Sample Attrition Research Articles

The Importance of Sample Attrition in Life Cycle Labor Supply Estimation

We examine the importance of possible nonrandom attrition to an econometric model of life cycle labor supply using both a Wald test comparing attriters to nonattriters and variable addition tests based on formal models of attrition. Estimates using the Panel Study of Income Dynamics show that nonrandom attrition is of little concern when estimating prime-age male labor supply because the effect of attrition is absorbed into fixed effects in labor supply. The wage measure and instrument set have much larger effects on the estimated labor supply function of prime-age men than how one adjusts for panel attrition.

An Analysis of Sample Attrition in Panel Data: The Michigan Panel Study of Income Dynamics

By 1989 the Michigan Panel Study on Income Dynamics (PSID) had experienced approximately 50 percent sample loss from cumulative attrition from its initial 1968 membership. We study the effect of this attrition on the unconditional distributions of several socioeconomic variables and on the estimates of several sets of regression coefficients. We provide a statistical framework for conducting tests for attrition bias that draws a sharp distinction between selection on unobservables and on observables and that shows that weighted least squares can generate consistent parameter estimates when selection is based on observables, even when they are endogenous. Our empirical analysis shows that attrition is highly selective and is concentrated among lower socioeconomic status individuals. We also show that attrition is concentrated among those with more unstable earnings, marriage, and migration histories. Nevertheless, we find that these variables explain very little of the attrition in the sample, and that the selection that occurs is moderated by regression-to-the-mean effects from selection on transitory components that fade over time. Consequently, despite the large amount of attrition, we find no strong evidence that attrition has seriously distorted the representativeness of the PSID through 1989, and considerable evidence that its cross-sectional representativeness has remained roughly intact.

An Analysis of the Impact of Sample Attrition on the Second Generation of Respondents in the Michigan Panel Study of Income Dynamics

Only a few studies have tried to estimate the trend in the elasticity of children's economic status with respect to parents' economic status, and these studies produce conflicting results. In an attempt to reconcile these findings, we use the Panel Study ...

The Importance of Sample Attrition in Life Cycle Labor Supply Estimation

We examine the importance of possible non-random attrition to an econometric model of life cycle labor supply including joint nonlinear taxation of wage and interest incomes and latent heterogeneity. We use a Wald test comparing attriters to nonattriters and variable addition testing based on formal models of attrition. Results from the Panel Study of Income Dynamics are that non-random panel attrition is of little concern for prime-aged male labor supply estimation because the effect of attrition is absorbed into the fixed effects. Attrition is less econometrically influential than research design decisions typically taken for granted; the wage measure or instrument set has a much greater impact on the estimated labor supply function of prime-aged men than how one includes panel attrition.

A REVIEW OF RECENT MULTIPLE SCLEROSIS PROPHYLACTIC CLINICAL TRIALS

BACKGROUND We reviewed four prophylactic treatments for multiple sclerosis recently described in the literature: copolymer-1, interferon β-1b, cladribine, and oral methotrexate. Six problem areas are the subject of this commentary: 1) disease scales, 2) steroid use, 3) definition of exacerbation or relapse, 4) previous symptomatic/radiologic history, 5) magnetic resonance imaging, and 6) design variables: sample size, attrition rate, and power. REVIEW SUMMARY The studies were chosen because of their importance in understanding therapeutic responses in multiple sclerosis patients. Aspects to be considered were disease scales, steroid use, previous history, definition of exacerbation, magnetic resonance imaging, and statistical data. The design of each trial was slightly different. Data from each study were analyzed independently and collectively. The aforementioned variables were compared, and shortcomings were elucidated. Discrepancies between entry criteria and outcome measures were identified. The frequency, preparation, and duration of steroid use was contrasted. Relevant magnetic resonance imaging parameters and outcome criteria were examined. Comparison of total sample size, attrition rate, power, a error rate, and predicted detection of difference were culled, and future design models were explored. CONCLUSIONS This article reviews four current therapeutic trials involving multiple sclerosis patients. Differences and pitfalls of each study are compared. Statistical analysis among the trials is compared. The aim of this paper is to alert readers to the variability of current trials by direct comparison and to suggest design parameters for future trials.

The course of depression in the elderly: a longitudinal community-based study in Australia.

We report the outcome of depressive states after 3-4 years in a community sample of the elderly. A sample of 1045 persons aged 70+ years in 1990-1 was re-interviewed after 3.6 years. Mortality (21.7%) and refusal or non-availability (10.4%) were higher in those who initially had had a diagnosis or symptoms of depression. Of those with an ICD-10 depressive episode in 1990-1, 13% retained that diagnosis. Of those who were not depressed initially only 2.5% had become cases. Depression was unrelated to age or apolipoprotein E genotype. The best predictors of the number of depressive symptoms at follow-up was the number at Wave 1, followed by deterioration in health and in activities of daily living, high neuroticism, poor current health, poor social support, low current activity levels and high service use. Depressive symptoms at Wave 1 did not predict subsequent cognitive decline or dementia. Non-random sample attrition is unavoidable. ICD-10 criteria yield more cases than other systems, while continuous measures of symptoms confer analytical advantages. Risk factors for depressive states in the elderly have been further identified. The prognosis for these states is favourable. At the community level, depressive symptoms do not seem to predict cognitive decline, as they do in referred series.

Panel Attrition and the Correcting Method for its Biases

This paper aims to analyze the sample attrition problem peculiar to a panel survey, especially related to multi-wave attrition biases. The paper also intends to develop a correcting method for those attrition biases and empirically examines an applicability of this method to shopping destination choice behavior. First of all, attrition models are calibrated for several waves in Kofu shopper panels in order to identify the difference in decision structure between the first attrition and the following-step attritions. Secondly, the ordered logit model corresponding to multi-wave attrition process is formulated, and the weighting method for shopping destination choice behavior with those attrition biases is also developed. As a result, it is found from an empirical application that the proposed method is useful for correcting those biases.

Worksite fitness and exercise programs: a review of methodology and health impact.

To examine the methodology of worksite fitness and exercise programs and to assess their effect on health-related fitness, cardiac risk factors, life satisfaction and well-being, and illness and injury. The 52 studies reviewed cover English-language literature for the period from 1972 to 1994, as identified by a search of the Cumulative Index Medicus, Medline, the Canadian Sport Documentation Centre's "Sport Discus," computerized bibliography, and my own files. Reports were divided into five controlled experimental studies, 14 quasi-experimental studies with matched controls (one reported in abstract), and 33 other interventions of varied quality. Methodologic problems include difficulty in allowing for Hawthorne effects, substantial sample attrition, and poor definition of the intervention (exercise or broad-based health promotion). Findings are analyzed by specific fitness and health outcomes. Program participants show small but favorable changes in body mass, skinfolds, aerobic power, muscle strength and flexibility, overall risk-taking behavior, systemic blood pressure, serum cholesterol, and cigarette smoking. Claims of improved mood state are based heavily on uncontrolled studies. Quasi-experimental studies suggest reduced rates of illness and injury among participants, but seasonal and year-to-year differences in health weaken possible conclusions. Participation in worksite fitness programs can enhance health-related fitness and reduce risk-taking behavior, but population effect is limited by low participation rates.

Mental health services outcome evaluation.

There is an urgent need for pertinent outcome information. Relevance for decision makers must take priority over scientific rigor. However, a review of computer-identified outcome evaluation reports from community service settings, during the past 5 years, suggests that much more has been said than has been done. Although relatively heterogeneous in scope, these studies focused on the effects of community support services for adults with persistent and severe mental illness; traditional outpatient services have been neglected. Studies are characterized by multidimensional, standardized outcome assessment, and nonequivalent comparison group and single cohort designs. Randomized designs, with usual services as the control condition, were feasible in some situations. Inadequate sample size and attrition continue to be method problems.

Detecting and Correcting Attrition Bias in Longitudinal Family Research

Conducting longitudinal research has been encouraged among family scholars as a way to better understand family process and to measure changes in family structure and relationships over time. When doing longitudinal research, a common problem is the loss of sample members between the first wave of data collection and subsequent waves. Some subjects move between data points and cannot be located, whereas others refuse to continue their participation. Some members of the original sample, especially if it includes older adults, will have died or have become too disabled to participate further in the study. These losses represent an attrition of the original sample over time. These various forms of attrition of a sample result in a potential threat of bias if those who drop out have unique characteristics such that the remaining sample ceases to be representative of the original sample. This bias, known as attrition bias, is assumed to exist when there are significant differences between the initial and later samples. Because the loss of sample members between the first wave of data collection and subsequent waves is relatively common, researchers in longitudinal studies should check for attrition bias. Unchecked bias may be the major threat to longitudinal research (Markides, Dickson, & Pappas, 1982; Norris, 1985). Larzelere and Klein (1987) listed sample attrition as one of the three major problems that researchers face in analyzing longitudinal data. Despite the threat of attrition bias, however, it has received little attention in longitudinal family research. Indeed, few family researchers address this issue when reporting the results of their analyses of longitudinal data. There are two ways attrition can bias a sample. Firs, it may alter the characteristics of the sample, making it no longer representative of the original sample. Because the results of later waves of data are not generalizable to the original population that was sampled, the external validity of the study is threatened. For example, a longitudinal study examining the effects of divorce on the mental health of former spouses may fail to retain those subjects who have become too depressed to respond to the questionnaire. The nonparticipation of this subsample could bias the findings towards a minimization of depression as an outcome of divorce. The second way that selective attrition can bias longitudinal data is by altering the covariance of variables (Goudy, 1985; Norris, 1987). This problem occurs when the underrepresentation of some groups in the longitudinal sample leads to correlations between variables that are different than the true correlations in the original sample. Referring back to the previous example, the underrepresentation of depressed people in the second wave of the study may distort the statistical relation between length of marriage and depression. This poses a potentially severe threat to theory building because of the effect of these distortions on internal validity. TESTING FOR ATTRITION BIAS The most common method for detecting attrition bias in the characteristics of the sample is to use t tests to compare those subjects who responded to all waves of the study with those who dropped out after only one wave. The means of important variables from the first wave, such as socioeconomic status (SES), age, marital satisfaction, and depression, are compared between the two samples to determine if the differences are statistically significantly different. Significant differences in the means of one or more variables indicates the existence of attrition bias. Differences in categorical variables, such as race and marital status, can be determined by using the chi-square statistic. Logit analysis is a second method for detecting differences in characteristics of the sample. A logit equation is developed to estimate the probability that each first wave respondent will participate in later waves. …

Increased Duration and Sample Attrition in New York City′s Rent Controlled Sector

This paper shows that between 1978 and 1987 tenants living in New York City′s strict rent controlled sector were less mobile than those living in the city′s moderate rent stabilized sector. However after taking into account differences in tenant characteristics across section, one cannot conclude that typical renters in the controlled sector would move more often if they lived in the other sector. Controlled sector tenants tend to be white and relatively old, which are characteristics associated with immobility. If is like that the 1971 Vacancy Decontrol Act, through attrition, divided tenants into the two sectors by their characteristics.

Economic Education for At-Risk Students: An Evaluation of Choices & Changes

This paper presents an evaluation of Choices & Changes, an economics education program designed to teach “at-risk” children in elementary and junior high school that they can control their future by making wise choices and investing in themselves. Using program-specific instruments and a national sample of approximately 1600 students, a controlled experiment was conducted to determine the effect of Choices & Changes on student learning and attitudes. The experimental data were analyzed using a simultaneous model estimated by 2SLS with correction for self-selection due to sample attrition between pre- and post-test observations. The results indicate that Choices & Changes had a positive effect on economic understanding for each grade level examined, ceteris paribus. Significant positive effects on attitudes were found in only one of the four student groups. These findings suggest that Choices & Changes is meeting its short-run objective of producing cognitive understanding of economic concepts and that the attitudes of at-risk students can be altered but may require additional intervention.

Earthquake Insurance: A Longitudinal Study of California Homeowners by Risa Palm (review)

Earthquake Insurance A Longitudinal Study of California Homeowners Risa Palm Boulder, CO: Westview Press, 1995 Reviewed by C h r ist in e M . R o d r ig u e Associate Professor, Department of Geography and Planning California State University, Chico CA 95929-0425 In Earthquake Insurance, Risa Palm offers an update ofAfter a Cali­ fornia Earthquake: Attitude and Behavior Change (Palm and Hodgson 1992). In this volume, she presents the results of a 1993 resurvey of participants in the earlier surveys of 1989 and 1990. The two books, then, report on an environmental perception study unique in querying respondents before the latest series of significant earth­ quakes in California (1989), after the Loma Prieta earthquake of 1989 (1990 resurvey), and after the 1992 implementation of a one-year state-mandated insurance policy covering small losses (1993 resurvey). Given the enormity of the seismic risk in urban California, Palm seeks to identify those factors predicting who among one subset of Californians—homeowners— will choose to reduce their hazard 191 192 APCG YEARBOOK • VOLUME 57 • 1995 vulnerability through insurance and structural and nonstructural miti­ gations. The original survey was mailed to a random sample of homeowners in two Bay Area counties (Contra Costa and Santa Clara) and two southern California counties (Los Angeles and San Bernar­ dino). The second and third surveys went to the same addresses. From these three cross-sectional surveys, a longitudinal sample was con­ structed of those individuals who had responded to all three surveys. Study responses were analyzed using logistic regression, a form of regression permitting the evaluation of ratio and interval scaled in­ dependent variables (e.g., age, income, and education) as they affect a dependent variable of binary scaling (e.g., insured vs. uninsured). With respect to noninsurance mitigations, Palm found that re­ spondents were more likely to adopt nonstructural mitigations (e.g., learning to shut off gas lines, buying an emergency kit, and strapping a water heater) than structural mitigations (e.g., foundation bolting or bracing cripple walls). Most, in fact, had adopted at least one such noninsurance mitigation by 1993. Few, however, are prepared to sur­ vive and take care of themselves in the critical first three days after a quake, despite extensive publicity. Palm found no consistency in the factors correlated with structural mitigations among the four coun­ ties, and only perceived risk helped predict the adoption of nonstructural mitigations. Those households who had adopted insur­ ance, though, were also likelier to have adopted some other mitigation in all counties except San Bernardino. Insurance as an earthquake mitigation has shown a consistent increase in all four counties during the three surveys; anywhere from over a third to over a half of all respondents in each county reported having insurance by 1993. For those who purchased such insurance, deciding factors included fear of a major quake, confidence in the insurance industry, little faith in government as a source of assis­ tance, and a belief that insurance costs (both premiums and deductibles) did not exceed the benefits. The households without RODRIGUE: Review of Earthquake Insurance 193 insurance cited its cost, their beliefs that a big quake would not sig­ nificantly damage their own homes, and a suspicion that the insurance industry could not honor claims after a serious disaster. Perceived vulnerability has increased in all four counties over the three surveys. Perception of vulnerability, however, proved un­ related to actual experience of earthquake damage, objective risk of the home site (i.e., proximity to a Special Studies Zone for surface fault ruptures), or socioeconomic and demographic characteristics. Respondents did perceive certain areas in California as more than usually vulnerable to quakes, namely, their own communities, places subjected to recent quakes, and the great population centers of Los Angeles and San Francisco. Though sample development and attrition prevent the longitudi­ nal analysis from being strictly representative of homeowners in the four counties, the longitudinal study does corroborate the three crosssectional studies. Longitudinal respondents report increases in their perceived vulnerability to quakes and have increased their use of earthquake insurance. The increased perception of vulnerability is strongly associated with insurance purchase, which proves also di­ rectly associated with both home value and income. The...

Statics and Dynamics of Income Distribution: An Introductory Survey

Abstract This article discusses some of the issues involved in making comparisons of income distributions using cross‐sectional data and in examining the dynamics of income distribution using longitudinal data. Any study of income distribution must consider the unit of analysis, the time period and the measure of income, and these are the focus of the article. Section 2, on annual incomes, begins with problems of intra‐household income sharing and the measure of income used, including household production. Comparisons among households of different size are discussed, involving the choice of equivalence scales, the use of decomposition analysis and more general dominance conditions. The role of value judgments in the use of equivalence scales is stressed. Section 3 is concerned with dynamic aspects of income distribution. It discusses methods of constructing synthetic cohorts for use in simulation exercises, the varieties of longitudinal data available, sample attrition and the unit of analysis. An appendix provides further references to the literature and data sources.

Benefits of a Parent Education and Support Program in the First Three Years

The family support movement (Zigler & Black, 1989) points to an increasing need among all American families for support, advice, and role models Weissbourd & Kagan, 1989). Consequently, many parent education and support programs have encouraged the participation of all families, without regard to specified risk. By targeting all families, rather than low-income or otherwise at-risk groups, parent education and family support programs have achieved the broad-based backing necessary to underwrite statewide programs such as those in Minnesota and Missouri (Weiss, 1989). However, particularly in times of severe financial constraints, evaluations of community-based family support and education programs are essential to determine what works, how it works, and for whom it works. This study describes an evaluation of a parent education and support program offered to all parents of newborn children regardless of socioeconomic status in three independent school districts. Potential benefits to the parents and the children were examined longitudinally over the 3-year course of the program. The Context The Parents As Teachers (PAT) program was first implemented in Texas in three pilot sites. The Mental Health Association in Texas requested an outside evaluation of the pilot sites, thus providing an opportunity to study outcomes of this community-based parent education and support program. The PAT curriculum, originally based on the work of Burton White (White, 1990), focuses on educating and supporting parents as their child's first and most influential teacher in language, cognitive, social-emotional, and motor development. The heart of the program is the home visit, in which educators provide information and guidance to parents to enhance the child's physical, social, and intellectual development. In these personalized home visits, the parent educators, who are certified by to the PAT National Center, are required to follow a curriculum of developmental learning activities, model the activities with the child, and discuss age-appropriate expectations, parenting issues, and practices. Although the parent educators follow the same curriculum with each family, they are trained to individualize their strategies to suit each family's needs. Parent educators emphasize the parents' role as the primary decision makers for their children. In addition to the home visits, the PAT program involves parents in group meetings with other parents to share parenting experiences and other topics of interest. A purpose of these group meetings, as well as the home visits, is to reduce the stress and increase the pleasures of parenting. The program also includes periodic screening of the children in general development, language, hearing, and vision. Finally, the PAT educators are prepared to refer families to other community services, if needs for services beyond those offered by PAT become evident. The programs are typically sponsored by school districts, and thus strengthen parents' support and awareness of the schools within their community before their children actually reach school age. PAT programs have also been adapted for sponsorship and delivery in some corporate day care centers. Based on a nondeficit model of service delivery, the program assumes that most, if not all, parents benefit from informal and formal support systems for child rearing (Powell, 1988). Thus, when initiated in three pilot sites in Texas, the PAT program was open to all families with children from 0 to 3 years of age, regardless of specific need; emphasis was placed on recruiting parents of newborns who planned to stay with the program over its 3-year duration. Methodological Concerns There are multiple challenges associated with conducting evaluation studies of parent education and support programs (Gray & Wandersman, 1980; Seitz, 1987), including translating the goals of the program into researchable questions, the use of comparison groups, sample size, and sample attrition when assessing long-term intervention. …

Cognitive?behavioral group treatment for social phobia: Effectiveness at five-year followup

Empirical studies of the behavioral or cognitive—behavioral treatment of social phobia have appeared with increasing frequency over the last decade, and there is reason for cautious optimism in the evaluation of treatment effectiveness. However, few studies have reported systematic followup data, and there is little information available about the durability of change in treated social phobics. We report on the followup evaluation of cognitive—behavioral group therapy (CBGT) for social phobia. Patients who received CBGT or a credible alternative treatment were recontacted after a period of 4.5 to 6.25 years and completed a battery of self-report questionnaires, an individualized behavioral test, and a structured interview with an independent assessor. Patients who received CBGT remained more improved than alternative treatment patients on measures from all assessment modalities. However, due to the long followup period, only a portion of the original study sample could be assessed, and these patients may have been less severely impaired than patients who did not participate in the long-term followup. Limitations of the current study and issues of sample attrition in the conduct of long-term followup studies are discussed.

The relationship between consumption and worker productivity: nutrition and economic approaches.

That better fed, healthier individuals work harder may seem intuitively appealing but there is surprisingly little rigorous evidence to support the hypothesis. There are numerous problems in establishing a causal link between production and consumption due to difficulties in measuring the productivity of workers, misinterpretation of simple correlations between production and consumption, and jointly determined production and consumption variables. The extent to which research findings can be generalized to larger populations often is jeopardized by sample attrition and sample selection. It is the purpose of this article to explain how these empirical and experimental difficulties arise, their implications for research results, how they may be overcome, and to access what is known about the consumption-production relationship. The field of nutrition contributes understanding of the biological relationship between energy intake and energy expenditure, necessary measurement skills, and clinical evidence supporting the consumption-production hypothesis. Economists contribute understanding of the larger social context in which the biological relationship takes place and empirical techniques necessary to overcome measurement and data problems.

Recovery from diffuse traumatic brain injury in Johannesburg: a concurrent prospective study.

Three hundred and twenty-six persons with light, moderate, or severe diffuse traumatic brain injury (TBI) were assessed during initial hospitalization in Johannesburg. One hundred and forty-four were followed up at approximately six months, 83 at 12 months, and 71 at 24 months posttrauma. The sample attrition rate was high. When patients could not be found, data were obtained from their caregivers. Recovery was measured in five different areas of functioning: family relationships, psychologic problems, activities of daily living (ADL), employment status, and physical status. Psychologic problems and family relationships worsened with time, but proficiency in ADL improved; physical and employment status remained relatively stable during the survey period. Severity of injury was significantly correlated with outcome in each functional area at some point in the follow-up period. Discrepancies between patients' and caregivers' perceptions of outcome were noted, particularly in the area of psychologic functioning. Caregivers tended to rate the recovery made by the patient as better than the outcome patients perceived themselves to have made. These differences in perception diminished with time. The multifactorial nature of recovery after TBI is indicated by the improvements seen in some functional areas and the worsening of outcome in other areas. These findings suggest that rehabilitation programs for persons with TBI should (1) include caregivers, (2) focus on the adjustment of both parties to their new circumstances, and (3) facilitate a convergence in their perceptions. Adequately defining recovery is a first step to estimating the prevalence of TBI, but necessitates further work. Epidemiology awaits a better understanding of the recovery process after TBI.

Attrition and Rotation in Panel Surveys

This paper draws on recent research into these issues in the commercial research sector; specifically, in the measurement of television viewing. Different analyses for detecting measurement errors associated with time-in-sample are described. Panel samples offer unique atrractions for the measurement of change. Estimates of net change have low sampling variation, while gross change and changes at the individual respondent level may also be measured. Their drawbacks, at least for panels of households or individuals, include low response rates (compared with single-interview surveys), sample attrition and and the possibility of conditioning. In such panels, attrition rates vary between demographic subgroups and this can give rise to sample imbalance and bias in estimation. Conditioning also gives rise to bias. Even in the absence of measurement error, differential rates of attrition demand an active replacement sampling policy if a representative panel is to be maintained. Sample rotation represents one method. An optimal method for introducing rotation is described. However, rotation incurs higher cost and sample turnover rates than a selective discard policy. As a result rotation has seen limited application to date.

Sample attrition bias in a longitudinal study of young people

Abstract Attrition analyses are reported from a longitudinal survey of young people conducted annually over the years 1980–1988. Not surprisingly, there was considerable sample attrition over the period so that the initial sample of 3,130 was reduced to 483 after eight years. This reflected an average annual retention rate of just under 80%. There was evidence of some selective bias in attrition, particularly during the early years, when those dropping out tended to be of lower academic ability, lower socioeconomic status, and from non-English backgrounds. Although most of the differences were slight, the drop-outs also exhibited greater externality of locus of control and less achievement motivation than those who stayed in, as well as being greater consumers of alcohol, cigarettes, and both non-prescribed and illicit drugs. These results are similar to those that have been reported from other comparable studies conducted in the United States. Implications for the interpretation of longitudinal survey da...