Disproportionate sampling for population subgroups in telephone surveys

Population studies often seek to examine phenomena in important population subgroups or to compare results among these and other subgroups. When subgroups of interest comprise a relatively small percentage of the population and acceptable subgroup member lists are not available to serve as sampling frames, it may be prohibitively expensive even by telephone to screen through a sample of the entire population. This paper considers some statistical effects of estimation from a class of two-stratum telephone sample designs where part of the frame with a higher subgroup concentration is disproportionately sampled compared to the rest of the frame. Using proportionate sampling as a reference, the relative impact of this disproportionate design is determined for nominal and effective sample sizes, where the latter are tied to the effect of variation in sample weights that occurs in disproportionately allocated samples. Findings are illustrated using two recent telephone surveys. Whereas nominal subgroup sample sizes may be improved by disproportionate sampling, we conclude that both the survey designer and analyst should use this type of design cautiously in telephone surveys.     

Household sampling through geocoded points and satellite view: A step-by-step approach to implement a spatial sampling method for demographic and health surveys in areas without population sampling frame and with limited resource settings

IntroductionSpatial sampling is increasingly used in health surveys as it provides a simple way to randomly select target populations on sites where reliable and complete data on the general population are not available. However, the previously implemented protocols have been poorly detailed, making replication difficult or even impossible. To our knowledge, ours is the first document describing step-by-step an efficient spatial sampling method for health surveys. Our objective is to facilitate the rapid acquisition of the technical skills and know-how necessary for its deployment.MethodsThe spatial sampling design is based on the random generation of geocoded points in the study area. Afterwards, these points were projected on the satellite view of Google Earth Pro™ software and the identified buildings were selected for field visits. A detailed formula of the number of points required, considering non-responses, is proposed. Density of buildings was determined by drawing circles around points and by using a replacement strategy when interviewing was unachievable. The method was implemented for a cross-sectional study during the April-May 2016 period in Cotonou (Bénin). The accuracy of the collected data was assessed by comparing them to those of the Cotonou national census.ResultThis approach does not require prior displacement in the study area and only 1% of identified buildings with Google Earth Pro™ were no longer extant. Most of the measurements resulting from the general census were within the confidence intervals of those calculated with the sample data. Furthermore, the range of measurements resulting from the general census was similar to those calculated with the sample data. These include, for example, the proportion of the foreign population (unweighted 8.9%/weighted 9% versus 8.5% in census data), the proportion of adults over 17 years of age (56.7% versus 57% in census data), the proportion of households whose head is not educated (unweighted 21.9%/weighted 22.8% versus 21.1% in census data).ConclusionThis article illustrates how an epidemiological field survey based on spatial sampling can be successfully implemented at low cost, quickly and with little technical and theoretical knowledge. While statistically similar to simple random sampling, this survey method greatly simplifies its implementation.     

New outcome rates for random sampling telephone surveys

BackgroundSurvey non-response rates are important quality indicators. Refusal rates can induce non-response bias in health survey estimates. However, comparisons across surveys highlight inconsistencies in the use of survey outcome categories and in the calculation of response rates. In this paper we discuss the relevance of these indicators and suggest other survey quality indicators.MethodsOutcome rates from two French random-digit dialing (RDD) telephone surveys are compared : the Nicolle survey on infectious diseases of 4112 individuals conducted in 2006, and the HIV knowledge, attitude, belief and practices (KABP) survey of 5071 individuals in 2004. Based on the same protocol, we describe in details the way the two RDD samples were drawn and how non-response rates were estimated.ResultsNon-response rates were different: 36% in Nicolle survey and 18% in KABP survey. However, the quantity of telephone numbers required to obtain one interview was higher in the KABP survey: 2.8 telephone numbers versus 2.1 in the Nicolle survey. The participation rates, aggregating together refusals, break-off and non-reachable numbers, were equivalent for the two surveys. This result occurred because of a greater proportion of unreached calls in the KABP surveys, which is not integrated into the non-response rates commonly used.ConclusionSurvey non-response rate is insufficient to estimate the quality of a survey. The need for other indicators has been previously stressed in the literature, notably with the adoption and utilization of the American Association for Public Opinion Research (AAPOR) standard definitions of four indicators. But these indicators are quite complex for evaluating non-response bias between surveys. In addition to the classical refusal rate, two other indicators are proposed in this paper: participation rate (number of complete interviews divided by the number of eligible and of unknown eligibility units) and a liking contact rate (number of unreachable units because of a long absence, break-off or non-answer divided by the number of eligible and of unknown eligibility units). The sum of these three indicators is equal to 100% and thus easier to manipulate when comparing surveys.     

复杂抽样调查数据实例分析

提出复杂抽样调查数据的分析思路和方法以及忽视权重和抽样设计时会出现的问题.文中以2002年中国居民营养与健康状况调查数据中高血压患病率的估算为例,分加权和不加权、考虑和不考虑整群设计特征的四种组合情况对数据进行分析.表明忽视权重的设置会影响点估计和标准误的估计,忽视对整群设计特征的考虑不仅会高估结果的精确度,还会得到地区间患病率有差异的假阳性结论.因此使用合理的统计方法分析复杂抽样调查数据非常有必要.     

Accurate sampling in general practice waiting room surveys: methodological issues

OBJECTIVES: This is a pilot study to test the validity of using one research assistant to train and support reception staff at five clinics to obtain waiting room data in general practice surveys. DESIGN: A research assistant trained reception staff at five randomly chosen general practices to administer a survey to all eligible women over a two-week period. Practices were audited daily by their appointment books and where possible by billing records to check total numbers of eligible women to determine the denominator of the sample. SETTING: Five metropolitan general practices in one divisional area. PARTICIPANTS: Twenty-five receptionists distributed surveys to 1,298 women. Main outcome measures: Number of eligible women missed at each clinic. RESULTS: The number of potential subjects missed by the reception staff ranged from nil to 18% with the appointment book audit. A second audit using billing records at three clinics revealed inaccuracies of up to 50%. The ability to accurately capture eligible patients for waiting room surveys depends on good administrative systems within the practice and motivated staff. CONCLUSIONS: It is important for general practice researchers to consider and account for eligible participants who are missed from the denominator with a waiting room survey method. Valid sampling using this method can be achieved in certain practices, however this may limit the generalisability of the findings.     

Epidemiologic studies utilizing surveys: accounting for the sampling design.

BACKGROUND. Since large-scale health surveys usually have complicated sampling schemes, there is often a question as to whether the sampling design must be considered in the analysis of the data. A recent disagreement concerning the analysis of a body iron stores-cancer association found in the first National Health and Nutrition Examination Survey and its follow-up is used to highlight the issues. METHODS. We explain and illustrate the importance of two aspects of the sampling design: clustering and weighting of observations. The body iron stores-cancer data are reanalyzed by utilizing or ignoring various aspects of the sampling design. Simple formulas are given to describe how using the sampling design of a survey in the analysis will affect the conclusions of that analysis. RESULTS. The different analyses of the body iron stores-cancer data lead to very different conclusions. Application of the simple formulas suggests that utilization of the sample clustering in the analysis is appropriate, but that a standard utilization of the sample weights leads to an uninformative analysis. The recommended analysis incorporates the sampling weights in a nonstandard way and the sample clustering in the standard way. CONCLUSIONS. Which particular aspects of the sampling design to use in the analysis of complex survey data and how to use them depend on certain features of the design. We give some guidelines for when to use the sample clustering and sample weights in the analysis.     

Effect of sampling frames on response rates in the WHO MONICA risk factor surveys

Sample surveys are used to investigate occurrence and determinants of diseases in populations. Their reliability is influenced by quality of sampling frame and response rate. We investigated relationship between sampling frame type and response rates and assessed their impact on non-response bias, using data from the WHO MONICA Project, where 37 centres in 20 countries conducted sample surveys, employing the best locally available sampling frame. Sampling frames fell into three categories: Population registers (PR), electoral registers (ER), and health care registers (HR). Response rate (rrs) was factored into components reflecting quality of sampling frame (contact rate cr) and characterizing willingness of sample members to participate (enrolment rate er). The mean quality score for the sampling frames was 92 for PR, 87 for HR and 85 for ER; they contributed on average 23, 20, and 26 to the respective non-response rates. For all frame types and both sexes the lowest quality score occurred in the age group 35–44, suggesting a reduced ability to track migration of a highly mobile population group. The patterns in the age/sex distribution of er indicate at least for males in PR and females in HR a potential for non-response bias. Estimation of non-response bias through an abbreviated questionnaire failed because of low item response. We found that contact rate characterizes sampling frame quality. For all frame types it had a major influence on response rate. It is likely that low er and low cr cause different kind of bias, requiring different measures to minimize their effects.for the WHO MONICA Project** Sites and key personnel of the WHO MONICA Project are found at http://www.ktl.fi/publications/monica/rr_sframe/appendix.htm     

Household health surveys in developing countries: could more use be made of them in planning?

Existing health information systems in Eastern Africa are basedon findings in patients visiting hospitals and clinics and onroutine administrative data. They do not reflect the healthsituation and the health care of the underserved, and do not,therefore, generate the information required for planning andprogramming health care for all. Household health surveys arepotentially very useful in filling important information gaps,but past surveys have been poorly adapted to existing healthcare systems; they have been expensive and difficult to replicate,and findings seem to have been rarely used in the planning process.After a brief presentation of current routine reporting, a fewplanning-oriented surveys conducted in Third World countriesare critically reviewed and presented from a planning perspective.It is concluded that household health surveys can be usefulcomponents of health information systems and could be used moreextensively. However, it is important to involve local healthplanners/managers in developing the survey design and the implementationplan, to keep costs at a replicable level and to estimate andreport survey utility and costs.     

单纯随机抽样设计在社区人群调查中的应用

目的 探讨单纯随机抽样设计在社区人群调查中的可行性及调查样本质量.方法 在杭州市下城区和拱墅区根据社区居民电子底册以单纯随机抽样方法 抽取居民户,对户内18～64岁个体采用KISH方法 随机抽取一名个体,两区各需完成500人的调查.结果 下城区抽取950户,完成调查511户(53.8%);拱墅区抽取1380户,完成调查506户(36.7%).两区因户内不符合年龄要求的个体、原户搬迁、社区集体拆迁、底册错误等导致的无应答分别为38.3%和43.5%;各种原因导致的户(或抽中个体)无应答(或拒答)分别为8.0%和19.9%.调查样本与随机抽样户人群的年龄、性别构成无差异.随机抽样户人群与杭州市市区人群性别构成无差异,但年龄结构偏大.结论 在地域相对局限的社区中,基于社区居民电子底册实施单纯随机抽样具备可行性,对调查员的入户时间提出要求,可保证调查样本对抽样框的代表性.     

Don't spin the pen: two alternative methods for second-stage sampling in urban cluster surveys

In two-stage cluster surveys, the traditional method used in second-stage sampling (in which the first household in a cluster is selected) is time-consuming and may result in biased estimates of the indicator of interest. Firstly, a random direction from the center of the cluster is selected, usually by spinning a pen. The houses along that direction are then counted out to the boundary of the cluster, and one is then selected at random to be the first household surveyed. This process favors households towards the center of the cluster, but it could easily be improved. During a recent meningitis vaccination coverage survey in Maradi, Niger, we compared this method of first household selection to two alternatives in urban zones: 1) using a superimposed grid on the map of the cluster area and randomly selecting an intersection; and 2) drawing the perimeter of the cluster area using a Global Positioning System (GPS) and randomly selecting one point within the perimeter. Although we only compared a limited number of clusters using each method, we found the sampling grid method to be the fastest and easiest for field survey teams, although it does require a map of the area. Selecting a random GPS point was also found to be a good method, once adequate training can be provided. Spinning the pen and counting households to the boundary was the most complicated and time-consuming. The two methods tested here represent simpler, quicker and potentially more robust alternatives to spinning the pen for cluster surveys in urban areas. However, in rural areas, these alternatives would favor initial household selection from lower density (or even potentially empty) areas. Bearing in mind these limitations, as well as available resources and feasibility, investigators should choose the most appropriate method for their particular survey context.     

Estimation tools for reducing the impact of sampling and nonresponse errors in dual-frame RDD telephone surveys

We discuss alternative estimators of the population total given a dual-frame random-digit-dial (RDD) telephone survey in which samples are selected from landline and cell phone sampling frames. The estimators are subject to sampling and nonsampling errors. To reduce sampling variability when an optimum balance of landline and cell phone samples is not feasible, we develop an application of shrinkage estimation. We demonstrate the implications for survey weighting of a differential nonresponse mechanism by telephone status. We illustrate these ideas using data from the National Immunization Survey-Child, a large dual-frame RDD telephone survey sponsored by the Centers for Disease Control and Prevention and conducted to measure the vaccination status of American children aged 19 to 35 months.     

Depression in geriatric population in urban slums of Mumbai

A community based study was conducted in an urban slum of Mumbai to assess the epidemiological factors associated with geriatric population and depression. A sample size of 196 was taken according to Lots quality technique, including all elderly above 60 years of age in the study area. Depressed elderly (using Geriatric Depression Scale) constituted 45.9% of the study population and was more in females (57.8%, p < 0.05). The significant variables associated with depression were poor socio-economic status, marital status, non-working or dependency and illiteracy (p < 0.05). Depressed elderly were more inclined towards substance abuse (58.13%), had disturbed sleep patterns (61.6%) and mostly suffered from acute/chronic illness (p < 0.05).     

Contraceptive prevalence in the slums of Rio de Janeiro

A community-based family planning operations research project was undertaken in selected low income communities of Rio de Janeiro; project field work began in February 1982. Prevalence data were collected and service delivery strategies were tested, including home visits promoting family planning, home distribution of condoms, and the introduction of community family planning depots. A high baseline contraceptive prevalence rate (CPR) of 70.1 percent was found for nonpregnant women currently in union, as well as substantial use of the private sector for contraceptive supply, despite the presence of free or subsidized sources within the communities. However, the most economically disadvantaged subgroups made the greatest use of the subsidized sources. The provision of additional service delivery sites may have contributed to a small increase in contraceptive prevalence noted over the life of the project; however, the high baseline CPR precluded a large increase in contraceptive use as a result of the program.     

Improved health outcomes in urban slums through infrastructure upgrading

The world is rapidly urbanizing with over half the population now living in urban areas. As the urban population grows, so does the proportion of these persons living in slums where conditions are deplorable. These conditions concentrate health hazards leading to higher rates of morbidity and mortality. This growing problem creates a unique challenge for policymakers and public health practitioners. While the Millennium Development Goals (MDGs) aim to address these conditions and standards for water and sanitation as well as pertinent health outcomes, little evidence on interventions exists to guide policymakers. Upgrades in slum household water and sanitation systems have not yet been rigorously evaluated to demonstrate whether there is a direct link to improved health outcomes. This study aims to show that slum upgrading as carried out in Ahmedabad, India, led to a significant decline in waterborne illness incidence. We employ a quasi-experimental regression model using health insurance claims (for 2001–2008) as a proxy for passive surveillance of disease incidence. We found that slum upgrading reduced a claimant’s likelihood of claiming for waterborne illness from 32% to 14% and from 25% to 10% excluding mosquito-related illnesses. This study shows that upgrades in slum household infrastructure can lead to improved health outcomes and help achieve the MDGs. It also provides guidance on how upgrading in this context using microfinance and a public–private partnership can provide an avenue to affect positive change.