ESRA 2019 Programme at a Glance
Evaluating and Analyzing Life History Survey Data 1
|Session Organisers|| Dr Mengyao Hu (University of Michigan)
Dr Brian Wells (University of California, Los Angeles)
Professor Jacqui Smith (University of Michigan)
|Time||Tuesday 16th July, 14:00 - 15:30|
There is an increasing multidisciplinary interest in understanding the long-term influences of early- and mid-life factors on late-life health and well-being outcomes. The knowledge gained about these time-varying complex relationships is critically important for designing policy intervention to reduce inequalities and foster healthy aging and well-being. One increasingly-popular method to obtain this information is to collect life history data using event history calendars. This approach has been applied in many large-scale surveys including the English Longitudinal Study of Ageing (ELSA) and the Survey of Health and Ageing in Europe (SHARE).
Recently, the U.S. Health and Retirement Study (HRS) collected life history data in their off-year mail surveys in 2015 and 2017. The U.S. Health and Retirement Study (HRS) is unique among large scale, high quality longitudinal surveys in representing the full range of the U.S. birth cohorts beginning with the first cohort born before 1923 through the late-Boomers born in 1960-1965. These cohorts have participated in the major demographic, economic and social transformation of the American people over much of the past century. The newly collected life history data fill in critical information about HRS respondents between birth and their entry into the HRS beginning in their 50s (70s for oldest cohort) and have great implications for conducting research and making effective policies that will promote healthy aging.
These aforementioned large-scale, representative life history surveys differ in their survey designs. For example, the HRS life history survey was conducted through mail surveys, and most other surveys were conducted through in-person interviews. Little research so far has been done to evaluate the pros and cons of different design conditions in life history surveys, the cognitive burden and usability of these surveys. There is also a growing need to evaluate the validity of such distantly recalled information, and the quality of life history data in the Total Survey Error framework.
We invite the submission of abstracts on 1) methodological evaluation on the quality of life history data such as examining the validity of life history measures, and evaluating nonresponse and recall errors; 2) statistical methods to identify, measure and control for errors in life history surveys; 3) developments in the statistical methods to analyze life history data; and 4) substantive research projects that links early- and mid-life factors to life course patterns and outcomes.
Keywords: Life History Survey, Event History Calendars, HRS Life History Survey
Agreement of Self-Reported and Administrative Data on Employment Histories in a German Cohort Study: A Sequence Analysis.
Dr Morten Wahrendorf (University Düsseldorf, Medical Faculty, Institute of Medical Sociology) - Presenting Author
Collecting life course data is increasingly common in social and epidemiological research, either through record linkage of administrative data or by collecting retrospective interview data. This paper uses data on employment histories collected through both strategies from Germany, compares the attained samples, and investigates levels of agreements of individual histories. We use data from the German Heinz Nixdorf Recall Study with information on employment histories collected retrospectively from 2011 until 2014 (N=3059). Administrative data from the German Institute for Employment Research (IAB) were linked to the survey data. After comparing respondents who provide self-reported histories with the subsample of the ones for which administrative data was available, we investigate the agreement of individual employment histories from the two sources (between 1975 and 2010) using sequence analyses. Almost all participants provided survey data on employment histories (97% of the sample), linkage consent was given by 93%, and administrative data was available for 63% of the participants. People with survey data were more likely to be female, to have a higher education, and to work self-employed and in the tertiary sector. The agreement of individual employment histories is high and similar across time, with a median level of agreement of 89%. Slightly lower values exist for women and people working in the tertiary sector, both having more complex histories. No differences exist for health-related factors. In conclusion, it is likely that missing consent and failed record linkage lead to sample differences; yet, both strategies provide comparable and reliable life course data.
How Cognitive Functioning Accounts for Inconsistent Reporting: A Case Study from the Health and Retirement Study
Mr Xinyu Zhang (University of Michigan) - Presenting Author
Dr Mengyao Hu (University of Michigan)
Professor Jacqui Smith (University of Michigan)
Measurement error is prevalent in retrospective surveys. The additional cognitive demands on respondents may reduce the accuracy of survey reporting resulting in biased survey statistics. Previous research on long-term memory finds that residences, schools, colleges, and work places, are conserved for over 50 years in healthy older adults. However, few studies investigate the association between cognitive status (normal vs CIND vs suspected dementia) and the likelihood of the consistency in reporting early life events. At study entry, the Health and Retirement Study (HRS) collected respondents’ answers about their early life, such as childhood health conditions. In later waves, the HRS Life History Mail Survey (LHMS) collected detailed information about respondents’ histories from birth to age 50, which can be compared to the HRS data to determine the consistency. The cohort sample design of the HRS also allows us to investigate subgroup differences in the consistency over time. We will examine if cognitive status differences remain after controlling for socioeconomic, depression, and physical health status.
Thinking Back to Your Childhood: How Accurate to Recall? Evidence from NCDS Age 61 Survey
Professor Alissa Goodman (Centre for Longitudinal Studies, UCL)
Mr Matt Brown (Centre for Longitudinal Studies, UCL)
Dr Darina Peycheva (Centre for Longitudinal Studies, UCL) - Presenting Author
Childhood circumstances have a significant impact on adult life and for this reason, many studies of adults collect information about the early years of life. For example, the English Longitudinal Study of Ageing (ELSA), the U.S. Health and Retirement Study (HRS) and the Survey of Health, Ageing and Retirement in Europe (SHARE), longitudinal studies following adults aged 50 and over have all administered life history questionnaires to collect information about life experiences prior to joining the study. However, little is known about the accuracy with which adults can recall information about their childhood, and whether some aspects of childhood can be more accurately recalled than others.
The National Child Development Study has been following the lives of over 17,000 people in Britain born in a single week of 1958. The tenth follow-up will commence in 2020 when participants will be aged 62 and will include a retrospective questionnaire about childhood circumstances. The majority of the questions were answered contemporaneously by participants themselves, parents or teachers during childhood follow-ups at ages 7, 11 and 16. The retrospective questionnaire covers many domains typically featured in life history questionnaires such as housing, parental employment and health, but also topics not typically featured because of doubts regarding the accuracy with which they can be recalled, such as behaviour and anxiety.
Retrospective responses will be compared with data gathered prospectively, providing a unique opportunity to explore the accuracy with which a wide range of childhood circumstances and experiences can be recalled.
This presentation will describe our approach to designing the retrospective questionnaire and will share some early findings from the pilot study on the accuracy with which childhood circumstances can be recalled.
Evaluating Item Nonresponse in Life History Calendar: An Analysis of Memory Effects
Dr Mengyao Hu (University of Michigan) - Presenting Author
Dr Roberto Melipillan (University of Michigan)
Professor Jacqui Smith (University of Michigan)
Mr Xinyu Zhang (University of Michigan)
Item nonresponse is a common result of memory effects in life history surveys. Specifically, respondents have different propensities to respond to life history questions, depending on their reporting motivations and recall difficulties of the questions. Event history calendar, or the life grid approach is commonly used to obtain retrospective data in life history studies. As indicated in previous literature, this approach can assist respondents’ memory retrieval and provide reliable information about sequences of behavior and experiences over the life course (e.g., Belli 2014). Despite its wide use, the important issue of item nonresponse in life grid questions have received little attention. One main reason for item nonresponse in life history data is memory effect (Auriat 1991). Autobiographical memory (AM) research has shown that there are two interconnected long-term memory systems: episodic memories of events from specific remote times in an individual’s life (e.g., first day at school); and semantic memories of the important facts and themes that define an individual’s life history (Conway, 2009). Episodic and semantic AM may introduce different levels of difficulty in retrieving memory and thus produce different levels of item missing data.
This study aims to examine the effects of both item-level predictors (e.g., types of AM memories required in each question) and respondent-level predictors (e.g., cognitive status, age, and socio-economic and health status) on the likelihood of providing item missing data in life grid questions. We analyzed data from the 2017 Health and Retirement Study (HRS) Life History Mail Survey (n = 3,844). To examine the patterns of missing data, we applied multilevel logistic regression. Our dependent variable is whether respondent provided missing or not to a question item. We examined the amount of clustering effect by respondent and assessed how much of the item-missing variance is introduced by type of AM.