persons with a s~cffi~ characteristic. • e average ~ing ~sed on inle~iew~g over 13 weeks. ~e pop~adon referred to is the civilian noninstitudonal pop~afion of the continental United States rather than that of one of ~e T~ Ar~s. An aggregate ra~er ~n a proportion or ra~e is • e statistic ~der obse~ation. ~casional varia- tions from ~is patte~ will ~ necessa~. As in, cared cartier, incoming reports are ~as~ through controls to insure that the data in- put to the computers is consistent with sample de- sign. pro~r~y c~. and ~pable of ~ing tabu- lated. A series of mechanical edits are carried out on the computers. These edits make the question- naire internally consistent, and adjust or account for item nonresponse. Step 3 -'~toeach record of an elementary unit(person, household, condition, and hospitalization) basic sampling inflation factors are inserted. This step takes account of all stages of sampling. The factor is the reciprocal of the combined sampling frac- tion which for a quarterly tabulation varies among Tab Areas from about 1-in-2,000 persons to about 1-in-t9,000 persons. [Sampling fractions for an- nual samples are one quarter of these numbers.] Step 4 "--~atistical theory demons~ates that a "ratio estimate" for any statistic is superior to an or- dinary "inflation estimate" if there is correlation bet~veen the numerator and the denominator of the ratio. Specifically, if Y' and X' are ordinary in- 16 flation estimates cf 2 characteris-dcs cf a ri~n, y and X, re~q:~ct/vely, and if fit.~ "true" tcLal X is tmo',vn in_~_~.r.Sendy, d~_=n the ratio e~rimate = .~ X is a l:_~tter e&rimate of Y than is provided there is correlation teL-.*;een Y' ~d In this form of estimate, th_~ quantity ~-~ becomes a calibration factor for the survey. This principle is utilized at 2 stages in the NHS. In the first instance it is used to reduce sam- pling variance between PSU's. Estimates of the 1950 population which would have been obtained from a complete enumeration of the 372 PSU's hut not other PSU's in the country were compared with official 1950 population counts for each of 120 color-residence classes. Resulting factors are shown in table A. In calculation, these factors are used in the following manner, the arithmetic being carried out automatically by the computers. Consider a eample record for a person who is white and who comes from an urban nonself-representing Standard Met- ropolitan Area in Geographic Region 1. All sample records for this person are multiplied by the fac- tor 1.075380. (See 1st line, 2d column of table A.) This brings the sample data into closer conform- ity with population controls for the universe, intro- duces only trivial, if any, bias into the estimate, and reduces sampling variance. NOTE: Steps 1 through 4 are carried out wee'kly, and provide a "deck of cards" (Uni- vac tape) of edited and adjusted sample data for each week of the 13 weeks of the quarter. The "scale" of data at this point is therefore 1/13th of universe totals. Weekly data are merged later into quarterly totals. Steps 5 and later apply tothe merged quarterly decks. Step 5 --]T~espite intensive follow-up efforts, reports on some households in the sample have not been received at the tabulation cutoff. In the first 2 quarters of operation the noninterview rate was 6 percent--1 percent refusal, and the rest for all other reasons, such as no one at home after re- peated call backs. For a sample household for which no interview is obtained, any estimating pro- cedure must necessarily impute values for each statistic for which measurement had been intended. Adjustment for noninterviews in the health survey is accomplished by a calculation which as- sumes that respondents within a particular seg- ment for a quarter represent the nonrespondents in that segment. In the rare instance in which less than half of a segment is interviewed, the nonin- terview adjustment is mcdified by evidence from T106653830

Table A. First-stage ratio estate fa=tors f~-r n~nself-repreEentimg* FgU's by residenae, col~, and section ~;onself-representing SMA' s geographic section I ..... I0 ..... ii ..... Urban ~rnite l~cn%~nite 1.0753~0 1.753515 .973243 .809129 .755966 .7Z4533 1.328674 .927637 1.210693 .509411 none • 973222 .869424 1.076720 1.027738 1.179743 1.179743 • 873733 .492896 none R~ral n~nfarm Rural farm '~hite .579098 1.345792 .723622 1.580769 1.640072 Nonself-representing no~n- SMA's geographic section I ..... 10 ..... ii ..... 1.000096 1.050026 1.101374 1.022545 1.109184 1.000335 .988710 .984943 .980173 1.043498 1.020053 .791543 .891068 1.476958 1.142936 1.649665 1.065752 1.044770 1.028117 .977126 .979703 1.203531 Non~hite .678533 .673733 1.048442 .927872 .664274 .633703 1.158533 1.158533 1.461507 .428340 .889450 .715170 .899394 1.349170 1.094866 1.874840 2.502347 2.502347 1.084175 1.243901 .883253 .912743 .873888 1.276623 1.027072 1.026288 1.013235 1.108361 .807592 .870792 .776500 1.250069 .873241 1.066578 .959741 .962977 .953182 .873888 1.011195 1.086449 1.016688 1.005968 1.005739 .978919 .991135 .991556 .919709 .997690 1.000277 1.013374 1.008986 1.004002 1.016431 .890752 1.006944 Non~-hite .488372 .721580 ,695719 1.580769 .383598 .745692 .991501 1.250069 .873241 1,003409 1.301490 .728575 ,871958 1.673347 1o156285 1.071351 .968411 .990630 1.199575 1.009057 IFirst-stage ratio estimate factors for each of 8 large sPparate ~bol.tlon or~as and for the self-repre- senting PSU's is 1,000000, reports over the entire Tab Area. An illustration of the process is given fora hypothetical Tab Area: House- holds Segment sched- number uled for inter- view i 6 2 6 3 8 4 4 Tab Area total 220 House- holds not inter- viewed 0 i 0 3 i0 Segment Excess adjust- non- ment inter- lacier views 1.0000 0 1.2000 0 1.0000 0 2.0000 2 2 Data for the 5 reports in segment 2 are multiplied by the factor 1.2000 sothat the 5 reports represent the 6 households intended for interview in the seg- ment. Segment 4 in the example is of the unusual type (where less than half the households in the segment were interviewed) which leads to a fur- r.her adjustment at the Tab Area level after a pre- liminary one has been made at the segment level. The Tab Area adjustment factor is the ratio of total households scheduled for interview to total households scheduled for interview less the "ex- cess" noninterviews; that is, the factor is 220/218 or 1.0092, in the example. Data for all reporting households in the Tab Area are multiplied by this factor to account for the 2-household "excess" of noninterviews. Advantages of the ratio-estimating process are exploited further by the introduction of a sec- ond calibrating or raNo factor which brings the 17 T108653831