Living Standard Measurement Surveys in Developing Countries: Some Sources of Error

ABSTRACT - The paper deals with error sources in Living Standard Measurement Surveys (LSMS) as these are often employed in developing countries. Drawing upon current literature from survey research in general and studies on survey research in developing countries in particular, we discuss the following error sources: problems to define the household unit errors imposed by use of key-informants, field translation ambiguity, reference period, problems and difficulties imposed by the interview setting. Evidence suggests that substantial errors occur in LSMSs in developing countries, but since it is rare that the survey data is thoroughly analyzed, the validity and reliability of the data are seldom questioned. As a summary we suggest some ideas for improvements.

Citation:

Ingeborg Astrid Kleppe and Kjell Gronhaug (1998) ,"Living Standard Measurement Surveys in Developing Countries: Some Sources of Error", in AP - Asia Pacific Advances in Consumer Research Volume 3, eds. Kineta Hung and Kent B. Monroe, Provo, UT : Association for Consumer Research, Pages: 229-235.

Asia Pacific Advances in Consumer Research Volume 3, 1998      Pages 229-235

LIVING STANDARD MEASUREMENT SURVEYS IN DEVELOPING COUNTRIES: SOME SOURCES OF ERROR

Ingeborg Astrid Kleppe, Norwegian School of Economics and Business Administration, Norway

Kjell Gronhaug, Norwegian School of Economics and Business Administration, Norway

ABSTRACT -

The paper deals with error sources in Living Standard Measurement Surveys (LSMS) as these are often employed in developing countries. Drawing upon current literature from survey research in general and studies on survey research in developing countries in particular, we discuss the following error sources: problems to define the household unit errors imposed by use of key-informants, field translation ambiguity, reference period, problems and difficulties imposed by the interview setting. Evidence suggests that substantial errors occur in LSMSs in developing countries, but since it is rare that the survey data is thoroughly analyzed, the validity and reliability of the data are seldom questioned. As a summary we suggest some ideas for improvements.

1. INTRODUCTION

Inequality in distribution of wealth among affluent and poor countries has for a long time been of major concern. Over the years a variety of activities and programs, ranging from immediate action to structural adjustment programs to foster long term economic growth have been initiated to enhance the standard of living and to reduce poverty. [To operationally define "poverty," a "poverty line" is often identified. See appendix A for more detailed discussion.]

In order to plan and implemnt activities to improve the standard of living, some diagnostic background information is needed. For example, since Drucker¦s (1958) seminal contribution on "Marketing and Economic Development", marketers have focused on marketing insights that may contribute to faster economic growth (for recent overview, see Taylor and Omwa 1994). Simple, diagnostic information revealing the "actual situation" is also needed to set goals as well as to measure program performance. A key instrument frequently used to gather such diagnostic information in the third world are Living Standard Measurement Surveys (LSMS), of which there are several different types.

In this paper we critically examine such surveys. In particular we focus on data collection procedures, the organization of data and underlying assumptions which influence data quality, and thus the usefulness of such surveys. The remaining part of the paper is organized as follows:

In the next section we report characteristics of LSMSs. We then describe the actual data collection procedures in such surveys. Examples included here are taken from countries in Sub-Saharan Africa (SSA), where the first author has substantial first-hand experience. Then in more detail we examine specific methodological problems related to unit of analysis, choice of respondent, instrumentation, and assumptions of and burdens put on the respondents. Finally, implications of our critical discussion are highlighted.

2. LIVING STANDARD MEASUREMENT SURVEYS

During the past 10 years the international community has undertaken several programs to provide data for macroeconomic analysis of poverty and to provide figures for national accounts and models used in planning. The United Nations Survey Capability Program (UNHCP), the World Bank Living Standard Measurement Survey (LSMS) and the Social Dimensions of Adjustment (SDA) Project are major efforts in this regard. For convenience these will all be referred to as LSMSs since their main goals are very similar, namely to improve data collection efforts at the household level in developing countries. The procedures underlying the collection and analysis of data are also very much the same. A typical LSMS is multi-purpose and complex. More precisely the purposes are:

* Diagnostic to map the actual situation, i.e. to measure distribution of welfare and level of poverty in households. Such information is needed to set goals and measure program output.

* To assess relationships. A key purpose is also to obtain information (data) which allows the assessment of potential impact prior to a program. Information about program and impact B if valid B allows for adequate design of programs, and thus economizes scarce resources.

* To get relevant information (data) allowing for detailed analyses B and thus understandingBof what constitutes and contributes to household welfare. (Grosh & Munoz, 1995, p. 1).

LSMSs typically use three kinds of questionnaires: (1) A questionnaire for the household where a key-informant is asked about many aspects of the household’s welfare, especially consumption, income, and the use of various social services, (2) A questionnaire designed for data collection of the local community. Here community leaders and groups representing the community are asked about the infrastructure and services available in the community. In addition a questionnaire asks vendors about price levels for selected items of importance to achieve comparable income and expenditure data across geographical regions within a country. It should also be noted tht LSMS surveys, to some extent, are adjusted to regions and purpose, and may as well include more questionnaires. From the above description it also follows that structured questionnaires are used to collect the data. It is assumed implicitly that this data collection procedure can B and willByield valid and reliable data (information).

In this paper we will limit ourselves and focus on households only. These questionnaires are very extensive and cover topics such as: a) household demographics; b) education; c) health; d) agricultural production; e) labor market participation; f) income; and g) consumption. For example, the 1992 National Sample Survey for Agriculture in Malawi consisted of 5 modules which required several visits to the same household and was conducted on a sample of 12,000 households. The 1996 Living Conditions Survey in Zambia has three questionnaires; a household questionnaire using the head of the household as informant; an individual questionnaire to all household members ages 12 years and above, and a child questionnaire for children aged 0 through 11 years using the mother as informant. This survey was conducted on a sample of 6000 households.

It is hard to find a sensible explanation why this tradition of gigantic multipurpose surveys has evolved. One explanation may be that the questionnaires have been designed by economists and statisticians with no special skills in questionnaire construction. Another explanation might be that the LSMSs to a large extent have been driven by the international donor community with little coordination and incentive for long term planning of statistical activities in each country.

Needless to say, such surveys also result in enormous data sets. Due to lack of research capacity little work has been done on the LSMSs in each individual country. The most common output on simple tabulations is produced by the National Statistical Office in each country. In an effort to utilize the data the World Bank has during the past 3-4 years produced Poverty Assessments and Profiles for 26 Sub-Saharan African Countries. The main objective for the World Bank reports is to determine the poverty status of the different populations based on data on income and expenditure. These analyses have been important in putting poverty on the agenda and in formulating policies and intervention programs. However, the World Bank reports have been criticized for being superficial and not understanding the idiosyncrasies of each individual country. In a review conducted by the Institute of Social Studies Advisory Service in the Netherlands (1996) it is concluded that little new insights have been acquired through these Poverty Assessments and Profile studies. The general findings across countries are that: a) rural people are more prone to be poor than people living in cities; b) the poor are less likely to live in households in which the household head is literate or attended school; c) the poverty status of female headed households varies across countries; d) poorer households are on average larger than non-poor households; e) the share of household income spent on food is higher for people living in rural regions than for people living in cities (Hanmer, Pyatt, and White; 1996). Critiques also claim that the World Bank assessments rely too much on "quantitative" data sources and money-metric measures of poverty.

The issues raised in this paper are based on survey research insights. The ultimate test of any survey instrument is the moment when the questions are presented to the respondent. In the field the survey instrument is usually out of the hands of those who designed the survey. From practice and research we know that there will always be problems in the fieldBthe question is how serious are these problems. In this paper we have selected a few obvious factors that can cause measurement errors, i.e.:

* Selection of respondent and the respondent’s capability to answer the questions on behalf of the unit of analysis

* The language used in the interview

* Difficulty of questions

* Length of the interview

* Interview setting

In addressing the various issues, countries in Sub-Saharan Africa are used as context.

3. THE USE OF HOUSEHOLDS AS THE UNIT OF ANALYSIS

The vast majority of people in Sub-Saharan Africa (SSA) live on small subsistence farms in rural areas. Although there is substantial variation in smallholder production and agricultural practices, one pattern seems to prevail across most countries in SSA. The typical smallholder wife is responsible for food production and domestic duties while the male smallholder owns the land and participates in marketing activities (B°e, 1997). This division of labor is usually very clear and rigid. The reality for most subsistence farm households is that they have very few assets and very few opportunities to earn income. Many households and communities have very limited market activities and consume most of their own production.

In a typical LSMS interviewers are instructed to select five households within a village and identify the head of the household as the key-informant. The correct listing of household members is crucial for the calculation of income/consumption pr. capita. The selection of the key-informant is also crucial in order to obtain information about the household.

Because of the temporary dislocation of households and individuals due to unstable political, social, and climatic environments, dealing with the household as the unit of analysis represents a major challenge.

Definitions of households

The instructions on who should be listed as a household member also vary across surveys. "Household members" who are not living in the household at the time of the interview, and "non relatives" who live in the household at the time of the interview are treated differently. For instance the enumerators’ manual for the 1992/1993 survey in Zambia says: "A household is a group of persons who normally eat and live together. These people may or may not be related by blood, but make common provision for food or other essentials for living and they have only one person whom they all regard as the head of the household. (Enumerators¦Instruction Manual 1992/93, p 5-7). The definition of household members goes on over two pages. In contrast the definition of a household in the enumerators manual for the 1992 Welfare Monitor Survey in Kenya is very short: "A person or a group of people living in the same compound (fenced or unfenced), answerable to the same head and sharing a common source of food and/or income. It is important to note the three elements of this definition namely: "Do they live in the same compound?", "Are they answerable to one head?", "Do they share common source of food and/or income?". If any of the answers to these questions is "No" then it is not one household." (Enumerators’ Manual, April-May 1994, p.5). In the Uganda manual the instruction tells the enumerators to define a household as a group if persons who live, cook and eat together or a single person who lives alone and eats independently (Manual of Instructions to field workers, 1992, p.10). The main criteria to define a household in the Uganda survey is that the people cook and eat together.

Studies also show that it is a practical problem to determine where the statistical unit begins and ends (Dubois, 1994). For example, the number of members in each household in the 1992 National Sample Survey of Agriculture in Malawi did not match across the survey modules. Peters’ (1994) study on smallholdersin Malawi documented that at any time many household members are in a transition from one household to the other. In Uganda the average household size in the 1989 Household Budget Survey was 5.45 compared to 4.76 in the 1992 Social Dimensions of Adjustment Survey. Appleton (1994) attributes this differential to a much looser definition of household members in the former survey. At the aggregate level of analysis this difference in household size also causes a much higher rise in consumption per capita than in consumption per household, since per capita consumption is calculated as the average consumption per adult equivalent in the household.

Key informant problems

Most survey manuals ask the enumerators to select "the head" of the household to be key-informant on behalf of the household. The head of the household is usually defined as the person all members of the household regard as the head, and who normally makes day-to-day decisions governing the running of the household. In most cases the household head is a man.

This headship variable is difficult in two ways. First the procedure for identifying the main decision maker in the household may be questioned, and second, to what extent the selected head or key-informant is the best informant for the purposes of the survey should be questioned. Handa (1994) who studied this question on data from the 1989 Jamaican Survey of Living Conditions found that the reported head is not necessarily the person who is in charge of the household consumption activities, but rather the oldest resident member of the household. The headship variable is also difficult because of unstable households. Peters’ (1994) study shows that single female heads of households often are in transition between two husbands. A household can be either de jure or de facto headed by a female. De facto means that the head is an unpartnered woman, while de jure usually implies that the female head of the household has a husband who is temporarily away from home.

Handa (1994) also found that female headed households (FHHs) and male headed households (MHHs) tend to represent different types of households. A male headed household frequently implies an intact couple, while a female headed household more frequently represents a single or unpartnered woman. This simple female-male dichotomy also hides important differences in household income, demographic composition, and intra-household resource allocation between the two groups of households. Partner headed households tend to show better outcomes for children on all indicators. The most important within group difference appears among male headed households, wherein unpartnered households display significantly worse child outcomes than partnered ones. This within group difference is not significant for female headed households. Based on this analysis Handa (1994) suggests that the absence of any female authority within the household is negatively associated with child welfare.

A general finding in the Sub-Saharan Africa surveys is that female headed households systematically report a larger share of total income/expenditure on food consumption and investment in human capital such as health and education. A possible explanation is that women report more accurately on these issues since such activities relate to their roles. However, the finding can also indicate that FHHs spent a larger share of their income on these items. Handa (1994) found that male respondents reported more consumption on luxury goods like cigarettes and beverages. These findings indicate that the use of key-informants can represent a systematic bias in reports of household consumption and expenditures.

The use of key informants has been discussed in research on households and organizations in developed countries from a perspective of response biases (Davis & Rigeaux 1974; Davis, Hoch, and Ragsdale 1989; Silk and Kalwani 1982; Phillips 1989). The general findings from this research are that key-informant responses tend to be vanity biased (represent the key-informant in a good ay), subjective (project their own feelings and preferences to other members of the unit), and erroneous since the key-informant is not informed on all aspects of the subject matter.

Such threats to valid studies may be even more prominent in developing countries. This is particularly so because the male/female roles are more distinct. Domestic duties such as child care, preparation of food and growing of food crops are almost entirely a female responsibility. It is thus obvious that the use of male respondents can cause substantial biases in records of household expenditures and consumption.

To use the man as key-informant could also be a problem with respect to intra-household allocation of welfare. It is a sad, but well known fact that families/households on the borderline of existence have to distribute their limited resources unevenly among household members. Girls’ education in particular suffers from this mechanism. Thus more than one informant is necessary to map intra-household allocation of resources.

4. QUESTIONNAIRE PROBLEMS

Evidence from research on surveys suggests that even very minor changes in the formulation of a question can produce significant effects. It has been repeatedly pointed out that it is not the interviewers who need to know the complete understanding of the question, but the respondents. Anything that the investigator wants the respondents to report should be included in the question itself, so that all respondents will be exposed to it (Fink, 1995). It is therefore recommended that a question(naire) should be self explanatory both to the respondent and enumerator. The LSMS practice of using extensive enumerator manuals and vaguely formulated questionnaires can lead to substantial differences in how questions are interpreted by the respondents, thus influencing their actual reporting.

Instruments used for survey interviewing may be broadly grouped into two categories; semi-structured schedule types of questionnaires and structured questionnaires with exact question wording. In the former the information required is summarized in a brief heading (such as age or number of children ever born) and the interviewer is free to ask any questions which he or she considers necessary to obtain this information. In the latter the exact question to be asked is given. The interviewer is only required to read it word for word to the respondent. If a questionnaire is written in a language other than that of the interview a structured approach will leave the interviewer to make the necessary translation into a language understood by the respondent.

Hardly any interview situation occurs where the respondent does not need any clarification at all. Studies on interviewer practice in developing countries show that structured questionnaires are not in fact used in a structured manner by interviewers, who often use their own questions and by doing so often change the questions formulated (See Scott et al., 1988). The more dependent the interview is on the interviewer, the more room for idiosyncratic interpretations, biases and random error in the reported data.

Multilingualism

Multilingualism presents a serious problem for survey work. Nearly all the countries in Sub-Saharan Africa are multilingual, in the sense that there is no single language which can be used for interviewing everyone. In most of the countries affected by this problem it has been customary to prepare survey questionnaires in a single language (often English or French) and the whole issue of translation is left in the hands of the interviewer. Available evidence points to the prevalence of translation errors in survey work, and indicates that greater control over wording of questions in the language of the interview is needed. Scott et al. (1988) conducted a field experiment to asses the magnitude of translation error in the Philippines and the Ivory Coast. By taping the interviewers in the interview setting they could count different types of error occurring in the translation. They developed codes according to severity of translation error (0), small change of wording, meaning preserved), (1) one response picked out or suggested, (2) alternatives not read out (or not all read out), (3) large change of wording, meaning preserved, (4) change of meaning, (5) qualifying phrase omitted or seriously distorted, (6) introduction omitted, and (7) introduction changed. Three types of survey instruments were used: schedule questionnaires, international language questionnaires, and local language questionnaires. Their main findings were that interviewing is more accurate with structured questionnaires than with schedules, and that structured local language questionnaires perform considerably better than field translation by the interviewers. In the Ivory Coast experiment field translation of the French questionnaire into BaoulF (local language) resulted in over four times as many serious errors as use of a verbatim BaoulF questionnaire. In the Philippines field translation of the English questionnaire into Tagalog ( local language) resulted in six times as many serious errors as when the same questionnaire was used for interviewing in English; for the other language Cubuani the figure was 10 times. These findings are hardly surprising. However, the Scott et al. (1988) study documents the magnitude of the problem.

Master versions of the questionnaires are often developed by people in the United Nations or The World Bank headquarters or by some international consultant. Combined with time pressure and lack of resources the language problem has not received the attention it deserves. The entire process of questionnaire design and identification of indicators requires a participatory approach where both the local user and the research community is involved in the process.

To make multipurpose questionnaires look shorter for most of the LSMSs a frequent practice is to organize the questionnaires in columns. Because of space the questions are phrased very briefly. Although questionnaires may seem structured in appearance they are not structured since 1) translation is required, 2) the questions are not properly operationalized, 3) the questions are not complete, and 4) instructions for the questions (intentions) are not spelled out.

In developed countries the use of structured questionnaires is dominant. In developing countries there appears to be a much greater use of schedules. Some survey workers also believe that exact wording is less critical in factual than in attitude questions, and conclude that schedule format is more acceptable in the former. Since developed countries make more frequent use of attitude surveys this might account for their greater use of structured questionnaires. Some statisticians also argue that rural or uneducated respondents cannot be questioned with short, direct predetermined questions; the interviewer must be allowed the liberty to approach the topic indirectly , or gradually, in the manner typical of villagers’ conversation (cf. Scott et al. 1988).

Appleton (1994) compared two surveys in Uganda, one using itemized and the other open questions to register consumption items. Open item lists resulted in much lower household expenditures and consequently a much higher incidence of poverty. The poverty differential in the two surveys could not be explained by economic and political factors. The items most likely to be omitted in the open item questionnaire were the more expensive luxury items such as cigarettes, eggs and soda. The rule of thumb to stick to one idea per question is also hard to realize in vaguely phrased schedule questions as each question in the questionnaires requires several follow-up questions. (see annex 2).

5. MEMORY DEMANDS

It is geneally recognized that the use of recall methods for collecting expenditure data for frequently purchased products are inadequate (Sudman & Ferber 1974). If survey researchers were guided entirely by a concern for valid descriptive data, they would focus on current, specific, and real events. It is increasingly apparent that questions related to memory tend to be difficult. Recalling an event or behavior can be specifically difficult if the decision was made almost mindlessly, if the event was so trivial that people hardly gave it a second thought, if questions refer to events that happened long ago, or if any of the questions require the recall of many separate events.

Usually the recall period for the consumption items varies. The recall period can span from one week to the last twelve months. Concepts like "season" are also used to frame a period. In surveys where the household is interviewed twice, the reference period for some expenditures would be the time between the two visits. In a typical LSMS the respondent has to relate to several reference periods. For example, the 1996 Zambia LSMS includes:

* employment out of agricultureBlast seven days

* visits to hospitalsBlast two weeks

* expenditure to agricultural inputsBlast 12 months/last season

* food expensesBlast month/last visit (two weeks)

* school feesBthis school year

* skills training past three months

* agricultural season 1995/1996

* income from non-farming business the last month

* non-regular allowances last month

* rent last month

* remittances last month

* transfer payments such as pension, grants, premium, interest on savingBlast month

* food consumptionBlast two weeks

This means that both the interviewer and the respondent cognitively have to shift from one reference period to the other during an interviewBin the Zambian case eight different reference periods are used in one interview. The Scott et. al (1988) study found that the words "since" and "during" caused trouble in translations, and that point of time was often confused with period of time.

Techniques to enhance validity of reporting past events

To enhance validity of the reporting of the past, five techniques has been recommended: a) bounded recall, b) narrowing the reference period, c) averaging, d) landmarks, e) cueing (Converse and Presser, 1986).

Bounded recall addresses over reportage due to "forward" telescoping. Evidence exists that if one asks respondents about events in the last six months, they may actually see #beyond’ that period and include events that happened earlier, which of course creates over reporting. When using bounded recall a baseline is established at first interview. Then in a subsequent panel interview, one asks about events or expenditures that have happened since the first interview. Both Neter and Waksberg (1964) and Sudman and Bradburn (1973) recommend the use of available records to improve memory expenditure surveys. The use of records generally controls for overstatements due to compression of time errors. Appleton (1994) suggests from his comparative study of two LSMSs in Uganda that records from the two-visit survey appear more correct than the one-visit survey.

It is recommended whenever logically feasible not to use longer reference periods than six months. For certain events one can narrow the time period of interest to the very immediate past, such as last week or yesterday. To determine whether this refers to "typical" or habitual experience, one can ask respodents to average the event or behavior by asking if this was a typical day week or month. The use of landmark events and cues to memory is still another strategy. The purpose of cues is to stimulate recall by presenting a variety of associations. Over the year consumption can vary substantially in many African settings. For instance during pre-harvest periods people may temporarily starve and recorded level of consumption may therefore be much lower than the rest of the year. One-shot-recordings on these issues are therefore very problematic for determination of absolute poverty levels.

TABLE 1

IMPLICATIONS OF DIFFERENT ERROR SOURCES IN LSMSs

Neter and Waksberg (1964) argue that the time period between (two) visits is the most reliable reference period to record household consumption and expenditures. Appleton (1994) argues that between visit estimates are more accurate than 30 day figures. However, findings are not conclusive since recall error may vary across different consumption items depending on frequency of purchase. In the Uganda case the published survey reports from the 1989 and 1992 surveys showed large rises in consumption between the two surveys. However, external validation of the data suggested that most of the reported increase could be attributed to questionnaire design and recall period used (Appleton, 1999).

Chaudhuri & Ravallion (1993) suggest that cross-sectional observation of consumption or income can correctly identify roughly three-quarters of the chronically poor in terms of either long-term income or consumption. However, it is obvious that the information on expenditure and consumption in the LSMSs is very demanding for the respondent. The respondent has to remember actual consumption by quantity or monetary value, or both. This task requires both a good memory and advanced computational skills. Some multipurpose surveys therefore simplify the welfare measure down to a question on how many meals were prepared yesterday or last week, and how many attended these meals.

6. THE INTERVIEW SETTING

Needless to say the facilities can be very unsatisfactory during the interviews in the field. The lay-out of the questionnaires is very crucial for the logistics of the interview setting. Some questionnaires are so complex that they require a stable table or desk in order for the enumerator to mark the correct code in the correct column and row (see annex 2). A finding from the 1992 Welfare Monitoring survey in Kenya illustrates this problem. The question "reasons for quitting secondary school" has eight answer categories depicted in a narrow column. Answers to this question should be recorded for each relevant household member. The enumerators must have had problems, since the first datacleaning showed pregnancy as the most common drop-out reason for boys. This mistake was easy to detect and could be sorted out in the data-cleaning process. Other mistakes are more difficult to detect.

Another obvious challenge is interviewer and respondent fatigue. Many people in SSA are sick or do not feel well at any time. This fact combined with very long and demanding questionnaires will cause fatigue. Interviewers in the 1992 survey in Uganda reported that they often could see the respondents’ eyelids dropping during the interview. The LSMS survey in Madagascar visited the same household several times with interviews that lasted for two to three hours. There is little doubt that such fatigue can dramatically reduce the quality of the reported data.

7. SOME IMPLICATIONS

In developed countries data bases on social and economic statistics are built over years. The use of large scaled LSMSs indicates the belief that it is possible to gather "everything" at one shot. The above discussion shows that this is possibly not the case, and that such large scaled studies result in data of ediocre quality. How can this practice be explained? One explanation is that easy access to international funding does not encourage survey managers to focus and set priorities. Because the data collected is only superficially analyzed, the data quality is seldom questioned.

The lack of incentives to set priorities and focus the surveys is a pitfall for the poverty monitoring activities in many SSA countries. Lack of adequate competence and capacity makes it worse. Small and meagerly funded National Statistical Offices undertake these enormous survey projects that they hardly can handle. For instance the National Statistical Office in Malawi spent more than one year in processing the data for one of the 5 modules in the 1992 National Sample Survey for Agriculture, three years after the data was collected. Similar stories can be heard from other countries.

Table 1 reports sources of error emphasized in the above discussion, and some suggestions of how such surveys can be improved.

Future surveys should also focus more on quality than quantity of data. Language related challenges are obvious in such surveys, indicating that substantial resources in the design of the survey instrument are needed. Also the use of multiple visits for each household is central in obtaining reliable measures on income and consumption. In collecting the data needed it is crucial not only to consider the respondents’ ability and capacity (and willingness) to give reliable and valid information, but also the actual setting for the collection. The actual approach should be designed by taking such constraining and biasing factors into account.

APPENDIX 1

ESTABLISHING "POVERTY LINE"

REFERENCES

Appleton, Simon (1994): "Changes in Poverty in Uganda 1989-92: Comparing Households Surveys Over Time", Working paper, Centre for the Study of African Economies, Oxford University

B°e, Turid (1997): "Share-Tenancy Within the Household Unit", Chr. Michelsens Institute, Norway, Unpublished paper.

Carmines, Edward G. & Richard A. Zeller (1979): "Reliability and Validity Assessment", Sage University paper No. 17

Chaduri, Shubham and Martin Ravallion (1994): "How Well Do Static Indicators Identify the Chronically Poor?", Journal of Public Economies, No.3, 367-394

Converse, J. M. and S. Presser (1986): "Survey Questions: Handcrafting the Standardized Questionnaire", Sage University Paper Series, No. 63

Dubois, Jean-Luc (1992): "Think Before MeasuringBMethodological Innovation for the Collection and Analysis of Statistical Data, Social dimensions of Adjustment in Sub-Saharan Africa", Working paper No. 7, Surveys and Statistics, World Bank

Farmer, Amy and Jill Tiefenthaler (1995): "Fairness concepts and the intra-household allocation of resources", Journal of Development Economics, 47, pp. 179-189

Fink, Arlene (1995): "How to Ask Survey Questions", The Survey Kit, 2, Sage Publications Ltd.

Glewwe, Paul (1990): "Improving Data on Poverty in the Third World", PRE Working paper No. 416, World Bank

Haddad, Lawrence and Ravi Kanbur (1992): "Intra-household inequality and the theory of Targeting", European Economic Review 36 , pp. 372-378

Handa, Sudhansu (1994), "Gender, Headship and Intra-household Resource Allocation", World Development, Vol. 22

Nisbett & Wilson (1977), "Telling more Than We Can Know: Verbal Reports on Mental Processes", Psychological Review, 84, 231-258

Ravallion, Martin, Gaurav Datt and Dominque van de Walle (1991). "Quantifying Absolute Poverty in the Developing World", Review of Income and Wealth, Series 37, Number 4, December

Ravallion, Martin (1993): "Poverty Comparisons", Harwood Academic Press, Fundamentals of Pure and Applied Economics

Scott, Chris, Martin Vaessen, Sidiki Coulibaly, and Jane Verrall(1988): "Verbatim Questionnaires Versus Field Translation of Schedules: An Experimental Study", International Statistical Review, 56, pp. 259-278

Sen, Amartya (1976): "Poverty: An ordinal Approach to Measurement", Econometrica, 44, March, No. 2

Sudman, Seymour and Robert Ferber ( 1971), "Experiments in Obtaining Consumer Expenditures by Diary Methods", Journal of the American Statistical Association, December, 66, (December), No 336

Sudman, Seymour and Norman M. Bradburn (1973): "Effects of Time and Memory Factors on Response in Surveys", Journal of the American Statistical Association, December, 68, No. 344

Taylor, Charles R. and Omura, Glenn, S. (1994), "An Evaluation of Alternative Paradigms of Marketing and Economic Development", Journal of Macromarketing, 2 (Fall), 6-20.

Urdy, Christopher (1996): "Gender, Agricultural Production, and the Theory of the Household", Journal of Political Economy, 104, No. 5.

Social Dimensions of Adjustment Priority 1992/93, Enumerators’ Instruction Manual , Lusaka, Zambia, Central Statistical Office.

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