Longitudinal studies, based on repeated observations of the same statistical units over time, represent an invaluable source for analysing the current state and the changes in human populations over time. Longitudinal studies traditionally comprehend panel studies, planned according to specific periodicity and time length, cohort studies based on people with shared experience (e.g. master's degree or first maternity) or characteristics at a particular time point (year of birth), retrospective studies based on different sources regarding past times. Victorian Britain used panel opinions to make better decisions in the nineteenth century. In the 1950s, we saw a lot of progress with panel studies, which were used to track client satisfaction with enterprises. Panel studies include a wide range of topics, including health, psychology, sociology, education, income, housing, and work experiences. In the 19th century, Victorian Britain collected panel opinions for better decisions. Also, the relevant price index survey can be considered a panel survey where the observational units are goods and services over time. Relevant examples at the EU and North-American levels are the National panel survey (ONS), the European labour force quarterly survey the American SIPPs survey. We cite even the Living Standard Measurement Study (LSMS) (World Bank, 2021). This survey provides information on health, access to essential services (water, etc.), risk of malnutrition, poverty status, etc., for over 50 developing countries. In this session, we'll look at three important longitudinal studies that have been undertaken around the world. The LSMS survey, the ONS Coronavirus COVID-19 Infection Survey, and the Longitudinal Social Data Development Program in Canada are all taken into consideration. In the Session, we'll go over many facets of this type of research. We shall go deeper into the problem of representativeness for the current time of data gathered through longitudinal observations and the concept of integrating sources for reconstructing longitudinal data. Furthermore, we will investigate some significant topics studied with longitudinal observation, such as the spread of infection and the transition to different conditions related to wellbeing.

SESSION ORGANIZATION

Chair. Gero Carletto, (World Bank). Discussant. Giorgio Alleva (University La Sapienza, Rome). Session organizer. Piero Demetrio Falorsi (University La Sapienza, Rome).

PRESENTATIONS

1.Improving the quality of survey estimates from longitudinal studies. 

Authors. Paolo Righi (Istat), Giulia Ponzini (World Bank), Piero Falorsi (University La Sapienza, Rome). 

Summary. This work focuses on the case of panels with a rotating sample design. This case represents a powerful hybrid solution for facing the sample erosion for deaths and movers and the impact of lack of sample representativeness for new births, migration flows. Moreover, the sample fatigue introduces an increasing measurement error. As the length of the panel surveys increases, there is an increasing interest, but also increasing challenges in preserving the quality of the panel sample estimates. The effect is particularly evident in long-run panels. A correct design, implementation, and use of a panel survey shall consider a set of methods to deal with these problems at different stages of the statistical process: The sampling design, the data collection, and the estimation. 

2. Developing a Proxy Labour Force Indicator in the Longitudinal Social Data Development Program (LSDDP).

 Authors. Abdelnasser Saïdi. (Chief – Statistical Integration Methods Division, Analytical Studies, Methodology and Statistical Infrastructure Field. Center of expertise in record linkage and Social Statistics administrative Databases. 

Summary. A proxy labour force indicator has been developed for the work domain of the Longitudinal Social Data Development Program (LSDDP) using microdata integration. The categories are defined in the same way as for the Labour Force Survey (LFS), namely: Employed (Category A), Unemployed (category B) or Out of Labour Force (category C). The purpose of this work is not to replace or compete with the LFS estimates, but to be able to derive a monthly proxy employment status (A,B,C) for each individual in the LSDDP population universe from the resulting algorithm. The microdata linkage required was performed using anonymized linkage keys and that quality assessments (MSE, classification error) have been conducted and compared to results from the LFS and the Census. It is anticipated that employment statistics could be improved for different subdomains of interest (e.g. age, sex/gender, census geography, economic region) in using small area estimation technique. We will discuss the results and limitations of this work. 

3. Looking back at the ONS Coronavirus (COVID-19) Infection Survey. 

Authors. Katie Davis. (Principal Methodologist, Methodology and Quality Directorate, Office for National Statistics).

Summary. The paper highlights several aspects of the ONS Coronavirus (COVID-19) Infection Survey, which has a longitudinal structure retesting participants to see how to swab positivity and antibody positivity change over time. The CIS represents a very relevant and almost unique experience for official statistics as it provides regular estimates of relevant parameters (infection rates, antibody positivity) related to the spread of the pandemic. It also allows more ad-hoc analysis into aspects such as a more detailed look into re-infections. 

4. Survival Modelling of Panel Attrition: A proposal with Application to Ethiopia’s HFPS Data 

Authors. Barbara Guardabascio (University of Perugia), Diego Zardetto (World Bank) 

Summary. Nonresponse and attrition are among the most significant problems for panel surveys, as they result in loss of data, decreased estimation efficiency, and increased risk of bias in research findings. Although nonresponse and attrition are sharply distinct phenomena conceptually, separating them rigorously can sometimes prove challenging in practice. To analyze in depth and disentangle these two phenomena, we propose an operational definition of attrition that strives to characterize attritors as units with a persistent non-respondent state over time. We use longitudinal data from the Ethiopia’s High Frequency Phone Survey to model our definition of attrition through survival analysis methods. The ultimate aim of our research would be to exploit the fitted survival model, after careful out-of-sample validation, to predict attrition hazards in similar ongoing panels, and optimize accordingly survey management decisions during data collection.

In this session, we'll look at three important longitudinal studies that have been undertaken around the world. The LSMS survey, the ONS Coronavirus COVID-19 Infection Survey, and the Longitudinal Social Data Development Program in Canada are all taken into consideration. In the Session, we'll go over many facets of this type of research. 
The first presentation focuses on the case of panels with a rotating sample design. This case represents a powerful hybrid solution for facing the sample erosion for deaths and movers and the impact of lack of sample representativeness for new births, migration flows. Moreover, the sample fatigue introduces an increasing measurement error. As the length of the panel surveys increases, there is an increasing interest, but also increasing challenges in preserving the quality of the panel sample estimates. The effect is particularly evident in long run panels. A correct design, implementation, and use of a panel survey shall consider a set of methods to deal with these problems at different stages of the statistical process: The sampling design, the data collection, and the estimation.
The second presentation illustrates how a proxy labour force indicator has been developed for the work domain of the Longitudinal Social Data Development Program (LSDDP) using microdata integration.  The purpose of this work is to derive a monthly proxy employment status for each individual in the LSDDP population universe from the resulting algorithm.  The microdata linkage required was performed using anonymized linkage keys and that quality assessments have been conducted and compared to results from the LFS and the Census.  It is anticipated that employment statistics could be improved for different subdomains of interest (e.g. age, sex/gender, census geography, economic region) in using small area estimation technique. We will discuss the results and limitations of this work.
The third presentation highlights several aspects of the ONS Coronavirus (COVID-19) Infection Survey (CIS), which has a longitudinal structure retesting participant to see how swab positivity and antibody positivity change over time.  The CIS represents a very relevant and almost unique experience for official statistics as it provides regular estimates of relevant parameters (infection rate, antibody positivity) related to the spread of the pandemic. It also allows more ad-hoc analysis into aspects such as a more detailed look into re-infections. 
The last presentation focuses on nonresponse and attrition that are among the most significant problems for panel surveys, as they result in loss of data, decreased estimation efficiency, and increased risk of bias in research findings. We analyse in depth and disentangle these two phenomena, and propose an operational definition of attrition that strives to characterize attritors as units with a persistent non-respondent state over time. We use longitudinal data from the Ethiopia’s High Frequency Phone Survey to model our definition of attrition through survival analysis methods. The ultimate aim of our research would be to exploit the fitted survival model, after careful out-of-sample validation, to predict attrition hazards in similar ongoing panels, and optimize accordingly survey management decisions during data collection.

Organiser: Dr piero falorsi 

Chair: Mr Calogero Carletto 

Speaker: Gero Carletto 

Speaker: Abdelnasser Saïdi 

Speaker: Mr Diego Zardetto