DATA EVALUATION AND METHODS RESEARCH

Multiplicity Studyof Marriages and Birthsin Israel

A description is given of an experimental study, the aims of whichwere to test the feasibility of conducting a multiplicity survey formarriages and births in Israel and to evaluate different ccluntingrules. The study included an evaluation survey from which esti-mates of the components of mean square error were obtained. Theresults indicate that an overall reduction of mean square error wasobtained by use of a multiplicity rule instead of a conventional rule.The substantial reduction in sampling error attain~d by the use ofmultiplicity was not offset by the slight increases in response biasand variance.

Series 2Number 70

DHEW Publication No. (HRA) 77-1344

U.S. DEPARTMENT OF HEALTH, EDUCATION, AND WELFAREPublic Health Service

Health Resources AdministrationNational Center for Health StatisticsHyattsville, Md, June 1977

Library of Congress Cataloging in Publication Data

Nathan, Gad.Multiplicity study of marriages and births in Israel.

(Vital and health statistics: Series 2, Data evaluation and methods research; no. 70)(DHEW publication; no. (HRA) 77-1344)

Prepared by G. Nathan, U. O. Schmelz, and J. Kenvin.Bibliography: p.Includes index.Supt. of Dots. no.: HE 20.6212:2/701. Israel–Statistics, Vital. I. Schmelz, Oskar, joint author. II. Kenvin, Jay, joint author.

III. United States. National Center for Health Statistics. IV. Title. V. Series: United States.National Center for Health Statistics. Vital and health statistics: Series 2, Data evaluationand methods research; no. 70. VI. Series: United States. Dept. of Health, Education andWelfare. DHEW publication; no. (HRA) 77-1344. [DN’L,A4: 1. Birth rate. 2. N’larriage-Statis-tics. 3. Vital statistics—Israel. 4. Evaluation studies. 5. Models, Theoretical. 6. Sociometrictechnics. W2AN148vb no. 70]RA409.U45 no. 70 [HB3659.18] 312’.07’23s 76-608362ISBN 0-8406 -0089-5 [301.32’072’31

For sale by tbe Superintendent of Documents, U.S. C+wernmemt Printing OfficeWashington, D.C. 20402

Stock NO.017-022-00559-O

NATIONAL CENTER FOR HEALTH STATISTICS

DOROTHY P. RICE, Director

ROBERT A. ISRAEL, Deputy Director

JACOB J. FELDMAN, Ph.D., Associate Director for Arudysis

GAIL F. FISHER, Associate Director for the Cooperative Health Statistics System

ELIJAH L. WHITE, Associate Director for Data Systems

JAMES T. BAIRD, JR., Ph.D., Associate Director for International Statistics

ROBERT C. HUBER, Associate Director for Management

MONROE G. SIRKEN, Ph.D., Associate Director for Mathematical Statistics

PETER L. HURLEY, Associate Director for Operations

JAMES M. ROBEY, Ph.D., Associate Director for Program Development

PAUL E. LEAVERTON, Ph.D., Associate Director for Research

ALICE HAYWOOD, Information Officer

OFFICE OF THE ASSOCIATE DIRECTOR FOR MATHEMATICAL STA17STlCS

MONROE G. SIRKEN, Ph.D., Associate Director

Vital and Health Statistics-Series 2-No. 70

DHEW Publication No. (HRA) 77-1344Library of Congress Catalog Card ~umber 76-608362

FOREWORD

The National Center for Health Statistics contracted with the Israel CentralBureau of Statistics to undertake methodological studies in Israel and in theAdministered Areas to evaluate the multiplicity y design of houschoId sample sur-veys for compiling vital statistics. This is a report of the evaluation project thatwas conducted in Israel.

This project involved a close and effective working relationship between thestaffs of the Central Bureau of Statistics and the National Center for HealthStatistics. Representing the Center, I served as the Project Officer, and Dr. SidneyGoldstein, Brown University, served as a demographic consultant to the project.Dr. U. O. Schmelz and Dr. Gad Nathan served as the Project Directors for theCentral Bureau of Statistics.

Acknowledgments are due to the United States-Israel Binaticmal ScienceFoundation for awarding a grant to the project. The Foundation’s support madeit possible to carry out the full study design despite the adverse effects of a rapidinflation and several devaluations of the Israel currency during the course of theproject.

Monroe G. Sirken, Ph.D.Associate Director for

hfathematical Statistics

...111

CONTENTS

Chapter 1. Introduction ........... .. ............... ..... ... ..... .. ........ ........ .......... .. .. ........... ...... ........

Chapter 2. The Basic Survey: Design and Execution ..... .... ............. ..... ........ .... .... ..... ... .2.1 Purpose and General Design ...................... .......... ..... ... .......... .... ......... ..... .. ...... ....2.2 The Pretest ........... ..... .......... ..... ..... ... ...... ................ ... ........ ..... .. .. ...... ..... ........ .......2.3 The Final Design .......... .. .......... ........ ... .. ......... ......... ....... ... .... .... ..... ..... ... .. ... ..... ....2.4 Execution of the Basic Survey ......... ........ ..... .......... ................ .. ............. ... .. ........

Chapter 3. The Evaluation Suxvey: Design and Execution .............. ..... ..... ........ .......... ..3.1 PmPose, General Design, and Pretest ........ .... ... ....... .... ... ... ..... .. ........ ..... ... ..... .....3.2 Final Design of the Evaluation Survey ................ .... . . .. ..... ... .. ........... ..... .. ... ..........3.3 Execution of the Evaluation Survey .............. ..... .............. .... ..... ..... ...... .. ........ .....

Chapter 4. Results ........ ..... ........ .......... ...... .... .......... ... .................. ..... .......... .................. ..4.1 Basic Results on Yields and Network Sizes ....... .......... ........... .... ..... ...... ..... ..........4.2 Components of Error and Comparisons With Demographic Data ......... ..... ... ........4.3 Distribution by Demographic Characteristics ................ ..... .......... .. ... .. ...... ..... .. ...4.4 Co-residence .... .. ....... ... ..... ....... ..... ... ........ .... ... ... ..... ... .. ..... ... .. ... ............... .... ....... .

Chapter 5. Discussion of Specific Points ...... .... ................ ... ............... ..... ... ..... ..... ...........5.1 Choice of Multiplicity Counting Rules ....... . ......... .... .... ... ..... ....... ... ..... ........... ......5.2 Operational Problems .............. .. ... ..... ............. .......... ................. ..... ...... ... ............5.3 Length and Position of Recall Period .... ...... .................. .............. ................ ..... ....5.4 Possibility of Application to Other Topics ....... .. ... .......... ..... ..... ..... ........ ... ...........

Chapter 6. Summary and Conclusions ................... ..... ........ ....... ..... ... ....... ..... ... ........ .. ....

Appendixes .... .................. .. ... .......... ..... .......... ........ ..... ... ............ ... ............... ..... ... ..... ..... ... .I. Bibliographical List ........... ........ .... ...... .... .... ... ....... .... ...... ... .. ..... ........ .................II. Multiplicity Survey Questionnaire ........ ...... ................ .. .. .. .......... ...... ... .. ... ..... .....III. Labour Force Survey Listing Form: Household MembersIV.

.... ............... ........ ........Record Format: Baaic Survey Processing

v..............................................................

Evaluation Survey Processing Form ..... ........ ..... ............... ........................ .... ..... .VI. The Model for Estimation of Error Components ..... .............. ...... .. ... ............. .....VII. Effect of Dependence on Evaluation Results .............. .......... ............ .. ......... ..... ..

...m

1

5568

12

14141517 ‘

2020222732

3737394143

45

4748495657596063

FIGURE

1. Questionnaire flow diagram-births .......... .................. ....... ... .................... ........ .......... . 11

LIST OF TEXT TABLES

A. Execution of the evaluation survey ....... ................... .... .. ... ..... ............. ... .. ..... ... ..... .... 16B. Reasons for noncompletion in the evaluation survey, by type of event ......... .......... .. 19c. Raw yields (number of events per household) and weighted yields (number of event

reports weighted by reciprocal of multiplicity), by type of event and counting rule.. 20D. Events by number of households in multiplicity network, harmonic mean of muMi-

plicities, and number of reportiug households, by multiplicity counting rule, typeof event, and type of report ......... ..... .... .... ..... ..... ..... .......... ........ ........ ..... ... .. ... ..... .. .... 21

v

E.F.G.

H.

J.

K.

L.

M.

N.o.P.

Q.

R.s.

Estimates of components of error, by counting rule, and current demographic data..Analysis of relative biases by household of reporting relative and counting rule ..... ...Relative efficiencies (compared to conventional survey), by sample size and count-ing rule ..... ................. ....... .... ............. ............ ..... ............................................ ...........Percent distribution of events, specific rates per 1,000 population, and estimatedrelative standard errors, by counting rule, type of event, and age of event person ......Percent distribution of events and estimated relative standard errors, by type ofevent, counting rule, and sex of event person ......................... ..... ............ ............ ......Percent distribution of events and estimated relative standard errors, by type ofevent, counting rule, and month of event ...... ..... .......... .. ...... ...... ..... ................... .. ...Percent distribution of events and estimated relative standard errors, by type ofevent, counting rule, and district of occurrence ......................... ...............................Underreporting and overreporting by relative’s household, by type of event andcharacteristics of event and of head of event household ...................... ... .............. ....Reporting discrepancies between event household and household of relative ...... .... ..Residence of mother’s mother (for births) ........... .... ...... ......... ........... .. .. .. .. ..... .. ..... ...Percent distribution of parents of married persons by residence of paxents, accord-ing to sex ........................ ........ ..... ......... ...... .... ..... ..... .... ........ ......... ... .. ................. ..... .Percent distributions of married persons’ parents residing outside event household,by co-residence with spouse and with children ........ .. ...................... ..... ............... .... ..Analysis of relative biases, by type of event, reporting relative, and counting rule ....Relative differences (between survey estimates and vital record estimates) and rel-ative standard errors for selected reference periods, by type of event and counting

“2325

27

28

28

29

30

313334

34

3538

rule ..... ....... . .. .... .. ........ ..... .. ..... ..... ..... .. .............. .... ......... .......... .. ..... ...................... ..... 42

SYMBOLS

Data not available ---------------------------------- ---

Category not applicable ------------------------- . . .

Quantity zero -------------------------------------- -

Quantity more than Obut less than 0.05----- 0.0

Figure does not meet standards ofreliabilityy or precision--------------------------– *

t

MULTIPLICITY STUDY OF MARRIAGESAND BIRTHS IN ISRAEL

Gad Nathan, Usiel O. Schmelz, and Jay Kentin, Israel central Bureau of Statistics

Chapter 1. INTRODUCTION

Some years ago, Dr. Monroe G. Sirken of theU.S. National Center for Health Statistics devel-oped a statistical model for a new type of house-hoId sample survey design-the design of house-hoId surveys with multiplicity. This type ofdesign appears to be particularity well suited forimproved estimation by means of a single-timehousehold sample survey of the number of vitalevents that occurred in a population. Improvedestimation as compared to a conventional house-hold survey is to be obtained through reductionof sampling error and of certain biases.

In the conventional household survey, ordyevents occurring to members of the particularhousehold interviewed are studied. The multi-plicity survey also covers events occurring inother households. The additional households tobe covered are defined by the specific rules a-dopted for the survey. For instance, the surveymay cover events to others living elsewhere inthe survey population who are relatives (parents,siblings, children, etc.) of members of the house-hold actually interviewed. Consequently, thesame event may be reported by more than onehousehold.

Increasing the number of households whichcan report an event obviously increases the num-ber of events reported by a given sample ofhouseholds, as compared to a conventional sur-vey. This does not automatically decrease sam-pIing variance in the same way as is achieved byincreasing the number of households in a con-ventional survey; in practice, however, samplingv~iance can be reduced in many cases. Further-

more, although some types of response errorsmay increase in a multiplicity survey due to thereporting of events by relatives, certain types ofcoverage bias encountered in a conventional sur-vey-those due to there being no household inthe survey population to report tm event–maybe reduced. For instance, it has been surmisedthat the use of multiplicity may increase thechances of locating deaths which occurred in sin-gle-person households.

In order to put this design into operation, itis necessary that each household sampIed report,in addition to the events that occurred to itsown residents:

(a) Events that occurred to a network ofspecified persons, such as relatives, livingelsewhere.

(b) The number of households containingpersons eligible to report an event ac-cording to the specific ruIes adopted forthe survey. This number is called the“multiplicity” of the event.

To estimate the totaI number of events of anytype studied in the population, each event re-ported in the survey is weighted by the recipro-czd of its multiplicity. This weighting is neededto adjust for the multiple number of householdseligibIe to report the same event. Thus eventswhich couId have been reported by a larger num- .ber of households make a sm41er contributionto the estimate than those which could havebeen reported by a smalIer number of house-

holds. While it is the number of different house-holds in the network that determines the muhi-plicity of each event, operationally it will oftenbe necessary to list all the persons in the net-work who are eligible to report and then ascer-tain in how many different households they re-side, as several relatives may be living together.

Like the conventional survey, the multiplic-ity survey collects information on various char-acteristics of each event covered and of the per-sons concerned, whether they live in the house-hold interviewed or elsewhere. Further informa-tion on the methodology and results of surveyswith multiplicity may be found in papers bySirken (1970) and Sirken and Royston (19 73).(See bibliographical list, appendix I.)

After several practical tests of the newmethod had b~en undertaken in the UnitedStates, negotiations took place in 1970 betweenthe U.S. National Center for Health Statistics(NCHS) and the Israel Central Bureau of Statis-tics concerning the feasibility of conducting amultiplicity survey for some types of vital eventsin Israel and in the Administered Areas. Israelseemed to offer several advantages for testingthe new design:

1. Israel has a very heterogeneous popula-tion with regard to type of culture, fam-ily structure, etc., including Jews of dif-ferent origin (from Europe, fromAsia-Africa) and of different duration ofstay in Israel and non-,Jews of various re-ligions. In addition, Arabs in the Admin-istered Areas differ from those in Israelin some respects.

2. Household surveys-the Labour ForceSurveys –c omprising a representativesample of the population are undertakenevery week and are regularly tabulatedfor each quarter of’ the year.

3. There exist reliable and detailed demo-graphic statistics for Israel, which can beused for assessing the results of the mul-tiplicity survey.

The Israel Central Bureau of Statistics gladlyaccepted the idea of undertaking .a trial of thenew multiplicity design in Israel and the Admin-istered Areas, with the twofold intention of col-

laborating generally in methodological progressand of acquiring a technique which might be ofassistance in its own work. In particular, deathreporting through routine notification is knownto be very deficient in the Administered Areas,and a successful application of the new tech-nique might remedy this lacuna.

Since the purpose of the multiplicity studywas to test a new survey design and the conse-quent innovations in fieldwork procedures, theproject consisted of two stages each for Israeland for the Administered Areas: the basic surveyand an evaluation study. The necessity of a spe-cial evaluation study, involving an additionalround of interviews, influenced some features ofthe basic survey (e.g., the need to collect de-taiIed addresses of relatives so as to make possi-ble the evaluation interview); it also influencedthe timetable and the budget of the entire proj-ect.

The topics of inquiry for the multiplicitysurveys in Israel and in the Administered Areaswere to be specific types of vital events (births,marriages, and deaths). The events that occurredduring 1 calendar year were to be reported.

Linking the fieldwork for the basic multip-licity survey in Israel to the continually con-ducted Labour Force Survey (LFS) had greatpractical advantages but ako imposed certainlimitations, which will be set out later in thisreport. In the Administered Areas the same sam-pling frame was used as for the current LFSthere, but the actual enumeration interviewswere eventually separated.

After the funds initially thought necessarycould be secured by NCHS, an agreement wasconcluded in summer 1971 between NCHS andthe IsraeI Central Bureau of Statistics. It pro-vided for the conduct of a multiplicity survey onvital events in Israel and the Administered Areas, ‘including special evaluation studies.

The actual implementation of this projectwas beset with numerous difficulties and delays,particularly for the following reasons:

1. By the time the U.S. funds became avail-able, somewhat belatedly, and the con-tract could be signed, the Israel CentralBureau of Statistics already had had tomake active preparations for taking a fia-

2

2.

3.

tional census ofheld eventually

population and housing,dunrm Mav,lune 1972.

The census taxed to tie u~m~st the Bu-reau’s professional and organizational re-sources. The two collaborating institu-tions therefore agreed that commence-ment of work on the multiplicity surveyshould be postponed until some timeafter the census enumeration so that thepkmning stage for the survey would notstart before fall 1972. This postponeddate of commencement was adhered to.

In recent years Israel has experiencedrapid inflation and several devaluationsof the local currency. The budgetary es-timates for the multiplicity survey weredrawn up in 1970 and formulated in Is-raeli currency. As time went on, theseamounts became clearly insufficient. Asa result of financial stringency, repeatedcuts had to be made in the survey pro-gram. Eventually it was found that thesurvey could be concluded onIy with thehelp of an additional grant from theUnited States-Israel BinationaI ScienceFoundation, which itself became avail-able later than anticipated, entailing fur-ther postponements of operations.

The war of October 1973 and its pro-tracted aftermath seriously affected themultiplicity surveys. The major hostil-ities occurred during the time betweenthe pretests and the fulI-scale implemen-tation of the multiplicity survey in Is-raeL As a consequence of the consider-able number of deaths caused by the warto the Jewish population and the recentdate of these losses, one of the plannedtopics of inquiry—deaths of relatives—had to be hastily exchanged in the ques-tionnaires for interviewing in Israel byanother topic unaffected by such sademotions. Furthermore, the massive andprolonged callups strongly affected theavailable staff at headquarters, in the re-gional offices, and among the rank andfile of the enumerators. AN this resultednot only in delays but especiaHy in re-grettable losses of quality. With a staff of

Itbirths

enumerators consisting largely of newand inexperienced workers, many ofwhom soon dropped out and had to berepIaced by other fresh recruits, andwith a depleted staff at the regional levelfor training and supervision of the newinterviewers, the Israel Lab our ForceSurvey, to which the basic survey of themultiplicity study was linked, failed toachieve the high completion rates cus-tomary before the war. On the otherhand, it maybe a cause for some gratifica-tion that the survey could be conductedat all, with substantial results, despitethe disruptions produced by the suddenoutbreak of war.

had orig-hdly been intended to studyand deaths among Tews in Israel; births

u.

among non-Jews in Israel (deaths among themare too few to be studied through a smaI1 sampleof the population); and births, child deaths, andother deaths (the two latter with appropriatelydifferent multiplicity rules) among Arabs in theAdministered Areas. After the pretests the sur-vey of non-Jews in Israel had to be dropped forreasons of economy that will be set out morefully in a later chapter of this report. And, asindicated above, questions about deaths to Jewsin Israel were dropped as a consequence of the1973 war. Mamages were substituted as a sec-ond topic of inquiry, along with births.

The years for which respondents were askedto report events were eventually 1973 for Israel,where the fieldwork for both basic survey andevaluation study was carried out in the first halfof 1974, and 1974 for the Administered Areas,where the fieldwork took place at the beginningof 1975.

Several ewduation strategies were utilized inthe surveys. Both in Israel and the AdministeredAreas, a round of special evaluation interviewswas held with relatives of some of the personsenumerated in the basic survey (the latter havingprovided the names and addresses of these rela-tives). This made it possible to compare reportson the same event from two households belong-ing to the same multiplicity network. For bothbirths and marriages in Israel and for births inthe Administered Areas, aggregative comparisons

3

could also betistics. In the

made from the available vital sta-pretest stage in Jsrael only, vital

events reported by routine notification were re-investigated through household interviews. Thisapproach, which checks the accuracy of subse-quent reporting in a household interview againstthe current notification but is not directly aimedat evaluating the multiplicity y technique, waseventually discarded.

Though undertaken by the same organiza-tion, the Central Bureau of Statistics, within theframework of the same contract, the basic sur-veys and evaluation studies for Israel, on the onehand, and for the Administered Areas, on theother, are to be viewed as distinct operations.The populations are demographically and cultur-ally very different (as reflected in differences ofcontents and layout of the respective multiplic-ity questionnaires), the scope and quality of thealready available vital statistics documentation isdifferent, and all fieldwork for the two territo-ries was conducted separately. Also, the resultsof applying the multiplicity technique weremarkedly cliff erent. Hence, each operation re-quires a separate report.

The following chapters of this report willdeal with the multiplicity survey in Israel only,excluding the Administered Areas. The designand execution of the basic survey and of theevaluation study are described in detail in chap-ters 2 and 3, respectively. The integrated resultsof both these stages are presented and discussedin chapter 4 in terms of testing the applicationof the multiplicity technique under Israeli condi-

tions. Also in chapter 4 are data ‘on co-residenceof relatives, a byproduct of the analysis of themultiplicity network of relatives and the distri-bution of relatives according to households. Inchapter 5 selected points from the experiencegathered in Israel are considered, and some re-marks are proffered on matters that may deservethought and exploration in future applicationsof the multiplicity technique. The report termi-nates with a summary (chapter 6) and a series ofappendixes containing a bibliographical list, thequestionnaires used, and other documentation.

In one respect, the Israeli multiplicity studyhas proved the obvious: the superiority of acomplete current registration of vital events oversubsequent household surveys, in which the re-spondents are not always the persons most di-rectly concerned and have to rely on memory,facing particular difficulties in the dating ofevents which occurred at the turn of two calen-dar years. But this is almost beside the point.

The reaI purpose of the Israeli multiplicitysurvey and its evaluation program has been tocompare the efficiency of this new techniquewith a conventional household survey. As ap-plied in the Israeli survey, the multiplicity tech-nique did indeed greatly reduce the total meansquare error, primarily through reduction of thesampling variance. The findings with regard tobiases were less clear-cut. This problem, as wellas that of response errors and of the interestingdivergencies that emerge when different multi-plicity rules are put into operation, will be setout and discussed in later chapters of the report.

—ooo —

4

2.1

ity

Chapter 2. THE BASIC SURVEY: DESIGN AND EXECUTION

Purpose and General Design

The major purpose of the study of multiplic-techniques was methodological rather than

substantive; This was particul~ly the case forIsrael, where adequate data on vitzd events areobtained on a current basis from vh.-hdy com-plete registration, so there is no reason to’ expectthat data from a sample survey, however sophis-ticated, would afford an improvement. The ex-istence of reliable data from a well developedregistration system does, however, provide astandard against which the results of the multi-plicity study may be checked.

The purposes of the basic survey in Israelwere primarily to test the feasibility of usingmultiplicity techniques in a population which isin general relatively advanced but which is akovery heterogeneous with respect to culturalbackground, and secondly to serve as a basisfor evaluating the multiplicity technique inthe specified population. Thus it was necessaryto test the ability and willingness of members ofvarious groups to report events which occurredto relatives, to inform enumerators of a few de-tails concerning the events, and to report theaddresses of relatives of the person to whom theevent occurred (the event person). This lastpoint served both as the means for determiningthe number of households in which the relativesof the event person lived—that is, the multiplic-ity of the event—and as the means for selectingthe sample for the evaluation study, to be dis-cussed later. It should be noted, however, thatthe multiplicity could have been determined bydirect questions without obtaining addresses ofrelatives.

In addition, it was necessary to assess thedifficulties involved in technical aspects of a sur-vey with multiplicity, such as questionnaire for-mulation, training and organization of fieldstaff, and coding and processing.

Finally, the basic survey was designed toserve as a basis for evaluating the efficiency ofdifferent multiplicity “counting rules.” The“counting rule” of a multiplicity survey is de-fined as the rule which links each event to the

households of relatives who sho~d report it, andit is determined by the network of relatives cov-ered. The efficiency of different multipIicit ycounting rules relative to the efficiency of a con-ventional survey was to be measured by the totalmean square error and its components—overre-porting and underreporting biases, sampling vari-ance, and response variance. The detailed basicsurvey estimates were to be compared with thecurrent demographic estimates based on vitalregistration. Reports of events in the basic sur-vey were also to serve as a framework for theevaluation survey sample and for the compari-sons needed to evaluate the error components.

In order to achieve these goals, the surveywas originally designed to cover as much of thepopulation of Israel as possible, both to simplifycomparisons with current demographic estimatesand to represent the heterogeneity of the popu-lation. Since the basic unit of inquiry was to bethe household (with one respondent answeringfor all members), the population of kibbutzimand institutions was excluded from the surveypopulation. These populations are generaIIy in-cluded in sociodemographic szunple surveys inIsrael (e.g., the Labour Force Survey). However,information is not obtained directly from theperson but from records available in the officeof the institution or the kibbutz. Information onvital events occurring to a person’s relatives isnot usually available from such basic personalrecords. The high cost of special individual inter-views for the multiplicity survey weighed in fa-vor of excluding these subpopulations,a espe-cially since for Jews they include only about 5percent of the population. The Bedouin tribes inthe south were excluded for simikw reasons.

According to the origimd pkms, the surveypopulation included the total resident de jurepopulation, according to the Bureau’s regulardefinitions, excluding residents of kibbutzimand institutions and the Bedouin population ofthe south. As expkined below, a decision to in-

aAn ~temative method that includes events to per-sons in institutions is proposed in section 5.2.

5

elude only the Jewish population, in the surveywas subsequently reached.

An important design decision related to top-ics to be included in the multiplicity survey.Since a comparison with reliable, current demo-graphic estimates was required, it was decidedfrom the start to limit the field of possibilities tobasic vital events of relatively high frequency–births, deaths, and marriages. The initial de-sign, as tried in the pretest, included births forJews and for non-Jews and deaths for ,Jews only.Subsequently, the basic survey covered birthsand marriages. The recall period was fixed as 1calendar year to attain sufficient frequencies, onthe one hand, and to reduce recall error associ-ated with long recall periods, on the other.

A basic design decision for multiplicity sur-veys is on the extent of the reporting networkconnected to each event—the counting rule.Wider counting rules, while usually reducingsampling variance, may increase reporting biasand place an added burden on the fieldwork. Ahierarchical series of counting rules was requiredto study the relative efficiency of each rule.Another requirement was that the counting rideselicit sufficient frequencies of events. This canbe attained by considering the relative positionof an event in the life cycle and ensuring thatrelatives of appropriate different generations areincluded in the network (e g., both siblings andparents of married persons, both siblings andsons and daugh~ers of deceased persons).

It was decided to limit multiplicity reportingon births to the female line, primarily to ensurea feasible burden on the field staff. Thereforereports on births to women in the householdand to daughters and sisters of women in thehousehold were requested. However, for deaths(the frequency of which is much lower), reportswere requested in the pretest on spouses, par-ents, siblings, and children of all persons in thehousehold. Similarly, in the final survey, thecounting rule for marriages was all persons in thehousehold and children and siblings of all per-sons in the household.

The feasibility of carrying out a large-scalemultiplicity study in Israel was enhanced by thefact that a quarterly Labour Force Survey (LFS)is carried out on a current basis. A concise de-scription of the survey can be found in the pub-

lication of(1975). The

the Central Bureau of Statisticssurvev, besides its main purpose of

providing data on ‘the labor force and-its ‘charac-teristics, serves as an omnibus for collecting ad-ditional data on the labor force and other topics.The LFS is a quarterly survey with a self-repre-senting probability sample on a rotating panelbasis, with four panels each quarter. The totalsample size per quarter since ,January 1974 hasbeen about 12,000 households. (Previously itwas half this size.) At an early stage of pkmning,the LFS was found to be a suitable vehicle forcarrying out the basic multiplicity survey.

The general design of the basic survey was toadd to the LFS questionnaire a multiplicityquestionnaire, which included questions onevents to the specified relatives and the ad-dresses of relatives who could have reportedthese events, in order to determine the multiplic-ity of each event. Before fixing the final detailedbasic design, described in section 2.3, a pretestwas carried out, as described in the followingsection.

2.2 The Pretest

‘The objective of the pretest for the basicsurvey was fundamentally to test the feasibilityof the general design, described above, on asmall scale. Specifically the pretest was designedto test the attitudes and reactions, both ofhousehold members and of field staff, to ques-tions relating to events, especially to thosewhich occurred to relatives outside the house-holds, and to questions on the existence andwhereabouts of relatives. Additional objectiveswere to test out the proposed procedures andquestionnaires, to get rough measures of durd-tions of interviews, and to serve as a lbasis for apretest of the evaluation survey. The pretest 01’the evaluation survey, which comprised an evalu-ation of’ reports from the basic survey pretestand an evaluation of reporting on events regis-tered in vital records, is described in chapter 3.

The topics covered in the pretest were thoseoriginally planned for the basic survey-birthsfor Jews and non-Jews and deaths for ,Jews only.The multiplicity counting rules, outlined above,were limited to the female line for births; i.e.,including reports on births to daughters and sis-

6

ters of females in the household as well as birthsto women in the household. This was done be-cause the frequency of births is sufficiently highthat even limited multiplicity counting ruleswere expected to provide high yields, and it wasnot possible to burden enumerators with toomany reported events. For deaths, however (andsubsequently for mai-riages), a wider multiplicitycounting rule was required, both in order to at-tain sufficient yields for analysis and to increasethe probability of there being an eligible surviv-ing relative. Thus reports were requested ondeaths of spouses, parents, sons, daughters,brothers, and sisters of ill persons in the house-hold irrespective of sex.

For both types of events multiplicity report-ing was hierarchical within the interviewedhousehold. This means that events reported by arelative of higher order (e.g., by the mother ofthe event person) were not to be reported againby a relative of lower order residing in the samehousehold (e.g., by a sister of the event person).The order adopted for births was: to women inthe household, to daughters living elsewhere,and to sisters living elsewhere (if not previouslyreported as daughters). That for deaths was: tospouse, to parents, to children, and to siblings,again each category excluding deaths already re-ported. For the pretest an attempt was made toinclude reports also on births which occurred towomen belonging to the household during thereference period (the year 1972) who had mean-while died.

The questionnaire administered in the pre-test comprised three basic parts. The first wasthe regtdar LFS questionnaire (see Central Bu-reau of Statistics (1975)), which included a list-ing of all persons in the household, with basicdemographic characteristics, and a detailed laborforce questionnaire for each person aged 14 andover. The second was the multiplicity screening

questionnaire booklet, on which was recorded,separately for births and for deaths, whetherevents occurred to specified relatives during thereference year. The last part comprised event re-ports, which were used only in the case that anevent was reported on the screening question-naire. A separate sheet was completed for eachevent reported. It contained demographic dataon the event (month of occurrence, sex of the

child born or of the deceased, age of mother orof deceased, and residence of deceased at thetime of the event). It also incIuded a list of alIrelatives who could have reported the event, to-gether with a complete address for each.

Originally, two different procedures wereconsidered for the completion of the event re-ports. According to the first alternative, the e-numerator was to complete all three parts of thequestionnaire-including the event report, in thecase that an event was reported-during a singlevisit, with czdlbacks as necessary. Since therewere apprehensions that househoId membersmight not be able, at the time of the enumera-tor’s visit, to report fulIy and accurately on thedetails of events which occurred to relatives out-side the household, and especially on addressesof reIatives, an alternative of mail-back for eventreports was considered. Under this procedurethe enumerator was to complete the LFS ques-tionnaire and the screening questionnaire on hisinitial visit. If an event was reported, an eventreport was to be Ieft with the household forself-completion and mailed back by the house-hold respondent. This was to be complementedby enumerator followup for nonresponse. Aspkmned, the first alternative was tested first(with mail-back, however, for households notcontacted after three visits). Since this altern-ative provided relatively good results (whereasvery few questionnaires were received by maiIfrom households not contacted), it was decidedto use this procedure without testing the mail-back procedure.

For technicaI reasons, it was not feasible tocarry out the pretest of the basic survey as anaddition to the regular LFS. Furthermore, a lessexpensive (i.e., highly clustered) sample was re-quired with only a Iimited number of enumera-tors. Thus the pretest was conducted on the ba-sis of a judgment sample of six 1972 Census of

Population and Housing enumeration dis&-icts(ED’s) for Jews and six for non-Jews. For theJewish population twcl ED’s were selected fromJerusalem (one from a high and one from a lowsocioeconomic area), one each from Haifa andfrom Tel Aviv, and two from an urban localityin the Tel Aviv conurbation. For the non-Jewishpopulation two ED’s were selected from each ofthree localities-Nazareth, a Moslem village, and

. a Druze village. For each ED about 10 addressesof households with at least 3 persons each whichdid not participate in the census 20 percent sam-ple stage were selected.

The pretest of the basic survey was carriedout in May and June 1973 by eight enumerators.They had had previous experience in the LFSand were given a supplementary 4-hour trainingperiod, including practice interviews, afterstudying written instructions and completinghome exercises.

In general, the pretest indicated that the de-sign proposed was feasible and could be usedwith only minor modifications. Completionrates were high—questionnaires were completedfor 60 out of 66 Jewish addresses and for 60 outof 61 non-Jewish addresses. Noncompletion wasprimarily due to not locating addresses or house-holds (four cases). It should be noted that theaddresses had been recorded about 1 year earlierin the 1972 census. Only one absence (aftertkree callbacks) and one refusal were recordedfor Jewish households.

The average duration of interview was 8.0minutes for Jewish households and 11.4 minutesfor nonjewish households. Households withevents required, on the average, only about 5minutes more interviewing time than householdswithout events (11.4 versus 5.4 minutes for Jewsand 14.1 versus 9.6 minutes for non-Jews). In-terview time varied considerably by locality, pri-marily due to variation in household size. Al-though exact numerical comparisons were notpossible, it seems that interview duration forhouseholds without events was hardly greaterthan that for the regular LFS interview.

Finally, the pretest clearly indicated that re-spondents were, in general, both willing and able

. to report on events occurring to relatives. Thiswas in contrast to strong prior apprehensions onthis point on the part of the field staff. How-

ever, some difficulties were encountered in re-porting on deaths and, in a few cases where noadult female was available for interview, onbirths. Moreover, respondents found difficultyin reporting exact street addresses of relatives.Even though they often knew the location ex-actly and how to reach it, they could not pro-vide the exact name or numerical identificationof the address. In fact, less than half the ad-

dresses reported were complete. This lpoint wasfurther borne out in the evaluation survey pre-test, in which about 20 percent of complete ad-dresses reported by relatives were not located.

The major design modifications resultingfrom the pretest related to the forma,t and theformulation of the questionnaire, to the cover-age of events to be reported, and to the report-ing of addresses of relatives. The questionnaireformat was changed to that of a booklet inwhich both the screening questionnaire and theeve n t reports were bound together. Slightchanges in the wording of the questicms and inthe questionnaire flow were aIso made on thebasis of the pretest experience. Event-reportingcoverage was limited to reports of events tomembers of the household at the time of theinterview and to their relatives, irrespective oftheir situation at the time the event occurred.Thus, for instance, births which occurred abroadin 1973 (the reference period for the final sur-vey) to new immigrants were included, whilebirths in Israel to women who had emigratedbefore the interview were not included. Finally,attempts were made to increase the usefulness ofaddresses reported for relatives by adding hus-band’s first name for married women, by record-ing telephone numbers (at home anc[ place ofwork), and by placing more emphasis on-addressreporting in enumerator training. The final de-sign, incorporating these and other modifica-tions, is described more fully in the. followingsection.

2.3 The Final Design

Two major changes in the final survey designwere made between the termination of the pre-test in ‘june 1973 and the execution of the basicsurvey in the first quarter of 1974. The first wasa change in topics–the replacement of cieaths by

marriages. This was primarily a result of theYom Kippur war in October 1973 and its after-math. The subject of death was considered toosensitive for inclusion as a topic so shortly afterthe war with its tragic losses. The topic of mar-riages was chosen to repIace it because multiplic-ity counting rules similar to those for birthscould be used without further pretesting, and

the event has sufficient frequency to be usefulfor analysis by the sampIe size envisaged.

The second major change concerned the sur-vey population—the limitation of the basic sur-vey in Israel to the Jewish population only. Thiswas done primarily out of budgetary considera-tions. Differences between the Jewish and thenon-Jewish population would have necessitatedseparate analyses for the two subpopulationsand therefore a large relative sample size fornon-Jews. In addition, the regular LFS sampleincludes only about 800 non-Jewish householdsper quarter. While this is sufficient to estimateprevalent labor force characteristics, it is smaIlfor estimation of relatively rare events such asbirth and deaths and certainly would not pro-vide a sufficient number of cases for evaluation.The use of an additional quarter to increase thenumber of events would result in too much vari-ation of the time period between the interviewand the reference year (19 73). Furthermore, theparallel study in the Administered Areas wouldprovide some insight on the applicability of mul-tiplicity methods to a non-,Jewish population, al-beit less used to contact with modern conditionsthan the non-Jewish population of Israel.

The final design thus followed the generaloutline in section 2.1 with the changes and themodifications resulting from the pretest conclu-sions described in the previous section. The pop-ulation of potential event-reporters (members ofthe event household and their specified rela~ives)was finally defined as the regular resident de jweJewish population (strictly speaking, membersof households whose head was Jewish) at thetime of the survey (February-March 1974), ex-cluding residents of kibbutzim and institutions,according to the standard definitions of theLFS.

The events on which reports were requestedwere all births and marriages which occurred tomembers of the above population in 1973. Re-ports were requested on births even if the babyhad since died and on maniages even if they hadsince terminated. Persons currently belonging tothe survey population were included even if atthe time of the event they did not belong to thesurvey population (e.g., if they then residedabroad or in a kibbutz). However, events whichoccurred to persons who were in the survey pop-

ulation at the time of the event but not at thetime of the interview (e.g., births to women whohad moved abroad or to a kibbutz since thebirth) were not included. The same rules ap-plied, obviously, to the network of relatives onwhom reports were received. Although this doesnot conform exactly to the definition used forregular demographic estimates of vital events,the differences were thought to be small andcounterbakmcing. While exact proof of this wasnot available from the survey, the results didgive an indication that these differences were in-deed very small.

The reporting relatives were finally definedas mothers and sisters of women who gave birthand as parents and sibIings of persons who mar-ried (in that hierarchical order). This facilitatedanalysis of the survey by the following a.lterna-tive counting rules.

1.

2.

3.

Conventional rule-based only on reportsof events to members of interviewedhousehold (each with multiplicity of1).

Restricted multiplicity rule–based on re-ports of events to members of the inter-viewed household, to their children (formarriages), or to daughters of women in

the household (for births). The multiplic-ity for this rule is the number of dif-

ferent households in which the eventperson and the person’s parents (ormother) reside.

Full multiplicity rule-based on reportsof events in the interviewed household,marriages occurring to children and sib-Iings, or births occurring to daughtersand sisters of persons in the household.The multiplicity for this rule is the totzdnumber of different households recordedin the multiplicity network.

Tt shouh-1 be noted that the definitions of

relationship included only biological relationswith at least one common parent (half-siblings)and legal adoptions, but not stepchildren orstep-sibIings (with no common parent). Further-

9

more, itevent is

should be noted that for marriages thedefined as “a person marry ing, ” so a

single marriage is counted as two events occur-ring to two partners (with different multiplicitynetworks).

The sample design of the basic survey, asmentioned above, was on the basis of a selectedpart of the regular LFS in the first quarter of1974. The LFS is a self-representing probabilitysample selected as a systematic random samplefrom apartment tax lists in all large urban locali-ties (the certainty stratum) and in a stratifiedP.P.S. sample of smaller localities. The basic fea-tures of the sample design can be found in Cen-tral Bureau of Statistics (1975), although itshould be noted that the sample size has mean-while been doubled, with the result that the cer-tainty stratum’s proportion has been increasedconsiderably. The sample is subdivided into fourpanels and into 13 enumeration weeks, each partbeing a self-representing probability sample. Forthe multiplicity study sample eight enumerationweeks (the 5th through the 12th) for three outof the four panels were used, providing a grosssample size of about 5,000 households. This wasabout the same size as originally planned, via all13 weeks and all four panels of a quarter, whenthe LFS sample was half its present size. Sinceinstitutions and kibbutzim, where clustering ef-fects may be high, were not included, the samplecould be regarded for all practical purposes as asimple random sample without replacement.

The LFS questionnaire comprises a listing ofall household members and their basic demo-graphic characteristics, which include year andmonth of birth, year of present marriage, and ifthe present marriage is the first one (appendixIII). This listing was used as the basis for screen-ing in the multiplicity survey, and the data oncharacteristics was an aid in this procedure. Thesupplementary multiplicity section included,separately for births and for marriages, a screen-ing questionnaire and a set of event reports, allbound in a single booklet (appendix II). The reg-ular LFS enumerators interviewed for the multi-plicity study as an additional topic after comple-

tion of the regular LFS questionnaire. The re-spondent for the multiplicity study, as for theLFS, could be any available household memberaged 14 and over. This was usually the house-

hold head or his wife. More stringent respondentrules could be employed, but in this case the useof a rule different from that of the LFS wouldhave increased costs considerably and might noteven have been possible within the existing inter-viewing arrangements.

In the screening questionnaire all ever-mar-ried women under 50 in the household werelisted (Q. 1) and it was ascertained if any ofthem gave birth during the year 1973 (Q. 2).Similarly, all household members who got mar-ried in 1973 were identified from the LFS listing(Q. 13). Then, all ever-married women aged 34and over were listed (Q. 4) and asked if they hadmarried daughters living elsewhere in Israel (Q.5) and, if so, if any of them gave birth in 1973(Q. 6). Similarly, aJl ever-married persons aged34 and over were listed (Q. 14) and asked if theyhad sons or daughters living elsewhere in Israelwho married in 1973 (Q. 15). Finally, all womenand girls in the household whose mother did notreside there were listed (Q. 8) and asked if theyhad sisters living elsewhere in Israel (Q. 9). Foreach sister listed they were asked if she gavebirth in 1973 (Q. 10). All persons who did nothave a parent residing in the household werelisted (Q. 17) and asked if they had siblings liv-ing elsewhere in Israel who married in 1973 (Q.18). The screening procedure is shown (forbirths) in diagrammatic flow form in figure 1.

Only if an event occurred to a member ofthe survey household or to a specified relativewas it necessary to complete an event report.This included basic demographic data on theevent (sex of event person, month of occur-rence, age of mother or of married person, andlocality of wedding) and a listing of the multi-plicity network (the woman who gave birth andher mother and sisters, or the person who gotmarried and his or her parents and siblings). Theaddress of each listed relative was recorded, andfrom this the total number of separate house-holds in which the members of the networklived (i.e., the multiplicity) was determined.Supplementary information (telephone numbers “at home and at work, first name of husband,

etc.) was requested, both in order to determineunequivocally which persons lived in the samehousehold and, especially, for location of rela-tives in the evaluation survey.

10

r—

1“

I

I

I

I

I

I

t––

I

I

I

I

I

t

.

I IQ. 1– Women eligible to be

crueried on births in

I..basic survey household I

+ Yes

—— —I form’ I

Q. 4- Women eligible to be queriedon births to deuahters

*

I Q. 7 I—— .1 Child listing

>

form

Q. 8– Women eligible to be queriedon births to sisters livingelsewhere

——. —.— ——— ——

Q. 9- Existence of sisters1

I

Iabour

Force

Suwey

Household

Listing

Form

Figure 1. Questionnaire flow diagram–births.

11

2.4 Execution of the Basic Survey

Although the execution of the basic surveyfollowed, fundamentally, the planned design de-scribed above, field operations encountered se-vere difficulties due to the special conditions ofthe first quarter of 1974. This was primarily dueto the fact that, although the Yom Kippur warhad officially terminated 2 months earlier, therewas still a war of attrition and a high degree ofmobilization of reserves. This severely affectedthe availability of experienced field staff at alllevels, both directly, for men, and indirectly, formarried women who found it difficult to workbecause of family problems. The problem wasaggravated by the fact that, as scheduled, theLFS sample was doubled in the first quarter of1974, while the training and recruitment of thenecessary additional field staff, scheduled forthe last quarter of 1973, was held up due to thewar. All this resulted in a high turnover rate offield staff and a high proportion of new andinexperienced enumerators. In addition, the highrate of mobilization caused dislocations in thesurvey population since many families whosehead was mobilized left their regular residencetemporarily to stay with parents or other rela-tives.

The fieldwork was carried out by about 30LFS enumerators, who received a supplementaryfull day’s training session for the multiplicitystudy, with practice interviews, after reading thespecial enumerators’ manual and completing ex-ercises at home. Supervision was carried out inthe Bureau’s three regional field offices by theregular LFS field supervisors, who also checkedand hand edited the multiplicity questionnaires.Due to the field-staffing problems mentionedabove, the field timetable by enumeration weekscould not Jalways be strictly adhered to andsome interviews had to be postponed to the firstpart of the second quarter. In particular, the in-terviews of some 60 households in Jerusalemwere postponed and carried out together withthe evaluation survey. Overall, enumeration forthe basic survey was carried out from January27 through April 10.

The primary factor affecting completionrates in the LFS was difficult field conditions.The following completion rate data, which show

this, are synthesized fromsources, with some minor

reports from variousadjustments and im-

putations:

(1) Total addresses sent to field (Jewishlocalities, excluding kibbutzim andinstitutions) : . . . . . . . . . . . . . . . . . . . . . 5,185

(2) Total LFS questionnaires received: . . . . . 4,197(3) Less non~ewish households in Jewish

localities: . . . . . . . . . . . . . . . . . . . . . . . 92(4) Jewish households for multiplicity

survey -(2) - (3): . . . . . . . . . . . . . . . . . . 4,105(5) Multiplicity questionnaires completed: . . 4,084(6) LFS completion rate-(2)/(1): . . . . . . . . . 0.81(7) Multiplicity survey completion rate–

(5)/( 4): . . . . . . . . . . . . . . . . . . . . . . . . . 0.99(8) Es&nX~;~ overall completion rate–

. . . . . . . . . . . . . . . . . . . . . . . . . 0.81

It should be noted that the LFS completionrate of 0.81 (line 6) expresses losses due to both“zero cases” (vacant dwelling units, commercialunits, etc.) and to genuine nonresponse (ad-dresses not located, residents not at home, re-fusals, etc.). Unfortunately, the work pressureon the field staff did not enable a detailed reporton the breakdown of noncompletion by cause.However, prewar LFS overall completion ratesfor 1972 through 1973, which are nc~t strictlycomparable to those for the specific multiplicitystudy population, were about 0.86-0.88, andnoncompletion was approximately equally di-vided between zero cases and nonresponse. Sincethe same sample frameworks were used, the in-crease in noncompletion must be assigned pri-marily to increased nonresponse. The specificmultiplicity survey completion rate (line 7) waspractically maximal and, as anticipated by theresults of the pretest, it was possible to completem ult ip Iicity questionnaires for virtually allhouseholds in which the LFS interview could becompleted. Overall completion rates varied con-siderably between field districts, from 0.68 inthe Jerusalem district to 0.86 in the Haifa dis-trict. This was primarily a function of differen-tial difficulties in recruitment of experiencedfield staff and in supervision in different areas.

The completed questionnaires were handedited in the district field offices (wiith seriousedit-failures returned to the fieId) and finally atthe center in Jerusalem. The most frequent errorfound was noncompletion of questiorls 12 and20 (totals of events reported). This item could

12

be completed in the office, but noncompletionlimited the possibility of consistency checks.

Negative answers to questions 2, 6, 10, 15,and 18 (reports on events) were also occasion-ally omitted. A few imputations were made atthe central editing stage, mostly on the basis ofcomparisons with the household listing form, forevents which should have been reported by theinterviewed household. In addition, five reportson events to persons not in the survey popula-tion at the time of interview were withdrawnfrom processing at the central editing stage, inaccordance with the definition of the event-re-porting population in section 2.3. On the otherhand, of the 1,553 events finally processed (887births and 666 marriages), 11 events (all mar-riages) were to persons in the survey populationat the time of interview but not at the time ofthe event.

Questionnaireswere coded directlv

of households with eventsfor punching, one record for.

each event, with common fields for data onhouseholds with more than one event. The rec-ord format is given in appendix IV. The punchedrecords underwent automated logicaI checks andediting and in some cases missing data and edit-failures were corrected or imputed. Basic com-putations were made by computer to determine,separately for births and for marriages, multi-plicity vahes for the three counting rules, thetotal weighted contribution of each householdaccording to the three rules, and the co-residencecodes. The tabulations on which the results pre-sented in chapter 4 are based were then com-pleted, primarily by the standard SPSS program-ing package.

13

3.1

Chapter 3. THE EVALUATION SURVEY: DESIGN AND EXECUTION

Purpose, General Design,and Pretest

A major component in the study of the use-fulness of the multiplicity technique is knowl-edge of the efficiency of multiplicity estimatesfor different counting rules as compared withconventional estimates. The efficiency dependson components of sampling error, response vari-ance, and response. bias. Basic parameters neces-sary for this purpose are measures of the extentof overreporting and underreporting of eventswhich occurred during the reference year andthe extent to which households report correctlyon demographic details of the basic survey.These can be determined by means of an inde-pendent set of reports on the same events.

Thus it was decided at the earliest planningstage to include a substantial evaluation surveyas an integral part of the multiplicity study. Thepurposes of the evaluation study were to provideestimates for rates of overreporting and under-reporting of events by different degrees of rela-tives, to estimate the reliability of multiplicityreporting, and to provide some measures of re-sponse errors in reports on the demographiccharacteristics of the events reported. The pa-rameters thus measured were to be synthesized,by means of a theoretical model explained inbrief in appendix VI. This makes possible com-parison of the efficiency of estimates for differ-ent counting rules as measured by the totalmean square error and its components—samplingvariance, response variance, and response bias.

Initially, two different evaluation techniqueswere considered. The first, which itself consistsof two alternatives, relied on the virtu~ com-pleteness in recording of vital events by the reg-ister in Israel and was to be based on a compari-son between replies of respondents in a multi-plicity survey and the data in the vital records.The evaluation could be done by starting outwith reports on events from the basic survey andsearching the appropriate vital record for com-parison. This, however, would obviously not al-low for any measure of underreporting in thebasic survey. Furthermore, the efficiency of a

measure of overreporting would strongly dependon the possibility of complete matching, whichis extremely difficult due to changes of addressand incomplete address repo~ing. This alterna-tive was therefore dropped from considerationbefore the pretest stage. The other alternativewas to use vital records as a source of events andthen to send interviewers with the multiplicityquestionnaire to event households in order toascertain if the event was correctly reported.While this approach to register evaluation couldonIy measure underreporting by members ofevent households (together with a measure ofresponse errors in their reports on demographiccharacteristics), it was tried out in the pretest. Itwas not included in the fired design for reasonsspecified below.

The second evaluation technique considered,and finally adopted, was a field evaluation basedon reports of events in the basic survey. For asample of events reported in the basic survey,interviews in households of relatives (as reportedin the basic survey) were to be carried out bythe same procedure and with the same question-naire as used in the basic survey. This was in-tended to be, as far as possible, an independentreplication rather than an improved procedure.A comparison of responses on the two question-naires thus obtained would enable measures ofdifferences in reporting of the same event inmultiplicity reporting and in demographic dataon the event. In order to determine which reportwas correct in the case of a difference, one ofthe households was to be the event household sothat its LFS household listing could be used foradjudication. It should be mentioned at thisstiage that reports on events with multiplicityone (i.e., no other household of relatives wasqualified to report the event) could not be evalu-ated by this alternative.

To test the feasibility of these alternativeevaluating techniques and their integration withthe basic survey procedure, a pretest was carriedout in conjunction with the basic survey pretestdescribed in section 2.2.

For the register evaluation, addresses ofevents from live-birth notifications amd from

14

death notifications were selected for specificmonths in 1972. There were 24 addresses of

Jewish households and 35 addresses of non-jew.Ish households. For each event, the address wasgiven to the enumerator and a photocopy of thenotification retained for comparison. Since theregister evaluation pretest was carried out to-gether with the basic survey pretest during Mayand June 1973, the interviewers could not knowwhich addresses were for the basic survey andwhich for the evaluation survey.

Interviews were completed for 19 out of the* 24 Jewish addresses (1 household not located

and 4 absences or refusals) and for 26 out of the35 non-Jewish addresses (6 households not lo-cated and 3 absences). All the non-Jewish house-holds and all but three of the Jewish householdsreported the event. A comparison of demo-graphic characteristics indicated high discrep-ancies for age (of mother or of deceased), lowerdiscrepancies for month of event, and no dis-crepancies for sex.

For the field evaluation pretest, addresses ofrelatives were selected from network listings forevent reports of the basic survey households andof the register evaluation. Interviewing was car-ried out by the same enumerators and accordingto the same procedures as those of the basicsurvey and register evaluation pretests, as a con-tinuing integrated process, in order to minimizethe knowledge of the interviewers that a reporton an event was expected. Out of 52 Jewishhouseholds, only 39 were interviewed (10 notlocated and 3 absences or refusals), and out of50 non-Jewish households, 44 were interviewed(2 not located and 4 other reasons for noncom-pletion). The low completion rate for Jews wasprimarily a result of poor address reporting, evenfor addresses judged to be complete at the fieldevaluation sample selection stage. On the basisof certain decision rules (to be specified fully inthe following section) cases of overreporting andunderreporting could be identified, and compari-sons of demographic data on the events could bemade. The numerical results are not significantbecause of the small size, but the field evalua-tion pretest did indicate that this approachcould serve its purpose if the problem of incom-plete address reporting (as described in section2.2) could be reduced.

Thus, results of the evaluation pretest indi-cated that both the fieId evaluation and the reg-ister evaluation techniques were feasible formeasuring errors. However, the limitations ofthe register evaluation-its lack of direct connec-tion with the basic survey and the possibility ofdiscovering only underreporting by event house-holds–made it imperative to consider it only asa possible supplement to a field evaluation. Pri-marily for budgetary considerations, the registerevaluation was finally dropped and all effortconcentrated on the field evaluation.

3.2 Final Design of theEvaluation Survey

The final design of the ewduation survey wasbasically that tried out in the field evaluationpretest. It was based on a sample of addresses ofrelatives reported in the basic survey. Althoughthe interviewing was to be carried out by thesame LFS enumerators empIoyed for the basicsurvey, this design involved a widely scatteredsarnpIe not derived directly from the LFS sam-ple, as was the basic survey. Taking into consid-eration the budgetary Imitations and especiallyhigh costs that this design involved, a maximalsample size of ordy about 500 households couldbe considered. Originally it was pkmned to usehalf for the evaluation of reports on births andhaIf for the evaluation of reports on marriages.However, practical Imitations, primarily due tothe small number of mmriages with usable ad-dresses, made it necessary to change the distribu-tion to about 200 marriages and about 300births.

The design ensured that for each pair of re-ports on the event being compared–one fromthe basic survey and one from the evaluationsurvey-one of the reports would be for theevent household and one for the household of arelative. This was done to enable use of the in-formation in the LFS household listing of theevent household for reconciliation in the case ofa discrepancy between the two reports. In orderto represent both degrees of relationship (par-ents and siblings) for each type of event, it wasnecessary to select four subsamples of addressesfor each type of event, as follows:

15

Report in basicsurvey by:

1. Event household

2. Event household

3. Mother’s mother/parentof married person

4. Mother’s sister/sibling ofmarried person

Evaluation by

household of:

Mother’s mother/parentof married person

Mother’s sister/sibling ofmarried person

Event household

Event household

The intention was to sample approximatelyequal numbers of addresses from each of theabove groups. This had to be modified some-

what in the execution stage (see table A), be-cause of differential frequencies of addressesavailable and the necessity to complete the eval-uation survey within a short time after the endof the basic survey.

Addresses which were incomplete or in out-lying localities were excluded. With these limita-tions, sampling was basically on a quota basis inthe sequence in which questionnaires reachedthe central office. However, since LFS enumera-tion weeks are self-representing random samples,the evaluation samples can be considered as ap-proximations of simple random samples within .each group. It should be pointed out again thatonly events reported with multiplicity of morethan one were represented in the evaluation sur-vey and that no attempt was made to evaluate

Table A. Execution of the evaluation survey

I IAddressessent to

Addresses Quest ion-sent to naires

Event evaluated, where event initially surveydivision field completedreported, and where event evaluated

(1) (2) (3)

Births

Total------------------------------

+Reported in event household; evaluated inhousehold of mother’s mother------------

Reported in event household; evaluated inhousehold of mother’s sister------------Reported in household of mother’s mother;evaluated in event household ------------

Reported in household of mother’s sister;evaluated in event household ------------

Marriages

Total------------------------------

Reported in event household; evaluated inparent’s household ----------------------

Reported in event household; evaluated insibling’s household---------------------

Reported in parent’s household; evaluatedin event household ----------------------

Reported in sibling’s household;evaluated in event household ------------

95

65

96

68

208

57

46

48

57

71

55

93

64

182

45

34

48

55

60

45

78

51

140

42

27

34

37

:ompletionrate,

(3)+(2)

(1,)

0.83

0.85

0.82

0.84

0.80

0.93

0.79

0.71

,0.6,7

16

nonreports or reports with multiplicity y one.Some evaluation of these events could have beendone based on a comparison with the LFS list-ing. However, under certain assumptions (spec-ified in appendix VI) the possibility of estimat-ing the error components without bias is notaffected by excluding these events, so they werenot evaluated.

The addresses seIected were to be tran-scribed on LFS listing forms and given to LFSenumerators who had participated in the basicsurvey. They were to complete the LFS listingform and the multiplicity questionnaires (butnot the LFS questionnaires) according to thesame procedure as in the basic survey.

The evaluation survey questionnaires were tobe matched with the relevant basic survey ques-tionnaires and the two reports in each pair-onefrom an event household and one from thehousehold of a relative–compared. Where nec-essary, the following criteria for assessing overre-porting and underreporting were used in con-junction with the LFS listing form:

1.

2.

Events reported in both the basic surveyand ewduation survey were to be con-sidered correctly reported true events.The status of events not reported in theevaluation survey was to b; adjudicatedon the basis of the LFS listing form ofthe event household (mother’s house-hold for births; married person’s house-hold for marriages).(a)

(b)

If the eve~t ‘was not listed in theevent household or was Iisted for adifferent year, it was to be treated asa nonevent, overreported by the ba-sic survey household.If the event was listed in the eventhousehoId as occurring in the correctyear (even if marked as deleted at alater visit, due to death or divorce),it was to be treated as a true event,underreported by the evaluation sur-vey household.

This adjudication rule implicitly assumes in-dependence between event reporting and theLFS listing, which is not always obtained inpractice. The possibility of dependence and its

effects on the evaluation results are discussed inappendix VU.

Comparisons of the two reports on the mul-tiplicity network and on demographic data forthe same event couId be made at the same time.

On the basis of a theoretical model, undercertain simplifying assumptions, basic evaluationparameters (such as overreporting and underre- -porting rates for ,different relatives and multi-plicity reporting response variance) could be es-timated. The parameters could be synthesized–together with measures from the basic survey—to provide, under the assumptions of the model,estimates of the total mean square error and itscomponents (e.g., sampling variance, responsevariance, and reporting bias) for each of thethree counting rules considered. FuU details on.the model used and the methods of estimationare given in Nathan (1976), and a nontechnicaldescription of the modeI is given in appendix VI.

3.3 Execution of theEvaluation Survey

Selection of the sample events for the evalua-tion survey began during the first week of thebasic survey fieldwork, at the end of January1974. Multiplicity questionnaires from the basicsurvey arrived at the center in Jerusalem to-gether with their respective LFS questionnaires.Both questionnaires had already undergone aninitial edit in one of the three regional fieIdoffices. The multiplicity questionnaires werescanned, and aI1 events reported were listed on aspecial form intended to serve as the frameworkfor the selection of events for the evaluation sur-vey sample. Separate forms were used for birthsand marriages. In addition to the questionnaireserial number, the form provided for the listingof the number of usable addresses (compIete andnot in outIying Localities) for each type of report(i.e., report by the event household, by thehousehold of a parent, or by the househoId of asibling).

Since most LFS questionnaires reached thecenter within 6 or 8 days of the date of inter-view, it was possible to obtain almost a full setof the completed questionnaires taken during agiven enumeration week within the week there-

17

after. The listingwas followed for 5the survey was in

procedure described aboveof the 8 weeks during whichthe field. The listed events

were then scanned and the evaluation surveysample selected from them according to the de-sign set out in section 3.2, separately for birthsand for marriages for each of the four reportgroups. The actual distribution of addresses bytype of event and report group is shown in col-umn (1) of table A.

The addresses selected from the special list-ing form were sent to the Survey Division of theBureau for processing. There were pressures imp-osed by the need for speed, the general condi-tions still pertaining with respect to field staff(section 2.4), and the requirement that the eva.l-uation survey be carried out by the regular LFSenumerators together with their regular LFSwork. Consequently, only 87 percent of the se-lected addresses could be sent to the field. Forthe 465 addresses finally sent to the field, 374completed questionnaires were received (80 per-cent), slightly more for births, and slightly lessfor marriages (see columns (3) and (4) of tableA).

As before, the relatively low completionrates are primarily attributable to the Yom Kip-pur war. Evaluation survey interviewing was car-ried out during April and May 1974, when theeffects of the war and its aftermath were stillfelt. Even though many men had been demobi-lized by that time, some families still had notreturned to their own homes or had left the ad-dresses at which they had been listed in the basicsurvey. Although in many instances the inter-viewers were able to locate them, this movementtended to lower the efficiency of the interview-ing process. In addition, the turnover’ of enumer-ators was still acute, particularly in Jerusalem,where none of the evaluation survey enumera-tors had had experience with the basic survey.

The design of the evaluation survey imposedan extra burden on the enumerators. In the basicsurvey, the survey household was reached via theregular sampling procedure of the LFS, in whichenumeration assignments are clustered geograph-ically. However, in the evaluation survey, the e-numerator was given a set of scattered addressesin addition to his regular LFS assignment. Whilethis problem was recognized by the supervisors,

shortage of personnel limitedcould adiust the workloads.

the amount they

From table A it is seen that for births thereis little variation in completion rates for differ-ent types of reports (all close to the average levelof 0.83). However, for marriages considerablevariation exists, although it must be pointed outthat small numbers are involved. Completionrates were lowest (0.67-0.7 1) when the initialreport was from the household of a parent orsibling of a newly married person and the evalua-tion was in the household of the newly marriedcouple. This is probably because newly marriedcouples tend to be highly mobile and their ad-dresses may not yet be well known to relatives.In addition, newly married men are often youngand were likely to have been kept in the army along time, so their wives might still have beenliving with relatives at the time of the evaluationsurvey. On the other hand, the higher comple-tion rate when evaluation was done in the house-hold of parents of newly married perscms (0.93),is probably due to the fact that members of theolder generation tend to be less mobile and theiraddresses are perhaps better known than the ad-dresses of siblings are. It should, however, bekept in mind with respect to the completionrates that the household members listed in thebasic survey as reported persons (i.e., personsabout whom questions were asked on events tothemselves or to their relatives) were not neces-sarily the actual respondents (i.e., persons whoreported on the addresses of network members).This is so since the usual LFS respondent ruleswere followed, by which the respondent couldbe any household member aged 14 and over.

In table B frequencies are presented of thereasons given by enumerators for noncomple-tion. For both births and marriages, almost two-thirds of the households for which question-naires were not filled out were not located (lines1 and 2 of table B). More than half of,the totalnoncompletion was due to inability to find anaddress because of inadequate information fur-nished in basic survey households. Temporaryabsence was the reason for about a quarter ofthe total noncompletion, indicating the highmobility of households already described.

As indicated earlier, the evaluation surveywas intended to be an independent replication

18

Table B. Reasons for noncompletion in theevaluation survey, by type of event

Reason fornoncompletion

All reasons-

1. Address notfound-------

2. Householdhas moved---

3. Absentteni-porarily-

4. Refused------5. Interviewer

error-------

IITotal Births Mar-riages

tempted to limitthiswork, itis not clear to

effect during the field-what extent estimates of

response en-ors were affected.After preliminary editing in the field offices,

the evaluation survev questionnaires were re-ll——t———

. .turned to the center. where the four basic docu-

91 49 I 42,

49 30

8 3

23 119 4

2 1

19,

5

125

1

of thebasic survey inthe sense that independentreports on the same event were to be obtained.However, some of the enumerators may havebeen aware that an event had been reported inthe basic survey for the network to which anevaluation household belonged.of them assumed that an eventcurred, and probed or settledwith that in mind. Although

PossibIy, somehad in fact oc-doubtful casessupervisors at-

ments foreachevent—two multiplicityquestion-nairesand two LFS household listing forms—were brought together. Data from these sourceswere transferred to a single Evaluation SurveyProcessing Form (see appendix V) to enable easycomparison of the data from the two surveys.On the basis of the decision rules, detailed in theprevious section, the event reporting status wascoded: correctly reported in both surveys; oroverreported or underreported in one of the sur-veys (and correctly reported in the other). Itshould be pointed out that most of the overre-porting discovered was due to reports on eventswhich occurred outside the reference year(1973). Further editing and a few correctionsand imputations were carried out at this stage.Final processing of the evaluation survey wasmanual, by means of the Evaluation SurveyProcessing Form, on the basis of the model andmethods described by Nathan (1976) and, innontechnical form, in appendix VI.

—ooo —

19

Chapter 4. RESULTS

4.1 Basic Results on Yields and Network Sizes household) and of weighted yields (sum of eventreports weighted by the reciprocal of their mul-

The basic survey results are presented in tiplicity) for each of the three counting rulestables C and D. Table C gives the distributions of considered. The counting rules are: the conven-raw yields (number of events reported per tional rule (events occurring to members of in-

Table C. Raw yields (number of events per household) and weighted yields (number ofevent reports we%ghted by reciprocal of multiplicity), by type of event and countingrule

Type of yield

Raw yield

Total ------------

0---------------- ------1--------------------------------------------:-----------------------4----------------------

Weighted yield

Total ------------

0---------------- ------0.01-0.19--------------0.20-0.29--------------0.30-0.39--------------0.40-0.49--------------

0.50-0.59--------------0.60-0.69---------------0.70-0.79--------------0.80-0.89--------------0.90-0.99--------------

1.00-1.49--------------1.50-1.99--------------2.00-2.99--------------3.00+------------------

Conven-tional1

(1)

4,084

3,7583242

4,084

3,758

324

2

Restrictedmulti-plicity2

(2)

Births

4,084

3,5215;;

2

4,084

3,521

459

9383

Counting rule

Fullmulti-

plicity

(3)

4,084

3,3186579712

4,084

3,318

1%1;;

2502981

48

;

:onven-:ionall

(1)

4,084

3,963

11:

i

4,084

3,963

4

11:1

Restrictednmlt i-

plicityz

(2)

Marriages

4,084

3,785177119

:

4,084

3,785

1;

159

;

;;

111

Fullmulti-

plicitys

(3)

4,084

3,581364117202

4,084

3,581

1::1006

1071213224

39145

lEvents occurring to members of interviewedhouseholds only.zEvents ~ccurring to members of interviewedhouseholds and to their offsPring*3Events occurring to members of interviewedhouseholds, to their offspring, and to

their siblings.

20

terviewed households only, each event having aweight of 1); the restricted multiplicity rule(events occurring to members of interviewedhouseholds and to their offspring); and the fullmukiplicit y rule (events occurring to membersof interviewed households, to their offspring,and to their siblings). Multiple reports of birthsby event households are seen to be negligible,whereas, of course, practically all reports byevent households on persons marrying are dou-ble, The increase in report yields for widercounting rules is primarily in single reports, bothfor births and for marriages. Besides the obviousconcentration at zero, weighted yields are con-

centrated at wdues of 1.0 and 0.5 for the re-stricted muRiplicity rule, corresponding to eventpersons with no parent (or mother) outside thehousehold and to event persons with parents (ormother) in another household, respectively. Theweighted yields are dispersed over a wide rangeof intermediate values for the fulI multiplicityn.de, with concentrations at values 1/5, 1/3, 1/2,2/3, and 1. Weighted yields greater than 1 areobtained only if more than one event of thesame type is reported by the same household.For instance, a birth to a woman in”the house-hold without mother or sisters in the surveypopulation (multiplicity 1) and a birth to her

Table P. Events by number of households in multiplicity network, harmonic mean of multiplicities, andnumber of reporting househoIds, by multiplicity counting rule, type.of event, and type of report

Number of households in network Har-monicmeanof

nulti-plic-ities

2.75

!Wmber>f re-port-ing~ouse-~olds

887

Multiplicity count-ing rule, type ofevent, and type of

report Total 1 2 3 4 5 6 7 II8 9 10FULL MULTIPLICITY

Births

Total----------

Percent

21.4

15.917.331.4

15.3

12.5L2.420.3

...

...

...

...

9.5

:::L5.4

9.9

;::17.4

...

...

...

...

...

...

100.0 5.0 I 26.8 27.6 5.3 3.3 1.0 0.1—

By event household---BY mother------------By sisters-----------

100.0100.0100.0

100.0

100.0100.0100.0

100.0

100.0100.0

100.0

12.5 32.3(1.1) 4;.;

.

28.425.628.7

24.5

22.527.624.1

0.9

:::

5.9

2.1

1;::

...

0.3

n

3.2

2.12.25.0

...

...

...

...

...

...

2.272.633.83

2.79

328266293

6661.4

0.1

2.0

Marriages

Total---------- 4-6.2 26.3By event household---By parent------------By siblings----------

15.4 33.7(2.2) 43.2

5.8

D.8

2.;

...

2.192.614.17

1.75

240185241

594

2.;3.7

...

RESTRICTEDMULTIPLICITY

Births

t

14.6 85.4

23.8 76.2(3.4) 96.6

Total----------

By event household---By mother------------

1.621.93

1.68

328266

425

.

3.3

. . .

. . .

. . .

. . .

. . .

. . .

,..,..

,..

. . .

. . .

. . .

Marriages

Total----------*

By event household---By parent------------

100.0100.0

33.3 65.4(3.2) 90.8

1.25.9

1.511.98

. . .

. . .. . .. . .

. . .

. . .,..,..

. . .

. . .

NOTE: Numbers enclosed in parentheses are errors. According to counting rules, cell should be empty.

21

daughter who hves elsewhere but does not havesisters living elsewhere (multiplicity 2) wouldgive a total weighted yield of 1.5. The largerdecrease in sampling variance due to multiplicityfor births than for marriages can already be in-ferred from the differences between the distribu-tions of weighted yields (primarily because ofthe concentration of weighted yields for theconventional rule at 2 for marriages as againstthe concentration at 1 for births).

Table D gives the distribution of networksizes by type of report for the two multiplicityrides and the harmonic mean of the multiplic-ities. The harmonic mean is the average contri-bution of a reporting household to the sampletotal of weighted events. Obviously the networksizes for reports from households of parents andsiblings are higher than those report ed directlyby the interview households, since their networksizes should be at least 2. However, elimin